Datasets:

Hendrik-a

/

arxiv

like 0

an important class of observables for gravitational waves consists of the fluxes of energy , momentum and angular momentum carried away by them and are well understood for weak gravitational waves in minkowski background . in de sitter background , the future null infinity , , is space - like which makes the meaning of these observables subtle . a spatially compact source in de sitter background also provides a distinguished null hypersurface , its _ cosmological horizon _ , . for sources supporting the short wavelength approximation , we adopt the isaacson prescription to define an effective gravitational stress tensor . we show that the fluxes computed using this effective stress tensor can be evaluated at and match with those given by ashtekar et al at .

1612.09511

weak gravitational field of a spatially compact source is identified as a perturbation about a background space - time which is a solution of the einstein equation in the source free region . in the presence of a positive cosmological constant , the background space - time is the de sitter space - time . unlike the minkowski background for the zero cosmological constant , de sitter space - time has different patches eg the global patch , a poincare patch and a static patch . in the cosmological context , a poincare patch is appropriate which is what we focus on . a solution at the _ linearized level _ , valid throughout the poincare patch and extending to the future null infinity @xmath2 , is available in @xcite . however the space - like character of the @xmath0 poses challenges for defining energy , momentum and their fluxes . let us recall that the cleanest articulation of ` infinity ' arises in the conformal completion of physical space - times . conformal completion preserves the light cone structure of the physical space - time and captures the notion of approaching infinity along many directions as well as with different speeds . it naturally identifies boundary components , @xmath3 where time - like and null geodesics ` terminate ' . the causal nature of these boundary components is determined by the asymptotic form of ` source - free ' equations : @xmath4 are null when @xmath5 and space - like for @xmath6 ( time - like for @xmath7 ) . these boundary components serve to define out - going ( in - coming ) fields as those solutions of the asymptotic equations that have suitably finite limiting values on @xmath8 . it is then a result that the weyl tensor of out - going fields evaluated along out - going null geodesics , has a definite pattern of fall - off in inverse powers of an affine parameter along the geodesics ( the peeling - off theorem ) @xcite . this enables one to identify the leading term as representing gravitational radiation ( far field of a source ) , in a coordinate invariant manner . it is conveniently described in terms of the weyl scalars which are defined with respect to a suitable null tetrad . when @xmath2 is null , a null tetrad at a point @xmath9 is uniquely determined ( modulo real scaling and rotation ) by the tangent vector @xmath10 of an outgoing null geodesic reaching @xmath11 , _ and _ the null normal , say @xmath12 at @xmath11 normalized so that with @xmath13 . clearly as the null geodesic changes its direction , @xmath14 changes but not @xmath15 and hence the weyl scalar @xmath16 remains unchanged . its non - zero value can be taken as showing the presence of gravitational radiation . this feature is lost when the @xmath2 is space - like . now the null vector @xmath12 , with @xmath13 , is chosen to be in the plane defined by @xmath10 and the ( time - like ) normal @xmath17 . clearly , as @xmath10 changes , so does @xmath12 and then _ no _ weyl scalar is invariant . an invariant characterization of gravitational radiation is no longer available @xcite . to test if a notion of radiation based on asymptotic behaviour is physically useful , one still needs suitable definitions of flux of energy - momentum , angular momentum in terms of the fields which will then correlate physical quantities being radiated away from a spatially compact source . fortunately , in the poincare patch , there are seven globally defined killing vectors , all space - like in the vicinity of @xmath0 , and corresponding conserved quantities can be expected . one of the definition of such conserved quantities is based on the covariant phase space framework @xcite . in the context of the linearised theory , it exploits the phase space structure of the space of solutions and defines a manifestly gauge invariant and conserved ` hamiltonian ' corresponding to every _ isometry _ of the background space - time . although defined on each space - like hypersurface of the poincare patch , the simplest expressions results for evaluation at @xmath2 . being defined as an integral over a hypersurface , it is also referred to as a _ flux_. thus we see that the null infinity provides a characterization of the radiation ( up to origin dependence for @xmath18 ) which can transport energy - momentum - angular momentum away from a spatially compact gravitating source . one other implication of this characterization at infinity is that the energy / momentum / angular momentum which _ is _ received at the future null infinity is _ necessarily lost _ from the source under the assumption of _ no incoming radiation _ boundary condition . it is this last implication that follows also for the cosmological horizon of a spatially compact source : energy / momentum / angular momentum that crosses the cosmological horizon is also lost forever from the source . note that the worldlines of different components of a spatially extended ( but compact ) source must reach the same point on @xmath0 to maintain a _ finite _ physical separation among them . it follows that that the cosmological horizon for a spatially compact source is well defined : it is the past light cone of the common point on @xmath0 where the source world tube converges . equally well , _ any observer _ who remains at a finite physical distance from the compact source for all times , must necessarily lie within the cosmological horizon i.e. within the static patch . we would like to explore to what extent and under what conditions may we regard the cosmological horizon as a `` substitute '' for the _ future null infinity_. it is obvious at the outset that the out - going null geodesics emanating from the source intersect the cosmological horizon at a _ finite _ value of any affine parameter and it can be chosen to be 1 by a suitable normalization . such a normalized affine parameter equals the ratio of the physical distance from the source to @xmath19 . all spatially compact sources may be taken to be within a sphere of radius @xmath20 . furthermore , only sources varying over cosmological time scales , will have comparable wavelengths . in other words , most sources producing gravitational waves would have wavelengths far smaller than @xmath21 and any wave crossing the horizon may be taken to be a ` far zone field ' . cosmological horizon being a null hypersurface , we do have the weyl scalar @xmath22 defined on it , which is independent of the _ direction _ of null geodesics meeting the horizon . for sources which are sufficiently rapidly varying ( relative to the scale set by the cosmological constant ) , there is an alternative identification of gravitational waves as _ ripples on a background _ within the so called _ short wave approximation _ @xcite . furthermore , it is possible to define an _ effective gravitational stress tensor _ , @xmath23 for the ripples . for vanishing @xmath24 , it is gauge invariant , conserved and symmetric . for non - zero @xmath24 it is _ not _ gauge invariant but the gauge violations are suppressed by powers of @xmath25 . it is very convenient to have such a stress tensor to define and compute fluxes of energy and momenta carried by the ripples across any hypersurface . we will use this to show that for the retarded solution given in @xcite , the flux of energy - momentum across the cosmological horizon exactly equals the corresponding flux across the @xmath0 and also equals the flux computed in @xcite . like - wise , the instantaneous power received at infinity matches with that crossing the horizon . this is our main result . the paper is organised as follow . in section [ solution ] , we recall the solution at the linearised level for which the fluxes will be evaluated . section [ energy ] is divided into five subsections . in subsection [ covariant ] we summarise the covariant phase space framework and in subsection [ effectivestresstensor ] , we discuss the isaacson prescription adapted to the presence of the cosmological constant . since this constructs a symmetric , conserved and gauge invariant ( to _ leading order _ ) stress tensor , we get conserved currents for each killing vector of the background space - time . in the subsection [ fluxcomputations ] , we present computations of the energy flux across various hypersurfaces . in particular we show that the fluxes across the out - going null hypersurfaces are zero , implying for example , that the energy propagation is sharp . subsection [ momentumfluxes ] contains the fluxes for the momentum and the angular momentum . in the subsection [ cosmologicalhorizon ] , we discuss its applications and show that the quadrupole power can be computed at the cosmological horizon . the final section [ finalsection ] concludes with a discussion . 0.3 cm weak gravitational fields are understood as perturbations about a background specified in the form , @xmath26 . the background @xmath27 is chosen to be a solution of the source free einstein equation with a positive cosmological constant . the einstein equation for @xmath28 , expanded to first order in @xmath29 , gives the linearised einstein equation for @xmath30 . the _ physical perturbations _ are understood as the equivalence classes of solutions @xmath30 , with respect to the _ gauge transformations _ : @xmath31 . in terms of the trace reversed combination @xmath32 , the linearised equation takes the form , @xmath33 + \frac{\lambda}{3}\left[\tilde{h}_{\mu\nu } - \tilde{h}\bar{g}_{\mu\nu}\right ] ~ = ~ 8\pi t_{\mu\nu } \label{lineqn } \ ] ] where , @xmath34 . the gauge freedom is exploited subsequently to simplify the equation . in the present context , the background space - time is taken to be the _ poincare patch _ of the de sitter space - time ( see figure [ desitterpenrose ] ) which admits a conformally flat form of the background metric in coordinates ( @xmath35 , @xmath36\ , ~~ \eta \in ( - \infty , 0 ) ~~,~~ h : = \sqrt{\frac{\lambda}{3 } } . \label{conformallyflat } \ ] ] . the line bd does not belong to the chart . the line ab denotes the _ future null infinity _ , @xmath0 while the line ae denotes the _ cosmological horizon _ , two constant @xmath37 space - like hypersurfaces are shown with @xmath38 . the two constant @xmath39 , time - like hypersurfaces have @xmath40 . the two dotted lines at 45 degrees , denote the out - going null hypersurfaces emanating from @xmath41 on the world line at @xmath42 , through the source . the region aed is the _ static patch _ admitting the stationary killing vector , @xmath43 of eqn . ( [ stationarykilling]).,scaledwidth=60.0% ] the conformally flat form leads to a great deal of simplification . the future null infinity is approached as @xmath44 while the @xmath45 corresponds to the flrw singularity . the killing equation in the poincare patch is given by , @xmath46 there are seven globally defined killing vectors on the poincare chart , corresponding to energy , 3 momenta and 3 angular momenta @xcite . they are given by ( up to constant scaling ) : @xmath47 we focus on the _ time translation vector field _ , @xmath48 it is time - like in the static patch ( hence generating ` time translations ' ) and space - like beyond the cosmological horizon in the poincare patch . in particular it is space - like and tangential to @xmath49 and null on the cosmological horizon . in this chart , the de sitter dalembertian can be conveniently expressed in terms of the minkowski dalembertian leading to , @xmath50 -~\left(\frac{2\lambda}{3}\right)\left[\tilde{h}_{\mu\nu } - \eta_{\mu\nu } \tilde{h}_{\alpha\beta}\eta^{\alpha\beta}\right ] \nonumber \\ & & -~ \left\{\left(\partial_{\mu}b_{\nu } + \partial_{\nu}b_{\mu } - \eta_{\mu\nu}\partial^{\alpha}b_{\alpha}\right ) + \frac{2}{\eta}\left(\delta_{\mu}^0b_{\nu } + \delta_{\nu}^0b_{\mu}\right)\right\ } ~~,~~ \omega^2 : = \frac{1}{h^2\eta^2 } \ . \label{lineqngaugefixed } \ ] ] the equation for the tensor components can be decoupled by imposing a gauge condition , @xmath51 . the equation ( with source included ) simplifies to @xcite , @xmath52 _ from now on in this subsection , the tensor indices are raised / lowered with the minkowski metric . _ it turns out to be convenient to work with new variables , @xmath53 . all factors of @xmath54 _ and _ @xmath24 drop out of the equations and @xmath55 satisfies @xcite , @xmath56 under the gauge transformations generated by a vector field @xmath57 , the @xmath55 transform as , @xmath58 it is further shown in @xcite that the residual invariance is exhausted by setting @xmath59 . the gauge condition ( [ chigauge ] ) then implies that _ fully gauge fixed solutions _ , satisfy the spatial transverse , traceless condition , tt for short : @xmath60 and thanks to the decoupled equations , it suffices to focus on the equation ( [ lineqnchi ] ) for @xmath61 . the _ retarded solution _ is given by , @xmath62 the spatial integration is over the matter source confined to a compact region and is finite . the second term in the eqns . ( [ radiativesoln ] ) is the _ tail term_. _ physical solutions _ are the tt part of the solution above . we generally suppress the @xmath63 label and restore it in the final expressions . for @xmath64 , we can approximate @xmath65 . this allows us separate out the @xmath66 dependence from the @xmath67 . in the first term , this leads to the spatial integral over @xmath68 while in the second term , we can interchange the order of integration , again leading to the same spatial integral . @xmath69 the spatial integral of @xmath70 can be simplified using moments . this is done through the matter conservation equation . to define these moments , introduce the orthonormal tetrad @xmath71 and denote the frame components of the source stress tensor as : @xmath72 . define moment variable @xmath73 . two sets of moments are defined as functions of @xmath37 , or of @xmath74 defined through @xmath75 as , @xmath76 in terms of these , the retarded solution is given by , @xmath77 ~ , \\ g_{ij}(\eta_{ret } ) & : = & - 2h\left [ { \cal l}^2_t{q}_{ij } + h{\cal l}_t{q}_{ij } + h{\cal l}_t{\bar{q}}_{ij } + h^2 \bar{q}_{ij}\right ] ~ \hspace{1.5cm}\mbox{and , } \\ \hat{g}_{ij } & : = & \left . -2h^2\left[\mathcal{l}q_{ij } + h\bar{q}_{ij}\right]\right|_{-\infty } \ ] ] all moments are evaluated at the retarded @xmath78 , @xmath79 and @xmath80 denotes the lie derivative with respect to the time translation killing vector . on the moments , it is given by , @xmath81 in equation , the first term is the contribution of the so called sharp term while the second and the third terms denote the tail contributions . the tail contribution has separated into a term which depends on retarded time , @xmath82 only , just as the sharp term does , and the contribution from the history of the source is given by the limiting value at @xmath83 . this expression is valid _ as a leading term for @xmath64_. for future use in section [ fluxcomputations ] , we display the derivatives of @xmath84 . since @xmath84 depends on @xmath85 only through @xmath39 , we need only the derivatives with respect to @xmath37 and @xmath39 . on functions of @xmath86 , @xmath87 and we can replace the r - derivatives in favour of @xmath88derivatives . hence , @xmath89 consider the space @xmath90 of a class of solutions of the full einstein equation , satisfying stipulated boundary condition . at each point of this space , the linearised solutions provide tangent vectors . under certain conditions , it is possible to define a _ pre - symplectic form _ on the tangent spaces . every infinitesimal diffeomorphism of the space - time , with suitable asymptotic behaviour , induces a vector field on @xmath90 . some of these lie in the kernel of the pre - symplectic form and constitute ` gauge directions ' while the remaining ones constitute ( asymptotic ) symmetries shared by the stipulated class of solutions . modding out by the gauge directions ( null space of the pre - symplectic form ) , one imparts a symplectic structure to the space of solutions , now denoted as @xmath91 . under favourable conditions , the vector fields on @xmath92 corresponding to the asymptotic symmetries descend to @xmath93 and generate infinitesimal _ canonical transformations_. their generating functions , or ` hamiltonians ' , are candidates for representing energy , momenta , angular momenta etc @xcite . in @xcite , this strategy is applied to the space of fully gauge fixed solutions of the linearised equation and we summarise it below . since linearization is always done about a background , space - time transformations which leave the background invariant i.e. isometries of the background , leave the covariant phase space itself invariant . in the present context , the hamiltonians corresponding to the 7 isometries are the proposed definitions of energy , linear momentum and angular momentum . explicitly , @xmath90 denotes the solutions of the equation ( [ gaugefixedlinearised ] ) together with the gauge fixing conditions ( [ gaugecondns ] ) : @xmath94 @xmath95 since the linearised equation is a hyperbolic equation , suitable classes of solutions can be chosen by suitably restricted initial data on @xmath96constant ( cauchy ) surfaces of the poincare patch . for two elements @xmath97 , define the two form @xmath98 on @xmath90 by @xcite , @xmath99 where , @xmath100 denotes the cosmological slice defined by a constant value @xmath37 . the integrals is actually independent of the hypersurface . also , a label ` tt ' signifying that the @xmath101 s are _ transverse , traceless _ , is suppressed . this definition of the symplectic form however , is not convenient for going to @xmath2 by taking the limit @xmath44 . it can be translated in terms of the _ electric part _ of the perturbed weyl tensor , @xmath102 = { { \textstyle \frac{1 } { 2 h \eta^{2}}}}(\partial^2_{\eta } \chi_{ij } + \eta~\nabla^{2}\chi_{ij } ) = { { \textstyle \frac{1 } { 2 h \eta}}}(\partial^2_{\eta } - { { \textstyle \frac{1}{\eta}}}\partial_{\eta})\chi_{ij}$ ] , leading to @xcite , @xmath103 for a killing vector @xmath104 of the de sitter background , we have the tangent vector @xmath105 , generating a canonical transformation . the corresponding hamiltonian function is given by , @xmath106 for the time translation killing vector @xmath43 , @xmath107 , so that @xmath108 . furthermore , @xmath109 and the corresponding hamiltonian , now referring to it as the _ energy _ @xmath110 , is obtained as , @xmath111 this integral is independent of the choice of @xmath100 and is conveniently performed on @xmath112 . the killing vector @xmath113 also has a smooth limit to @xmath2 , @xmath114 which is tangential to @xmath115 . the equation simplifies to , @xmath116 here we have integrated by parts and used the vanishing of the weyl tensor at @xmath117 for de sitter metric @xcite . now using @xmath118 , @xmath119^{tt } ~(t^{m}\partial_{m } \chi_{ij})^{tt}~\delta^{ik } \delta^{jl } \label{energy } \ ] ] in the last line we have used equation of motion for @xmath84 in source free region and restored the @xmath63 label @xcite . both @xmath120 and @xmath121 have smooth limit on @xmath122 . now for the approximated solution given in ( [ radiativesoln1 ] ) , the energy flux turns out to be given by @xcite , @xmath123\delta^{ik}\delta^{jl } , \ ] ] where , the ` radiation field ' , @xmath124 on @xmath2 is given by @xmath125^{tt}(t_{ret } ) \ , \ ] ] with the overdot denoting the lie derivative @xmath126 . the instantaneous _ power _ received on @xmath0 at ` @xmath127 ' is given by , @xmath128(-r(\tau))\ . \ ] ] for completeness , we also give the expressions for the momentum and angular momentum fluxes @xcite : @xmath129^{tt } \label{angularmomentumflux } \ ] ] the momentum flux is zero because the integrand is linear in @xmath130 ( parity odd ) and in the angular momentum flux , the second factor is proportional to the tail term . in the previous subsection we saw a definition of _ total energy _ of radiation field of compactly supported sources in equation ( [ energyatinfinity ] ) . the radiated power , received at infinity , is given in equation ( [ poweratinfinity ] ) . in this subsection we recall an alternative framework , based on a ` short wavelength expansion ' @xcite , for a restricted class of sources but with the benefit of a symmetric , conserved , suitably gauge invariant _ effective gravitational stress tensor_. conceptually , the framework is somewhat different from perturbation about a _ fixed _ , given background solution . it envisages construction of a class of solutions for which there exists a coordinate system in which the metric components display two widely separated temporal / spatial scales of variation . the slowly varying ( or long wavelength ) component is taken as the _ background _ component and the fast ( or short wavelength ) component whose amplitude is small compared to that of the background , is identified as the _ ripple _ component . these statements are manifestly coordinate dependent , but existence of a coordinate system with sufficiently large domain admitting such an identification , itself is a physical property . the calculational scheme is again iterative but now allows for both the background and the ripple components to be corrected . to make such a separation , an _ averaging scheme _ is introduced . it splits the einstein equation into two separate , coupled equations for the background and the ripple . these equations provide a definition of the effective gravitational stress tensor . let @xmath131 denote the length scale of variation of the background and @xmath132 the length scale of the ripple with @xmath133 . the length scale @xmath134 of the spatially compact source satisfies @xmath135 . for example , if the waves are produced by a source with a characteristic frequency @xmath136 . the ` time ' used in identifying the characteristic frequency is either the time coordinate in a class of coordinate systems or in more favourable cases the time coordinate provided by a stationary killing vector of the background geometry . in the present context of de sitter background having a stationary killing vector @xmath137 , such a situation can be easily envisaged . returning to a general context , one begins with an expansion of the form @xmath138 and writes the einstein equation as , @xmath139 introduce an averaging over an intermediate scale @xmath14 , @xmath140 which satisfies the properties : ( i ) average of odd powers of @xmath141 vanish and ( ii ) average of space - time divergence of tensors are sub - leading . the average of course leaves the @xmath131-scale variations intact , in particular average of @xmath28 equals @xmath27 . for simplicity , we will assume that the average of matter stress tensor is zero i.e. it has only @xmath132-scale variations . taking the average of the above equation gives , @xmath142 notice that @xmath143 which is quadratic in @xmath141 , _ can _ have @xmath131-scale variations and hence non - zero average . its @xmath132-scale variations are suppressed by a power of @xmath29 relative to @xmath144 . we note that @xmath145 and use @xmath146 . subtracting the eqn . ( [ backgroundeqn0 ] ) from the eqn.([expandedeqn ] ) we get ( to order @xmath29 ) , @xmath147 in terms of first order einstein tensor , this is exactly the linearized equation we had before . the equation ( [ backgroundeqn0 ] ) can likewise be rearranged and now includes an additional source term for the equation determining the background metric @xmath27 . the rearranged equations take the form , @xmath148 \label{rippletensordef } \ ] ] the equation ( [ rippleeqn ] ) is exactly the same linearised einstein we had before for the weak field @xmath30 and every term of it has a scale of variation @xmath132 . however , the equation ( [ backgroundeqn ] ) for the background is different . although it has terms of order @xmath149 , every term has a scale of variation @xmath131 . if we now recognise that for @xmath150scale variation , @xmath151 and @xmath152 is taken to be of the same order as @xmath29 , then the effective stress tensor which has a leading term of the form @xmath153 , is of the order @xmath154 and is thus included in the equation . the effective stress tensor defined in equation ( [ rippletensordef ] ) is manifestly symmetric and is covariantly conserved since divergence of the right hand side of ( [ backgroundeqn ] ) vanishes identically . modulo the subtleties of its tensor nature and gauge invariance under @xmath155 , it is well qualified to be identified as gravitational stress tensor . an averaging procedure constructing a tensor has been given by isaacson @xcite . to verify gauge invariance of @xmath23 to leading order in @xmath29 , we need to be more explicit about the gauge transformations @xcite . recall that the gauge transformation involves derivatives of @xmath156 and for a consistency with the background plus ripple split , the gauge transformation should also be restricted to preserve it . there are two possibilities for the generator : ( i ) @xmath157 is comparable with @xmath141 and slowly varying , and ( ii ) @xmath157 is order @xmath158 but is rapidly varying so that its derivative becomes order @xmath141 . the gauge transformation of @xmath23 has terms of the form @xmath159 ( which vanish identically for minkowski background ) . for the @xmath157 of type ( i ) , the average vanishes since the enclosed quantity is rapidly varying and for @xmath157 of type ( ii ) , the averaged quantity is order @xmath29 . but @xmath23 itself is @xmath160 and hence _ gauge invariance of @xmath23 is ensured to the leading order _ @xcite . the calculational scheme begins by _ selecting _ a solution of the background equation ( [ backgroundeqn ] ) ignoring the effective stress tensor . this is just the vacuum einstein equation with cosmological constant . in the present context , we choose the maximally symmetric , de sitter space - time . using this in the ripple equation ( [ rippleeqn ] ) , determines linearised solutions , @xmath30 after a suitable gauge choice and choice of boundary conditions . in the present context , the focus is on the retarded solution . this solution gives the effective gravitational stress tensor of equation ( [ rippletensordef ] ) . this is fed back into the background equation to construct a new , corrected background . this will still have have @xmath161scale variations since the effective stress tensor varies on the scale @xmath131 . the corrected background is used in the ripple equation to get corrections to the ripple and so on . since the equations for the background and the ripple have consistent scales of variations , the background and ripple characters are maintained . within this scheme , the effective gravitational stress tensor finds its justification . once we have a conserved and symmetric stress tensor , for every killing vector of the background , we get a conserved _ current _ and the corresponding conserved _ charge_. it turns out that the energy momentum computed using the effective gravitational stress tensor for ripples over minkowski background , agrees with the quadrupole formula obtained by other methods , thereby strengthening the interpretation . as an aside , we make a numerical estimate comparing the effective stress tensor and the cosmological constant . typical estimate for the wave amplitude ( dimensionless ) at ligo detectors is around @xmath162 or smaller . the highest frequency , @xmath163 , detectable is about @xmath164 hz @xmath165 meter@xmath166 in the @xmath167 units . in the geometrized units we are using , @xmath168 meter@xmath169 . although tiny , this is very much comparable to the cosmological constant ! thus , unlike the minkowski background which would be inconsistent with the sourcing by gravitational waves , the de sitter background at least has a chance of being self consistent . to evaluate the effective stress tensor , we need to carry out the expansion of ricci tensor to order @xmath149 . in terms of the trace - reversed @xmath170 , this is given by , @xmath171 + \left [ - \frac{1}{2 } b_{\mu}b_{\nu } \right ] \nonumber \\ & & + \left[\tilde{h}\left\ { \frac{1}{4}\left(\bar{\box}\tilde{h}_{\mu\nu } - \frac{1}{2}\bar{\nabla}_{\mu}\bar{\nabla}_{\nu}\tilde{h}\right ) - \frac{\lambda}{6}\tilde{h}_{\mu\nu } \right\}\right ] + \mbox{total derivative terms . } \ ] ] the expression has been arranged in a suggestive form in the anticipation that it will be evaluated for gauge fixed solution . the second group of terms containing @xmath172 , which eventually vanish when the _ synchronous gauge _ condition , @xmath173 is imposed . likewise , the third group of terms vanish when the trace free condition , @xmath174 , is imposed . the total divergence terms are not displayed as they will be negligible under averaging . only the first group of terms , with the sum over @xmath175 restricted to sum over spatial indices @xmath176 , survive for gauge fixed solutions . the ricci scalar , @xmath177 is obtained as , @xmath178 next , the @xmath179 terms are eliminated using the source - free linearized equation ( [ lineqn ] ) , since we are interested in the ripple stress tensor . substitution leads to , @xmath180 \right\rangle \ ] ] this expression reduces to the stress tensor for the minkowski background by taking @xmath181 and dropping the last two terms . however , for the ripple , @xmath182 . the connection terms in the covariant derivatives are order @xmath183 . hence , to the leading order in @xmath184 , all the terms without _ derivatives of the ripple _ , can be dropped and we are back to the same expression for the minkowski background . notice that the leading term has no @xmath29 . this is further simplified in the conformal coordinates by substituting @xmath185 and also expressing the covariant derivatives in terms of the minkowski derivatives . once again , dropping the derivatives of the conformal factor and using the synchronous gauge and the trace - free condition on the field , we write the gauge invariant stress tensor for ripples , @xmath186 let us compute the @xmath187 from the solution given in . as is well known , to a leading order in @xmath188 , the tt part can be obtained approximately by an algebraic projection and @xmath189 . these two expressions being not equal is important for the angular momentum flux . ] : @xmath190 we need the derivatives of @xmath191 . the @xmath192 passes through the @xmath193projector and so does @xmath194 . for spatial derivatives , @xmath195 , we get a contribution from the projector , @xmath196 . \ ] ] while on @xmath84 the spatial derivative is @xmath197 . hence , @xmath198 using the derivatives of @xmath84 given in , we can write , @xmath199\chi_{kl } ) . \ ] ] the first term is proportional to a null vector . the second term is proportional to the space - like , radial vector . the third is again a space - like vector . both the second and the third terms are down by a power of @xmath39 relative to @xmath84 and therefore also relative to the first term . _ when _ the second and the third terms can be neglected , the ripple stress tensor corresponds to an _ out - going null dust _ with energy density proportional to @xmath200 . we will see later in the calculation of the fluxes that for energy and momentum , the second and the third terms can be neglected . however for flux of angular momentum , the third term is crucial . given _ any _ symmetric , conserved stress tensor , for every killing vector of the background space - time , @xmath57 , the current @xmath201 , is covariantly conserved . such a current is defined to within a constant multiple since the killing vector is defined to within a constant multiple , including a sign . an integral curve of a _ time - like _ killing vector can represent an observer and for a future directed worldline , the observer measures , say , the matter energy density to be @xmath202 which is _ positive _ for normal matter . the observer also expects the energy momentum current representing the flow of energy - momentum to obey special relativity i.e. should be a vector which is time - like / null and future directed . thus @xmath203 must also hold ( our metric signature is ( - , + , + , + ) ) . this implies that the energy - momentum _ current _ , @xmath204 , must be defined with sign : @xmath205 . this fixes the _ sign _ for a future directed time - like killing vector . the normalization can only be done for _ one _ killing observer and this is chosen as per the context . we adopt this definition for the time translation killing vector @xmath206 . for the time translation killing vector field , @xmath207 involves only the @xmath37 and @xmath39 derivatives since @xmath208 . these pass through the @xmath193projector . for the space translation along @xmath209 direction , we have @xmath210 which does act on the @xmath193projector . the @xmath211 . in the present context where derivatives of the ripple dominate over ( ripple / r ) , the derivative of the projector can be neglected . for generators of rotation however the situation is different . once again we get two term from the @xmath195 , but now the @xmath212 and we can no longer neglect the derivative of the projector . with these understood , we write the the corresponding currents , @xmath213 . note that the gravitational stress tensor has been defined as a covariant rank 2 tensor and hence there is the factor of @xmath214 since the index @xmath215 has been raised . @xmath216 notice that for the energy and momentum currents ( [ energycurrent ] , [ momentumcurrent ] ) , both fields have the tt label whereas for the angular momentum current , the second factor does _ not _ have the tt label . the entire contribution to the angular momentum current comes from the derivative of the @xmath193projector . the contribution from the derivative of the field vanishes since the field ( without the projector ) is spherically symmetric . in all these equations we may use @xmath217 from . we note in passing that _ if _ the @xmath218 can be neglected compared to @xmath219 , then the currents corresponding to the generators of time and space translations , both become _ proportional _ to the vector @xmath220 which is null and @xmath221 is also future directed . both energy and momentum propagate along this direction . finally we note the integral form of the conservation equation . let @xmath222 denote a space - time region with a boundary @xmath223 . then it follows that , @xmath224 where @xmath225 is the oriented volume element of the boundary @xmath223 . in the next subsection we evaluate the _ energy flux _ , @xmath226 , for various hyper - surfaces , @xmath227 s . these , together with the conservation equation ( [ generalconservationeqn ] ) will be used to relate power received at @xmath0 to that crossing the cosmological horizon . in the following subsection , we will present the fluxes for momentum and angular momentum . we present flux calculations for three classes of hypersurfaces : ( a ) hypersurfaces of constant physical radial distance , ( b ) space - like hypersurfaces of constant @xmath37 and ( c ) the out - going and in - coming null hypersurfaces . constant hypersurfaces which are time - like for @xmath228 , null for @xmath229 and space - like for @xmath230 . the two 45 degree out - going null hypersurfaces intersecting the @xmath1 and @xmath0 in the spheres at @xmath231 , bound a space - time region . the figure on the right shows the space - like hypersurfaces with constant value of @xmath37 . the fluxes across the out - going null hypersurfaces turn out to be zero signifying sharp propagation of energy - momentum and angular momentum . hence the energy flux across the portion of the horizon bounded by the spheres at @xmath232 equals the flux across the portion of the future infinity bounded by the spheres at @xmath233.,scaledwidth=100.0% ] these hypersurfaces are time - like , null and space - like according as the physical distance being less than , equal to and greater than the physical distance to the cosmological horizon , namely @xmath234 . they are spanned by the integral curves of the killing vector @xmath43 . this killing vector is special because in the static patch , it is time - like and its integral curves represent killing observers . denoting @xmath235 , in general , its integral curves are given by @xmath236 . evidently , along each curve , @xmath237 is constant . this also represents the _ physical radial distance _ , @xmath238 . we compute the flux across the hypersurface @xmath239 , defined by @xmath240 . this surface is coordinatized by the killing parameter @xmath127 and the usual spherical angles @xmath241 represented by the unit vectors @xmath242 . these hypersurfaces are topologically @xmath243 and their embedding is given by , @xmath244 with @xmath245 . the induced metric is given by @xmath246 . this has lorentzian signature for @xmath228 ( inside the static patch ) , is degenerate for @xmath229 ( the cosmological horizon ) and euclidean signature for @xmath230 ( beyond the cosmological horizon ) . the measure factor for the non - null cases is given by @xmath247 while on the cosmological horizon it is given by @xmath248 . here @xmath249 denote the ` transverse ' coordinates @xmath241 . in the non - null case , the unit normal is given by @xmath250 . here @xmath251 for _ time - like _ @xmath239 ( @xmath228 ) and @xmath252 for _ space - like _ @xmath239 ( @xmath230 ) . on the cosmological horizon , we _ choose _ the normal to be : @xmath253 , so that @xmath254 is future directed . introduce @xmath255 , so that the normal for non - null cases is expressed as @xmath256 . note that the @xmath12 is the same for the space - like and the null hypersurfaces , @xmath257 . for the time - like hypersurface , the @xmath12 points in the opposite direction . however , the induced orientation on @xmath239 is also reversed as the hypersurface changes from being space - like to being time - like . hence , in _ all cases , @xmath258 _ , @xmath259 the minus sign in front of the hypersurface integral is because the orientation defined by the killing parameter and the angles , is _ negative _ relative to that defined by the @xmath39 and the angles . the @xmath260 is absorbed in @xmath261 . the minus sign in the last parentheses is due to the definition @xmath262 . in the second line , we have also used @xmath263 valid on @xmath239 . substituting for the effective stress tensor , the expression within the braces becomes , @xmath264 in the above , @xmath84 is the tt part of the solution given in eqns . ( [ cosmogwamplitude],[liederivative ] ) . the tt projection introduces angle dependence in the @xmath191 , however equation ( [ fluxeqn0 ] ) needs only @xmath37 and @xmath39 derivatives . eliminating @xmath265 in eqn . ( [ aveintegrand ] ) using equation , we write , @xmath266\big\ { \big\ } \label{physicalflux } \ ] ] in arriving at this conclusion we have used the approximated expression valid for ( source dimension)/(distance to the source ) @xmath279 for the solution . this is consistent with the assumption that the @xmath280 . the source being rapidly changing , @xmath281 , it follows that @xmath282 . hence we dropped @xmath283 terms . effectively , this sets @xmath284 and @xmath285 takes a simple quadratic form @xmath286 . we should also note that the effective stress tensor is gauge invariant only to the leading order in @xmath287 and the flux is quadratic in the @xmath124 . hence , consistency requires that @xmath124 should also be used only to order @xmath288 . the full flux integral is independent of @xmath274 and is non - zero . however , the @xmath289 corresponds to the _ line _ @xmath42 and the hypersurface degenerates to a time - like line . the limit should therefore be zero . our answer is non - zero because we used the approximate solution which is singular as @xmath290 . to evaluate the limit , we we should use the form of the solution near @xmath291 . the @xmath292 is then finite as @xmath290 and then the explicit factors of @xmath293 ( or @xmath294 ) in gives vanishing flux . the @xmath274 independence of the full flux integral in particular means that the total flux across @xmath0 equals the total flux across the cosmological horizon , @xmath1 . @xmath295 the hypersurface @xmath296 defined by @xmath297 is a cosmological slice @xmath298 . it is space - like , with a normal @xmath299 $ ] . the hypersurface is topologically @xmath300 . choosing the @xmath301 coordinates on the hypersurface , the embedding is given by @xmath302 the induced metric is given by @xmath303 giving @xmath304 . denoting @xmath305 , the hypersurface integral is given by , @xmath306\mathcal{r}_{mn}^{tt}\mathcal{r}^{mn}_{tt } \label{etaflux } \ ] ] by the same reasoning as before , we have dropped the @xmath307 term in the last equation . in the limit @xmath308 with @xmath309 , the hypersurface becomes @xmath0 and the integration measure becomes @xmath310 . the limit @xmath311 is thus finite . as noted earlier , the hypersurface integral when expressed in terms of the killing parameter , has a minus sign due to the reversal of the induced orientation . the measures ( positive ) themselves are related as @xmath312 , leading to @xmath313 and we get , @xmath314 and this matches exactly with the flux defined by the covariant phase space method when the limits of integration on the killing parameter are extend to @xmath315 . there are two families of _ future directed _ null hypersurfaces given by @xmath316 , see figure ( [ fluxfig ] ) . for @xmath317 , these 45 degree lines in the penrose diagram are parallel to the cosmological horizon while for @xmath318 , the lines are parallel to the null boundary of the poincare patch . we refer to these as the _ in - coming _ ( @xmath319 ) and _ out - going _ ( @xmath318 ) null hypersurfaces . the parameter @xmath320 labels members of these families . the null normals of these families are of the form @xmath321 , where @xmath322 is to be chosen suitably and should be _ negative _ for future directed hypersurfaces . choosing coordinates @xmath323 on a null hypersurface , its embedding may be taken as @xmath324 with identity mapping of the angles . here @xmath132 is an affine parameter of the null geodesics generating the null hypersurfaces . the induced metric is obtained as @xmath325 . note that the orientation of the hypersurfaces , relative to that defined by @xmath301 is the same for the out - going hypersurfaces and opposite for the in - coming hypersurfaces . the hypersurface integral is then given by ( @xmath326 ) , @xmath327 ( h^2\eta^2)\gamma ( -n^{\mu}t_{\mu\nu}t^{\nu } ) ~~ , \\ n^{\mu}t_{\mu\nu}t^{\nu } & = & - h\left ( -t_{\eta\eta}\eta - t_{\eta j}r\hat{x}^j + \epsilon\hat{x}^it_{i\eta}\eta + \epsilon r \hat{x}^it_{ij}\hat{x}^j \right ) \nonumber \\ & = & -\frac{h}{32\pi}\left ( -\eta\partial_{\eta}\chi_{mn}\partial_{\eta}\chi^{mn } + ( \epsilon\eta - r)\partial_{\eta}\chi_{mn}\partial_r\chi^{mn } + \epsilon r \partial_r\chi_{mn}\partial_r\chi^{mn}\right ) \nonumber \\ & = & -\frac{h}{32\pi}(1 + \epsilon)(r - \eta)\langle\partial_{\eta}\chi^{tt}_{mn}\partial_{\eta}\chi_{tt}^{mn}\rangle + o(r^{-2 } ) \\ & = & -\frac{h}{32\pi}(1 + \epsilon)(r - \eta)\left[\frac{4}{r^2}\frac{\eta^2}{(\eta - r)^2}\right]\langle\mathcal{r}^{tt}_{mn}\mathcal{r}_{tt}^{mn}\rangle + o(r^{-2 } ) \\ \therefore \int_{\sigma_{(\epsilon , \sigma)}}d\sigma_{\mu}j^{\mu}_t & = & \epsilon\int^{\lambda_2}_{\lambda_1}d\lambda\int_{s^2}d^2s\left[\gamma h\right]\left [ \frac{(1 + \epsilon)}{8\pi}\frac{\eta^2}{\eta - r}\right ] \langle\mathcal{r}^{tt}_{mn}\mathcal{r}_{tt}^{mn}\rangle \label{nullflux } \ ] ] as before , we have dropped the @xmath283 terms from @xmath328 . it is immediately clear that the flux through the out - going null hypersurfaces ( @xmath274 or @xmath39 increase along these ) vanishes . in the @xmath319 family , only the cosmological horizon is of interest . for this we have @xmath329 and we _ choose _ the factor @xmath330 so the null normal matches with the killing vector ( @xmath322 is negative as desired for future orientation ) and the affine parameter @xmath132 matches with the killing parameter @xmath127 . with this choice , the flux in eqn . ( [ nullflux ] ) matches with that given in eq . ( [ rhoflux ] ) for @xmath229 . thus , once again , the full flux through cosmological horizon is exactly same as that of @xmath331 hypersurfaces . _ remarks : _ all three calculations consistently have the same @xmath332 $ ] factor , with integrals oriented along the stationary killing vector . it is surprising at first that the flux through @xmath333 hypersurfaces is zero , which indicates sharp propagation of the energy , even though the retarded solution has a tail contribution . this can be seen more directly as follows . let us recast eqn . as @xmath334 where we have neglected the @xmath335 terms and have used @xmath336 . now in taking the @xmath88derivative , contribution of the tail term in cancels out , leaving only the contribution from the sharp term : @xmath337 for the same three classes of hypersurfaces , we present the momentum and angular momentum fluxes . we already have the measures for these hypersurfaces as well as the currents given in ( [ momentumcurrent ] , [ angularmomentumcurrent ] ) . the full fluxes , only to the leading order in @xmath188 , are given by , @xmath338\left[\frac{1}{h\eta_0}\right](1 , \vec{0})j^{\mu } \\ \sigma_{(\epsilon , \sigma ) } & : & -\epsilon \int_{\lambda_1}^{\lambda_2}d\lambda\int_{s^2}d^2s \left[\frac{r^2\gamma}{|h\eta_0|^{2}}\right](1 , \epsilon\hat{x}_i)j^{\mu } \ ] ] _ momentum fluxes : _ the momentum current is given by , @xmath339 ^ 2\langle\mathcal{r}^{mn}_{tt}~ \mathcal{r}^{tt}_{mn}\rangle(\hat{x}_j)(1 , -\hat{x}^i ) \ ] ] dotting with the @xmath340 produces a rotational scalar and the average is a rotational scalar too . then the angular integration with @xmath341 vanishes , in all three cases . hence , the momentum flux is zero across the three classes of hypersurfaces . _ angular momentum fluxes : _ replacing @xmath342 , we can write the angular momentum current as , @xmath343 \left[\epsilon_{jmn}\hat{x}^m\hat{x}^k\langle\mathcal{r}^{nl}_{tt}~\chi_{kl}\rangle\right ] ( 1 , \hat{x}^i ) \ ] ] the fluxes then take the form , @xmath344 \\ \sigma_{\eta_0 } & : & \frac{1}{8\pi h^2|\eta_0| } \int_0^{\infty}dr\frac{r}{r -\eta_0 } \int_{s^2}d^2s \left[\epsilon_{jmn}\hat{x}^m\hat{x}^k \langle\mathcal{r}^{nl}_{tt}~\chi_{kl}\rangle\right ] \\ \sigma_{(\epsilon , \sigma ) } & : & -\epsilon\frac{1 + \epsilon}{8\pi } \int_{\lambda_{1}}^{\lambda^2}d\lambda ( -\gamma ) \frac{r\ \eta } { \eta -~r}\int_{s^2}d^2s \left[\epsilon_{jmn}\hat{x}^m\hat{x}^k \langle\mathcal{r}^{nl}_{tt}~\chi_{kl}\rangle\right ] \ ] ] consider the average . the function enclosed in averaging is product of the @xmath193projector containing angular dependence and a function having dependence on @xmath345 . the averaging can then be split into averaging over a sphere and averaging over a cell in the @xmath345 plane . thus , we write , @xmath346 \left[\langle\mathcal{r}^{rs}\chi_{kl}\rangle\right ] \\ & = & \frac{1}{15}\left[3(\delta^n_r\delta^l_s + \delta^n_s\delta^l_r ) - 2\delta^{nl}\delta_{rs}\right ] \left[\langle\mathcal{r}^{rs}\chi_{kl}\rangle\right ] \\ & = & \frac{2}{15}\langle\left(3 \mathcal{r}^{nl}\chi_{kl } - \left(\mathcal{r}^{rs}\delta_{rs}\right)\chi^n_{~l}\right)\rangle \ ] ] the remaining factor of @xmath347 also generates @xmath348 after subsequent angular integration leading to , @xmath349 & = & \frac{8\pi}{15}\epsilon_{jmn}\langle\mathcal{r}^{nl}\chi_{l}^{~m}\rangle \ ] ] the averaging now involves only a @xmath345 cell . the remaining integration in the flux expressions above , can be expressed in terms of the killing parameter @xmath127 and then they all take the same form _ provided _ for @xmath296 we consider the @xmath350 and for the null hypersurface we choose the cosmological horizon , @xmath1 ( @xmath351 : @xmath352 the radiation field @xmath353 is given in equation ( [ radiationfield ] ) but without the @xmath63 label and , @xmath354(\eta_{ret } ) \nonumber \\ & & + 2h^2\left[\dot{q}_{lm } + h\bar{q}_{lm}\right](\eta_{ret } ) - 2h^2\left[\dot{q}_{lm } + h\bar{q}_{lm}\right](-\infty ) \ . \ ] ] this flux does _ not _ have a finite limit to @xmath0 due to the _ tail term _ in @xmath355 and does _ not _ match with the flux given by @xcite . it is finite along the @xmath1 though . it does _ not _ match with the correct angular momentum flux in the flat space limit as well and it is well known @xcite that the isaacson effective stress tensor does not suffice to capture the flux of angular momentum . the sharp propagation property still holds in the sense that the flux across out - going null hypersurface is zero . in the next section we restrict to the energy fluxes and see two applications of the conservation equation and the sharp propagation property . in the previous subsection , we assembled fluxes through various hypersurfaces , all having the topology @xmath356 . we considered @xmath357 to be a finite interval and also the cases with @xmath358 . in all cases , the energy flux integral had the form , @xmath359 \mathcal{r}_{mn}^{tt}\mathcal{r}^{mn}_{tt } \ . \ ] ] by judicious choices of hypersurfaces comprising the boundary @xmath360 of a space - time region @xmath361 , we can relate different fluxes using the conservation equation . the sharp propagation of energy comes in very useful . we note two of its implications . \(1 ) the flux across two hypersurfaces @xmath362 and @xmath363 _ can not _ be equal , see the right side figure of ( [ fluxfig ] ) . let @xmath38 . let @xmath362 meet the @xmath42 line at @xmath364 . let the out - going null hypersurface through @xmath364 intersect the @xmath363 in a @xmath365 at @xmath366 with the radial coordinate being @xmath367 . the three hypersurfaces @xmath362 , the out - going null hypersurface and the hypersurface @xmath363 bounded by the sphere at @xmath366 enclose a space - time region , @xmath368 . by the conservation equation , the sum of the fluxes through these bounding hypersurfaces must vanish . but the flux through the out - going null hypersurface vanishes as shown before . hence the fluxes through @xmath362 and the partial hypersurface @xmath363 between @xmath366 and @xmath369 , must be equal . however , this leaves the contribution of the flux through the ` remaining ' portion of the @xmath363 hypersurface between @xmath370 and @xmath366 . hence the result . alternatively , one can also see this explicitly by writing the full flux through the two hypersurfaces using the expression given in equation and matching the integrands along the out - going null hypersurface . evidently , the full flux through @xmath371 is also not equal to that through @xmath0 . physically this is understandable since the hypersurface at a later value of @xmath37 receives energy emitted _ after _ the earlier value of @xmath37 . the null infinity of course records _ all _ the energy emitted by the source and so does the cosmological horizon . we also conclude that the total flux at @xmath0 computed by ashtekar et al matches with that given in equation for @xmath372 . note that unlike the spatial slices @xmath100 , all hypersurfaces @xmath373 intercept all the emitted energy . \(2 ) the sharp propagation of energy can also be used to infer the _ instantaneous emitted power_. consider two out - going null hypersurfaces intersecting the cosmological horizon in spheres with radii @xmath374 and @xmath375 . the same hypersurfaces intersect the null infinity at corresponding spheres at @xmath376 and @xmath377 , see the left side figure in ( [ fluxfig ] ) . for @xmath378 , we have @xmath379 and @xmath380 . by the conservation equation and sharp propagation , the flux integral over the portion bounded by the spheres @xmath381 on @xmath0 and the flux integral over the portion bounded by the spheres @xmath232 on the @xmath1 , are equal . taking @xmath382 , the integral becomes @xmath383 the integral over the sphere at @xmath374 . the _ emitted power _ is then defined by dividing the flux integral by @xmath384 and taking the limit . thus we get the instantaneous power as : @xmath385 this matches with the definition given by ashtekar et al @xcite and is manifestly positive . since the power defined at @xmath0 in equation is gauge invariant and that defined at @xmath1 in equation using the effective stress tensor are equal , we conclude that the power computed using the effective stress tensor is gauge invariant . thus , the quadrupole power can be computed at the cosmological horizon . we have dealt with two aspects namely the role of the cosmological horizon and the use of effective gravitational stress tensor in the limited context of rapidly changing , distant sources . as noted in the introduction , the cosmological horizon is unambiguously defined for a spatially compact source . this follows because worldlines with finite physical separation at every @xmath37 must converge to @xmath386 , the point @xmath387 of figure ( [ desitterpenrose ] ) . if @xmath357 denotes the physical radial distance corresponding to the radial coordinate difference @xmath388 , then @xmath389 . to maintain @xmath390 to be finite as @xmath44 , we must have @xmath391 near @xmath386 . this identifies @xmath388 with @xmath392 or @xmath393 . thus , the worldlines approach @xmath386 along the @xmath394constant hypersurfaces . the cosmological horizon is then the past lightcone of @xmath386 . the same argument also shows that any observer , who remains at finite physical distance away from the source must remain confined within the cosmological horizon . furthermore , neither any such observer , nor the source has any access to energy / momentum which has crossed the horizon . hence cosmological horizon does share physically relevant properties with the future infinity . incidentally , any future directed causal curve reaching @xmath0 also registers on @xmath1 . further support for the role of cosmological horizon as future null infinity comes from the computations of the energy momentum fluxes . for these , we employed the effective gravitational stress tensor and showed that the fluxes defined at @xmath0 by more geometric methods , also matched with those computed at the horizon . in particular the quadrupole power can be evaluated at the horizon . the effective gravitational stress tensor , although limited to short wavelength regimes ( which covers most common sources ) , provides a convenient picture of energy momentum flows much like those for matter . there is a shortcoming of the effective stress tensor - it does not capture the angular momentum flux correctly . a clearer understanding of this failure is lacking at present . it should be noted that definition of fluxes is not necessarily _ unique_. apart from a definition being well defined , its ` correctness ' should be tested in conjunction with the definition of the bondi - type quantities having a loss formula relating to flux . the observation that the cosmological horizon is a killing horizon and hence an isolated horizon should be helpful in this regard . h. j. de vega , j. ramirez and n. sanchez , generation of gravitational waves by generic sources in de sitter space - time , _ phys . d _ , * 60 * , 044007 , ( 1999 ) , [ arxiv : astro - ph/9812465 ] . a. ashtekar , b. bonga and a. kesavan , asymptotics with a positive cosmological constant : iii . the quadrupole formula , _ phys . rev . d _ , * 92 * , 10432 ( 2015 ) , arxiv:1510.05593 . g date and sk . jahanur hoque , gravitational waves from compact sources in de sitter background , _ phys . d _ , * 94 * , 064039 ( 2016 ) , [ arxiv:1510.07856 ] . r sachs , gravitational waves in general relativity vi . the outgoing radiation condition , _ proc . a _ , * 264 * , 309 - 338 , ( 1961 ) . r penrose , zero rest - mass fields including gravitation : asymptotic behaviour , _ proc . a _ , * 284 * , 159 - 203 , ( 1963 ) . a. ashtekar , l. bombelli , and o. reula , covariant phase space of asymptotically flat gravitational fields ; in : mechanics , analysis and geometry : 200 years after lagrange , edited by m. francaviglia and d. holm ( north holland , amsterdam 1991 ) . a. ashtekar , b. bonga and a. kesavan , asymptotics with a positive cosmological constant : ii . linear fields on de sitter space - time , _ phys . * 92 * , 044011 ( 2015 ) , [ arxiv:1506.06152 ] ; r. a. isaacson , gravitational radiation in the limit of high frequency i : the linear approximation and geometrical optics , _ phys . rev . _ , * 166 * , 1263 , ( 1968 ) ; r. a. isaacson , gravitational radiation in the limit of high frequency ii : non - linear terms and the effective stress tensor , _ phys _ , * 166 * , 1272 , ( 1968 ) ; charles w. misner , kip s. thorne , and john archibald wheeler . _ gravitation _ , macmillan , ( 1973 ) . a. ashtekar , b. bonga and a. kesavan , asymptotics with a positive cosmological constant : i. basic framework , _ class . _ , * 32 * , 025004 , ( 2014 ) , [ arxiv:1409.3816 ] ; y choquet - bruhat , construction de solutions radiatives approchees des equations deinstein , _ comm . _ , * 12 * , 16 - 35 , ( 1969 ) . eric poisson and clifford m. will , _ gravity newtonian , post - newtonian , relativistic _ , cambridge univ . press , ( 2014 ) .

we calculate the masses of baryons containing one , two , or three heavy quarks using lattice qcd . we consider all possible combinations of charm and bottom quarks , and compute a total of 36 different states with and . we use domain - wall fermions for the up , down , and strange quarks , a relativistic heavy - quark action for the charm quarks , and nonrelativistic qcd for the bottom quarks . our analysis includes results from two different lattice spacings and seven different pion masses . we perform extrapolations of the baryon masses to the continuum limit and to the physical pion mass using heavy - hadron chiral perturbation theory including and finite - volume effects . for the 14 singly heavy baryons that have already been observed , our results agree with the experimental values within the uncertainties . we compare our predictions for the hitherto unobserved states with other lattice calculations and quark - model studies . = 1

1409.0497

baryons containing heavy quarks are interesting both from the theoretical and experimental points of view . because the bottom and charm quark masses are greater than the intrinsic energy scale of qcd , approximate heavy - quark flavor and spin symmetries constrain the spectrum and dynamics of heavy baryons @xcite . singly charmed and singly bottom baryons exhibit a similar spectrum of excitations of the light degrees of freedom . interactions with the spin of the heavy quark , and hence the hyperfine splittings , are suppressed by @xmath3 . a particularly interesting symmetry emerges for doubly heavy baryons : in the large - mass limit , the two heavy quarks are expected to form a point - like diquark that acts like a single heavy antiquark , and the light degrees of freedom behave as in a heavy - light meson @xcite . the ratio of hyperfine splittings of doubly heavy baryons ( with two equal heavy - quark flavors ) and singly heavy mesons is predicted to approach the value @xmath4 in the heavy - quark limit @xcite . finally , triply heavy baryons can be viewed as baryonic analogues of heavy quarkonia , making them very interesting systems to study in effective field theories and perturbative qcd @xcite . the masses of all low - lying singly charmed baryons with @xmath0 and @xmath1 , and of most of their singly bottom partners , are well known from experiments @xcite . in this sector , the most recent discoveries are the @xmath5 @xcite , and a state that is likely the @xmath1 @xmath6 @xcite ; the @xmath7 an @xmath8 remain to be found . the @xmath5 masses reported by d@xmath9 @xcite and cdf @xcite are inconsistent with each other , but a recent more precise measurement by the lhcb collaboration @xcite agrees with the cdf result . in contrast to the singly heavy baryons , the arena of doubly and triply heavy baryons remains experimentally unexplored to a large extent , with the only possibly observed state being the @xmath10 . the discovery of the @xmath10 was reported by the selex collaboration @xcite , but subsequent searches for this state by the focus @xcite , babar @xcite , belle @xcite , and lhcb collaborations @xcite returned negative results . nevertheless , there is still potential for discoveries of doubly and triply heavy baryons at the lhc @xcite and perhaps also at the coming generation of spectroscopy experiments at besiii @xcite , belle ii @xcite , and panda @xcite . lattice qcd can predict the masses and other properties of heavy baryons from first principles , and can help resolve experimental controversies such as those surrounding the @xmath5 and @xmath11 . for the doubly and triply heavy baryons , which may remain beyond the reach of experiments at the present time , lattice qcd results can also serve as a benchmark for other theoretical approaches , such as quark models and perturbative qcd . complete control over all sources of systematic uncertainties , including the nonzero lattice spacing and finite lattice volume , unphysical values used for the quark masses , any approximations made for the heavy quarks , as well as excited - state contamination in the correlation functions , is essential in both of these contexts . most lattice calculations of heavy baryon masses that have been published to date are still lacking in some of these aspects . the earliest studies @xcite were performed in the quenched approximation , removing the effects of sea quarks to reduce the computational cost but at the expense of connection to experiment . the first unquenched calculations were reported in refs . @xcite . since then , additional unquenched calculations have been performed with various choices of lattice actions for the light and heavy quarks @xcite ; reviews can be found in refs . @xcite . in this paper , we present the first lattice qcd determination of singly , doubly , and triply heavy baryon masses that includes both charm and bottom quarks in any combination , and also achieves good control over all major sources of systematic uncertainties . our calculation includes dynamical up , down , and strange quarks implemented with a domain - wall action @xcite , and is performed at two different lattice spacings and seven different values of the up / down quark mass corresponding to pion masses as low as 227(3 ) mev . because the masses of the charm and bottom quarks are not small in units of the lattice spacing , special heavy - quark actions are needed for them to avoid large discretization errors . we use a relativistic heavy - quark action @xcite for the charm quarks and improved nonrelativistic qcd @xcite for the bottom quarks . details of the actions and parameters are given in sec . [ sec : actions ] . the interpolating fields we use for the heavy baryons and our methodology for fitting the two - point functions are described in sec . [ sec:2pt ] . we extrapolate the results for all baryon masses to the physical pion mass and the continuum limit as explained in sec . [ sec : extrap ] . for the singly and doubly heavy baryon masses , heavy - hadron chiral perturbation theory at next - to - leading order is used to fit the light - quark mass dependence and to remove the leading finite - volume effects . because some of our data sets use valence light - quark masses lower than the sea - quark masses , we use the partially quenched extension of heavy - hadron chiral perturbation theory @xcite . for the singly heavy baryons , we generalize the expressions given in ref . @xcite to include hyperfine splittings . the final results for the baryon masses and mass splittings are presented in sec . [ sec : finalresults ] , which also includes a detailed discussion of the systematic uncertainties . we conclude in sec . [ sec : conclusions ] with a comparison of our results to the literature . in this work , we performed the euclidean path integral using ensembles of gauge field configurations generated by the rbc and ukqcd collaborations @xcite . these ensembles include the effects of dynamical up- , down- and strange quarks , implemented with a domain - wall action @xcite . the quark fields in this action depend on an auxiliary fifth dimension with extent @xmath12 . four - dimensional quark fields , for which the low - energy effective field theory obeys an exact lattice chiral symmetry in the limit @xmath13 , are obtained in terms of the quark fields at the boundaries @xmath14 and @xmath15 @xcite . the rbc and ukqcd collaborations chose the iwasaki gauge action @xcite , which , compared to the standard wilson or symanzik gauge actions , reduces the residual chiral symmetry breaking of the domain - wall action at finite @xmath12 @xcite . for hadron spectroscopy calculations , the primary benefit of approximate chiral symmetry is the smallness of @xmath16 discretization errors . here we selected four different ensembles of gauge fields : two ensemble with lattice size @xmath17 and lattice spacing @xmath18 fm ( in the following referred to as `` coarse '' ) , and two ensembles with lattice size @xmath19 and lattice spacing @xmath20 fm ( in the following referred to as `` fine '' ) @xcite . all ensembles have @xmath21 , and the domain - wall height is @xmath22 . the residual chiral symmetry breaking can be quantified by the residual additive quark - mass renormalization , @xmath23 . the coarse ensembles have @xmath24 , while the fine ensembles have @xmath25 @xcite . we work in the isospin limit @xmath26 ; this means that our results for the baryon masses should be considered as isospin - averaged values . we computed domain - wall light and strange quark propagators with various quark masses @xmath27 and @xmath28 as shown in table [ tab : params ] , leading to eight different data sets in total . for the four data sets ` c104 ` , ` c54 ` , ` f43 ` , and ` f63 ` , the valence quark masses are equal to the sea quark masses . the data sets ` c14 ` , ` c24 ` , and ` f23 ` have @xmath29 in order to achieve lighter valence pion masses , and the data set ` c53 ` has @xmath30 to enable interpolations to the physical strange quark mass . we used about 200 gauge configurations from each ensemble , and computed domain - wall propagators for multiple source locations on each configuration . the resulting total numbers of light / strange propagator pairs in each data set are given in the last column of table [ tab : params ] . to further increase the statistical precision of our calculations , we computed hadron two - point functions propagating both forward and backward in euclidean time and averaged over these . .[tab : params]properties of the gauge field ensembles @xcite and of the light / strange quark propagators we computed on them . here , @xmath31 and @xmath32 are the numbers of lattice points in the spatial and temporal directions , @xmath33 is the gauge coupling , @xmath34 and @xmath35 are the light and strange sea quark masses , and @xmath36 is the lattice spacing ( determined in ref . the valence quark masses used for the calculation of the light and strange quark propagators are denoted by @xmath27 and @xmath28 . the corresponding valence pion and @xmath37 masses are denoted as @xmath38 and @xmath39 . the @xmath37 is an artificial @xmath40 state that is defined by treating the @xmath41 and @xmath42 as different , but mass - degenerate flavors . this state is useful as an intermediate quantity to tune the strange - quark mass @xcite ; its mass at the physical point has been computed precisely by the hpqcd collaboration and is @xmath43 @xcite . in the last column of the table , @xmath44 is the number of pairs of light and strange quark propagators computed in each data set . [ cols="^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^",options="header " , ] no other dynamical lattice qcd calculations have been published so far for mixed charm - bottom baryons ( results of a quenched lattice calculation can be found in ref . we therefore compare our lattice qcd results for the masses of these baryons to predictions from potential models , qcd sum rules , and other continuum - based approaches . these comparisons are shown in fig . [ fig : mixeddoublyheavycomp ] for the @xmath45 , @xmath46 , @xmath47 , @xmath48 , @xmath49 , and @xmath50 , and in fig . [ fig : triplyheavymixedcomp ] for the @xmath51 , @xmath52 , @xmath53 , and @xmath54 . it is evident that the mass predictions in the literature cover ranges far wider than our uncertainties . we hope that our lattice qcd results provide a useful benchmark for future studies of these interesting systems . comparison of our lattice qcd results for the @xmath45 , @xmath46 , @xmath47 , @xmath48 , @xmath49 , and @xmath50 baryon masses with estimates based on continuum methods , including quark models and qcd sum rules @xcite . from refs . @xcite ( silvestre - brac , 1996 ) and @xcite ( ghalenovi _ et al . _ , 2014 ) , we show results for multiple different choices of the interquark potentials . note that the bag - model calculation of ref . @xcite ( he _ et al . _ , 2004 ) predicts @xmath55 and @xmath56 , and the qcd sum rule calculation of ref . @xcite ( tang _ et al . _ , 2012 ) predicts @xmath57 and @xmath58 , both rather unusual . the sum - rule calculation of ref . @xcite ( zhang _ et al . _ , 2008 ) gives extremely large hyperfine splittings @xmath59 and @xmath60 [ our results for the hyperfine splittings are @xmath61 , @xmath62 ; the @xmath47 mass from ref . @xcite is beyond the upper limit of the plot . ] comparison of our lattice qcd results for the masses of triply heavy charm - bottom baryons with estimates based on continuum methods , including quark models , qcd sum rules , and perturbative qcd @xcite . from refs . @xcite ( silvestre - brac , 1996 ) and @xcite ( ghalenovi _ et al . _ , 2014 ) , we show results for multiple different choices of the interquark potentials . from ref . @xcite , which used the static three - quark potential from perturbative qcd , we show the nnlo results from table 16 . not shown in this plot are the results of the qcd sum - rule calculation of ref . @xcite , which are far lower than all other results . ] the chiral function @xmath64 results from the evaluation of a one - loop self - energy diagram , where the internal baryon has a mass difference @xmath65 from the external baryon , in heavy - hadron chiral perturbation theory . we write it as the sum of the infinite - volume part and a finite - volume correction , @xmath66 the infinite - volume part is given by @xmath67 where @xmath68 \nonumber.\ ] ] here , we use a renormalization scheme in which the real part of @xmath69 vanishes in the chiral limit @xcite . our definition of @xmath69 differs from the one used in ref . @xcite by the term @xmath70 . an approximate expression for the finite - volume correction is given by @xcite @xmath71 where @xmath72 , @xmath73 , @xmath74 , and @xmath75 + \frac{1}{u m l } \bigg [ \frac{1}{\sqrt{\pi } } \left ( \frac{9z}{4 } - \frac{z^{3}}{2}\right ) + \left(\frac{z^{4}}{2}-2\,z^2\right)e^{(z^{2 } ) } \big [ 1 - { \mathrm{erf}}(z)\big ] \bigg ] \\ & & -\frac{1}{(u m l)^2 } \bigg [ \frac{1}{\sqrt{\pi}}\left ( -\frac{39z}{64 } + \frac{11z^{3}}{32 } -\frac{9z^{5}}{16 } + \frac{z^{7}}{8 } \right ) -\left ( -\frac{z^{6}}{2 } + \frac{z^{8}}{8}\right ) e^{(z^{2 } ) } \big [ 1 - { \mathrm{erf}}(z)\big ] \bigg ] + \mathcal{o}\left ( \frac{1}{(u m l)^{3}}\right ) , \nonumber\end{aligned}\ ] ] with @xmath76 we thank the rbc and ukqcd collaborations for making their gauge field configurations publicly available . sm was supported by the u.s . department of energy under cooperative research agreement contract number de - fg02 - 94er40818 . ko and zsb were supported by the u.s . department of energy through grant number de - fg02 - 04er41302 and through grant number de - ac05 - 06or23177 under which jsa operates the thomas jefferson national accelerator facility . zsb also acknowledges support by the jsa jefferson lab graduate fellowship program . wd was supported by the u.s . department of energy early career research award de - sc0010495 and the solomon buchsbaum fund at mit . this work made use of high - performance computing resources provided by xsede ( supported by national science foundation grant number oci-1053575 ) and nersc ( supported by u.s . department of energy grant number de - ac02 - 05ch11231 ) . a. v. manohar and m. b. wise , `` heavy quark physics , '' camb . monogr . part . * 10 * ( 2000 ) 1191 . j. krner , m. kramer , and d. pirjol , `` heavy baryons , '' http://dx.doi.org/10.1016/0146-6410(94)90053-1[prog . * 33 * ( 1994 ) 787868 ] , http://arxiv.org/abs/hep-ph/9406359[arxiv:hep-ph/9406359 ] . m. j. savage and m. b. wise , `` spectrum of baryons with two heavy quarks , '' http://dx.doi.org/10.1016/0370-2693(90)90035-5[phys . b * 248 * ( 1990 ) 177180 ] . n. brambilla , a. vairo , and t. rsch , `` effective field theory lagrangians for baryons with two and three heavy quarks , '' http://dx.doi.org/10.1103/physrevd.72.034021[phys . d * 72 * ( 2005 ) 034021 ] , http://arxiv.org/abs/hep-ph/0506065[arxiv:hep-ph/0506065 ] . n. brambilla , j. ghiglieri , and a. vairo , `` the three - quark static potential in perturbation theory , '' http://dx.doi.org/10.1103/physrevd.81.054031[phys . d * 81 * ( 2010 ) 054031 ] , http://arxiv.org/abs/0911.3541[arxiv:0911.3541 [ hep - ph ] ] . f. j. llanes - estrada , o. i. pavlova , and r. williams , `` a first estimate of triply heavy baryon masses from the pnrqcd perturbative static potential , '' http://dx.doi.org/10.1140/epjc/s10052-012-2019-9[eur . j. c * 72 * ( 2012 ) 2019 ] , http://arxiv.org/abs/1111.7087[arxiv:1111.7087 [ hep - ph ] ] . n. brambilla , f. karbstein , and a. vairo , `` symmetries of the three - heavy - quark system and the color - singlet static energy at next - to - next - to - leading logarithmic order , '' http://dx.doi.org/10.1103/physrevd.87.074014[phys . d * 87 * ( 2013 ) 074014 ] , http://arxiv.org/abs/1301.3013[arxiv:1301.3013 [ hep - ph ] ] . collaboration , j. beringer _ et al . _ , `` review of particle physics ( rpp ) , '' http://dx.doi.org/10.1103/physrevd.86.010001[phys . d * 86 * ( 2012 ) 010001 ] . collaboration , v. abazov _ et al . _ , `` observation of the doubly strange @xmath77 baryon @xmath78 , '' http://dx.doi.org/10.1103/physrevlett.101.232002[phys . * 101 * ( 2008 ) 232002 ] , http://arxiv.org/abs/0808.4142[arxiv:0808.4142 [ hep - ex ] ] . collaboration , t. aaltonen _ et al . _ , `` observation of the @xmath5- baryon and measurement of the properties of the @xmath79- and @xmath5- baryons , '' http://dx.doi.org/10.1103/physrevd.80.072003[phys . rev . d * 80 * ( 2009 ) 072003 ] , http://arxiv.org/abs/0905.3123[arxiv:0905.3123 [ hep - ex ] ] . collaboration , s. chatrchyan _ et al . _ , `` observation of a new @xmath79 baryon , '' http://dx.doi.org/10.1103/physrevlett.108.252002[phys . * 108 * ( 2012 ) 252002 ] , http://arxiv.org/abs/1204.5955[arxiv:1204.5955 [ hep - ex ] ] . collaboration , r. aaij _ et al . _ , `` measurement of the @xmath80 , @xmath81 and @xmath78 baryon masses , '' http://dx.doi.org/10.1103/physrevlett.110.182001[phys . rev . * 110 * ( 2013 ) 182001 ] , http://arxiv.org/abs/1302.1072[arxiv:1302.1072 [ hep - ex ] ] . collaboration , m. mattson _ et al . _ , `` first observation of the doubly charmed baryon @xmath82 , '' http://dx.doi.org/10.1103/physrevlett.89.112001[phys . * 89 * ( 2002 ) 112001 ] , http://arxiv.org/abs/hep-ex/0208014[arxiv:hep-ex/0208014 ] . collaboration , a. ocherashvili _ et al . _ , `` confirmation of the double charm baryon @xmath83 via its decay to @xmath84 , '' http://dx.doi.org/10.1016/j.physletb.2005.09.043[phys . b * 628 * ( 2005 ) 1824 ] , http://arxiv.org/abs/hep-ex/0406033[arxiv:hep-ex/0406033 ] . s. ratti , `` new results on @xmath85-baryons and a search for @xmath86-baryons in focus , '' http://dx.doi.org/10.1016/s0920-5632(02)01948-5[nucl . * 115 * ( 2003 ) 3336 ] . collaboration , b. aubert _ _ , `` search for doubly charmed baryons @xmath10 and @xmath87 in babar , '' http://dx.doi.org/10.1103/physrevd.74.011103[phys . rev . d * 74 * ( 2006 ) 011103 ] , http://arxiv.org/abs/hep-ex/0605075[arxiv:hep-ex/0605075 ] . collaboration , r. chistov _ et al . _ , `` observation of new states decaying into @xmath88 and @xmath89 , '' http://dx.doi.org/10.1103/physrevlett.97.162001[phys . * 97 * ( 2006 ) 162001 ] , http://arxiv.org/abs/hep-ex/0606051[arxiv:hep-ex/0606051 ] . collaboration , r. aaij _ et al . _ , `` search for the doubly charmed baryon @xmath10 , '' http://dx.doi.org/10.1007/jhep12(2013)090[jhep * 1312 * ( 2013 ) 090 ] , http://arxiv.org/abs/1310.2538[arxiv:1310.2538 [ hep - ex ] ] . m. a. doncheski , j. steegborn , and m. stong , `` fragmentation production of doubly heavy baryons , '' http://dx.doi.org/10.1103/physrevd.53.1247[phys . d * 53 * ( 1996 ) 12471252 ] , http://arxiv.org/abs/hep-ph/9507220[arxiv:hep-ph/9507220 ] . chen and s .- z . wu , `` production of triply heavy baryons at lhc , '' http://dx.doi.org/10.1007/jhep08(2011)144 , 10.1007/jhep09(2011)089[jhep * 1108 * ( 2011 ) 144 ] , http://arxiv.org/abs/1106.0193[arxiv:1106.0193 [ hep - ph ] ] . g. chen , x .- wu , j .- w . zhang , h .- y . and h .- b . fu , `` hadronic production of @xmath11 at a fixed - target experiment at the lhc , '' http://dx.doi.org/10.1103/physrevd.89.074020[phys . d * 89 * ( 2014 ) 074020 ] , http://arxiv.org/abs/1401.6269[arxiv:1401.6269 [ hep - ph ] ] . d. asner , t. barnes , j. bian , i. bigi , n. brambilla , _ et al . _ , `` physics at bes - iii , '' int . j. mod . a * 24 * ( 2009 ) s1794 , http://arxiv.org/abs/0809.1869[arxiv:0809.1869 [ hep - ex ] ] . t. aushev , w. bartel , a. bondar , j. brodzicka , t. browder , _ et al . _ , `` physics at super b factory , '' http://arxiv.org/abs/1002.5012[arxiv:1002.5012 [ hep - ex ] ] . collaboration , m. lutz _ et al . _ , `` physics performance report for panda : strong interaction studies with antiprotons , '' http://arxiv.org/abs/0903.3905[arxiv:0903.3905 [ hep - ex ] ] . c. alexandrou , a. borrelli , s. gsken , f. jegerlehner , k. schilling , _ et al . _ , `` beautiful baryons from lattice qcd , '' http://dx.doi.org/10.1016/0370-2693(94)90985-7[phys . b * 337 * ( 1994 ) 340346 ] , http://arxiv.org/abs/hep-lat/9407027[arxiv:hep-lat/9407027 ] . collaboration , k. bowler _ _ , `` heavy baryon spectroscopy from the lattice , '' http://dx.doi.org/10.1103/physrevd.54.3619[phys . d * 54 * ( 1996 ) 36193633 ] , http://arxiv.org/abs/hep-lat/9601022[arxiv:hep-lat/9601022 ] . collaboration , c. michael and j. peisa , `` maximal variance reduction for stochastic propagators with applications to the static quark spectrum , '' http://dx.doi.org/10.1103/physrevd.58.034506[phys . d * 58 * ( 1998 ) 034506 ] , http://arxiv.org/abs/hep-lat/9802015[arxiv:hep-lat/9802015 ] . a. ali khan , t. bhattacharya , s. collins , c. davies , r. gupta , _ et al . _ , `` heavy light mesons and baryons with @xmath77 quarks , '' http://dx.doi.org/10.1103/physrevd.62.054505[phys . rev . d * 62 * ( 2000 ) 054505 ] , http://arxiv.org/abs/hep-lat/9912034[arxiv:hep-lat/9912034 ] . r. woloshyn , `` hyperfine effects in charmed baryons , '' http://dx.doi.org/10.1016/s0370-2693(00)00165-9[phys . b * 476 * ( 2000 ) 309315 ] , http://arxiv.org/abs/hep-ph/0002088[arxiv:hep-ph/0002088 ] . r. lewis , n. mathur , and r. woloshyn , `` charmed baryons in lattice qcd , '' http://dx.doi.org/10.1103/physrevd.64.094509[phys . rev . d * 64 * ( 2001 ) 094509 ] , http://arxiv.org/abs/hep-ph/0107037[arxiv:hep-ph/0107037 ] . n. mathur , r. lewis , and r. woloshyn , `` charmed and bottom baryons from lattice nrqcd , '' http://dx.doi.org/10.1103/physrevd.66.014502[phys . d * 66 * ( 2002 ) 014502 ] , http://arxiv.org/abs/hep-ph/0203253[arxiv:hep-ph/0203253 ] . collaboration , j. flynn , f. mescia , and a. s. b. tariq , `` spectroscopy of doubly charmed baryons in lattice qcd , '' jhep * 0307 * ( 2003 ) 066 , http://arxiv.org/abs/hep-lat/0307025[arxiv:hep-lat/0307025 ] . w . chiu and t .- h . hsieh , `` baryon masses in lattice qcd with exact chiral symmetry , '' http://dx.doi.org/10.1016/j.nuclphysa.2005.03.090[nucl . * a755 * ( 2005 ) 471474 ] , http://arxiv.org/abs/hep-lat/0501021[arxiv:hep-lat/0501021 ] . h. na and s. gottlieb , `` heavy baryon mass spectrum from lattice qcd with 2 + 1 flavors , '' pos * lat2006 * ( 2006 ) 191 , http://arxiv.org/abs/hep-lat/0610009[arxiv:hep-lat/0610009 ] . h. na and s. a. gottlieb , `` charm and bottom heavy baryon mass spectrum from lattice qcd with 2 + 1 flavors , '' pos * lat2007 * ( 2007 ) 124 , http://arxiv.org/abs/0710.1422[arxiv:0710.1422 [ hep - lat ] ] . h. na and s. gottlieb , `` heavy baryon mass spectrum from lattice qcd with 2 + 1 dynamical sea quark flavors , '' pos * lattice2008 * ( 2008 ) 119 , http://arxiv.org/abs/0812.1235[arxiv:0812.1235 [ hep - lat ] ] . t. burch , c. hagen , c. b. lang , m. limmer , and a. schfer , `` excitations of single - beauty hadrons , '' http://dx.doi.org/10.1103/physrevd.79.014504[phys . d * 79 * ( 2009 ) 014504 ] , http://arxiv.org/abs/0809.1103[arxiv:0809.1103 [ hep - lat ] ] . w. detmold , c .- j . d. lin , and m. wingate , `` bottom hadron mass splittings in the static limit from 2 + 1 flavour lattice qcd , '' http://dx.doi.org/10.1016/j.nuclphysb.2009.03.019[nucl . * b818 * ( 2009 ) 1727 ] , http://arxiv.org/abs/0812.2583[arxiv:0812.2583 [ hep - lat ] ] . r. lewis and r. woloshyn , `` bottom baryons from a dynamical lattice qcd simulation , '' http://dx.doi.org/10.1103/physrevd.79.014502[phys . d * 79 * ( 2009 ) 014502 ] , http://arxiv.org/abs/0806.4783[arxiv:0806.4783 [ hep - lat ] ] . lin , s. d. cohen , n. mathur , and k. orginos , `` bottom - hadron mass splittings from static - quark action on 2 + 1-flavor lattices , '' http://dx.doi.org/10.1103/physrevd.80.054027[phys . d * 80 * ( 2009 ) 054027 ] , http://arxiv.org/abs/0905.4120[arxiv:0905.4120 [ hep - lat ] ] . l. liu , h .- w . lin , k. orginos , and a. walker - loud , `` singly and doubly charmed @xmath90 baryon spectrum from lattice qcd , '' http://dx.doi.org/10.1103/physrevd.81.094505[phys . d * 81 * ( 2010 ) 094505 ] , http://arxiv.org/abs/0909.3294[arxiv:0909.3294 [ hep - lat ] ] . s. meinel , `` prediction of the @xmath91 mass from lattice qcd , '' http://dx.doi.org/10.1103/physrevd.82.114514[phys . d * 82 * ( 2010 ) 114514 ] , http://arxiv.org/abs/1008.3154[arxiv:1008.3154 [ hep - lat ] ] . s. meinel , `` excited - state spectroscopy of triply - bottom baryons from lattice qcd , '' http://dx.doi.org/10.1103/physrevd.85.114510[phys . d * 85 * ( 2012 ) 114510 ] , http://arxiv.org/abs/1202.1312[arxiv:1202.1312 [ hep - lat ] ] . c. alexandrou , j. carbonell , d. christaras , v. drach , m. gravina , _ et al . _ , `` strange and charm baryon masses with two flavors of dynamical twisted mass fermions , '' http://dx.doi.org/10.1103/physrevd.86.114501[phys . rev . d * 86 * ( 2012 ) 114501 ] , http://arxiv.org/abs/1205.6856[arxiv:1205.6856 [ hep - lat ] ] . r. a. briceno , h .- w . lin , and d. r. bolton , `` charmed - baryon spectroscopy from lattice qcd with @xmath92=2 + 1 + 1 flavors , '' http://dx.doi.org/10.1103/physrevd.86.094504[phys . d * 86 * ( 2012 ) 094504 ] , http://arxiv.org/abs/1207.3536[arxiv:1207.3536 [ hep - lat ] ] . collaboration , y. namekawa _ et al . _ , `` charmed baryons at the physical point in 2 + 1 flavor lattice qcd , '' http://dx.doi.org/10.1103/physrevd.87.094512[phys . d * 87 * ( 2013 ) 094512 ] , http://arxiv.org/abs/1301.4743[arxiv:1301.4743 [ hep - lat ] ] . p. perez rubio , `` charmed baryon spectroscopy from lattice qcd for @xmath93 flavours , '' pos * confinementx * ( 2012 ) 141 , http://arxiv.org/abs/1302.5774[arxiv:1302.5774 [ hep - lat ] ] . m. padmanath , r. g. edwards , n. mathur , and m. peardon , `` spectroscopy of triply - charmed baryons from lattice qcd , '' http://arxiv.org/abs/1307.7022[arxiv:1307.7022 [ hep - lat ] ] . s. basak , s. datta , a. lytle , m. padmanath , p. majumdar , _ et al . _ , `` hadron spectra from overlap fermions on hisq gauge configurations , '' pos * lattice2013 * ( 2013 ) 243 , http://arxiv.org/abs/1312.3050[arxiv:1312.3050 [ hep - lat ] ] . c. alexandrou , v. drach , k. jansen , c. kallidonis , and g. koutsou , `` baryon spectrum with @xmath94 twisted mass fermions , '' http://arxiv.org/abs/1406.4310[arxiv:1406.4310 [ hep - lat ] ] . k. orginos , `` charmed and bottom baryon spectrum from lattice qcd , '' econf * c0906083 * ( 2009 ) 01 . r. lewis , `` bottom and charmed hadron spectroscopy from lattice qcd , '' http://dx.doi.org/10.1063/1.3647208[aip conf . * 1374 * ( 2011 ) 581584 ] , http://arxiv.org/abs/1010.0889[arxiv:1010.0889 [ hep - lat ] ] . lin , `` review of baryon spectroscopy in lattice qcd , '' chin . j. phys . * 49 * ( 2011 ) 827 , http://arxiv.org/abs/1106.1608[arxiv:1106.1608 [ hep - lat ] ] . d. b. kaplan , `` a method for simulating chiral fermions on the lattice , '' http://dx.doi.org/10.1016/0370-2693(92)91112-m[phys . b * 288 * ( 1992 ) 342347 ] , http://arxiv.org/abs/hep-lat/9206013[arxiv:hep-lat/9206013 ] . v. furman and y. shamir , `` axial symmetries in lattice qcd with kaplan fermions , '' http://dx.doi.org/10.1016/0550-3213(95)00031-m[nucl . phys . * b439 * ( 1995 ) 5478 ] , http://arxiv.org/abs/hep-lat/9405004[arxiv:hep-lat/9405004 ] . y. shamir , `` chiral fermions from lattice boundaries , '' http://dx.doi.org/10.1016/0550-3213(93)90162-i[nucl . phys . * b406 * ( 1993 ) 90106 ] , http://arxiv.org/abs/hep-lat/9303005[arxiv:hep-lat/9303005 ] . a. x. el - khadra , a. s. kronfeld , and p. b. mackenzie , `` massive fermions in lattice gauge theory , '' http://dx.doi.org/10.1103/physrevd.55.3933[phys . d * 55 * ( 1997 ) 39333957 ] , http://arxiv.org/abs/hep-lat/9604004[arxiv:hep-lat/9604004 ] . p. chen , `` heavy quarks on anisotropic lattices : the charmonium spectrum , '' http://dx.doi.org/10.1103/physrevd.64.034509[phys . rev . d * 64 * ( 2001 ) 034509 ] , http://arxiv.org/abs/hep-lat/0006019[arxiv:hep-lat/0006019 ] . s. aoki , y. kuramashi , and s .- tominaga , `` relativistic heavy quarks on the lattice , '' http://dx.doi.org/10.1143/ptp.109.383[prog . theor . * 109 * ( 2003 ) 383413 ] , http://arxiv.org/abs/hep-lat/0107009[arxiv:hep-lat/0107009 ] . s. aoki , y. kayaba , and y. kuramashi , `` a perturbative determination of mass dependent @xmath95 improvement coefficients in a relativistic heavy quark action , '' http://dx.doi.org/10.1016/j.nuclphysb.2004.07.017[nucl . * b697 * ( 2004 ) 271301 ] , http://arxiv.org/abs/hep-lat/0309161[arxiv:hep-lat/0309161 ] . n. h. christ , m. li , and h .- w . lin , `` relativistic heavy quark effective action , '' http://dx.doi.org/10.1103/physrevd.76.074505[phys . d * 76 * ( 2007 ) 074505 ] , http://arxiv.org/abs/hep-lat/0608006[arxiv:hep-lat/0608006 ] . lin and n. christ , `` non - perturbatively determined relativistic heavy quark action , '' http://dx.doi.org/10.1103/physrevd.76.074506[phys . d * 76 * ( 2007 ) 074506 ] , http://arxiv.org/abs/hep-lat/0608005[arxiv:hep-lat/0608005 ] . collaboration , y. aoki _ et al . _ , `` nonperturbative tuning of an improved relativistic heavy - quark action with application to bottom spectroscopy , '' http://dx.doi.org/10.1103/physrevd.86.116003[phys . d * 86 * ( 2012 ) 116003 ] , http://arxiv.org/abs/1206.2554[arxiv:1206.2554 [ hep - lat ] ] . b. thacker and g. p. lepage , `` heavy quark bound states in lattice qcd , '' http://dx.doi.org/10.1103/physrevd.43.196[phys . d * 43 * ( 1991 ) 196208 ] . g. p. lepage , l. magnea , c. nakhleh , u. magnea , and k. hornbostel , `` improved nonrelativistic qcd for heavy quark physics , '' http://dx.doi.org/10.1103/physrevd.46.4052[phys . d * 46 * ( 1992 ) 40524067 ] , http://arxiv.org/abs/hep-lat/9205007[arxiv:hep-lat/9205007 ] . b. c. tiburzi , `` baryon masses in partially quenched heavy hadron chiral perturbation theory , '' http://dx.doi.org/10.1103/physrevd.71.034501[phys . d * 71 * ( 2005 ) 034501 ] , http://arxiv.org/abs/hep-lat/0410033[arxiv:hep-lat/0410033 ] . t. mehen and b. c. tiburzi , `` doubly heavy baryons and quark - diquark symmetry in quenched and partially quenched chiral perturbation theory , '' http://dx.doi.org/10.1103/physrevd.74.054505[phys . d * 74 * ( 2006 ) 054505 ] , http://arxiv.org/abs/hep-lat/0607023[arxiv:hep-lat/0607023 ] . collaboration , y. aoki _ _ , `` continuum limit physics from 2 + 1 flavor domain wall qcd , '' http://dx.doi.org/10.1103/physrevd.83.074508[phys . d * 83 * ( 2011 ) 074508 ] , http://arxiv.org/abs/1011.0892[arxiv:1011.0892 [ hep - lat ] ] . y. iwasaki , `` renormalization group analysis of lattice theories and improved lattice action . 2 . four - dimensional nonabelian @xmath96 gauge model , '' report no . y. iwasaki and t. yoshie , `` renormalization group improved action for @xmath97 lattice gauge theory and the string tension , '' http://dx.doi.org/10.1016/0370-2693(84)91500-4[phys . b * 143 * ( 1984 ) 449 ] . y. aoki , t. blum , n. christ , c. cristian , c. dawson , _ et al . _ , `` domain wall fermions with improved gauge actions , '' http://dx.doi.org/10.1103/physrevd.69.074504[phys . d * 69 * ( 2004 ) 074504 ] , http://arxiv.org/abs/hep-lat/0211023[arxiv:hep-lat/0211023 ] . s. meinel , `` bottomonium spectrum at order @xmath98 from domain - wall lattice qcd : precise results for hyperfine splittings , '' http://dx.doi.org/10.1103/physrevd.82.114502[phys . d * 82 * ( 2010 ) 114502 ] , http://arxiv.org/abs/1007.3966[arxiv:1007.3966 [ hep - lat ] ] . collaboration , c. davies , e. follana , i. kendall , g. p. lepage , and c. mcneile , `` precise determination of the lattice spacing in full lattice qcd , '' http://dx.doi.org/10.1103/physrevd.81.034506[phys . d * 81 * ( 2010 ) 034506 ] , http://arxiv.org/abs/0910.1229[arxiv:0910.1229 [ hep - lat ] ] . collaboration , r. dowdall _ et al . _ , `` the upsilon spectrum and the determination of the lattice spacing from lattice qcd including charm quarks in the sea , '' http://dx.doi.org/10.1103/physrevd.85.054509[phys . rev . d * 85 * ( 2012 ) 054509 ] , http://arxiv.org/abs/1110.6887[arxiv:1110.6887 [ hep - lat ] ] . m. a. shifman and m. voloshin , `` on production of @xmath99 and @xmath100 mesons in @xmath101 meson decays , '' sov . j. nucl . phys . * 47 * ( 1988 ) 511 . g. p. lepage and p. b. mackenzie , `` on the viability of lattice perturbation theory , '' http://dx.doi.org/10.1103/physrevd.48.2250[phys . d * 48 * ( 1993 ) 22502264 ] , http://arxiv.org/abs/hep-lat/9209022[arxiv:hep-lat/9209022 ] . t. hammant , a. hart , g. von hippel , r. horgan , and c. monahan , `` radiative improvement of the lattice nrqcd action using the background field method and application to the hyperfine splitting of quarkonium states , '' http://dx.doi.org/10.1103/physrevlett.107.112002[phys . lett . * 107 * ( 2011 ) 112002 ] , http://arxiv.org/abs/1105.5309[arxiv:1105.5309 [ hep - lat ] ] . g. t. bodwin , e. braaten , and g. p. lepage , `` rigorous qcd analysis of inclusive annihilation and production of heavy quarkonium , '' http://dx.doi.org/10.1103/physrevd.55.5853 , 10.1103/physrevd.51.1125[phys . d * 51 * ( 1995 ) 11251171 ] , http://arxiv.org/abs/hep-ph/9407339[arxiv:hep-ph/9407339 ] . collaboration , e. follana _ et al . _ , `` highly improved staggered quarks on the lattice , with applications to charm physics , '' http://dx.doi.org/10.1103/physrevd.75.054502[phys . d * 75 * ( 2007 ) 054502 ] , http://arxiv.org/abs/hep-lat/0610092[arxiv:hep-lat/0610092 ] . w. detmold , c .- j . d. lin , and s. meinel , `` axial couplings and strong decay widths of heavy hadrons , '' http://dx.doi.org/10.1103/physrevlett.108.172003[phys . * 108 * ( 2012 ) 172003 ] , http://arxiv.org/abs/1109.2480[arxiv:1109.2480 [ hep - lat ] ] . w. detmold , c .- j . d. lin , and s. meinel , `` calculation of the heavy - hadron axial couplings @xmath102 , @xmath103 , and @xmath104 using lattice qcd , '' http://dx.doi.org/10.1103/physrevd.85.114508[phys . d * 85 * ( 2012 ) 114508 ] , http://arxiv.org/abs/1203.3378[arxiv:1203.3378 [ hep - lat ] ] . w. detmold , c .- j . d. lin , s. meinel , and m. wingate , `` @xmath105 form factors and differential branching fraction from lattice qcd , '' http://dx.doi.org/10.1103/physrevd.87.074502[phys . d 87 , * 074502 * ( 2013 ) ] , http://arxiv.org/abs/1212.4827[arxiv:1212.4827 [ hep - lat ] ] . c. morningstar and m. j. peardon , `` analytic smearing of @xmath97 link variables in lattice qcd , '' http://dx.doi.org/10.1103/physrevd.69.054501[phys . d * 69 * ( 2004 ) 054501 ] , http://arxiv.org/abs/hep-lat/0311018[arxiv:hep-lat/0311018 ] . g. p. lepage , `` simulating heavy quarks , '' http://dx.doi.org/10.1016/0920-5632(92)90228-k[nucl . proc . suppl . * 26 * ( 1992 ) 4556 ] . m. b. wise , `` chiral perturbation theory for hadrons containing a heavy quark , '' http://dx.doi.org/10.1103/physrevd.45.r2188[phys . d * 45 * ( 1992 ) 21882191 ] . g. burdman and j. f. donoghue , `` union of chiral and heavy quark symmetries , '' http://dx.doi.org/10.1016/0370-2693(92)90068-f[phys . b * 280 * ( 1992 ) 287291 ] . cheng , c .- y . cheung , g .- lin , y. lin , _ et al . _ , `` heavy quark symmetry and chiral dynamics , '' http://dx.doi.org/10.1103/physrevd.46.1148 , 10.1103/physrevd.55.5851[phys . rev . d * 46 * ( 1992 ) 11481164 ] . l. cho , `` heavy hadron chiral perturbation theory , '' http://dx.doi.org/10.1016/0550-3213(93)90263-o[nucl . phys . * b396 * ( 1993 ) 183204 ] , http://arxiv.org/abs/hep-ph/9208244[arxiv:hep-ph/9208244 ] . j. hu and t. mehen , `` chiral lagrangian with heavy quark - diquark symmetry , '' http://dx.doi.org/10.1103/physrevd.73.054003[phys . d * 73 * ( 2006 ) 054003 ] , http://arxiv.org/abs/hep-ph/0511321[arxiv:hep-ph/0511321 ] . c. w. bernard and m. f. golterman , `` partially quenched gauge theories and an application to staggered fermions , '' http://dx.doi.org/10.1103/physrevd.49.486[phys . d * 49 * ( 1994 ) 486494 ] , http://arxiv.org/abs/hep-lat/9306005[arxiv:hep-lat/9306005 ] . s. r. sharpe and n. shoresh , `` physical results from unphysical simulations , '' http://dx.doi.org/10.1103/physrevd.62.094503[phys . d * 62 * ( 2000 ) 094503 ] , http://arxiv.org/abs/hep-lat/0006017[arxiv:hep-lat/0006017 ] . c. bernard and m. golterman , `` on the foundations of partially quenched chiral perturbation theory , '' http://dx.doi.org/10.1103/physrevd.88.014004[phys . d * 88 * ( 2013 ) 014004 ] , http://arxiv.org/abs/1304.1948[arxiv:1304.1948 [ hep - lat ] ] . m. lscher , `` volume dependence of the energy spectrum in massive quantum field theories . scattering states , '' http://dx.doi.org/10.1007/bf01211097[commun . math . * 105 * ( 1986 ) 153188 ] . m. lscher , `` two - particle states on a torus and their relation to the scattering matrix , '' http://dx.doi.org/10.1016/0550-3213(91)90366-6[nucl . phys . * b354 * ( 1991 ) 531578 ] . g. colangelo , s. drr , a. jttner , l. lellouch , h. leutwyler , _ et al . _ , `` review of lattice results concerning low energy particle physics , '' http://dx.doi.org/10.1140/epjc/s10052-011-1695-1[eur . j. c * 71 * ( 2011 ) 1695 ] , http://arxiv.org/abs/1011.4408[arxiv:1011.4408 [ hep - lat ] ] . b. blok , j. krner , d. pirjol , and j. rojas , `` spectator effects in the heavy quark effective theory , '' http://dx.doi.org/10.1016/s0550-3213(97)00202-2[nucl . phys . * b496 * ( 1997 ) 358374 ] , http://arxiv.org/abs/hep-ph/9607233[arxiv:hep-ph/9607233 ] . e. gregory , c. davies , e. follana , e. gamiz , i. kendall , _ et al . _ , `` a prediction of the @xmath106 mass in full lattice qcd , '' http://dx.doi.org/10.1103/physrevlett.104.022001[phys . rev . lett . * 104 * ( 2010 ) 022001 ] , http://arxiv.org/abs/0909.4462[arxiv:0909.4462 [ hep - lat ] ] . s. borsanyi , s. drr , z. fodor , c. hoelbling , s. katz , _ et al . _ , `` ab initio calculation of the neutron - proton mass difference , '' http://arxiv.org/abs/1406.4088[arxiv:1406.4088 [ hep - lat ] ] . r. roncaglia , d. lichtenberg , and e. predazzi , `` predicting the masses of baryons containing one or two heavy quarks , '' http://dx.doi.org/10.1103/physrevd.52.1722[phys . d * 52 * ( 1995 ) 17221725 ] , http://arxiv.org/abs/hep-ph/9502251[arxiv:hep-ph/9502251 ] . b. silvestre - brac , `` spectrum and static properties of heavy baryons , '' http://dx.doi.org/10.1007/s006010050028[few body syst . * 20 * ( 1996 ) 125 ] . d. ebert , r. faustov , v. galkin , and a. martynenko , `` mass spectra of doubly heavy baryons in the relativistic quark model , '' http://dx.doi.org/10.1103/physrevd.66.014008[phys . d * 66 * ( 2002 ) 014008 ] , http://arxiv.org/abs/hep-ph/0201217[arxiv:hep-ph/0201217 ] . v. kiselev , a. likhoded , o. pakhomova , and v. saleev , `` mass spectra of doubly heavy omega @xmath107 baryons , '' http://dx.doi.org/10.1103/physrevd.66.034030[phys . rev . d * 66 * ( 2002 ) 034030 ] , http://arxiv.org/abs/hep-ph/0206140[arxiv:hep-ph/0206140 ] . he , k. qian , y .- b . ding , x .- q . shen , `` evaluation of spectra of baryons containing two heavy quarks in bag model , '' http://dx.doi.org/10.1103/physrevd.70.094004[phys . d * 70 * ( 2004 ) 094004 ] , http://arxiv.org/abs/hep-ph/0403301[arxiv:hep-ph/0403301 ] . c. albertus , e. hernandez , j. nieves , and j. verde - velasco , `` static properties and semileptonic decays of doubly heavy baryons in a nonrelativistic quark model , '' http://dx.doi.org/10.1140/epja/i2007-10364-y , 10.1140/epja / i2008 - 10547 - 0[eur . j. a * 32 * ( 2007 ) 183199 ] , http://arxiv.org/abs/hep-ph/0610030[arxiv:hep-ph/0610030 ] . a. martynenko , `` ground - state triply and doubly heavy baryons in a relativistic three - quark model , '' http://dx.doi.org/10.1016/j.physletb.2008.04.030[phys . b * 663 * ( 2008 ) 317321 ] , http://arxiv.org/abs/0708.2033[arxiv:0708.2033 [ hep - ph ] ] . w. roberts and m. pervin , `` heavy baryons in a quark model , '' http://dx.doi.org/10.1142/s0217751x08041219[int . j. mod . a * 23 * ( 2008 ) 28172860 ] , http://arxiv.org/abs/0711.2492[arxiv:0711.2492 [ nucl - th ] ] . a. bernotas and v. simonis , `` mixing of heavy baryons in the bag model calculations , '' http://arxiv.org/abs/0801.3570[arxiv:0801.3570 [ hep - ph ] ] . zhang and m .- q . huang , `` doubly heavy baryons in qcd sum rules , '' http://dx.doi.org/10.1103/physrevd.78.094007[phys . d * 78 * ( 2008 ) 094007 ] , http://arxiv.org/abs/0810.5396[arxiv:0810.5396 [ hep - ph ] ] . f. giannuzzi , `` doubly heavy baryons in a salpeter model with ads / qcd inspired potential , '' http://dx.doi.org/10.1103/physrevd.79.094002[phys . d * 79 * ( 2009 ) 094002 ] , http://arxiv.org/abs/0902.4624[arxiv:0902.4624 [ hep - ph ] ] . l. tang , x .- h . yuan , c .- f . qiao , and x .- q . li , `` study of doubly heavy baryon spectrum via qcd sum rules , '' http://dx.doi.org/10.1088/0253-6102/57/3/15[commun . * 57 * ( 2012 ) 435444 ] , http://arxiv.org/abs/1104.4934[arxiv:1104.4934 [ hep - ph ] ] . z. ghalenovi , a. a. rajabi , s .- x . qin , and d. h. rischke , `` ground - state masses and magnetic moments of heavy baryons , '' http://dx.doi.org/10.1142/s0217732314501065[mod . a * 29 * ( 2014 ) 1450106 ] , http://arxiv.org/abs/1403.4582[arxiv:1403.4582 [ hep - ph ] ] . m. karliner and j. l. rosner , `` baryons with two heavy quarks : masses , production , decays , and detection , '' http://arxiv.org/abs/1408.5877[arxiv:1408.5877 [ hep - ph ] ] . w. ponce , `` heavy quarks in a spherical bag , '' http://dx.doi.org/10.1103/physrevd.19.2197[phys . d * 19 * ( 1979 ) 2197 ] . p. hasenfratz , r. horgan , j. kuti , and j. richard , `` heavy baryon spectroscopy in the qcd bag model , '' http://dx.doi.org/10.1016/0370-2693(80)90906-5[phys . b * 94 * ( 1980 ) 401 ] . j. d. bjorken , `` is the @xmath108 a new deal for baryon spectroscopy ? , '' http://dx.doi.org/10.1063/1.35379[aip conf . * 132 * ( 1985 ) 390403 ] . m. tsuge , t. morii , and j. morishita , `` baryon spectrum in the potential model with relativistic corrections , '' http://dx.doi.org/10.1142/s0217732386000191[mod . lett . a * 1 * ( 1986 ) 131 ] . y. jia , `` variational study of weakly coupled triply heavy baryons , '' http://dx.doi.org/10.1088/1126-6708/2006/10/073[jhep * 0610 * ( 2006 ) 073 ] , http://arxiv.org/abs/hep-ph/0607290[arxiv:hep-ph/0607290 ] . a. bernotas and v. simonis , `` heavy hadron spectroscopy and the bag model , '' http://dx.doi.org/10.3952/lithjphys.49110[lith . j. phys . * 49 * ( 2009 ) 1928 ] , http://arxiv.org/abs/0808.1220[arxiv:0808.1220 [ hep - ph ] ] . g . wang , `` analysis of the triply heavy baryon states with qcd sum rules , '' http://dx.doi.org/10.1088/0253-6102/58/5/17[commun . * 58 * ( 2012 ) 723731 ] , http://arxiv.org/abs/1112.2274[arxiv:1112.2274 [ hep - ph ] ] . zhang and m .- q . huang , `` deciphering triply heavy baryons in terms of qcd sum rules , '' http://dx.doi.org/10.1016/j.physletb.2009.02.056[phys . b * 674 * ( 2009 ) 2835 ] , http://arxiv.org/abs/0902.3297[arxiv:0902.3297 [ hep - ph ] ] . a. walker - loud , `` topics in effective field theory for lattice qcd , '' http://arxiv.org/abs/hep-lat/0608010[arxiv:hep-lat/0608010 ] . d. arndt and c. d. lin , `` heavy meson chiral perturbation theory in finite volume , '' http://dx.doi.org/10.1103/physrevd.70.014503[phys . d * 70 * ( 2004 ) 014503 ] , http://arxiv.org/abs/hep-lat/0403012[arxiv:hep-lat/0403012 ] . w. detmold and c. d. lin , `` twist - two matrix elements at finite and infinite volume , '' http://dx.doi.org/10.1103/physrevd.71.054510[phys . d * 71 * ( 2005 ) 054510 ] , http://arxiv.org/abs/hep-lat/0501007[arxiv:hep-lat/0501007 ] .

we derive an effectively three - dimensional relativistic spin precession formalism . the formalism is applicable to any spacetime where an arbitrary timelike reference congruence of worldlines is specified . we employ what we call a _ stopped _ spin vector which is the spin vector that we would get if we momentarily make a pure boost of the spin vector to stop it relative to the congruence . starting from the fermi transport equation for the standard spin vector we derive a corresponding transport equation for the stopped spin vector . employing a spacetime transport equation for a vector along a worldline , corresponding to spatial parallel transport with respect to the congruence , we can write down a precession formula for a gyroscope relative to the local spatial geometry defined by the congruence . this general approach has already been pursued by jantzen et . al . ( see e.g. jantzen , carini and bini 1997 _ ann . phys . _ * 215 * 1 ) , but the algebraic form of our respective expressions differ . we are also applying the formalism to a novel type of spatial parallel transport introduced in jonsson ( 2006 _ class . quantum grav . _ * 23 * 1 ) , as well as verifying the validity of the intuitive approach of a forthcoming paper ( jonsson 2007 _ am . journ . phys . _ * 75 * 463 ) where gyroscope precession is explained entirely as a double thomas type of effect . we also present the resulting formalism in explicit three - dimensional form ( using the boldface vector notation ) , and give examples of applications . + + pacs numbers : 04.20.-q , 95.30.sf

0708.2533

in special and general relativity the spin of a gyroscope is represented by a four - vector @xmath0 . assuming that we move the gyroscope without applying any torque to it ( in a system comoving with the gyroscope ) , the spin vector will obey the fermi transport equation @xmath1 here @xmath2 is the four - velocity of the gyroscope . for a trajectory in a given spacetime , and a spin vector specified at some point along this trajectory , we can integrate ( [ rall ] ) to find the spin at any point along the trajectory . the fermi transport equation is however deceivingly simple since we have not inserted explicitly the affine connection coming from the covariant differentiation . also , even when we have a flat spacetime and inertial coordinates ( so that the affine connection vanishes ) the equation is more complex than you might think . as an example we consider motion with fixed speed @xmath3 along a circle in the @xmath4-plane , with an angular frequency @xmath5 . letting the groscope start at @xmath6 at the positive @xmath7-axis , we get a set of coupled differential equations @xmath8 here @xmath9 and @xmath10 is the spatial part of @xmath0 . for initial conditions @xmath11 the solutions ( see @xcite p. 175 - 176 ) can be written as @xmath12 + ( \gamma-1)\sin[\omega \gamma t ] \sin[\omega t]\right ) \quad \quad \label{hm1}\\ s^y&=&s \left(\sin[(1-\gamma)\omega t ] -(\gamma-1)\sin[\omega \gamma t]\cos[\omega t]\right ) \label{hm2}\\ s^z&=&0 , \quad \quad s^0=-s r \omega \gamma \sin[\omega \gamma t ] \label{hm3}\end{aligned}\ ] ] looking at @xmath13 and @xmath14 , we note that ( written in the particular form above ) the first terms in respective expression corresponds to a rotation around the @xmath15-axis , but then there is also another superimposed rotation with time dependent amplitude . to find this solution directly from the coupled differential equations that are the fermi equations , seems at least at first sight quite difficult , even for this very symmetric and simple scenario . to get a simpler formalism we may consider , not the spin vector @xmath0 itself , but the spin vector we _ would _ get if we momentarily would stop the gyroscope ( relative to a certain inertial frame ) by a pure boost ( i.e. a non - rotating boost ) this object we will call the _ stopped _ spin vector . while being a four - vector it is effectively a three - dimensional object ( having zero time component in the inertial frame in question ) and we will show that the spatial part of this object undergoes pure rotation with a constant rate for the example of motion along a circle in special relativity . knowing that there is a simple algebraic relation between the stopped and the standard spin vector , the stopped spin vector can be used as an intermediate step to easily find the standard spin vector . there is however also a direct physical meaning to the stopped spin vector , apart from being the spin vector we would get if we stopped the gyroscope . the stopped spin vector _ directly _ gives the spin as perceived in a comoving system , see section [ tttt ] for further discussion on this . in this article we will also consider more general reference frames than inertial ones . for instance we will consider a rotating and accelerating reference frame . this allows us to apply the formalism , via the equivalence principle , to describe in a simple three - dimensional manner how a gyroscope orbiting for instance a rotating black hole will precess relative to the stationary observers . in figure [ fig1 ] we illustrate how a gyroscope spin vector precesses relative to a vector parallel transported with respect to the spatial geometry . given a reference congruence of timelike worldlines , we first derive a general spacetime transport equation for the stopped spin vector ( stopped relative to the congruence in question ) . we then consider a spacetime equation corresponding to _ spatial _ parallel transport with respect to the spatial geometry defined by the congruence . for the case of a rigid congruence , we easily derive such a transport law . considering a shearing congruence we use the formalism derived in @xcite . having both the transport equation for the stopped spin vector and the equation for parallel transport , we can put them together and thus get an equation for how fast the stopped spin vector precesses relative to the local spatial geometry connected to the reference congruence . as is the case for the inertial congruence , we will see that the precession corresponds to a simple law of three - rotation . the general scheme as outlined here has already been pursued by jantzen et . ( see @xcite ) , although the angle of approach and the algebraic formalisms are different . the explicit use of the three - dimensional formalism of this paper also appears novel . this article is complementary to a companion paper @xcite , where the formalism of relativistic spin precession in three - dimensional language is derived in a very intuitive manner . this paper verifies , through a more formal derivation , the result of @xcite for the particular case of a rigid congruence as assumed in @xcite . let us denote the local four - velocity of our reference congruence by @xmath16 . we introduce a _ stopped _ spin vector @xmath17 as the spin vector that we get if we make a pure boost of the spin vector such that it is at rest with respect to the local congruence line . in figure [ fig2 ] we illustrate in 2 + 1 dimensions how the two spin vectors are related to each other . it follows readily from the lorentz transformation that we get the stopped vector by removing the @xmath16-part of @xmath0 , and shortening the part parallel to the spatial direction of motion by a @xmath18-factor . note that the resulting stopped vector is not in general parallel to the spatial part of @xmath0 . letting @xmath19 be a normalized vector orthogonal to @xmath16 in the direction of motion , we can express the stopped spin vector as @xmath20 s^\alpha .\end{aligned}\ ] ] here we have adopted the spatial sign convention @xmath21 as we will throughout the article . knowing a little about thomas precession we may guess that for the simple case of motion along a circle in an inertial frame as discussed earlier , there is a simple law of three - dimensional rotation for this object . indeed in the following discussion we will show this , and at the same time consider the effects of rotation coming from having non - inertial reference frames ( connected to @xmath16 ) . we also need an explicit expression for the standard spin vector in terms of the stopped spin - vector @xmath17 . the relationship between the two vectors follows readily from the lorentz - transformation : @xmath22 \label{s1b } .\end{aligned}\ ] ] this we may now insert into the fermi transport equation to derive an expression for the _ stopped _ spin vector . in this section we consider gyroscope transport relative to an arbitrary reference congruence @xmath16 . for a spin vector @xmath0 transported along a worldline of four - velocity @xmath2 , we have the fermi transport law @xmath23 using ( [ s1a ] ) in ( [ fermi ] ) readily yields @xmath24 .\end{aligned}\ ] ] we need now the inverse of @xmath25 to get an explicit transport equation for the stopped spin vector . through a general ansatz . the ansatz is of the form @xmath26 . ] , we find @xmath27 that this is indeed the inverse of @xmath25 is easy to verify we are free to add terms containing @xmath28 , since these anyway die when multiplied by @xmath29 . if we instead would have defined @xmath30 we would get the inverse @xmath31 . here the perfect symmetry in @xmath32 and @xmath33 is transparent . there however does not appear to be any particular advantages of this gauge . ] . so we have @xmath34{{k^{-1}}^\nu}_\mu .\end{aligned}\ ] ] here we have the desired expression . in [ simp ] we expand and simplify this to find @xmath35_\perp -t_\alpha \left [ \frac{d}{d\tau}\left(u^\mu + \eta^\mu \right)\right]_\perp \right ) \nonumber\\ & & + \eta^\mu \bar{s}^\alpha \frac{d\eta_\alpha}{d\tau } .\end{aligned}\ ] ] by the perpendicular sign @xmath36 we here mean that we should select only the part orthogonal to both @xmath19 and @xmath16 . note that @xmath37 means covariant differentiation along the gyroscope worldline . equation ( [ kott1 ] ) then tells us how the stopped spin vector deviates from a parallel transported vector relative to a freely falling system . in fact we notice from the antisymmetric form of ( [ kott1 ] ) that ( excepting the @xmath16 term ) it corresponds to a spatial rotation ( see section [ gotland ] for a more detailed argument ) . that seems very reasonable since it insures that the norm of the stopped spin vector will be constant ( consider the rotation with respect to a freely falling system locally comoving with the congruence ) . we also see that only if @xmath38 changes along the gyroscope worldline , with respect to a freely falling system , do we get a net rotation relative to this freely falling system . introducing the wedge product defined by @xmath39 and the projection operator @xmath40 , we can put ( [ kott1 ] ) in a more compact form @xmath41_\perp \right ) .\end{aligned}\ ] ] incidentally we may note that , as regards @xmath19-components within the bracketed expression , we do not need the @xmath36 sign . any @xmath19 components within the bracketed expressions will cancel due to the anti - symmetrization as is easy to see . we however keep the @xmath36 sign to indicate orthogonality to @xmath16 . the simple form of ( [ kottfinal ] ) appears to be novel . while we have yet to put the formalism in its final form , some applications and discussion may be useful already at this point for the simple case of an inertial reference congruence in special relativity . as a particular example , consider a flat spacetime with an inertial congruence . for this case it is not hard to show , see e.g @xcite , that the spatial curvature of a trajectory depends on the four - acceleration as @xmath42_\perp=\gamma^2 v^2\frac{n_\alpha}{r } .\end{aligned}\ ] ] here @xmath43 is the spatial curvature and @xmath44 is a normalized four - vector , orthogonal to the inertial congruence @xmath16 , pointing in the direction of spatial curvature . using this in ( [ kottfinal ] ) we get @xmath45 as we will see in the following section , this differential equation corresponds to a three - dimensional rotation . choosing inertial coordinates adapted to the inertial congruence in question so that @xmath46 , @xmath47 and @xmath48 we get from ( [ kottfinal2 ] ) @xmath49 .\end{aligned}\ ] ] the expression within the brackets is a vector triple product and we may write it as a double cross product . letting @xmath50 we get @xmath51 rather than using @xmath52 we could use local time @xmath53 ( the time as experienced by observers at rest relative to the inertial congruence in question ) in which case we get a gamma factor less on the right hand side . @xmath54 this is the famous thomas precession , in stopped spin vector three - formalism . introducing @xmath55 as the precession vector , around which the stopped spin vector rotates , we can alternatively write ( [ aye ] ) as @xmath56 looking at ( [ ayea ] ) component - wise , it is a set of coupled differential equations , just like the standard fermi equations . unlike the fermi - equations however , the new equations correspond to a simple law of rotation ( precession ) . as a specific example we may consider , as in the introduction , the precession of a gyroscope transported at constant speed @xmath3 around a circle of radius @xmath43 in the @xmath57 plane . assuming a motion with a clockwise angular velocity @xmath58 , the counterclockwise angular velocity @xmath59 for the precession of the stopped spin vector is then according to ( [ ayeb ] ) given by @xmath60 consider then for instance the net precession after one lap . the local time per lap is simply @xmath61 and hence the net precession angle ( in radians ) around the plane normal is given by @xmath62 . if the circular motion is counter - clockwise , the precession is clockwise and vise versa . we can also trivially find the solution for the standard ( projected ) spin vector for the case of circular motion with constant speed with initial conditions as listed in the example in the introduction . we know that the standard ( projected ) spin vector is related to the stopped spin vector through a lengthening of the stopped spin vector in the forward direction of motion @xmath63 by a @xmath18-factor . we have then @xmath64 using the notation of the previous subsection we have then trivially for the case at hand @xmath65 using these expressions in ( [ hpp ] ) we immediately get the desired solution . using elementary rules for manipulating the trigonometric functions we can write it in the form of ( [ hm1])-([hm3 ] ) . if we are interested also in @xmath66 , it is given by the orthogonality of the standard spin vector and the four - velocity as @xmath67 . note that by use of the stopped spin vector formalism there is effectively no differential equation solving involved for this simple case . for motion in a circle with constant velocity , the fermi equation can be solved without use of the stopped spin vector formalism , although the solution is a bit complicated . what about if we consider motion with constant velocity along some other curve , say a part of a parabola or some more irregular curve ? then the fermi equation would likely appear to be very complicated to solve analytically in the general case . using the method with the stopped spin vector the solution can however trivially be found for arbitrary curves . first let us state a small theorem that we will then easily prove . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the stopped spin vector of a gyroscope transported with constant speed @xmath3 along a smooth curve in a spatial plane in a flat spacetime will rotate a net clockwise angle around the normal of the plane given by @xmath68 where @xmath69 is the net counterclockwise turning angle of the tangent direction of the curve . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ note that the parameter @xmath69 may be larger than @xmath70 . for a simple closed curve ( one that is not crossing itself ) , assuming the gyroscope to be transported once around the curve , we have @xmath71 . this theorem is easily proven by dividing an arbitrary smooth curve into infinitesimal segments within which we can consider the local curvature radius to be constant . letting @xmath5 denote the counter - clockwise angular velocity of the forward direction of motion @xmath63 ( so @xmath72 ) , we have according to ( [ how ] ) the clockwise angular velocity as @xmath73 . thus the net angles of the gyroscope precession and the turning of the forward direction of the curve , along the segment in question , are related through @xmath74 . adding up the precession contributions from all the segments of the curve we get @xmath75 thus the theorem is proven . note that while the motion is assumed to be in a plane , the spin vector may point off the plane . we can draw a conclusion from the above proven theorem ( also knowing that there is a simple algebraic relation between the stopped and the standard spin - vector ) that can be expressed in terms of the standard spin vector , without reference to the stopped spin vector . consider then a smooth simple closed curve and let a certain point along this curve be the initial position for the gyroscope . for given initial spin vector , initial direction of motion and constant speed @xmath3 , the final spin vector ( after one lap around the loop ) is independent of the shape of the loop as illustrated in figure [ xf1 ] . but of course , the theorem is stronger than this . given an arbitrary , not necessarily closed but smooth curve along which we transport the gyroscope with constant velocity , we can trivially find the standard spin vector at any point along the curve . we take the spatial part of the initial spin vector and shorten the part parallel to the direction of motion by a @xmath18-factor to form the initial stopped spin vector . for any given curve @xmath76 we then calculate the initial direction of the curve together with the direction of the curve at the point in question . then , modulo a winding number times @xmath77 we only get the turning angle @xmath69 up to a term @xmath78 , where @xmath79 is an integer , from the local quantity @xmath80 , we can trivially find the corresponding @xmath69 and thus through ( [ hh ] ) the corresponding stopped spin vector at the point in question . lengthening the parallel part of the stopped spin vector by a factor @xmath18 , we get the spatial part of the standard spin vector at the point in question . if we are interested in the zeroth component of the standard spin vector it is given by @xmath81 . thus solving a possibly very complicated differential equation is reduced to performing a few algebraic steps . for motion in a plane where the velocity is not constant , the procedure is analogous to that described in section [ concon ] except that we need to integrate ( a single integral which may or may not be complicated to solve analytically ) to find @xmath82 . for the most general motion , not necessarily confined to a plane and with a speed that may vary , it is however not just a matter of ordinary integration . given an arbitrary motion @xmath83 along a smooth curve we can however solve a differential equation , given by ( [ ayea ] ) and ( [ ayeb ] ) , for @xmath84 . likely this differential equation will be simpler to solve than the fermi equation . to gain further intuition on the meaning of the stopped spin vector it may be useful to explore how it is related to other vectors of physical interest connected to the gyroscope spin . in particular we may consider the gyroscope intrinsic angular momentum , and the momentary direction of the gyroscope axis as perceived in the reference system in question ( where the observers are integral curves of @xmath16 ) . consider then a gyroscope moving along a straight line in the @xmath4-plane in special relativity ( using inertial coordinates ) with constant speed . the gyroscope axis is assumed to lie in the plane of motion and to be tilted somewhere between the forward and the sideways direction . in 2 + 1 dimensions we can easily visualize the worldsheet of the gyroscope central axis as well as various vectors of interest , see figure [ xf2 ] . we note that there are ( at least ) three different spatial directions of relevance for the gyroscope . it is easy to realize ( length contraction ) that the direction of the gyroscope axis is simply related to the direction of the stopped spin vector through a gamma factor . given any of these directions the other two can thus easily be found . furthermore one can show , at least for an idealized scenario as considered in [ spinapp ] , that the the intrinsic angular momentum , that we will denote @xmath85 , is in fact given by @xmath86 . the various vectors involved are illustrated in figure [ xf3 ] . consider a spatial vector @xmath87 that connects the base of the gyroscope to the tip of the gyroscope , as perceived in the reference system connected to @xmath16 . we understand that this vector evolves according to a simple rule of rotation given by ( [ aye ] ) modulated by a contraction by a factor of @xmath18 in the direction of motion . it is a short exercise to show that this means that @xmath87 in fact obeys a rather compact differential equation @xmath88.\end{aligned}\ ] ] we can perform a corresponding analysis for the projected spin vector to find @xmath89.\end{aligned}\ ] ] the equations for the stopped spin vector can be written in the form @xmath90 from ( [ rall5 ] ) , letting @xmath91 , we readily find @xmath92 - \gamma^2 v \frac{dv}{dt } { \bf s}{_{{\textrm{\scriptsize l}}}}.\end{aligned}\ ] ] comparing the four differential equations we see that they are all quite compact , although the equation for the stopped spin vector , corresponding to a pure rotation , is more likely to be simple to solve ( as we have seen for the example of motion on a circle ) . one might argue that the object of physical interest is the intrinsic ( spin ) angular momentum of the gyroscope which is given by @xmath93 , or perhaps the observed direction of the gyroscope central axis . from this point of view the stopped spin vector is in a sense a means to an end . by using the stopped spin vector as an intermediate step we can find the solutions to otherwise quite complicated differential equations for the objects of physical interest . from a mathematical point of view this is certainly sufficient to motivate the use of the stopped spin vector . there is however more to the stopped spin vector than this . in particular we note that the stopped spin vector _ directly _ gives us the spin as perceived in a comoving system . for instance , if the stopped spin vector is at a @xmath94 angle with respect to the forward direction so it will be with respect to a system comoving with the gyroscope . this is contrary to the standard spin vector which only gives the spin direction with respect to the comoving system after a lorentz transformation . consider the following example . a gyroscope is suspended inside a satellite such that no torque is exerted on the gyroscope as seen from the satellite . the satellite is assumed to be orbiting along some predetermined smooth simple closed curve , on a plane in special relativity . ] , using it s jet engines to stay on the path . suppose then that we wish to measure , from the satellite , the precession angle of the gyroscope ( as predicted by relativity ) after a full orbit ( or maybe several full orbits ) . we note that the direction of the gyroscope relative to the satellite itself is not a good measure . assuming that we have a couple of fixed stars , we can however use the direction of these stars ( as perceived from the satellite ) as guidelines to set up a reference system within the satellite for this scenario the stopped spin vector is exactly the physical object that we are interested in . it exactly represents the gyroscope direction relative to the star - calibrated reference system of the satellite . thus if the stopped spin vector turns a certain angle , that is precisely the turning angle of the gyroscope relative to the star - calibrated reference system of the satellite . while we are here focusing on spinning gyroscopes , it should also be noted that the formalism of the stopped spin vector is immediately applicable to describe the resulting rotation of _ any _ object which has zero _ proper _ ( comoving ) rotation . in conclusion , the stopped spin vector may be used as an intermediate step to simplify the calculation of the evolution of the intrinsic angular momentum ( spin ) of a gyroscope , or the perceived direction of the gyroscope axis . the stopped spin vector is however also of direct physical importance since it gives us the spin as perceived in a comoving system . so far we have only given examples that apply to flat spacetime , and inertial reference frames . as we will see in the following sections the stopped spin vector can be just as useful also for curved spacetimes and non - inertial reference frames . the transport equation ( [ kottfinal ] ) tells us how the stopped spin vector deviates from a vector that is parallel transported with respect to the spacetime geometry . this by itself is however not really what we are after if the reference congruence is non - inertial . to get a truly three - dimensional formalism , we in stead want an expression telling us how fast the stopped spin vector deviates ( rotates ) from a vector that is parallel transported with respect to the _ spatial _ geometry determined by the congruence . as is demonstrated in @xcite and in @xcite , it is possible derive a _ spacetime _ transport law corresponding to a _ spatial _ parallel transport . for the simple , and perhaps most useful , case of a rigid congruence the issue is sufficiently simple that we will briefly review it in the coming subsection . suppose then that we have a rigid congruence with nonzero acceleration @xmath95 , nonzero rotation tensor @xmath96 but with vanishing expansion - shear tensor @xmath97 , the acceleration vector @xmath98 , the shear tensor @xmath99 and the rotation tensor @xmath100 . furthermore we employ what we denote the expansion - shear tensor @xmath101 . ] . in figure [ fig3 ] we show an illustration of the spacetime transport of a vector orthogonal to the congruence . it is easy to show that in the coordinates of a freely falling system @xmath102 , locally comoving with the congruence , the velocity of the congruence points ( assuming vanishing @xmath97 ) is to first order given by @xmath103 knowing that the velocity of the congruence is zero to lowest order , relative to the inertial system in question , we need not worry about length contraction and such . it is then easy to realize that the proper spacetime transport law of a spatial vector @xmath104 corresponding to standard spatial parallel transport is @xmath105 here @xmath106 can easily be determined from the orthogonality of @xmath104 and @xmath16 follows ( differentiate @xmath107 along the gyroscope worldline ) that @xmath108 . contracting both sides of ( [ init ] ) by @xmath109 gives @xmath110 . ] . here we have then a spacetime transport equation corresponding to spatial parallel transport , for the case of a non - shearing ( non - expanding ) congruence . for a more complicated congruence that is shearing and expanding , it is not quite so obvious how to define the spatial parallel transport . indeed as discussed in e.g @xcite and @xcite , there are several ways of doing this . we will here follow the approach of @xcite , and consider two different such parallel transports . these transports are connected to two different ways of defining a spatial curvature for a test particle worldline , with respect to the congruence @xmath111_\perp = [ a^\mu]_\perp+2v ( { \omega^\mu}_\alpha t^\alpha + [ { \theta^\mu}_\alpha t^\alpha]_\perp ) + v^2 \frac{n{_{{\textrm{\scriptsize ps}}}}^\mu}{r{_{{\textrm{\scriptsize ps } } } } } \\[3 mm ] \textrm{new : } & & \frac{1}{\gamma^2 } \left[\frac{du^\mu}{d\tau}\right]_\perp = [ a^\mu]_\perp+2v { \omega^\mu}_\alpha t^\alpha + v^2 \frac{n{_{{\textrm{\scriptsize ns}}}}^\mu}{r{_{{\textrm{\scriptsize ns } } } } } .\end{aligned}\ ] ] here @xmath112 and @xmath113 are the curvature and the curvature direction that we get if we project the the spacetime trajectory down along the congruence onto a local timeslice ( orthogonal to the congruence at the point in question ) . the suffix ps stands for projected straight. the curvature @xmath114 and the curvature direction @xmath115 are defined with respect to deviations from a certain ( new ) notion of a spatially straight line . the latter is defined as a line that with respect to variations in the projected curvature , leaves the integrated spatial distance ( as defined by the congruence ) unaltered ( to first order in the variation ) . as it turns out , a straight line with respect to this definition , has in general a non - zero projected curvature when the congruence is shearing . the suffix ns stands for new - straight. this particular curvature is connected to fermat s principle , and optical geometry @xcite . for brevity we let the suffix s denote either ps , or ns. introducing @xmath116 , @xmath117 we can then express both curvatures jointly as @xmath118_\perp & = & [ a^\mu]_\perp+2v ( { \omega^\mu}_\alpha t^\alpha + c{_{{\textrm{\scriptsize s}}}}[{\theta^\mu}_\alpha t^\alpha]_\perp ) + v^2 \frac{n{_{{\textrm{\scriptsize s}}}}^\mu}{r{_{{\textrm{\scriptsize s}}}}}.\end{aligned}\ ] ] from these two curvature measures one can introduce corresponding equations for spatial parallel transports @xcite . a joint expression for the parallel transport of a vector @xmath104 is given by @xmath119 here @xmath120 is the covariant derivative along the ( gyroscope ) worldline in question . notice that for vanishing shear expansion tensor , the two transports both correspond to ( [ init ] ) . having defined two types of parallel transport according to ( [ ps ] ) , we can define corresponding covariant differentiations along a curve as @xmath121 these derivatives then tells us how fast a vector deviates from a corresponding parallel transported vector ( momentarily parallel to the vector in question ) . substituting @xmath122 and using ( [ kottfinal ] ) we get the equations for how fast the stopped spin vector precesses relative to a spatially parallel transported vector ( of the two types ) . first we however rewrite ( [ kottfinal ] ) . we saw in the preceding section how the kinematical invariants of the congruence entered naturally in the definition of spatial parallel transport . we can also expand @xmath123 , in the transport equation ( [ kottfinal ] ) for the stopped spin vector , in terms of the kinematical invariants of the congruence . first of all we have @xmath124 also we know that ( see e.g @xcite p. 566 ) @xmath125 using ( [ katta ] ) , we have then @xmath126 using this together with ( [ jcurve ] ) in ( [ kottfinal ] ) , also adding the proper @xmath16-term enabling the removal of the projection operator in ( [ kottfinal ] ) , we readily find @xmath127 + \eta^\mu \bar{s}^\alpha \frac{d\eta_\alpha}{d\tau } .\end{aligned}\ ] ] notice that we have omitted the perpendicular signs ( @xmath36 ) on @xmath128 and @xmath129 since these objects are already orthogonal to @xmath16 and any @xmath19 components die due to the anti - symmetrization . now it is time to put together the results of the preceding two sections . what we want is the net rotation of the stopped spin vector relative to a spatially parallel transported vector . using ( [ delettps ] ) and ( [ transproj ] ) ( setting @xmath130 ) , we then readily find @xmath131 .\end{aligned}\ ] ] here @xmath116 and @xmath117 . so this gives us how fast a gyroscope stopped spin vector deviates from a corresponding ( spatially ) parallel transported vector . in particular considering the expression in a freely falling system locally comoving with the congruence , we understand that the expression within the brackets on the right hand side is simply the effective rotation tensor relative to the spatial geometry . it could be practical with an expression corresponding to ( [ finp ] ) but where the proper four - acceleration is explicit . using ( [ kottfinal ] ) , ( [ transproj ] ) and ( [ deta ] ) we readily find @xmath132_\perp { \hspace{-0.5mm}}{\hspace{-0.5mm}}{\hspace{-0.5mm}}{\hspace{-0.5mm}}+ { \hspace{-0.5mm}}\gamma a_\alpha { \hspace{-0.5mm}}+ { \hspace{-0.5mm}}\gamma v \omega_{\alpha \rho}t^\rho { \hspace{-0.5mm}}+ { \hspace{-0.5mm}}(2\gamma c{_{{\textrm{\scriptsize s } } } } { \hspace{-0.5mm}}-{\hspace{-0.5mm}}1)\frac{\gamma { \hspace{-0.5mm}}+ { \hspace{-0.5mm}}1}{\gamma v}\theta_{\alpha \rho}t^\rho\right ] \nonumber\\ & & \hspace{-0 mm } -\gamma { \omega^\mu}_\alpha \bar{s}^\alpha .\end{aligned}\ ] ] notice that the expression for the four - acceleration here ( naturally ) is independent of what curvature measure that we use . still ( [ fin2p ] ) depends on what curvature measure we are using ( manifesting itself in the occurrence of @xmath133 ) assuming non - zero @xmath128 , since the transport laws for the two types of spatial parallel transport differs . we can rewrite ( [ finp ] ) and ( [ fin2p ] ) as purely three - dimensional equations . for any specific global labeling of the congruence lines ( i.e. any specific set of spatial coordinates adapted to the congruence ) we can locally choose a time slice orthogonal to the congruence so that @xmath134 . this then uniquely defines the three - vector @xmath84 at any point along the gyroscope trajectory . analogous to what we did in going from ( [ kottfinal2 ] ) to ( [ e2 ] ) , for a set of vectors @xmath17 , @xmath19 and @xmath104 orthogonal to the congruence , we let @xmath135 . of two three - vectors @xmath136 and @xmath137 is @xmath138^{-\frac{1}{2 } } \epsilon^{ijk}a_j b_k$ ] where the indices have been lowered with the local three - metric ( again assuming local coordinates orthogonal to the congruence ) . notice that in general ( for congruences with rotation ) there are no global time - slices that are orthogonal to the congruence . the local three - metric corresponding to local orthogonal coordinates is however well defined everywhere anyway . for a shearing ( expanding ) congruence it will however be time dependent ( whatever global time slices we choose ) . ] also we let @xmath139 in coordinates locally comoving with the congruence , we have @xmath140 , where @xmath141 $ ] and @xmath142 is @xmath143 , @xmath144 or @xmath145 for @xmath146 being an even , odd or no permutation of @xmath147 respectively . ] . for simplicity , let us assume that the congruence has vanishing shear and expansion . for this case the two different approaches to spatial curvature radius coincide and we will drop any instances of subscripts ns or ps. introduce then @xmath148 , where @xmath149 and @xmath53 are the inertial coordinates of a system locally comoving with the congruence naturally transforms as a three - vector . ] . also denoting the acceleration of the reference congruence relative to an inertial system locally comoving with the reference congruence by @xmath150 , we get from ( [ fin2p ] ) corresponds to covariant differentiation with respect to the three - metric . ] @xmath151\times { \bf v } \right ) \times { \bf \bar{s}}-\gamma { \mbox{\boldmath$\omega$ } } \times { \bf \bar{s } } .\end{aligned}\ ] ] this is a perfect match with the result of the intuitive derivation performed in @xcite . analogously we may study ( [ finp ] ) for the particular case of vanishing shear , thus considering a rigid congruence . the three - dimensional version of this equation then becomes @xmath152 \times \bar{\bf s } .\end{aligned}\ ] ] we may simplify this expression a bit by introducing @xmath153 , where @xmath154 and @xmath36 means parallel respectively perpendicular to @xmath63 . also using @xmath50 we readily find @xmath155 \times \bar{\bf s } .\end{aligned}\ ] ] again this is a perfect match with the intuitive formalism of @xcite . letting @xmath156 denote local time for an observer comoving with the congruence we can write ( [ fin3 ] ) and ( [ fin5 ] ) respectively as @xmath157 here @xmath158 is given by ( [ fin3 ] ) @xmath159 this form is practical if the gyroscope is freely falling , in which case @xmath160 . alternatively we can get @xmath158 from ( [ fin5 ] ) @xmath161 this form is practical if the gyroscope follows some predetermined path while being acted on by forces . from the simple relation ( see section [ spax ] ) between the stopped spin vector and the projected spin vector and the gyroscope axis respectively , we can use the law of rotation for the stopped spin vector to derive corresponding differential equations for @xmath10 and * x * @xmath162 + { \mbox{\boldmath$\omega$}}_{{{\textrm{\scriptsize e } } } \parallel } \times [ { \bf s}]_\perp + { \mbox{\boldmath$\omega$}}_{{{\textrm{\scriptsize e } } } \perp } \times \left ( \frac{1}{\gamma } [ { \bf s}]_\parallel + \gamma [ { \bf s}]_\perp \right ) \label{combo1}\\ \frac{d { \bf x}}{d\tau_0}{\hspace{-0.5mm}}{\hspace{-0.5mm}}{\hspace{-0.5mm}}{\hspace{-0.5mm}}{\hspace{-0.5mm}}&=&{\hspace{-0.5mm}}{\hspace{-0.5mm}}{\hspace{-0.5mm}}{\hspace{-0.5mm}}-\gamma^2 \frac{d { \bf v}}{d\tau_0 } \left[{\bf x } \cdot { \bf v } \right]+ { \mbox{\boldmath$\omega$}}_{{{\textrm{\scriptsize e } } } \parallel } \times [ { \bf x}]_\perp + { \mbox{\boldmath$\omega$}}_{{{\textrm{\scriptsize e } } } \perp } \times \left ( \gamma [ { \bf x}]_\parallel + \frac{1}{\gamma } [ { \bf x}]_\perp \right ) . \label{combo2}\end{aligned}\ ] ] here we have for brevity introduced @xmath163 note that the @xmath164 entering ( [ combo1 ] ) and ( [ combo2 ] ) is the velocity derivative relative to the reference frame ( not relative to a freely falling frame ) . we note that these differential equations are more complicated than the ones for the stopped spin vector . we conclude that if we are interested in @xmath10 or @xmath87 , it is likely wise to first solve the equation for the stopped spin vector and then ( as in section [ revhep ] ) use the result to find @xmath10 or @xmath87 . as a first example of how one may use the derived formalism , consider a train moving along a straight spatial line in some static geometry . in fact , to be specific , we may consider the train to be moving relative to an upwards accelerating platform in special relativity . on the train we have suspended a gyroscope so that there are no torques acting on it in the comoving system . we thus consider gyroscope motion along a straight line , with respect to a non - rotating but accelerating reference frame . letting @xmath165 and @xmath166 , ( [ fin5 ] ) is immediately reduced to @xmath167 we understand that a gyroscope initially pointing in the forward direction will tip forward as depicted in figure [ fig12 ] . note that the stopped spin vector with respect to the platform corresponds precisely to the spin vector as perceived relative to the train . for example , if the stopped spin vector points @xmath94 down from the horizontal direction , the gyroscope as seen from the train points @xmath94 down from the horizontal direction . to express the gyroscope precession with respect to coordinates comoving with the train we therefore just let @xmath168 in ( [ perfa ] ) and we have the precession explicitly in terms of the time @xmath52 on the train . relative to the train , the gyrocope thus precesses at a steady rate given by @xmath169 . this means that the train in fact has a proper rotation , but more on this is given in @xcite . we can parameterize the gyroscope trajectory by the distance @xmath170 along the platform rather than the time @xmath53 . then ( [ perfa ] ) can be expressed as @xmath171 assuming the train velocity to be low , the tipping angle per distance traveled is thus independent of the velocity . we have simply @xmath172 . thus on a stretch of length @xmath173 we get a net rotation @xmath174 @xmath175 if we want to express @xmath173 and @xmath176 and in si units we must divide the right hand side by @xmath177 ( expressed in si units ) . setting @xmath178 and @xmath179 we get simply @xmath180 this is quite a small angle , and we understand that the relativistic effects of gyroscope precession for most cases here at earth are small . notice how simple this calculation was in the stopped spin vector three - formalism . the equations ( [ fin3 ] ) and ( [ fin5 ] ) both describe how the gyroscope rotates with respect to a coordinate frame that is parallel transported with respect to the spatial geometry . suppose then that we consider motion in the equatorial plane of some axisymmetric geometry . as a specific example we may want to know the net rotation of the gyroscope relative to its initial configuration after a full orbit ( not necessarily a circular orbit ) . we must then take into consideration that a parallel transported frame in general will be rotated relative to its initial configuration after a complete orbit . we can however introduce a new reference frame , that rotates relative to local coordinates spanned by @xmath181 and @xmath182 , in the same manner as a parallel transported reference frame does on a plane . in other words , if we for instance consider a counterclockwise displacement ( @xmath183,@xmath184 ) , then relative to the local vectors @xmath181 and @xmath185 , the new reference frame should turn precisely @xmath186 clockwise . such a reference frame would always return to its initial configuration after a full orbit . to find the rotation of the new reference frame with respect to a parallel transported frame , we first investigate how a vector that is parallel transported with respect to the curved axisymmetric geometry rotates relative to the local coordinates spanned by @xmath181 and @xmath182 . the line element for a two - dimensional axisymmetric spatial geometry can be written in the form - terms , we can not simply select the spatial terms ( without @xmath187 ) to get the spatial line element . there is however an effective spatial geometry also for this case . we may derive the form of this geometry by for instance sending photons back and forth between nearby spatial points and checking the proper time that passes . ] @xmath188 as depicted in figure [ fig14 ] we can imagine an embedding of the geometry , where we cut out a small section and put it on a flat plane . what we want is an expression for how much a vector that is parallel transported , for example along the depicted straight dashed line , rotates relative to the local coordinates @xmath181 and @xmath182 . we understand that the rotation angle corresponds to the angle @xmath174 as depicted . using the notations introduced in figure [ fig14 ] we have simply @xmath189 eliminating @xmath190 and using @xmath191 it follows readily that @xmath192 so this tells us how a parallel transported vector turns relative to the local @xmath181 and @xmath182 , for a small displacement in @xmath193 and @xmath194 . on a flat plane , the corresponding expression to ( [ rdone ] ) is of course simply @xmath195 a reference frame that with respect to @xmath181 and @xmath182 rotates as if we had a flat geometry would then according to ( [ rdone ] ) and ( [ rdonex ] ) rotate relative to a parallel transported reference frame with an angular frequency ( never mind the sign for the moment ) @xmath196 note also that we have @xmath197 thinking about the sign for a second , we realize that with respect to the would - be - flat reference frame , a parallel transported reference frame will have a rotation given by @xmath198 knowing that infinitesimal rotations can simply be added ( to lowest order ) , using ( [ katt ] ) together with ( [ rjx2 ] ) and letting @xmath199 , we get the gyroscope rotation relative to the would - be - flat-grid as ) a little by introducing a @xmath200 sign in the @xmath201-term . if @xmath181 , which by definition is taken to point away from the center of symmetry , points in the direction of increasing @xmath194 , we choose the positive sign , otherwise the negative sign should be chosen . note that @xmath201 is zero for radii where the sign changes , so there is no discontinuity in @xmath202 . ] @xmath203 we can integrate this equation to find the net precession of a gyroscope transported along any path in the axisymmetric geometry . as a particular application of ( [ katt2 ] ) , we can consider the net precession of a gyroscope transported along some closed orbit . since the would - be - flat reference frame returns to its original configuration after a full turn , we just integrate the effects of the infinitesimal rotations following from ( [ katt2 ] ) to calculate the net turn . notice however that to do this straightforwardly , we need @xmath202 in the coordinate base of the reference frame ( i.e. the would - be - flat frame ) . in general we however only have @xmath202 in the coordinates adapted to the stationary observers . for most cases where we would be interested in motion in an axisymmetric geometry , like motion in the equatorial plane of a kerr black hole , this however presents no problem . then all rotations are in the plane of motion and the rotation vector @xmath202 is constant ( in the @xmath204-direction ) in the coordinate basis of the reference frame . notice that the @xmath53 implicitly entering these equations in the @xmath202 is the proper time for a stationary observer . if we instead want to express the precession in global ( schwarzschild ) time , we just multiply by the time dilation factor . even assuming all rotations to be in the plane of motion , we must still in general integrate to get the net precession of the gyroscope , we understand that time dilation , @xmath43 , @xmath205 , @xmath3 and @xmath206 all depends on @xmath207 . assuming all rotations to be in the plane of motion it is effectively a single ( scalar ) integral ( of the net rotation angle around the @xmath15-axis ) . ] . for the particular case of circular motion with constant speed , assuming the time dilation ( i.e. the lapse ) , @xmath205 and @xmath208 to be independent of @xmath193 ( as is the case for the equatorial plane of a kerr black hole ) , there is however no need to integrate at all since all the terms of ( [ katt2 ] ) are constant . the result follows immediately , assuming that we know @xmath205 , @xmath209 and @xmath210 . incidentally it follows from ( [ pr1 ] ) and ( [ pr2 ] ) that @xmath211 for circular motion . the reference background ( fixed to the stars ) around a spinning planet , like the earth , is both accelerating and curved . also there is frame dragging due to the planet rotation , giving a non - zero rotation of the stationary reference observers . if we have the spacetime metric , we can easily find @xmath205 , @xmath209 and @xmath210 . if we do not have an exact spacetime metric however , as is the case for a spinning planet , we need some approximate method ( like the post - newtonian approximation ) to estimate @xmath205 , @xmath209 and @xmath210 . once this is done , assuming the approximation to be valid , ( [ katt2 ] ) gives an accurate description of the precession even considering relativistic speeds . in the case of a rotating ( kerr ) black hole , we do know the metric , and the precession relative to the stationary observers can readily be calculated . notice that within the ergosphere , there are no stationary ( timelike ) observers . still we can in principle use the formalism of this paper also within the ergosphere . to do this we consider coordinates that rigidly rotate around the black hole sufficiently fast to be timelike in the region in question . indeed for the particular case of circular motion there is a paper @xcite , that uses this technique . as a simple example , consider a freely falling gyroscope ( @xmath212 ) orbiting in the equatorial plane of a schwarzschild black hole . using the static observers as our reference congruence , we have @xmath213 . then it follows from ( [ rjx3 ] ) that we have @xmath214 the schwarzschild line element in the equatorial plane is given by @xmath215 here @xmath216 . for convenience we now consider large @xmath194 , so that @xmath217 is small . we have then the acceleration of the freely falling frames @xmath218 ( counted positive in the inwards direction ) . it follows readily , using ( [ katt ] ) , that for this case we have @xmath219 for the large @xmath194 in question the velocities are low and we may set @xmath220 . using ( [ ospace ] ) together with @xmath221 and @xmath222 in ( [ yesrotny ] ) gives @xmath223 this result was derived by w. de sitter in 1916 ( although in a quite different manner than that described here , see @xcite p. 1119 ) . we may note that one third of the net effect comes from the spatial geometry . using a little bit of newtonian mechanics it is easy to derive that for a satellite orbiting the earth at a radius @xmath224 , inserting the proper factor of @xmath225 to handle si - units , ( [ h3 ] ) amounts to @xmath226 knowing that the exterior metric of the earth is approximately schwarzschild , we have then an approximation of the effective rotation vector for a gyroscope orbiting the earth . we can refine this approximation by considering an appropriate non - zero @xmath205 , as discussed earlier . note that , as discussed in section [ tttt ] , the derived precession is the precession with respect to a star - calibrated reference system on the satellite . in @xcite p. 1117 - 1120 , a similar explicitly three - dimensional formalism of spin precession is derived . it is only valid in the post - newtonian regime however . the precession given by ( [ katt2 ] ) is however exact ( assuming an ideal gyroscope ) . for instance , considering the above example of free circular motion in a static geometry , we can easily calculate the exact expressions for @xmath176 and @xmath3 , and thus express the gyroscope precession rate arbitrarily close to the horizon . we have seen how we in a covariant manner can derive an effectively three - dimensional spin precession formalism in a general spacetime . in particular the simple form of ( [ kottfinal ] ) seems novel . in @xcite p. 1117 a similar approach is taken where they consider only the standard spin vector , but expressed relative to a _ boosted _ set of base vectors . they however only apply it to the post - newtonian regime . as mentioned earlier , jantzen et . ( @xcite ) have already pursued the same general idea , although the specific approach and final form of the results differ . in particular they have not employed the explicit 3-dimensional formalism . while the general formalism is derived assuming a general congruence , it seems to have its greatest use as a simple three - dimensional formalism assuming a non - shearing congruence . then we have a fixed spatial geometry and the spatial parallel transport is unambiguous . for this particular case , the derived three - dimensional formalism verifies the result of the intuitive derivation of @xcite . we have also given examples of how the three dimensional formalism can be used to easily find results of physical interest . in the expansion of the second term within the brackets of ( [ done ] ) there will according to ( [ kk ] ) be terms of the type @xmath227 . these can be rewritten in terms of @xmath228 and @xmath229 since we have @xmath230 dealing with @xmath228 rather than @xmath227 is convenient since the former readily can be expressed in terms of spatial curvature and velocity changes relative to the congruence , see @xcite . also @xmath228 has a direct physical relevance . using the identity @xmath231 , it is then easy to derive an alternative form of @xmath25 @xmath232 using this in the second term within the brackets of ( [ done ] ) we have @xmath233 .\end{aligned}\ ] ] as we expand this expression there will be terms containing @xmath16 , @xmath28 and @xmath234 . these we will disregard for the following reasons . terms containing @xmath28 will anyway die when multiplied by @xmath29 ( as they are in ( [ done ] ) ) . terms containing @xmath16 we will disregard since we for the moment only are interested in @xmath235 . when we have a neat expression for this we can find the @xmath16-part a posteriori using the orthogonality of @xmath29 and @xmath28 . we will disregard terms of the type @xmath234 since we know that these must cancel anyway for @xmath29 to stay normalized ( as we know it must by construction of the fermi transport and the relation to the stopped spin vector ) . for the particular case where @xmath236 momentarily , it follows that any net term of the form @xmath237 in the right hand side of ( [ done ] ) must vanish . since the parameter @xmath238 does not depend on @xmath239 it follows that it must vanish entirely . the point is that the form of ( [ come ] ) is such that , when expanded it can be written as a sum of tensors of the type @xmath240 . letting the suffix @xmath36 indicate that only the part orthogonal to both @xmath16 and @xmath19 should be selected , each such term can be written in the form @xmath241_\perp)(t_\alpha t^\sigma b_\sigma+ [ b_\alpha]_\perp)$ ] . adding up the resulting terms of the type @xmath234 ( including the terms of this type coming from the first term within the brackets of ( [ done ] ) ) into a single term @xmath242 we _ know _ that @xmath238 must be zero . ] . note however that in principle , we should contract with the inverse @xmath243 ) that the effect of contracting @xmath244 with a contravariant vector is that it shortens the @xmath19-component of the vector by a @xmath18-factor , while the rest of the on - slice ( orthogonal to @xmath16 ) part of thee vector is unaffected . ] , before disregarding the terms of the described types ( see ( [ done ] ) ) . the form of the inverse is however such that we can carry out the effective cancellations prior to applying the inverse we could not cancel @xmath16 terms directly within the brackets of ( [ done ] ) . that the inverse is not containing any such terms is a benefit of the particular gauge choice in choosing @xmath245 where we had a freedom to include any terms containing @xmath109 . ] . we then readily find @xmath246_\perp t^\mu+ t_\alpha \left[\frac{d u^\mu}{d\tau } + \frac{d \eta^\mu}{d\tau } \right]_\perp \right ) .\end{aligned}\ ] ] by the perpendicular sign @xmath36 we here mean that we should select only the part orthogonal to both @xmath19 and @xmath16 . by @xmath247 we indicate that the equality holds excepting terms of the type @xmath16 , @xmath28 and @xmath234 . in an analogous manner we readily find for the first term within brackets of ( [ done ] ) @xmath248_\perp .\end{aligned}\ ] ] now use ( [ part1 ] ) and ( [ part2 ] ) in ( [ done ] ) . shorten the @xmath19 components by a @xmath18 factor ( according to the effect of the inverse ) , and neglect the @xmath16-term . we readily find @xmath249_\perp -t_\alpha \left [ \frac{d}{d\tau}\left(u^\mu + \eta^\mu \right)\right]_\perp \right ) .\end{aligned}\ ] ] now that we have this compact expression we may also find the @xmath16 term that we earlier omitted . from orthogonality , @xmath250 , follows that @xmath251 which gives @xmath252_\perp -t_\alpha \left [ \frac{d}{d\tau}\left(u^\mu + \eta^\mu \right)\right]_\perp \right ) \nonumber\\ & & + \eta^\mu \bar{s}^\alpha \frac{d\eta_\alpha}{d\tau } .\end{aligned}\ ] ] so here we have the transport equation for the stopped spin vector ( giving the rotation relative to inertial coordinates ) . as an idealized scenario we consider a special relativistic gyroscope which we model as an isolated system of point particles undergoing four - momentum conserving internal collisions . following the discussion in @xcite p. 87 - 90 , we define the angular momentum tensor with respect to the spacetime origin as @xmath253 here the summation runs over events @xmath254 and four - momenta @xmath255 for the various particles at a particular time slice @xmath256 . the ( pauli - lubanski ) spin vector can be written as @xmath257 here @xmath258 is the four - velocity of the center of mass and @xmath259 is the levi - civita tensor ( density ) where @xmath260 . furthermore we introduce an angular momentum four - vector @xmath261 , where @xmath262 is the standard ( relativistic ) angular momentum three - vector , with respect to our reference coordinates . defining @xmath16 as a purely time directed normalized vector with respect to the reference coordinates , we can write @xmath263 letting @xmath264 be the velocity of the center of mass , @xmath18 the corresponding gamma factor and setting @xmath265 with respect to the reference coordinates , we can decompose the four - velocity of the center of mass as @xmath266 . using this in ( [ britt ] ) together with ( [ bratt ] ) , it follows that @xmath267 it is a short exercise to show that in three - formalism this amounts to @xmath268 here @xmath262 is the net angular momentum of the system of point particles , @xmath269 is the center of mass ( center of energy ) , @xmath18 is the gamma factor for the velocity of the center of mass , @xmath270 is the net relativistic three - momentum and @xmath10 is the spatial part of the spin vector . note that the intrinsic angular momentum is not given by @xmath10 but by @xmath93 . note incidentally also that there is a difference between the center of mass and the _ proper _ center of mass ( see @xcite p. 87 - 90 ) . as pointed out e.g. in @xcite , the gyroscope center of mass does not in general lie on the gyroscope central axis . a real gyroscope moving under the influence of forces is neither ( simply ) consisting of point particles nor is it isolated . a more detailed analysis would likely assume a general energy momentum tensor @xmath271 and allow for external forces acting on the elements of the gyroscope . for the purposes of this article the simple derivation outlined above will however suffice . + + * references * +

let be a sparse random polynomial system . this means that each has fixed support ( list of possibly non - zero coefficients ) and each coefficient has a gaussian probability distribution of arbitrary variance . we express the expected number of roots of inside a region as the integral over of a certain * mixed volume * form . when , the classical mixed volume is recovered . the main result in this paper is a bound on the probability that the condition number of on the region is larger than . this bound depends on the integral of the mixed volume form over , and on a certain intrinsic invariant of as a subset of a toric manifold . polynomials with real coefficients are also considered , and bounds for the expected number of real roots and for the condition number are given . the connection between zeros of sparse random polynomial systems , khler geometry , and mechanics ( momentum maps ) is discussed . * keywords : * mixed volume , condition number , polynomial systems , sparse , random . * 2000 math subject classification : * 65h10 , 52a39 . +

math0012104

a complexity theory of homotopy algorithms for solving dense systems of polynomial equations was developed in @xcite . ( see also ( * ? ? ? * ch . the number of steps for these homotopy algorithms was bounded in terms of a condition number and the bzout number . one of the main features of that theory was unitary invariance : the roots of a dense system of polynomial equations are points in projective space , and all the invariants of the theory are invariant under the action of the unitary group . however , unitary action does not preserve sparse coefficient structure . in this paper , we give the one - distribution of the roots of random sparse polynomial systems . we also bound the probability that the condition number of a random sparse polynomial system is large . the roots of a sparse polynomial system are known to belong to a certain toric variety . however , in order to obtain the theorems below , we needed to endow the toric variety with a certain geometrical structure , as explained below . the main insight comes from mechanics , and from symplectic and khler geometry . let @xmath6 be an @xmath7 matrix , with non - negative integer entries . to the matrix @xmath6 we associate the convex polytope @xmath8 given by the convex hull of all the rows , @xmath9 , of @xmath6 : @xmath10 here , we use the notation @xmath11 to denote the dual of a vector space @xmath12 . there are deep reasons to write @xmath8 as a polytope in dual space , as the reader will see later on . assume that @xmath13 . then we can associate to the matrix @xmath6 the space @xmath14 of polynomials with support contained in @xmath15 . this is a linear space , and there are many reasonable choices of an inner product in @xmath14 . let @xmath16 be a diagonal positive definite @xmath17 matrix . its inverse @xmath18 is also a diagonal positive definite @xmath17 matrix . this inverse matrix defines the inner product : @xmath19 the matrix @xmath16 will be called the _ variance matrix_. this terminology arises when we consider random normal polynomials in @xmath14 with variance @xmath20 for the @xmath21th coefficient . we will refer to these randomly generated functions as _ random normal polynomials _ , for short . we will also produce several objects associated to the matrix @xmath6 ( and to the variance matrix @xmath16 ) . the most important one for this paper will be a khler manifold @xmath22 . this manifold is a natural `` phase space '' for the roots of polynomial systems with support in @xmath6 . it is _ the _ natural phase space for the roots of systems of random normal polynomials in @xmath23 . more explicitly , let @xmath24 ( which , as a real manifold , happens to be an @xmath25-fold product of cylinders ) . let @xmath26 denote coordinatewise exponentiation . then we will look at the preimages of the roots of a polynomial system by @xmath27 . we leave out roots that have one coordinate equal to zero and roots at infinity . the differential 2form @xmath28 corresponds to the pull - back of the canonical 2form in a suitable veronese variety ( see section [ symplectic ] ) . systems where all the polynomials have the same support are called _ unmixed_. the general situation ( _ mixed _ polynomial systems ) , where the polynomials may have different supports , is of greater practical interest . it is also a much more challenging situation . we shall consider systems of @xmath25 polynomials in @xmath25 variables , each polynomial in some inner product space of the form @xmath29 ( where @xmath30 and each @xmath31 and each @xmath32 are as above ) . in this realm , a mathematical object ( that we may call a _ mixed manifold _ ) seems to arise naturally . a mixed manifold is an @xmath33tuple @xmath34 where for each @xmath35 , @xmath36 is a khler manifold . mixed manifolds do * not * have a natural canonical hermitian structure . they have @xmath25 equally important hermitian structures . however , they have one natural volume element , the _ mixed volume form _ , given by @xmath37 as explained in @xcite , the volume of @xmath38 relative to the mixed volume form is ( up to a constant ) the mixed volume of the @xmath25tuple of polytopes @xmath39 . we extend the famous result by bernshtein @xcite on the number of roots of mixed systems of polynomials as follows : [ gen - bernshtein ] let @xmath40 and @xmath41 be as above . for each @xmath42 , let @xmath43 be an ( independently distributed ) normal random polynomial in @xmath44 . let @xmath3 be a measurable region of @xmath38 . then , the expected number of roots of the polynomial system @xmath45 in @xmath46 is @xmath47 when each @xmath43 is dense with a variance matrix @xmath32 of the form : @xmath48 the volume element @xmath49 becomes the bzout number @xmath50 times the pull - back to @xmath51 of the fubini - study metric . we thus recover shub and smale s stochastic real version of bzout s theorem @xcite . @xmath52 the general unmixed case ( @xmath53 , @xmath54 ) is a particular case of theorem 8.1 in @xcite . this is the only overlap , since neither theorem generalizes the other . on the other hand , when one sets @xmath55 , one recovers bernshtein s first theorem . the quantity @xmath56 is precisely the _ mixed volume _ of polytopes @xmath40 ( see @xcite for the classical definition of mixed volume and main properties ) . a version of theorem [ gen - bernshtein ] was known to kazarnovskii @xcite and khovanskii . in @xcite , the supports @xmath31 are allowed to have complex exponents . however , uniform variance ( @xmath57 ) is assumed . his method may imply this special case of theorem [ gen - bernshtein ] , but the indications given in @xcite were insufficient for us to reconstruct a proof . the idea of working with roots of polynomial systems in logarithmic coordinates seems to be extremely classical , yet it gives rise to interesting and surprising connections ( see the discussions in @xcite ) . let @xmath58 , and let @xmath59 . a _ root _ of @xmath2 will be represented by some @xmath60 . ( properly speaking , the root of @xmath2 is @xmath61 ) . in this discussion , we assume that the `` root '' @xmath62 is non - degenerate . this means that the derivative of the _ evaluation map _ @xmath63 with respect to the variable in @xmath51 at the point @xmath64 has rank @xmath65 . we are then in the situation of the implicit function theorem , and there is ( locally ) a smooth function @xmath66 such that for @xmath67 in a neighborhood of @xmath2 , we have @xmath68 and @xmath69 . the condition number of @xmath2 at @xmath70 is usually defined as @xmath71 this definition is sensitive to the norm used in the space of linear maps between tangent spaces @xmath72 . in general , one would like to use an operator norm , related to some natural hermitian or riemannian structure on @xmath73 and @xmath38 . in the previous section , we already defined an inner product in each coordinate subspace @xmath74 , given by the variance matrix @xmath32 . since the evaluation function is homogeneous in each coordinate , it makes sense to projectivize each of the coordinate spaces @xmath74 ( with respect to the inner product @xmath75 ) . alternatively , we can use the fubini - study metric in each of the @xmath74 s . by doing so , we are endowing @xmath73 with a fubini - like metric that is scaling - invariant . we will treat @xmath73 as a multiprojective space , and write @xmath76 for @xmath77 . another useful metric in @xmath78 is given by @xmath79 each of the terms in the sum above corresponds to the square of the sine of the fubini ( or angular ) distance between @xmath1 and @xmath80 . therefore , @xmath81 is never larger than the hermitian distance between points in @xmath82 , but is a correct first - order aproximation of the distance when @xmath83 in @xmath78 . ( compare with ( * ? ? ? 12 ) ) . while @xmath73 admits a natural hermitian structure , the solution - space @xmath51 admits @xmath25 possibly different hermitian structures , corresponding to each of the khler forms @xmath84 . in order to elucidate what the natural definition of a condition number for mixed systems of polynomials is , we will interpret the condition number as the inverse of the distance to the _ discriminant locus_. given @xmath85 , we set : @xmath86 and we set @xmath87 as the space of degenerate polynomial systems in @xmath88 . since the fiber @xmath88 inherits the metric structure of @xmath73 , we can speak of the distance to the discriminant locus along a fiber . in this setting , theorem 3 in @xcite becomes : [ condnumber ] under the notations above , if @xmath89 is a non - degenerate root of @xmath2 , @xmath90 there are two interesting particular cases . first of all , if @xmath53 and @xmath54 , we obtain an equality : let @xmath53 and @xmath54 , then under the hypotheses of theorem [ condnumber ] , @xmath91 we can also obtain a version of shub and smale s condition number theorem ( theorem 3 in @xcite ) for dense systems as a particular case , once we choose the correct variance matrices : let @xmath92 be positive integers , and let @xmath31 be the @xmath25-columns matrix having all possible rows with non - negative entries adding up to at most @xmath93 . let @xmath94 then , @xmath91 the factor @xmath95 in the definition of the variance matrix @xmath32 corresponds to the factor @xmath96 in the definition of the normalized condition number in @xcite . it scales the khler forms @xmath84 so that they are equal ( see remark [ recoverkostlan ] p. below ) . in the general mixed case , we would like to interpret the two `` minmax '' bounds as condition numbers related to some natural hermitian or finslerian structures on @xmath51 . see section [ finsler ] for a discussion . theorem [ condnumber ] is very similar to theorem d in @xcite , but the philosophy here is radically different . instead of changing the metric in the fiber @xmath88 , we consider the inner product in @xmath73 as the starting point of our investigation . theorem [ condnumber ] gives us some insight about reasonable metric structures in @xmath97 . as in theorem [ gen - bernshtein ] , let @xmath3 be a measurable set of @xmath51 . in view of theorem [ condnumber ] , we define a restricted condition number ( with respect to @xmath3 ) by : @xmath98 where the distance @xmath81 is taken along the fiber @xmath99 . although we do not know in general how to bound the expected value of @xmath100 , we can give a convenient bound for @xmath101 whenever @xmath3 is compact and in some cases where @xmath3 is not compact . the group @xmath102 acts on @xmath103 by sending @xmath104 into @xmath105 , for any @xmath106 . in more intrinsic terms , @xmath107 and the @xmath102-action commute . with this convention , we can define an intrinsic invariant of the mixed structure @xmath108 : [ intrinsic ] the _ mixed dilation _ of the tuple @xmath109 is : @xmath110 given a set @xmath3 , we define : @xmath111 provided the supremum exists , and @xmath112 otherwise . we will bound the expected number of roots with condition number @xmath113 on @xmath3 in terms of the mixed volume form , the mixed dilation @xmath114 and the expected number of ill - conditioned roots in the _ linear case_. the linear case corresponds to polytopes and variances below : @xmath115 & c^{{\mathrm{lin}}}_i = \left [ \begin{matrix } 1 \\ & 1 \\ & & \ddots & \\ & & & 1 \end{matrix } \right]\end{aligned}\ ] ] [ mixed - cond ] let @xmath116 be the probability that a random @xmath25variate linear complex polynomial has condition number larger than @xmath117 . let @xmath118 be the probability that @xmath119 for a normal random polynomial system @xmath2 with supports @xmath120 and variance @xmath121 . then , @xmath122 there are a few situations where we can assert that @xmath123 . for instance , [ unmixed : complex ] under the hypotheses of theorem [ mixed - cond ] , if @xmath124 and @xmath125 , then @xmath126 the dense case ( theorem 1 p. 237 in @xcite ) is also a consequence of theorem [ mixed - cond ] . we interpret @xmath116 as the probability that a random linear polynomial @xmath2 is at multiprojective distance less than @xmath127 from the discriminant variety @xmath87 . let @xmath128 be such that the following minimum is attained : @xmath129 without loss of generality , we may scale @xmath130 such that @xmath131 . in that case , @xmath132 the right hand term is the projective distance to the discriminant variety along the fiber , in the sense of @xcite . since we are in the linear case , this may be interpreted as the inverse of the condition number of @xmath2 in the sense of ( * ? ? ? * prop . 4 and remark 2 p. 250 ) . recall that each @xmath1 is an independent random normal linear polynomial of degree 1 , and that @xmath32 is the identity . therefore , each @xmath133 is an i.i.d . gaussian variable . if we look at the system @xmath2 as a random variable in @xmath134 , then we obtain the same probability distribution as in @xcite . then , using theorem 6 p. 254 _ ibid _ , we deduce that @xmath135 shub and smale showed in @xcite that the expected number of real roots , in the dense case ( with unitarily invariant probability measure ) is exactly the square root of the expected number of roots . unfortunately , this result seems to be very hard to generalize to the unmixed case . under certain conditions , explicit formul for the unmixed case are available @xcite . also , less explicit bounds for the multi - homogeneous case were given by @xcite . here , we will give a very coarse estimate in terms of the square root of the mixed volume : [ expected - real ] let @xmath3 be a measurable set in @xmath136 , with total lebesgue volume @xmath137 . let @xmath120 and @xmath121 be as above . let @xmath2 be a normal random real polynomial system . then the average number of real roots of @xmath2 in @xmath138 is bounded above by @xmath139 this is of interest when @xmath25 and @xmath3 are fixed . in that case , the expected number of positive real roots ( hence of real roots ) grows as the square root of the mixed volume . it is somewhat easier to investigate real random polynomials in the unmixed case . let @xmath140 be the probability that a linear random real polynomial has condition number larger than @xmath117 . [ unmixed : real ] let @xmath124 and @xmath125 . let @xmath141 be measurable . let @xmath2 be a normal random real polynomial system . then , @xmath142 \le e(u ) \ \nu_{\mathbb r}(n,{\varepsilon})\ ] ] where @xmath143 is the expected number of real roots on @xmath3 . notice that @xmath143 depends on @xmath16 . even if we make @xmath144 , we may still obtain a bound depending on @xmath16 . steve smale provided valuable inspiration for us to develop a theory similar to @xcite for sparse polynomial systems . he also provided examples on how to eliminate the dependency upon unitary invariance in the dense case . the paper by gromov @xcite was of foremost importance to this research . to the best of our knowledge , @xcite is the only clear exposition available of mixed volume in terms of a wedge of differential forms . we thank mike shub for pointing out that reference , and for many suggestions . we would like to thank jean - pierre dedieu for sharing his thoughts with us on newton iteration in riemannian and quotient manifolds . also , we would like to thank felipe cucker , alicia dickenstein , ioannis emiris , askold khovanskii , eric kostlan , t.y . li , martin sombra and jorge p. zubelli for their suggestions and support . this paper was written while g.m . was visiting the liu bie ju center for mathematics at the city university of hong kong . he wishes to thank cityu for the generous support . let @xmath145 be a manifold . a @xmath146form on @xmath145 is said to be non - degenerate if and only if for all @xmath147 , the only vector @xmath148 such that for all @xmath149 , @xmath150 is the zero vector . a _ symplectic form _ on @xmath145 is a closed , non - degenerate @xmath146form @xmath151 on @xmath145 . in that case , @xmath152 is said to be a symplectic manifold . let @xmath145 be a complex manifold . ( we assume that @xmath145 is given with a certain maximal holomorphic atlas ) . if @xmath153 is a chart of @xmath145 , and @xmath154 , then we define @xmath155 such that the following diagram commutes : @xmath156 this is well - defined for each @xmath157 in @xmath145 . indeed , if two charts @xmath12 and @xmath158 overlap at @xmath157 , then @xmath159 is holomorphic so its derivative exists and commutes with multiplication by @xmath160 . the map @xmath161 is called the _ canonical complex structure _ of @xmath145 . ( the complex structure may depend on the holomorphic atlas . we assume a canonical holomorphic atlas of @xmath145 is given ) . note that @xmath162 is the identity on @xmath163 . let @xmath145 be a complex manifold , with complex structure @xmath107 . let @xmath151 be a symplectic form on @xmath145 ( considered as a real manifold ) . then @xmath151 is said to be a _ @xmath164form _ if and only if @xmath165 . a @xmath164 form @xmath151 corresponds to a symmetric form @xmath166 . we say that @xmath151 is strictly positive if and only if the corresponding symmetric form is positive definite for all @xmath167 . therefore , a strictly positive @xmath164form defines a riemann structure on @xmath145 . also , we can recover an hermitian structure on @xmath145 by setting @xmath168 . the triple @xmath169 is said to be a khler manifold when @xmath145 , @xmath151 and @xmath107 are as above . we identify @xmath170 to @xmath171 and use coordinates @xmath172 . the _ canonical _ @xmath146form @xmath173 makes @xmath170 into a symplectic manifold . the natural complex structure @xmath107 is just the multiplication by @xmath174 . the triple @xmath175 is a khler manifold . [ ex : proj ] projective space @xmath176 admits a _ canonical _ @xmath146form defined as follows . let @xmath177 , and let @xmath178 = ( z^1 : \cdots : z^m ) \in \mathbb p^{m-1}$ ] be the corresponding point in @xmath176 . the tangent space @xmath179 } \mathbb p^{m-1}$ ] may be modelled by @xmath180 . then we can define a two - form on @xmath176 by setting : @xmath181 } ( u , v ) = \|z\|^{-2 } \omega_z ( u , v ) { \hspace{0.3cm}\text{,}}\ ] ] where it is assumed that @xmath182 and @xmath183 are orthogonal to @xmath184 . the latter assumption tends to be quite inconvenient , and most people prefer to pull @xmath185}$ ] back to @xmath186 by the canonical projection @xmath187 $ ] . it is standard to write the pull - back @xmath188}$ ] as : @xmath189 using the notation @xmath190 , and where @xmath191 denotes the pull - back by @xmath107 . projective space also inherits the complex structure from @xmath186 . then @xmath185}$ ] is a strictly positive @xmath164form . the corresponding metric is called _ fubini - study _ metric in @xmath186 or @xmath192 . @xmath52 some authors prefer to write @xmath193 instead of @xmath194 . the following notation is assumed : @xmath195 and @xmath196 . then they write @xmath197 as : @xmath198 let @xmath6 be an @xmath7 matrix with non - negative integer entries , and we associate every row @xmath199 of @xmath6 to the monomial @xmath200 . we also assume ( as in the introduction ) that the corresponding polytope ( the convex hull of all the rows ) is @xmath25-dimensional . also , as in the introduction , let @xmath16 be a diagonal positive - definite matrix ( that we called the variance matrix ) . the variance matrix was the matrix of the inner product in @xmath14 . let @xmath201 be the diagonal positive - definite matrix such that @xmath202 . the right - multiplication of some @xmath203 by @xmath204 makes the inner product canonical . the left - multiplication by @xmath201 is the pull - back of this operation in dual - space @xmath205 . we define the map @xmath206 from @xmath207 into @xmath170 : @xmath208 { \hspace{0.3cm}\text{.}}\ ] ] because @xmath201 is diagonal , @xmath209 is invariant by the natural action @xmath210 of @xmath211 on the variable @xmath212 . moreover , we still have the pairing @xmath213 . the variable @xmath214 is corresponds to @xmath215 i.i.d . gaussian variables . we can also compose with the projection into projective space , @xmath216 , @xmath217 { \hspace{0.3cm}\text{.}}\ ] ] when @xmath16 is the identity , the zariski closure of the image of @xmath218 is called the _ veronese variety_. the map @xmath218 is called the _ veronese embedding_. notice that @xmath218 is not defined for certain values of @xmath212 , like @xmath219 . those values are called the _ exceptional set_. the exceptional set is contained in the union of the planes @xmath220 . there is a natural symplectic structure on the closure of the image of @xmath218 , given by the restriction of the fubini - study 2form . we will see below ( lemma [ dimnondeg ] ) that @xmath221 has rank @xmath25 for @xmath222 , because the polytope of @xmath6 has dimension @xmath25 . thus , we can pull - back this structure to @xmath223 by : @xmath224 also , we can pull back the complex structure of @xmath176 , so that @xmath225 becomes a strictly positive @xmath164-form . therefore , the matrix @xmath6 defines a khler manifold @xmath226 . @xmath52 let @xmath227 . we will use coordinates @xmath62 for @xmath38 , where @xmath228 and @xmath229 . given @xmath230 , we define @xmath231 . for any matrix @xmath6 as in the previous example , we can pull - back the khler structure of @xmath232 to obtain another khler manifold @xmath22 . ( actually , it is the same object in logarithmic coordinates , minus points at `` infinity '' . ) an equivalent definition is to pull back the khler structure of the veronese variety by @xmath233 the fubini - study metric on @xmath186 was constructed by applying the operator @xmath234 to a certain convex function ( in our case , @xmath235 ) . this is a general standard way to construct khler structures . in @xcite , it is explained how to associate a ( non - unique ) convex function to any convex body , thus producing an associated khler metric . @xmath52 now a little bit of magic ... @xmath236 , where @xmath237 . both @xmath238 and @xmath107 commute with pull - back , so @xmath239 [ recoverkostlan ] the same is true for @xmath232 . a particular case should be mentioned here . unitary invariance played an important role in @xcite and in @xcite . let us now recover that invariance for dense polynomials . suppose that the rows of our matrix @xmath6 are the exponent vectors of all possible monomials of degree exactly @xmath240 in @xmath241 variables . let @xmath242 $ ] . we set @xmath243 . then , @xmath244 so @xmath225 is a multiple of the fubini - study metric , and we can actually extend @xmath225 to @xmath245 . let @xmath246 represent the polynomial @xmath247 . then evaluation corresponds to the pairing @xmath248 . the action of @xmath249 on @xmath250 induces an action on @xmath251 . all these actions are unitary , and the hermitian structure of the space of polynomials ( in the coordinates above ) is invariant under such actions . @xmath52 for the record , we state explicit formul for several of the invariants associated to the khler manifold @xmath22 . first of all , the function @xmath252 is precisely : + the terminology _ integral _ is borrowed from mechanics , and its refers to the invariance of @xmath253 by @xmath254-action ( see appendix [ momentum ] for more analogies ) . also , the gradient of @xmath253 is called the _ momentum map_. recall that the veronese embedding takes values in projective space . we will use the following notation : @xmath255 this is independent of the representative of equivalence class @xmath256 . now , let @xmath257 mean coordinatewise squaring and @xmath258 be the transpose of @xmath257 . the gradient of @xmath253 is then : + since @xmath259 is a well - defined real function , we may write its derivative as @xmath260 where @xmath261 is the projection operator @xmath262 . then the second derivative of @xmath253 is + using the relation @xmath263 , one obtains an expression for @xmath28 : + we still have to show that @xmath28 is a symplectic form . clearly , @xmath264 . the remaining condition to check is non - degeneracy . in view of formul [ secder ] and [ omega ] , this is a consequence of the following fact : [ dimnondeg ] let @xmath6 be a matrix with non - negative integer entries , such that @xmath8 has dimension @xmath25 . then @xmath265 is injective , for all @xmath228 . the conclusion of this lemma can fail only if there are @xmath228 and @xmath266 with @xmath267 . this means that @xmath268 this can only happen if @xmath269 is in the space spanned by @xmath270 , or , equivalently , @xmath271 is in the space spanned by @xmath272 . this means that all the rows @xmath273 of @xmath6 satisfy @xmath274 for some @xmath275 . interpreting a row of @xmath6 as a vertex of @xmath276 , this means that @xmath276 is contained in the affine plane @xmath277 . we can also write down the hermitian structure of @xmath51 as : + in general , the function @xmath278 goes from @xmath51 into projective space . therefore , its derivative is a mapping @xmath279 for convenience , we will write this derivative as a mapping into @xmath186 , with range @xmath280 . let @xmath261 be the projection operator @xmath281 then , + an immediate consequence of formula [ deriv ] is : [ lem - orthog ] let @xmath203 and @xmath230 be such that @xmath282 . then , @xmath283 in other words , when @xmath284 vanishes , @xmath285 and @xmath286 are the same up to scaling . notice that the hermitian metric is also @xmath287 finally , the volume element associated to @xmath6 is + the _ momentum map _ , also called _ moment map _ , was introduced in its modern formulation by smale @xcite and souriau @xcite . the reader may consult one of the many textbooks in the subject ( such as abraham and marsden @xcite or mcduff and salamon @xcite ) for a general exposition . in appendix [ momentum ] , we will explicitly show what lie group action @xmath288 is the momentum of , and what the associated hamiltonian dynamical system is . in this section we instead follow the point of view of gromov @xcite . the main results in this section are that [ convexity ] the momentum map @xmath288 maps @xmath38 onto the interior of @xmath8 . when @xmath288 is restricted to the real @xmath25-plane @xmath289 \subset { \ensuremath{\mathcal t^{n}}}$ ] , this mapping is a bijection . @xmath52 this seems to be a particular case of the atiyah - guillemin - sternberg theorem . however , technical difficulties prevent us from directly applying this result here ( see appendix [ momentum ] ) . [ volpres ] the momentum map @xmath288 is a volume - preserving map from the manifold @xmath22 into @xmath8 , up to a constant , in the following sense : if @xmath3 is a measurable region of @xmath8 , then @xmath290 we prove proposition [ volpres ] by first assuming proposition [ convexity ] . consider the mapping @xmath291 since we assume @xmath292 , we can apply proposition [ convexity ] and conclude that @xmath145 is a diffeomorphism . the pull - back of the canonical symplectic structure in @xmath293 by @xmath145 is precisely @xmath28 , because of formul [ secder ] and [ omega ] . diffeomorphisms with that property are called _ symplectomorphisms_. since the volume form of a symplectic manifold depends only of the canonical 2form , symplectomorphisms preserve volume . we compose with a scaling by @xmath294 in the first @xmath25 variables , that divides @xmath295 by @xmath296 , and we are done . symplectomorphisms are also known to preserve a few other invariants such as the symplectic width ( see @xcite ) . however , symplectomorphisms are not required to preserve the complex structure and therefore need not be isometries . however , it is explained in @xcite how to define a new complex structure in @xmath297 that will make the map @xmath145 a khler isomorphism , hence an isometry . @xmath52 before proving proposition [ convexity ] , we will need the following result about convexity . we follow here convexity theorem 1.2 in @xcite , attributed to legendre : if @xmath2 is convex and of class @xmath298 on @xmath136 , then the closure of the image @xmath299 in @xmath300 is convex . let @xmath301 be the set of covectors @xmath302 with the property that @xmath303 notice that @xmath301 is a convex subset of @xmath304 . geometrically , the planes in @xmath301 with @xmath305 minimal correspond to the _ envelope _ of the graph of @xmath2 . the set @xmath301 contains @xmath299 : for any given @xmath306 , we set @xmath307 . since @xmath2 is convex , @xmath308 for the converse , assume that there is @xmath309 not in the closure of @xmath310 . then there is some @xmath311 such that @xmath312 we define the following gradient vector field in @xmath136 : @xmath313 because the denominator is bounded below by @xmath127 , this field is well - defined and lispchitz in all of @xmath136 . let us fix an arbitrary initial condition @xmath314 , and let @xmath315 denote a maximal solution of the vector field . since the vector field has norm 1 , @xmath315 can not diverge in finite time and therefore @xmath315 is well - defined for all @xmath316 . now we look at the function @xmath317 . its derivative w.r.t . @xmath318 is @xmath319 . therefore , @xmath320 . we deduce from there that @xmath321 hence , @xmath322 , a contradiction . by replacing @xmath2 by @xmath253 , we conclude that the image of the momentum map @xmath288 is convex . the momentum map @xmath288 maps @xmath38 onto the interior of @xmath276 . indeed , let @xmath323 be a row of @xmath6 , associated to a vertex of @xmath276 . then there is a direction @xmath324 such that @xmath325 for some unique @xmath273 . we claim that @xmath326 . indeed , let @xmath327 , @xmath318 a real parameter . if @xmath328 is another row of @xmath6 , @xmath329 as @xmath330 . we can then write @xmath331 as : @xmath332^t c { \mathrm{diag}}\left [ \begin{matrix } \vdots \\ e^{t a \cdot v } \\ \vdots \end{matrix } \right ] { \hspace{0.3cm}\text{.}}\ ] ] since @xmath16 is positive definite , @xmath333 and @xmath334 where @xmath335 is the unit vector in @xmath336 corresponding to the row @xmath273 . it follows that @xmath337 when we set @xmath338 , we have @xmath339 on @xmath136 , so we have a local diffeomorphism at each point @xmath228 . assume that @xmath340 for @xmath341 . then , let @xmath342 . the function @xmath343 has the same value at @xmath344 and at @xmath345 , hence by rolle s theorem its derivative must vanish at some @xmath346 . in that case , @xmath347 and since @xmath348 , @xmath349 must vanish in some @xmath228 . this contradicts lemma [ dimnondeg ] . we can also give an interpretation of the derivative @xmath285 in terms of the momentum map ( see figure [ figmomentum ] ) . [ lem : geom ] @xmath350 where @xmath351 stands for @xmath352 , and where @xmath199 and @xmath288 are co - vectors . by formula [ deriv ] , @xmath353 hence its @xmath21-th coordinate is : @xmath354 bearing in mind that @xmath355 , we obtain an immediate consequence : [ usefulbound ] for all @xmath356 , @xmath357 @xmath358 in the setting of theorem [ gen - bernshtein ] , we can identify each space of polynomials @xmath359 to the ( co)vector space @xmath360 , endowed with the canonical inner product . the value of @xmath43 at @xmath361 is then precisely @xmath362 . more generally , we can define the _ evaluation map _ by @xmath363 \end{array } { \hspace{0.3cm}\text{.}}\ ] ] following @xcite , we look at the linearization of the implicit function @xmath364 for the equation @xmath365 . the _ condition matrix _ of @xmath366 at @xmath367 is @xmath368 where @xmath58 . above , @xmath369 is a linear operator from an @xmath25-dimensional complex space into @xmath207 , while @xmath370 goes from an @xmath371-dimensional complex space into @xmath207 . [ herecomesthewedge ] assume that @xmath372 . then , @xmath373 note that although @xmath374 is a complex - valued form , each wedge @xmath375 is a real - valued 2form . we compute : @xmath376 { \hspace{0.3cm}\text{,}}\ ] ] and hence @xmath377 also , @xmath378 { \hspace{0.3cm}\text{.}}\ ] ] therefore , @xmath379 \right|^2 { \hspace{0.3cm}\text{.}}\ ] ] we can now use lemma [ lem - orthog ] to conclude the following : + we can now write the same formula as a determinant of a block matrix : @xmath380\ ] ] and replace the determinant by a wedge . the factor @xmath381 comes from replacing @xmath382 by @xmath383 . given @xmath384 , we define @xmath88 as the space of @xmath385 such that @xmath386 vanishes . using ( * ? ? ? * theorem 5 p. 243 ) ( or proposition [ coarea2 ] p. below ) , we deduce that the average number of complex roots is : @xmath387 by lemma [ herecomesthewedge ] , we can replace the inner integral by a @xmath65form valued integral : @xmath388 since the image of @xmath389 is precisely @xmath390 , one can add @xmath25 extra variables corresponding to the directions @xmath391 without changing the integral : we write @xmath392 . since @xmath393 is equal to @xmath394 , the average number of roots is indeed : @xmath395 in the integral above , all the terms that are multiple of @xmath396 for some @xmath397 will cancel out . therefore , @xmath398 now , we apply the integral formula : @xmath399 to obtain : @xmath400 according to formul [ secder ] and [ omega ] , the integrand is just @xmath401 , and thus @xmath402 let @xmath230 and let @xmath403 . without loss of generality , we can assume that @xmath2 is scaled so that for all @xmath35 , @xmath404 . let @xmath405 be such that @xmath406 is singular at @xmath89 , and assume that @xmath407 is minimal . then , due to the scaling we chose , @xmath408 since @xmath409 is singular , there is a vector @xmath410 such that @xmath411 u = 0\ ] ] and hence @xmath412 u = 0 { \hspace{0.3cm}\text{.}}\ ] ] this means that @xmath413 let @xmath414 denote the matrix @xmath415 { \hspace{0.3cm}\text{.}}\ ] ] given @xmath416 , we obtain : @xmath417 we can then scale @xmath182 and @xmath183 , such that @xmath418 . under the assumptions above , @xmath419 is colinear to @xmath420 . assume that @xmath421 , with @xmath130 colinear and @xmath422 orthogonal to @xmath420 . as the image of @xmath389 is orthogonal to @xmath423 , @xmath130 is orthogonal to @xmath424 , so @xmath425 and hence @xmath426 . we can therefore replace @xmath419 by @xmath130 without compromising equality ( [ eqonv ] ) . since @xmath427 was minimal , this implies @xmath428 . we obtain now an explicit expression for @xmath419 in terms of @xmath183 : @xmath429 therefore , @xmath430 so we have proved the following result : [ lemonv ] fix @xmath183 so that @xmath418 and let @xmath405 be such that equation ( [ eqonv ] ) holds and @xmath427 is minimal . then , @xmath431 lemma [ lemonv ] provides an immediate lower bound for @xmath432 : since @xmath433 we can use @xmath418 to deduce that @xmath434 also , for any @xmath183 with @xmath418 , we can choose @xmath435 minimal so that equation ( [ eqonv ] ) applies . using lemma [ lemonv ] , we obtain : @xmath436 hence @xmath437 since this is true for any @xmath183 , and @xmath438 is minimal for all @xmath183 , we have @xmath439 and this proves theorem [ condnumber ] . the proof of theorem [ mixed - cond ] is long . we first sketch the idea of the proof . recall that @xmath88 is the set of all @xmath59 such that @xmath440 , and that @xmath87 is the restriction of the discriminant to the fiber @xmath88 : @xmath441 the space @xmath73 is endowed with a gaussian probability measure , with volume element @xmath442 where @xmath443 is the usual volume form in @xmath444 and @xmath445 . for @xmath3 a set in @xmath38 , we defined earlier ( in the statement of theorem [ mixed - cond ] ) the quantity : @xmath446 = \mathrm{prob } [ \exists ( p , q ) \in u : d_{\mathbb p}(f,\sigma_{(p , q ) } ) < { \varepsilon } ] { \hspace{0.3cm}\text{.}}\ ] ] the nave idea for bounding @xmath447 is as follows : let @xmath448 . we also define @xmath449 as the canonical projection mapping @xmath450 to @xmath82 , and set @xmath451 . then , @xmath452 with equality in the linear case . now we apply the coarea formula ( * ? ? ? * theorem 5 p. 243 ) to obtain : @xmath453 where @xmath454 stands for lebesgue measure in @xmath38 . again , in the linear case , we have equality . we already know from lemma [ herecomesthewedge ] that @xmath455 we should focus now on the inner integral . in each coordinate space @xmath74 , we can introduce a new orthonormal system of coordinates ( depending on @xmath89 ) by decomposing : @xmath456 where @xmath457 is the component colinear to @xmath424 , @xmath458 is the projection of @xmath1 to @xmath459 , and @xmath460 is orthogonal to @xmath457 and @xmath458 . of course , @xmath461 if and only if @xmath462 . also , @xmath463 it is an elementary fact that @xmath464 it follows that for @xmath403 : @xmath465 with equality in the linear case . hence , we obtain : @xmath466 with equality in the linear case . we can integrate the @xmath467 variables @xmath468 to obtain : [ assimpleasitgets ] @xmath469 with equality in the linear case . @xmath52 the domain of integration in proposition [ assimpleasitgets ] makes integration extremely difficult . in order to estimate the inner integral , we will need to perform a change of coordinates . unfortunately , the gaussian in proposition [ assimpleasitgets ] makes that change of coordinates extremely hard , and we will have to restate proposition [ assimpleasitgets ] in terms of integrals over a product of projective spaces . the domain of integration will be @xmath470 . translating an integral in terms of gaussians to an integral in terms of projective spaces is not immediate , and we will use the following elementary fact about gaussians : [ gaussians1 ] let @xmath471 be @xmath472-invariant ( in the sense of the usual scaling action ) . then we can also interpret @xmath473 as a function from @xmath474 into @xmath475 , and : @xmath476\in \mathbb p^{n-1 } } \varphi(x)d[x ] = \int_{x \in \mathbb c^{n } } \varphi(x ) \frac{e^{-\|x\|^2/2}}{(2\pi)^n } dx { \hspace{0.3cm}\text{,}}\ ] ] where , respectively , the natural volume forms on @xmath477 and @xmath478 are understood for each integral . now the integrand in proposition [ assimpleasitgets ] is not @xmath472invariant . this is why we will need the following formula : [ gaussians2 ] under the hypotheses of lemma [ gaussians1 ] , @xmath476\in \mathbb p^{n-1 } } \varphi(x ) d[x ] = \frac{1}{2n } \int_{x \in \mathbb c^{n } } \|x\|^2 \varphi(x ) \frac{e^{-\|x\|^2/2}}{(2\pi)^n } dx { \hspace{0.3cm}\text{.}}\ ] ] where , respectively , the natural volume forms on @xmath477 and @xmath478 are understood for each integral . @xmath479 @xmath480_0^{\infty } \right . + 2n \int_{r=0}^{\infty } \frac{e^{-|r|^2/2}}{(2\pi)^n } dr \right ) \varphi(\theta ) d\theta\ ] ] @xmath481 we can now introduce the notation : @xmath482 this function is invariant under the @xmath483-action @xmath484 . we adopt the following conventions : @xmath485 is the space spanned by coordinates @xmath486 and @xmath487 is its quotient by @xmath483 . we apply @xmath25 times lemma [ gaussians2 ] and obtain : [ multiprojective ] let @xmath488 . then , @xmath489 and in the linear case , @xmath490 now we introduce the following change of coordinates . let @xmath106 be such that the minimum in definition [ intrinsic ] p. is attained : @xmath491 without loss of generality , we scale @xmath492 such that @xmath493 . the following property follows from the definition of @xmath494 : @xmath495 assume now that @xmath496 . then there is @xmath497 , such that @xmath498 and @xmath499 ( assuming the scaling @xmath500 for all @xmath35 ) . setting @xmath501 and @xmath502 , we obtain that @xmath503 . @xmath504 at each value of @xmath35 , @xmath505 where @xmath506 denotes wilkinson s condition number of the linear operator @xmath507 . this is precisely @xmath508 . thus , @xmath509 thus , an @xmath127-neighborhood of @xmath510 is mapped into a @xmath511 neighborhood of @xmath512 . we use this property and equation ( [ wedge1 ] ) to bound : @xmath513 where @xmath514 is the jacobian of @xmath515 at @xmath516 . considering each @xmath389 as a map from @xmath207 into @xmath207 , the jacobian is : @xmath517 we will not use this value in the sequel . @xmath52 in order to simplify the expressions for the bound on @xmath518 , it is convenient to introduce the following notations : @xmath519 now equation ( [ wedge2 ] ) becomes : @xmath520 let @xmath89 be fixed . then @xmath521 together with density function @xmath522 , is a probability space . the expected number of roots in @xmath3 for a linear system is @xmath523 it is also @xmath524 . this holds for all @xmath3 , hence the volume forms are the same and @xmath525 this allows us to interpret the inner integral of equation ( [ wedge3 ] ) as the expected value of a product . this is less than the product of the expected values , and : @xmath526 because generic systems of linear equations have one root , we can also consider @xmath3 as a probability space , with probability measure @xmath527 . therefore , we can bound : @xmath528 the first parenthesis is @xmath529 . the second parenthesis is @xmath530 . this concludes the proof of theorem [ mixed - cond ] . we set @xmath531 , then @xmath532 . as in the complex case ( theorem [ gen - bernshtein ] ) , the expected number of roots can be computed by applying the co - area formula : @xmath533 now there are three big diferences . the set @xmath3 is in @xmath136 instead of @xmath38 , the space @xmath534 contains only real polynomials ( and therefore has half the dimension ) , and we are integrating the square root of @xmath535 . since we do not know in general how to integrate such a square root , we bound the inner integral as follows . we consider the real hilbert space of functions integrable in @xmath534 endowed with gaussian probability measure . the inner product in this space is : @xmath536 where @xmath537 is lebesgue volume . if @xmath538 denotes the constant function equal to @xmath345 , we interpret @xmath539 hence cauchy - schwartz inequality implies : @xmath540 by construction , @xmath541 , and we are left with : @xmath542 as in the complex case , we add extra @xmath25 variables : @xmath543 and we interpret @xmath544 in terms of a wedge . since @xmath545 we obtain : @xmath546 now we would like to use cauchy - schwartz again . this time , the inner product is defined as : @xmath547 hence , @xmath548 this time , @xmath549 , so we bound : @xmath550 let @xmath311 . as in the mixed case , we define : @xmath551 \\ & = & \mathrm{prob}_{f \in \mathcal f } \left [ \exists p \in u : \mathit{ev}(f;p)=0 \text{\ and \ } d_{\mathbb p } ( f , \sigma_{p } ) < { \varepsilon}\right]\end{aligned}\ ] ] where now @xmath552 . let @xmath553 . we also define @xmath554 to be the canonical projection mapping @xmath555 to @xmath556 and set @xmath557 . then , @xmath558 as before , we change coordinates in each fiber of @xmath559 by @xmath560 with @xmath457 colinear to @xmath561 , @xmath562 in the range of @xmath563 , and @xmath460 othogonal to @xmath564 and @xmath458 . this coordinate system is dependent on @xmath64 . in the new coordinate system , formula [ det : cond : matrix ] splits as follows : @xmath565 \right| \left| \det \left [ \begin{matrix } ( { dv_{a}}^{{\scriptscriptstyle\mathrm{i\!i}}})^1_1 & \hdots & ( { dv_{a}}^{{\scriptscriptstyle\mathrm{i\!i}}})^1_n \\ \vdots & & \vdots \\ ( { dv_{a}}^{{\scriptscriptstyle\mathrm{i\!i}}})^n_1 & \hdots & ( { dv_{a}}^{{\scriptscriptstyle\mathrm{i\!i}}})^n_n \\ \end{matrix } \right ] \right| dv \\ & = & \left| \det \left [ \begin{matrix } ( f^1_{{\scriptscriptstyle\mathrm{i\!i}}})_1 & \hdots & ( f^1_{{\scriptscriptstyle\mathrm{i\!i}}})_n\\ \vdots & & \vdots\\ ( f^n_{{\scriptscriptstyle\mathrm{i\!i}}})_1 & \hdots & ( f^n_{{\scriptscriptstyle\mathrm{i\!i}}})_n\\ \end{matrix } \right ] \right| \sqrt{\det dv_a^h dv_a}\end{aligned}\ ] ] the integral @xmath143 of @xmath566 is the expected number of real roots on @xmath3 , therefore @xmath567 \right| \ d\mathcal f^{\mathbb r}_{p } \ { \hspace{0.3cm}\text{.}}\end{gathered}\ ] ] in the new system of coordinates , @xmath568 is defined by the equation : @xmath569 = 0 { \hspace{0.3cm}\text{.}}\ ] ] since @xmath570 , @xmath571 this implies : @xmath572 ) < { \varepsilon } } } \frac{e^{-\sum_i \|f_{{\scriptscriptstyle\mathrm{i\!i}}}^i + f_{{\scriptscriptstyle\mathrm{i\!i\!i}}}^i\|^2/2}}{\sqrt{2\pi}^{\sum m_i } } \cdot \\ \cdot \left| \det \left [ \begin{matrix } ( f^1_{{\scriptscriptstyle\mathrm{i\!i}}})_1 & \hdots & ( f^1_{{\scriptscriptstyle\mathrm{i\!i}}})_n\\ \vdots & & \vdots\\ ( f^n_{{\scriptscriptstyle\mathrm{i\!i}}})_1 & \hdots & ( f^n_{{\scriptscriptstyle\mathrm{i\!i}}})_n\\ \end{matrix } \right ] \right| \ d\mathcal f^{\mathbb r}_{p } \ { \hspace{0.3cm}\text{.}}\end{gathered}\ ] ] we can integrate the @xmath573 variables @xmath468 to obtain : @xmath574 ) < { \varepsilon } } } \frac{e^{-\sum_i \|f_{{\scriptscriptstyle\mathrm{i\!i}}}^i\|^2/2}}{\sqrt{2\pi}^{n^2 } } \left| \det f_{{\scriptscriptstyle\mathrm{i\!i } } } \right|^2 \ d\mathbb r^{n^2 } { \hspace{0.3cm}\text{.}}\ ] ] this is @xmath143 times the probability @xmath575 for the linear case . let @xmath576 be a linear operator . here , we assume that @xmath577 has a canonical riemannian structure , and that @xmath578 has @xmath25 possibly different riemannian structures @xmath579 . we would like to interpret the quantities @xmath580 and @xmath581 in terms of an operator norm on @xmath12 . let @xmath582 denote the unit ball in @xmath577 , and let @xmath583 denote the unit ball in @xmath584 . note that @xmath585 while @xmath586 this should be compared to : @xmath587 one standard way to define norms is by choosing an arbitrary symmetric convex set , and equating that set to the unit ball . ( such norms are called * minkowski norms*. ) there are two immediate obvious choices : 1 . we can use @xmath588 as the unit ball . 2 . we can use @xmath589 as the unit ball . in the first case , we can endow @xmath51 with a @xmath590 finsler structure , while in the second case we can obtain a @xmath591 finsler structure . using @xmath589 would have the advantage of a known probabilistic bound for @xmath113 . however , @xmath588 seems to be more convenient for the study of polyhedral homotopy @xcite . finsler structures are legitimate ways to endow a non - riemannian manifold with a few familiar concepts . for instance , once we define a finsler structure @xmath592 , the length of a curve @xmath315 , @xmath593 $ ] is defined to be @xmath594 . a general discussion on finsler geometry can be found in ( * ? ? ? the proof of theorem [ condnumber ] strongly suggests that the geometry of mixed manifolds should be determined by a much more fundamental invariant , a norm in the space @xmath595 , which we can take to be either side of the following equality : @xmath596 there is a class of polynomial systems that are not unmixed , but nevertheless can be treated as if they were unmixed . for instance , in the dense case , the potentials @xmath597 , @xmath598 are all multiples of one another , therefore @xmath599 . the toric variety associated to those systems admits therefore a ( possibly singular ) hermitian structure . that structure is non - singular provided that the @xmath31 s satisfy delzant s condition @xcite ( see also appendix [ momentum ] below ) . roughly speaking , delzant s condition is an assertion about the angle cones of the minkowski sum of the @xmath600 . the objective of this section is to clarify the analogy between the geometry of polynomial roots and hamiltonian mechanics . the key for that analogy was the existance of a _ momentum map _ associated to convex sets . in the case the convex set is the support of a polynomial , that _ momentum map _ is also the momentum map associated to a certain lie group action , namely the natural action of the @xmath25-torus on the _ toric _ manifold @xmath51 : the @xmath25-torus @xmath601 acts on @xmath38 by @xmath602 where @xmath603 . this action preserves the symplectic structure , since it fixes the @xmath157-variables and translates the @xmath604-variables ( see formula [ 211 ] ) . also , the lie algebra of @xmath254 is @xmath136 . an element @xmath605 of @xmath136 induces an _ infinitesimal _ action ( i.e. a vector field ) @xmath606 in @xmath51 . this vector field is the derivation that to any smooth function @xmath2 associates : @xmath607 if we write @xmath608 , then this formula translates to : @xmath609 by formula [ omega ] . this vector field is hamiltonian : if @xmath610 is a solution of the equation @xmath611 then we can write @xmath612 where @xmath613 . this construction associates to every @xmath614 , the hamiltonian function @xmath613 . the term @xmath615 is a function of @xmath157 , with values in @xmath304 ( the dual of @xmath136 ) . in more general lie group actions , the momentum map takes values in the dual of the lie algebra , so that the pairing @xmath616 always makes sense . a lie group action with such an expression for the hamiltonian is called _ hamiltonian _ or _ strongly hamiltonian_. the theorem of atiyah , guillemin and sternberg asserts that under certain conditions , the image of the momentum map is the convex hull of the points @xmath199 . one of those conditions requires that the action should take place in a compact symplectic manifold ( as in ( * ? ? ? 1 ) or ( * ? ? ? * th . 4 ) ) or ( sometimes ) a compact khler manifold ( * ? ? ? * th 2 ) . we may consider a compactification of @xmath38 , such as the closure of @xmath617 . unfortunately , there are situations where this compactification is not a symplectic manifold , because the form @xmath28 vanishes in the preimage of one or more @xmath199 s ( thus differing from the assumptions of @xcite ) . in @xcite , a necessary and sufficient condition for the compactification of @xmath38 to be a symplectic manifold is given . namely , the polytope @xmath8 should be simple ( i.e. , every vertex should be incident to exactly @xmath25 edges ) and unimodular ( the integer points along the rays generated by each such @xmath25-tuple of edges should span @xmath618 as a @xmath619-module ) . if all the polytopes @xmath620 satisfy delzant s condition then we can construct a corresponding compactifaction @xmath51 and apply atiyah - guillemin - sternberg s theorem . another possibility is to blow - up the singularities , as explained in @xcite . if we do so , polytopes @xmath40 will be `` shaved : '' locally , the cone emanating from a vertex will be truncated by the intersection with a half - space with boundary infinitesimally close to a supporting hyperplane . from another point of view , the underlying normal fan @xcite will be refined . however , the relation between the original polynomial system and the new momentum map is not yet clear . this is an attempt to give a short proof of the coarea formula , in a version suitable to the setting of this paper . this means we take all manifolds and functions smooth and avoid measure theory as much as possible . 1 . let @xmath12 be a smooth riemann manifold , of dimension @xmath145 and volume form @xmath621 . 2 . let @xmath158 be a smooth riemann manifold , of dimension @xmath25 and volume form @xmath622 . 3 . let @xmath3 be an open set of @xmath12 , and @xmath623 be a smooth map , such that @xmath624 is surjective for all @xmath625 in @xmath3 . 4 . let @xmath626 be a smooth function with compact support contained in @xmath3 . then for almost all @xmath627 , @xmath628 is a smooth riemann manifold , and @xmath629 where @xmath630 is the volume element of @xmath631 and @xmath632 is the product of the singular values of @xmath624 . @xmath52 the proposition below is essentially theorem 3 p. 240 of @xcite . however , we do not require our manifolds to be compact . we assume all maps and manifolds are smooth , so that we can apply proposition [ coarea2 ] . 1 . [ ca1 ] let @xmath12 be a smooth @xmath145-dimensional manifold with volume element @xmath621 . [ ca2 ] let @xmath158 be a smooth @xmath25-dimensional manifold with volume element @xmath622 . [ ca3 ] let @xmath637 be a smooth @xmath145-dimensional submanifold of @xmath663 , and let @xmath664 and @xmath665 be the canonical projections from @xmath666 to its factors . [ ca4 ] let @xmath667 be the set of critical points of @xmath668 , we assume that @xmath667 has measure zero and that @xmath667 is a manifold . [ ca5 ] we assume that @xmath669 is regular ( all points in @xmath670 are regular values ) . [ ca6 ] for any open set @xmath671 , for any @xmath672 , we write : @xmath673 . we assume that @xmath674 is finite . every @xmath679 admits an open neighborhood such that @xmath668 restricted to that neighborhood is a diffeomorphism . this defines an open covering of @xmath680 . since @xmath680 is locally compact , we can take a countable subcovering and define a partition of unity @xmath681 subordinated to that subcovering .

the present - day orbit distribution of the neptune trojans is a powerful probe of the dynamical environment of the outer solar system during the late stages of planet migration . in this work , i conservatively debias the inclination , eccentricity , and libration amplitude distributions of the neptune trojans by reducing _ a priori _ unknown discovery and follow - up survey properties to nuisance parameters and using a likelihood - free bayesian rejection sampler for parameter estimation . using this _ survey - agnostic _ approach , i confirm that the neptune trojans are a dynamically excited population : at% confidence , the neptune trojans inclination width must be . for comparison and motivation purposes , i also model the jupiter trojan orbit distributions in the same basis and produce new estimates of their parameters ( jupiter trojan , , and ) . the debiased inclination , libration amplitude , and eccentricity distributions of the neptune trojans are nominally very similar to those of the jupiter trojans . i use these new constraints to inform a suite of simulations of neptune trojan capture by an eccentric , rapidly - migrating neptune from an initially dynamically - hot disk . these simulations demonstrate that if migration and eccentricity - damping timescales were short ( myr , myr ) , the disk that neptune migrated into _ must _ have been pre - heated ( prior to neptune s appearance ) to a width comparable to the neptune trojans extant width to produce a captured population with an inclination distribution width consistent with that of the observed population .

1409.6735

a small sample of neptune trojans has been accumulated by a variety of surveys ; however , inferences drawn from this sample about the intrinsic distributions of neptune trojan orbital properties have been limited and generally qualitative . the challenge inherent in extracting meaningful information from this sample is accurately determining the properties of the surveys that discovered them , and the properties of all surveys that , while sensitive to neptune trojans , did not discover any . in this work , i treat these unknown survey properties as nuisance parameters , and marginalize over them to extract as much useful information about the intrinsic orbital distributions of the neptune trojans as possible . of particular dynamical interest are the inclination , eccentricity , and libration amplitude distributions . these distributions encode information about the formation mechanism ( in - situ formation , chaotic capture , or other processes ) and post - formation evolution . several neptune trojans have remarkably high inclinations ( @xmath7 ) , even though surveys have by - and - large targeted fields near the ecliptic where objects on inclined orbits spend relatively little time . previous works have noted that this qualitatively indicates the existence of a large , poorly - sampled high - inclination neptune trojan population ( sheppard & trujillo 2006 , sheppard & trujillo 2010a ) . in this work , i simultaneously consider the inclination , eccentricity , and libration amplitude distributions , generate synthetic populations of neptune trojans defined by these distributions , then pass these synthetic populations through `` coverage functions : '' simplified observational filters that are treated as independent functions of heliocentric ecliptic latitude @xmath8 , and heliocentric longitudinal separation from the trojan libration centers @xmath9 ( libration centers located roughly @xmath10 from neptune ) , and inclination . the properties of these observational coverage functions are then marginalized over , effectively marginalizing the unknown properties of the surveys which discovered the trojans . like the jupiter trojans and other trans - neptunian populations , the inclination distribution is modeled as a brown s distribution ( @xmath11 ) . the libration amplitude @xmath12 and eccentricity @xmath13 distributions are both modeled as rayleigh distributions , motivated by the distributions of the jupiter trojans . the inclination , libration amplitude , and eccentricity distributions are all truncated at upper limits derived from stability constraints , requiring appropriate corrections to their proposal volume and probability density functions . all statistical analysis is performed in the conceptually simple yet analytically powerful `` approximate bayesian computation '' framework , described in section [ abc ] . l r r r r + name & @xmath14 & @xmath15 & @xmath16 & @xmath17@xmath18 + 2001 qr@xmath19 & 1.3@xmath20 & 25.5@xmath21@xmath20 & 0.57@xmath20 & 10.46@xmath20 + 2004 up@xmath22 & 1.4@xmath20 & 10.8@xmath23@xmath20 & 0.73@xmath20 & 10.24@xmath20 + 2005 tn@xmath24 & 25.0@xmath20 & 8.7@xmath25@xmath20 & 0.62@xmath20 & 8.51@xmath20 + 2005 to@xmath26 & 5.2@xmath20 & 9.2@xmath27@xmath20 & 1.62@xmath20 & 9.12@xmath20 + 2006 rj@xmath28 & 8.2@xmath20 & 6.3@xmath29@xmath20 & 7.99@xmath20 & 0.58@xmath20 + 2007 vl@xmath30 & 28.1@xmath20 & 14.2@xmath31@xmath20 & 11.25@xmath20 & 9.44@xmath20 + 2008 lc@xmath32 & 27.6@xmath20 & 16.4@xmath33@xmath20 & 2.80@xmath20 & 0.84@xmath20 + 2011 hm@xmath34 & 29.4@xmath20 & 9.8@xmath35@xmath20 & 2.60@xmath20 & 7.72@xmath20 + + @xmath36 : j2000 ecliptic inclination . @xmath37 : half - peak rms libration amplitude and @xmath38 uncertainty . @xmath18 : absolute value of j2000 heliocentric ecliptic longitude separation of object and nominal trojan center ; see eqn . [ lprime ] . @xmath39 : j2000 heliocentric ecliptic latitude . the minor planet center ( mpc ) lists nine neptune trojans ( six l4 , three l5 ) , but one of the l5 trojans is unstable and likely a recently captured centaur ( gladman et al . 2012 , horner et al . this object is therefore not considered to be reflective of the intrinsic inclination distribution of the ( putatively primordial ) neptune trojans . with the addition of the newly discovered l5 trojan 2011 hm@xmath34 ( parker et al . 2013 ) , the 8 known long - term stable trojans have ecliptic inclinations ranging from @xmath40 to @xmath41 , and heliocentric ecliptic latitudes at discovery ranging in amplitude from @xmath42 to @xmath43 . figure 1 illustrates these properties , and it is clear that trojans have generally been discovered at latitudes significantly lower than their inclinations , even though an object spends roughly 50% of their time at latitudes greater than 70% of their inclination . only the object 2006 rj@xmath28 was higher than its median latitude at the time of discovery . this indicates that it is likely that most surveys that discovered neptune trojans targeted the ecliptic , and were therefore strongly biased toward detecting low inclination objects , and yet discovered a surplus of high - inclination objects . the larger number of known l4 neptune trojans compared to l5 is likely an artifact of the l5 cloud being more poorly surveyed due to its current proximity to the galactic plane . these neptune trojans were discovered by a variety of surveys , performed at a variety of facilities and under varying conditions , and normally would not represent a sample from which estimating an intrinsic , debiased orbit distribution would be statistically advisable . however , using appropriate statistical care , we can make a conservative estimate of the range of plausible properties of the neptune trojan orbital distribution by marginalizing over the plausible volume of the unknown characteristics of all discovery surveys . this _ survey - agnostic _ approach can conceivably be applied to other populations , and since it is performed in a bayesian framework , the outcomes can be meaningfully combined with results from large , monolithic , well - characterized surveys such as des ( eg . , gulbis et al . 2010 ) , cfeps ( eg . , petit et al . 2011 ) and the ongoing _ _ outer solar system origins survey__. because of their small sample size and currently poorly - characterized orbit distributions , i consider the neptune trojans a useful demonstration population . the libration amplitudes listed in table 1 were generated with the same technique as parker et al . each object s motion was integrated with _ mercury6 _ ( chambers 1999 ) in the presence of the giant planets for 1 myr . 100 clones of each object were integrated , with initial state vectors centered on the jpl horizons solution , perturbed to populate the cartesian uncertainty manifold generated by fitting all ground - based observations of each object with the _ fit_radec _ and _ abg_to_xyz _ routines developed in association with bernstein & kushalani ( 2000 ) . libration amplitudes for each clone were measured by assuming that libration is sinusoidal and deriving the sinusoidal half - amplitude from the rms of the @xmath44 samples of the resonant angle over the entire 1 myr integration : @xmath45 which produces the more appropriate half - amplitude for scaling a sinusoidal model than the usual peak - to - peak definition of @xmath46 $ ] . the rms produces a value that better reflects the mean libration behavior , while defining the amplitude from peak to peak is sensitive to large , single - cycle excursions of the resonant argument . as such , the rms - defined amplitude is always smaller than the peak - to - peak definition . for all parameter estimation in this work i utilize a likelihood - free rejection sampler specifically , the `` approximate bayesian computation '' rejection ( abcr ) scheme first presented in pritchard et al . approximate bayesian computation ( abc ) is conceptually simple but statistically powerful , and has the primary advantage of not requiring the computation of any true likelihood value . i briefly outline this approach below , and refine its description later as merited by the specifics of each application . the literature on abc is well - developed , with much more sophisticated methods available than those employed here ; for a recent review see marin et al . ( 2011 ) . in the context of this work , i utilize abcr in order to approximate the posterior probability density functions of a set of parameters @xmath47 which define the properties of a model @xmath48 that describes an observed population of minor planets . given a metric of similarity @xmath49 that reflects the relative similarity of some set of properties of an observed sample @xmath50 and a synthetic sample @xmath51 , where the synthetic set is generated by a model @xmath52 , and where the parameters @xmath47 of @xmath48 are drawn from prior distribution @xmath53 : 1 . propose a new set of model parameters @xmath54 , drawing from the prior distribution @xmath55 for each parameter . 2 . generate a synthetic sample of observations @xmath51 from model @xmath48 with parameters @xmath54 . 3 . generate the similarity metric @xmath56 for this synthetic sample compared to the observed sample . 4 . if similarity metric @xmath56 is less than some cutoff value @xmath57 , keep @xmath54 as a successful trial . repeat steps 14 until sufficient number of trials have been successful . the distribution of parameters of the retained @xmath54 sample from successful trials approximates the posterior distributions of those parameters in the bayesian framework , given @xmath57 is sufficiently small and the number of trials is sufficiently large . in practice , i set @xmath57 dynamically by sampling many times and saving each sampled @xmath54 , then determining a value for @xmath57 which will retain less than some fraction of the total @xmath54 sample ( in this work , this acceptance rate was set between 0.1%0.01% , depending on the number of dimensions of the parameter space under consideration ) . the similarity metric @xmath56 is usually defined by some distance on a set of summary statistics of the properties of @xmath50 and @xmath51 , but implementations vary on a case - by - case basis . the use of summary statistics to define a metric of similarity is computationally advantageous for large datasets where other methods might be computationally prohibitive . if the summary statistics adopted are sufficient for the underlying distributions , then there is no information lost by using them . in this work , however , the data set is very small , and in the absence of a clear choice of sufficient summary statistics i instead utilize the two - sample anderson - darling test to determine a similarity metric ( the two - sample ad - test @xmath58 statistic ) . for multiple dimensions , the adopted metric @xmath56 is the sum of all one - dimensional , two - sample @xmath58 statistics over all dimensions , divided by the number of dimensions tested . qualitatively , this @xmath56 is small when all distributions are similar to each other , and large when one or more of the distributions are discrepant . in this implementation , the critical values of @xmath58 or @xmath56 and sample - size effects are accounted for as the abcr analysis effectively bootstraps these out . the adoption of a metric based on a collection of 1d 2-sample ad - tests is motivated by its use by the cfeps survey ( eg . , petit et al . 2011 ) , but by utilizing the abcr framework instead of simply adopting the lowest @xmath59-value inferred from each independent ad - test over all dimensions , this work reduces the risk of erroneously identifying a low @xmath59-value in any given application of the metric . the only other large population of trojan objects in the solar system belongs to jupiter . while evidently less populous than neptune s trojan swarms ( sheppard & trujillo 2006 ) , the jupiter trojans are much better sampled , and are nearly complete to @xmath60 . to motivate the functional forms of inclination , libration amplitude , and eccentricity distributions i will adopt for the neptune trojans , i consider those of the jupiter trojans as templates . to date , no population - wide orbit distribution models have been computed for the jupiter trojans . here i consider only the relatively complete @xmath60 sample , and model the inclination , libration amplitude , and eccentricity distributions . i find that the inclination distribution is very well - modeled by a truncated brown s distribution ( brown 2001 ) : @xmath61 while both the libration amplitudes are well - modeled by truncated rayleigh distributions defined by widths @xmath62 and @xmath63 and truncations @xmath64 and @xmath65 . for the jupiter trojans i hold the truncations of each distribution fixed at the maximum observed value of each parameter ; @xmath66 , @xmath67 , and @xmath68 . in the later neptune trojan analysis , these thresholds are defined by dynamical constraints . for consistency , i perform parameter estimation in the same abcr framework that i later apply to the neptune trojans , though i only consider one distribution at a time . in the abcr analysis , the parameters of interest ( @xmath69 , @xmath62 , and @xmath63 ) are sampled uniformly in their distributions means . that is to say , for each trial , @xmath70 is selected uniformly , and then mapped to @xmath69 given that the model inclination distribution is a brown s distribution . because the nominal functional form for the inclination , libration amplitude , and eccentricity distributions are truncated at some upper limit , their means are not trivial to compute . for example , the mean of a non - truncated rayleigh distribution is simply given by @xmath71 , but for a truncated rayleigh the mean is given by @xmath72 where @xmath73 is the truncation value ( see appendix [ raymeanappendix ] for a derivation and more details ) . the functional form for a truncated brown s distribution is similar ( see appendix [ brownsmeanappendix ] for a derivation and more details ) . proposing in the distribution mean has several advantages ; chiefly , it translates an infinite prior volume ( @xmath74 ) to a finite one , as the distribution mean is limited to a maximum value by the truncation of the distribution . for a truncated rayleigh distribution , the mean asymptotically approaches @xmath75 as @xmath76 . a truncated brown s distribution has similar asymptotic behavior ( see appendices [ raymeanappendix ] & [ brownsmeanappendix ] for details ) . for the jupiter trojans , the uniform - mean prior is not particularly important because the sample is so large and well - defined . however , when i later consider the more poorly - sampled neptune trojans , the utility of this prior becomes much more apparent . for each distribution , @xmath77 abcr trials were run , and the 1,000 best trials were retained to generate the posterior pdf for each parameter . figure [ jtpdf ] illustrates the @xmath78 sample of observed osculating inclination , libration amplitude , and osculating eccentricity , compared to the accepted model distributions , and also illustrates the posterior pdfs for @xmath69 , @xmath62 , and @xmath63 defined by the accepted 1,000 abcr trials . the observed jupiter trojan sample is very well modeled by the adopted functional forms for the orbit distribution ; a ks - test indicates peak @xmath59-values of 92% , 70% , and 98% for the inclinations , libration amplitudes , and eccentricities , respectively ; indicating that there is no statistically significant evidence for discrepancy between the model distributions and the observed sample and qualitatively indicating good agreement between the two . because of the lack of consistent information regarding their discovery surveys , a canada - france ecliptic plane survey ( cfeps)-style survey - simulator approach ( jones et al . 2006 ) is not feasible for the neptune trojans . instead , i adopt a set of basis functions informed by typical trans - neptunian object survey strategies which represent the coverage of the sky by all surveys which _ did discover _ or _ plausibly could have discovered _ neptune trojans . the parameters of these `` coverage functions '' are allowed to vary , and the discovery circumstances of each known neptune trojan are used to quantify how well a given choice of parameters reproduces them , given a set of assumed orbit distribution parameters . finally , the properties of these coverage functions are marginalized over , and conservative orbit distribution posterior probability density functions are determined . to model the _ a priori _ unknown survey properties , i assume that the probability of detecting a given trojan is essentially independent of its orbital phase save for effects of areal coverage in ecliptic latitude @xmath8 and longitude @xmath79 . i fold all the effects of survey depth ( which couples with the luminosity function ) and areal coverage into two nuisance `` coverage functions '' @xmath80 and @xmath81 . these represent the probability of any survey detecting a neptune trojan given the trojan s ecliptic latitude and longitude at the time of discovery . nominally , these coverage functions will ( 1 ) only depend on ecliptic latitude and longitude , ( 2 ) likely be monotonic away from the centers of the trojan clouds , and ( 3 ) have a functional range bracketed by the interval [ 0,1 ] and a domain over the entire sky . in addition to modeling survey coverage in ( @xmath82 ) , i also include the possibility of a direct bias against high - inclination objects in the mpc database . the cfeps survey found moderately higher inclination distribution widths @xmath83 for the 3:2 population than previous studies ; because cfeps carefully characterized their tracking biases in a self - consistent way , they suggested that earlier samples of objects with sufficiently accurate orbits to be confidently characterized as members of the 3:2 population may have had an unacknowledged bias against high inclinations . that such biases might exist is not surprising , and the possible sources of such biases were explored in jones et al . ( 2006 ) . because the surveys that contributed to this inclination - biased sample were also sensitive to neptune trojans , it is likely that a similar inclination bias also exists in the observed sample of neptune trojans . ( top , dotted regions ) and a gaussian inclination coverage function ( bottom , dotted regions ) , with parameters selected by procedure in text . the resulting biased model distributions ( solid regions ) excellently match the observed distribution , indicating that the general form of eqn . [ latcov ] is a good model for the shape of the amalgamated latitude coverage functions of all previous surveys . 1- and 2-@xmath84 ranges illustrated by dark and light regions , respectively.,scaledwidth=50.0% ] and @xmath85 after abcr analysis , and marginalized posterior pdfs for each parameter . , scaledwidth=50.0% ] . middle panel : inclination distribution width @xmath86 . bottom panel : inclination coverage function width @xmath87 . vertical dashed lines indicate approximate distribution modes ( @xmath88 , @xmath89 , and @xmath90 ) , estimated using a gaussian kernel density estimator . , scaledwidth=50.0% ] in order to determine the shape of the ecliptic latitude coverage function , i use the well - characterized 3:2 resonant population as a tracer . it is likely that similar ecliptic latitude and inclination biases apply to the 3:2 population as to the neptune trojan population ; most surveys that discovered 3:2 objects were capable of discovering neptune trojans as well , as both populations share broad inclination distributions ( and therefore both occupy a broad swath of ecliptic latitudes ) , similar on - sky rates of motion , and probably share similar luminosity functions . the intrinsic properties of the 3:2 population have been measured by the carefully calibrated cfeps survey ( petit et al . 2011 , gladman et al . 2012 ) . by considering the _ observed _ sample of ( non - kozai ) 3:2 population members ( provided by buie , private communication ) , and adopting the intrinsic inclination distribution measured by cfeps , it is possible to derive the longitudinally - averaged ecliptic coverage function that applied to the amalgam of all surveys that reported 3:2 discoveries to the minor planet center . i begin with an ansatz of an appropriate functional form for the ecliptic latitude coverage function @xmath80 : @xmath91 this form has substantial flexibility provided by three free parameters , but the 3:2 observations constrain them to be strongly correlated . in general , its behavior for small @xmath8 is driven by index @xmath92 and half - width @xmath93 , while for large @xmath8 the function flattens to a constant probability floor set by @xmath94 . the function is truncated above @xmath95 , where @xmath96 is the highest ecliptic latitude ( at discovery ) of any observed 3:2 object , since beyond this point the population provides little constraint on the properties of the coverage function . i then consider the functional form of a possible inclination bias . in this case , there is useful information in literature to inform our functional ansatz : namely , that cfeps and other surveys all found that the _ form _ of the 3:2 population s inclination distribution was well - modeled by a brown s distribution @xmath97 but they found significantly different widths . in that case , i determine that the plausible form of the inclination bias is a gaussian : @xmath98 so @xmath99 where @xmath100 . to map the best - fit cfeps inclination width @xmath101 to the best - fit des inclination width @xmath102 ( gulbis et al . 2010 ) , a gaussian inclination coverage function requires a width of @xmath103 . gladman et al . ( 2012 ) notes that of 24 3:2 objects detected by cfeps , four had inclinations over @xmath104 , while no des 3:2 objects had inclinations so high ; adopting a gaussian inclination coverage function width @xmath105 , the probability of the observed sample retaining an otherwise observable object on an @xmath106 orbit is less than 40% . at @xmath107 ( the highest 3:2 inclination observed by cfeps ) , this drops to roughly 7% . the goal of the following analysis is to utilize the 3:2 population as a `` backlight '' for the neptune trojans , illuminating the biases that likely apply to both populations . specifically , i do this by using the 3:2 population to inform the functional form of the coverage functions @xmath80 and @xmath108 , and to derive priors for the parameters of these coverage functions . to do this , i used the sample of all objects in the mpc which are confidently identified as non - kozai 3:2 objects by the des classification scheme ( marc buie , private communication ) . the observables i model are their inclination @xmath109 and their ecliptic latitude at discovery @xmath110 . i model their inclination distribution as a brown s distribution with width @xmath83 , and adopt the functional forms for @xmath80 and @xmath108 outlined above , and use abcr to back out the pdfs of @xmath83 , @xmath93 , @xmath92 , @xmath111 , and @xmath87 . given a sample of synthetic inclinations @xmath112 drawn from a model @xmath52 , synthetic ecliptic latitudes are generated by @xmath113 , where @xmath114 is a uniform random variate over @xmath115 . to define the prior on @xmath83 , i adopt the inclination distribution results for the plutino population from the cfeps analysis in gladman et al . ( 2012 ) ; i approximate the posterior probability distribution for the brown s distribution @xmath83 as a triangular pdf with minimum value @xmath116 , modal value @xmath117 , and maximum value @xmath118 . this triangular pdf reproduces the quoted mode from gladman et al . ( 2012 ) , 2.5% of its integrated likelihood below their quoted 95% lower confidence limit and 2.5% of its integrated likelihood above their 95% upper confidence limit . the parameter @xmath93 is drawn from a uniform prior over the 5th and 95th percentiles of @xmath110 ( @xmath119 ) . the parameter @xmath92 is drawn from a uniform prior over @xmath120 $ ] . the parameter @xmath111 was drawn from a uniform prior over @xmath121 $ ] . the parameter @xmath87 was defined as @xmath122 , where @xmath123 was drawn from a uniform prior over @xmath124 this effectively defines the prior on @xmath87 such that the value of @xmath125 is drawn from a uniform prior . given these priors , the model for generating @xmath126 and @xmath127 , and the observed samples of @xmath128 and @xmath110 , i perform @xmath129 abcr trials with @xmath56 defined as the mean of @xmath130 and @xmath131 , with equal sample sizes for the observed and model samples . the 1,000 trials which produced the smallest @xmath56 are reserved , and the distribution of @xmath132 , @xmath93 , @xmath92 , @xmath111 , and @xmath87 which defined these trials defines the pdfs of each parameter . given the simple population model , the proposed functional forms for @xmath80 and @xmath108 work remarkably well . figure [ plut_model ] illustrates the quality with which the model inclinations and ecliptic latitudes at discovery reproduce the observed distributions . figures [ plut_modelas ] and [ plut_modelothers ] illustrate the marginalized pdfs of each parameter for the plutino coverage function model that produced the excellent match illustrated in figure [ plut_model ] . these pdfs ( except for that of @xmath133 ) will become the priors for the coverage functions applied to the model neptune trojans in the next section . using a similar procedure to determine the form of the longitude coverage function is less informative , as there may be a bias present for targeting the center of the trojan clouds that is _ not _ present in the discovered and reported 3:2 population . since i am considering two trojan clouds , but am making the implicit assumption that both clouds have identical properties ( aside from some relative occupation normalization , which does not affect this model ) , the observable derived from longitude i consider is the absolute value of the separation from the mean trojan centers ( defined for zero libration amplitude ) , or @xmath134 i adopt a rollover function , normalized such that @xmath135 : @xmath136 this functional form also has two free parameters ( a turnover width @xmath137 and a turnover location @xmath138 ) . to consider coverage ranging from effectively uniform in longitude to sharply biased toward the center of the trojan swarms , i adopt a uniform distribution as the prior on @xmath137 , allowing it to range from @xmath139 ( uniform coverage ) to @xmath140 ( strongly favoring detection toward the core of the trojan clouds ) . @xmath138 is drawn from a uniform prior over @xmath141 $ ] , roughly twice the range of all observed neptune trojan @xmath9 . the functional forms adopted for the inclination and libration amplitude distributions are motivated by the distributions of the jupiter trojans and other trans - neptunian populations . the intrinsic inclination @xmath142 distribution is taken to be a truncated brown s distribution ( brown 2001 ) , @xmath143 similarly , the intrinsic libration amplitude distribution is taken to be a truncated rayleigh distribution , as was adopted for the distribution of libration amplitudes for the jupiter trojans ) , @xmath144 finally , the intrinsic eccentricity distribution is also taken to be a truncated rayleigh distribution , as was adopted for the eccentricity distribution of jupiter trojans . @xmath145 these distributions are clipped by regions of dynamical instability . for the inclination distribution , there are no stable orbits above @xmath146 ( zhou , dvorak & sun 2009 ) , and in the following analysis the inclination threshold is set at @xmath147 . for the libration amplitude distribution , the truncation is placed at @xmath148 as larger libration amplitudes are unstable ( nesvorn & dones , 2002 ) . eccentricities are conservatively limited to values below @xmath149 ( zhou , dvorak & sun 2010 ) . besides the physically - motivated fixed truncation locations , all of these these distributions are defined with one free parameter each . for a variety of reasons , the eccentricity distribution of the neptune trojans is not likely to be strongly biased by observational effects . the eccentricity of neptune trojans are limited by stability constraints to relatively low values , @xmath150 ; additionally , by definition , all neptune trojans have nearly identical semi - major axes . these two facts conspire to produce a relatively weak pericenter bias . i demonstrate this by considering the relative detection rate of the eccentricity extrema of the population using a simple monte carlo test : draw two large samples of @xmath151-magnitudes from a power - law luminosity function with slope @xmath92 and maximum value @xmath152 . one sample i preserve , and take it to represent the apparent magnitudes of objects on circular orbits @xmath153 . to the other sample i add a quantity @xmath154 where @xmath155 is drawn from a sample of keplerian orbits with unit semi - major axis , uniformly distributed mean anomaly , and non - zero , finite eccentricity this sample represents the apparent magnitudes of objects on eccentric orbits @xmath156 . i then define a detection probability function of similar form to real survey sensitivities , @xmath157,\ ] ] where @xmath158 for typical ground based surveys . amalgams of multiple surves tend to have broader turnovers still . i select a value for @xmath159 and set @xmath160 . the potential for bias is estimated by determining the ratio of the detection rate of objects on eccentric orbits to the detection rate of objects on circular orbits . the results of this simple analysis are illustrated in fig . [ ecc_bias ] ; for reasonable power - law slopes , the expected preference for eccentric orbits over circular ones never exceeds 20% over unity and for the most eccentric _ known _ neptune trojan , the preference never exceeds 10% . in addition , the neptune trojan luminosity function has been observed to break to very shallow slope at @xmath161 ( sheppard & trujillo 2010a ) . if a survey reaches fainter than this break , it effectively `` punches through '' the entire trojan population and is complete over all heliocentric distances occupied by them . the majority of discovered neptune trojans were found in surveys that reached substantially fainter than this ( six from sheppard & trujillo 2006 & 2010b , one from parker et al . 2013 ) . for the purposes of the present analysis , then , i proceed assuming that the eccentricity bias in the detected sample is weak and can be neglected . [ ecc_bias ] : black lines illustrate ratio of detection rates for @xmath162 orbits ( roughly the maximum stable neptune trojan eccentricity ) over circular orbits , while gray lines illustrate ratio of detection rates for @xmath163 orbits ( the maximum observed eccentricity ) over circular orbits . dashed line corresponds to a survey flux - sensitivity rollover width of @xmath164 magnitudes , and the solid line to a broader @xmath165 magnitude rollover width . , title="fig:",scaledwidth=50.0% ] as with the jupiter trojans , for the main parameters of interest ( @xmath69 , @xmath62 , and @xmath63 ) i propose uniformly in their distributions means . this becomes particularly important for the neptune trojans because of the inversion of an infinite prior volume to a finite one , allowing the consideration of very large widths in a meaningful way and without imposing an artificial cutoff to the proposal volume . the latitude and inclination coverage function parameters are drawn randomly from the retained sample of 1,000 plutino model trials . this preserves all correlations present in the posteriors of these parameters when informed by the observed plutino population . the longitude coverage function s inverse rollover width @xmath137 is drawn from a uniform prior over [ @xmath139 , @xmath140 ] , while @xmath138 is drawn from a uniform prior over @xmath167 $ ] , containing the range of all observed neptune trojan @xmath9 . given the orbit distributions and priors described in the previous sections , synthetic samples of neptune trojans can be generated . the model i use to generate the synthetic properties and discovery circumstances for the neptune trojans is described in detail in appendix b ; in brief , it takes as input the orbit distribution model parameters @xmath69 , @xmath62 , @xmath63 , @xmath168 , @xmath169 , and @xmath170 , and the coverage function forms and parameters , and returns a sample of synthetic ( @xmath171 ) that reflects both the model intrinsic distributions and the model biases simulated by the coverage functions . [ resultstable ] c c c c c c + parameter & mode & lower 68% & upper 68% & lower 95% & upper 95% + @xmath172 & 10@xmath20 & 9@xmath20 & 16@xmath20 & 7@xmath20 & 26@xmath20 + @xmath63 & 0.044 & 0.039 & 0.070 & 0.033 & 0.125 + + @xmath69 & 21@xmath20 & 17@xmath20 & 46@xmath20 & 13@xmath20 & 91@xmath20 + + @xmath69 & 17@xmath20 & 14@xmath20 & 26@xmath20 & 11@xmath20 & 46@xmath20 + parameter estimation was performed in the abcr framework . for each case under consideration , @xmath173 abcr trials were run , and the 10,000 best trials from each were retained to generate the marginalized posterior pdfs for each parameter under consideration . two cases were considered : in case ( 1 ) , i included the effect of the inclination coverage function informed by the plutinos . in case ( 2 ) , the inclination coverage function was ignored . this allows the effect of the putative inclination bias on the results to be quantified . range of accepted models . , scaledwidth=50.0% ] range of accepted models . , scaledwidth=50.0% ] , top panel illustrating the distribution when allowing a gaussian inclination bias , and bottom panel when such a bias is not allowed for . shaded regions illustrate 68% and 95% confidence intervals determined directly from distribution of accepted abcr samples ; solid vertical line marks mode estimated by a gaussian kernel density estimator , dashed vertical line marks distribution median . , title="fig:",scaledwidth=50.0% ] , top panel illustrating the distribution when allowing a gaussian inclination bias , and bottom panel when such a bias is not allowed for . shaded regions illustrate 68% and 95% confidence intervals determined directly from distribution of accepted abcr samples ; solid vertical line marks mode estimated by a gaussian kernel density estimator , dashed vertical line marks distribution median . , title="fig:",scaledwidth=50.0% ] , but for libration amplitude width @xmath62 . no distinction between models including or ignoring inclination bias , so only one pdf is illustrated.,scaledwidth=50.0% ] , but for eccentricity width @xmath174 . no distinction between models including or ignoring inclination bias , so only one pdf is illustrated.,scaledwidth=50.0% ] figure 7 illustrates the allowed intrinsic distributions of neptune trojan orbit properties and observables , while figure 8 illustrates the allowed apparent distributions of these same properties after the survey coverage functions are applied . the most drastic effect of these coverage functions is apparent in the distribution of latitude at discovery ; while the intrinsic distribution is substantially different from the observed sample , the biased distribution matches the observed sample very well ( much as it did for the 3:2 sample ) . with the inclination coverage function drawn from the plutino priors , the width of the neptune trojan inclination distribution is allowed to climb to very large values ( @xmath175 ) within the 95% confidence interval , effectively allowing solutions approaching a uniform @xmath176 distribution . the lower limit on the 95% confidence interval is @xmath177 , and the mode of the posterior pdf ( illustrated in figure 9 ) is @xmath178 . this is admittedly a large range ; but even so it is a formal definition of the uncertainty when adopting a brown s distribution for the inclination distribution , and can be utilized for studying the origins of the population . while their overall distribution is reasonably well - reproduced by the models adopted here , the observed neptune trojan inclinations appear somewhat bi - modal , perhaps suggesting structure more complex than the simple single brown s distribution used . preliminary tests did not indicate any statistical evidence for a more complex two - component brown s distribution ( similar to the superimposed hot and cold classical kuiper belt ) , but such a model can not be ruled out with the current sample . the libration amplitude distribution is somewhat better constrained at its upper end ; the maximum allowed value of @xmath179 within the 95% confidence interval is @xmath180 , which is nine degrees shy of the truncation limit of the libration amplitudes , indicating a strong requirement for the presence of the gaussian component of the rayleigh distribution . widths as low as @xmath181 are allowed , and the peak of the distribution , @xmath182 is very similar to that of the jupiter trojans ( @xmath183 ) . the posterior pdf for the accepted libration amplitude distribution @xmath179 parameter is illustrated in figure 10 . the eccentricity distribution is a more moderate case ; if a slightly lower eccentricity truncation were adopted ( eg . , 0.08 , nesvorny & vokrouhlicky 2009 ) the distribution would be consistent with @xmath184 . with the adopted truncation of @xmath185 , the maximum distribution width allowed in the 95% confidence interval is @xmath186 . the lower limit is @xmath187 , and the peak falls at @xmath188 , similar again to the jupiter trojans ( @xmath189 ) . the posterior pdf for the accepted eccentricity distribution @xmath63 parameter is illustrated in figure 11 . when the inclination coverage function is set to unity over all inclinations , and all other priors remain as before , the entire confidence interval for the inclination distribution width is shifted to lower widths . the upper limit of the 95% confidence interval becomes better constrained , at @xmath190 , significantly lower than the truncation inclination and indicating evidence for the presence of the gaussian component of the brown s distribution . the lower limit drops by a few degrees to @xmath191 , and the mode of the pdf ( illustrated in the lower panel of figure 9 ) decreases to @xmath192 , similar to the nominal width of the plutinos ( @xmath117 , gladman et al . 2012 ) and the jupiter trojans ( @xmath193 ) . all other distributions were unaffected , and retained identical posterior pdfs . cccccccccccc + i d & @xmath194 & @xmath195 & @xmath196 & @xmath197 & @xmath198 & capture eff . & @xmath199 captured & @xmath200 & @xmath62 & peak @xmath142 prob . & peak @xmath12 prob . + 1 & 26 & 17.820 & 10@xmath20 & 0.05 & 1@xmath201 & 3.20@xmath202 & 8 & @xmath203 & @xmath204 & 0.335 & 0.935 + 2 & 26 & 17.820 & 10@xmath20 & 0.1 & 1@xmath201 & 2.52@xmath205 & 63 & @xmath206 & @xmath207 & 0.544 & 0.972 + 3 & 26 & 17.820 & 10@xmath20 & 0.15 & 1@xmath201 & 1.74@xmath205 & 87 & @xmath208 & @xmath209 & 0.745 & 0.158 + 4 & 26 & 17.820 & 10@xmath20 & 0.15 & 1@xmath201 & 8.80@xmath202 & 22 & @xmath210 & @xmath211 & 0.984 & 0.425 + 5 & 26 & 17.820 & 10@xmath20 & 0.2 & 1@xmath201 & 6.00@xmath202 & 15 & @xmath212 & @xmath213 & 0.658 & 0.841 + 6 & 26 & 17.820 & 15@xmath20 & 0.15 & 1@xmath201 & 1.00@xmath205 & 25 & @xmath214 & @xmath215 & 0.739 & 0.837 + 7 & 26 & 17.820 & 2@xmath20 & 0.15 & 1@xmath201 & 3.32@xmath205 & 83 & @xmath216 & @xmath217 & 0.861 & 0.144 + 8 & 26 & 17.820 & 5@xmath20 & 0.15 & 1@xmath201 & 2.52@xmath205 & 63 & @xmath218 & @xmath219 & 0.926 & 0.461 + 9 & 28 & 19.191 & 10@xmath20 & 0.0086 & 1@xmath201 & 2.67@xmath220 & 2 & & & & + 10 & 28 & 19.191 & 10@xmath20 & 0.05 & 1@xmath201 & 1.20@xmath205 & 30 & @xmath221 & @xmath222 & 0.993 & 0.985 + 11 & 28 & 19.191 & 10@xmath20 & 0.10 & 1@xmath201 & 4.00@xmath220 & 1 & & & & + 12 & 28 & 19.191 & 10@xmath20 & 0.15 & 1@xmath201 & 5.40@xmath202 & 54 & @xmath223 & @xmath224 & 0.366 & 0.813 + 13 & 28 & 19.191 & 10@xmath20 & 0.15 & 1@xmath201 & 3.20@xmath202 & 8 & @xmath225 & @xmath226 & 0.912 & 0.608 + 14 & 28 & 18.531 & 10@xmath20 & 0.1 & 1@xmath201 & 5.20@xmath202 & 26 & @xmath227 & @xmath228 & 0.964 & 0.852 + 15 & 28 & 18.531 & 10@xmath20 & 0.2 & 1@xmath201 & 2.13@xmath202 & 16 & @xmath229 & @xmath230 & 0.273 & 0.884 + 16 & 28 & 17.870 & 10@xmath20 & 0.15 & 1@xmath201 & 1.60@xmath205 & 40 & @xmath231 & @xmath232 & 0.325 & 0.879 + 17 & 28 & 19.191 & 15@xmath20 & 0.15 & 1@xmath201 & 2.00@xmath202 & 10 & @xmath233 & @xmath234 & 0.896 & 0.705 + 18 & 28 & 19.191 & 2@xmath20 & 0.15 & 1@xmath201 & 7.20@xmath202 & 18 & @xmath235 & @xmath236 & 0.779 & 0.998 + 19 & 28 & 19.191 & 5@xmath20 & 0.15 & 1@xmath201 & 7.48@xmath202 & 56 & @xmath237 & @xmath238 & 0.940 & 0.599 + 20 & 28 & 18.531 & 8@xmath20 & 0.2 & 1@xmath201 & 8.00@xmath220 & 8 & @xmath239 & @xmath240 & 0.396 & 0.969 + 21 & 28 & 19.191 & 10@xmath20 & 0.15 & 1@xmath241 & 5.60@xmath202 & 28 & @xmath242 & @xmath243 & 0.378 & 0.381 + 22 & 28 & 19.191 & 2@xmath20 & 0.15 & 1@xmath241 & 3.60@xmath202 & 27 & @xmath244 & @xmath245 & 0.539 & 0.784 + 23 & 28 & 19.191 & 5@xmath20 & 0.15 & 1@xmath241 & 7.60@xmath202 & 19 & @xmath246 & @xmath247 & 0.619 & 0.839 + 24 & 28 & 19.191 & 10@xmath20 & 0.15 & 5@xmath201 & 5.40@xmath202 & 27 & @xmath248 & @xmath249 & 0.467 & 0.690 + 25 & 28 & 19.191 & 10@xmath20 & 0.15 & 1@xmath201 & 6.67@xmath220 & 5 & @xmath250 & @xmath251 & 0.616 & 0.764 + [ captureresults ] in order to determine the implications of the measured properties of the neptune trojan orbit distribution , i performed an exploratory suite of neptune trojan capture simulations . these simulations were similar to those of lykawka et al . ( 2009 ) , except that i include an initial eccentric epoch for neptune and consider a pre - excited disk . these simulations and their outcomes are detailed in the following section . these simulations were performed with a version of _ mercury6 _ ( chambers 1999 ) that includes an artificial , user - defined force designed to drive semi - major axis migration and eccentricity damping ( wolff et al . 2012 ) . the semi - major axis evolution followed an exponential form @xmath252 where @xmath253 is a given planet s semi - major axis at time @xmath254 in the integration , @xmath255 and @xmath256 represent the planet s initial and final semi - major axes , respectively , and @xmath198 is a characteristic timescale for migration . eccentricity damping was performed in a slightly more general way than has been implemented in previous works . i use an identical functional form as for semi - major axis migration , @xmath257 where @xmath258 is a given planet s eccentricity at time @xmath254 in the integration , @xmath259 and @xmath260 represent the planet s initial and final eccentricities , respectively , and @xmath198 is a characteristic timescale for eccentricity damping . this allows me to damp eccentricity asymptotically to some final , finite value . because i am interested in the long - term behavior and stability of captured trojans , i wish to replicate the final architecture of the giant planets orbits as closely as possible . as such , this ability to drive eccentricities to values near their current vaues ( instead of to zero ) is important . the initial configuration of the giant planets was motivated by the nice model and by the recent work of wolff et al . ( 2012 ) and dawson & murray - clay ( 2012 ) , which argued that in order to preserve the cold classical kuiper belt while populating the hot classical kuiper belt , neptune must have had a significant eccentricity ( @xmath261 ) at the start of its final epoch of smooth outward migration . this initial eccentricity must have damped away relatively quickly , with a characteristic timescale less than a few times @xmath262 years . they argue that the characteristic timescale for semi - major axis must have been longer than the characteristic timescale for eccentricity damping . in these simulations , neptune is started with a moderate initial eccentricity . however , in order to avoid early scattering interactions with uranus that would add an unwanted stochastic element to the outcome of these simulations , for many initial configurations neptune s eccentricity is not set as high as dictated by the limits of dawson & murray - clay ( 2012 ) given neptune s initial semi - major axis . the disks that neptune migrates into are composed of massless tracer particles . the disks are started with a brown s distribution for inclination , a rayleigh distribution for eccentricity , and a power - law semi - major axis distribution with index @xmath263 ( approximately producing a @xmath264 surface density distribution ) defined over the range of 22 au @xmath265 30 au . the width of the eccentricity distribution is scaled from of the width of the inclination distribution , such that @xmath266 . if any object is generated with a pericenter lower than saturn s initial semi - major axis , it is re - drawn . initial inclination widths between @xmath267 and @xmath268 were explored . the smallest disk considered was populated with 25,000 particles , while the largest was populated with 100,000 particles . two initial semi - major axes were considered ( @xmath26926 au and 28 au ) , and the initial semi - major axis of uranus was set in each case to sweep through a desired range of neptune : uranus period ratios . since uranus and neptune are not trapped in their 2:1 mmr today , it is unlikely that they ever passed through it . as such , due to their current proximity to the 2:1 , the parameter volume from which the system could migrate through _ decreasing _ period ratios is very small , and i only consider initial configurations where the neptune : uranus period ratio increased ( or remained constant ) with time . at one extreme , uranus is placed as close to its current location as possible without causing early - time interactions with the initial state of neptune . for the @xmath270 au simulation , this places uranus at its current semi - major axis . at the other extreme , uranus was placed such that over the course of the simulation , the neptune : uranus period ratio did not evolve . for equal @xmath198 , this is trivially achieved by setting @xmath271 . several intermediate cases were explored as well . all giant planets other than neptune experienced no eccentricity migration ; their initial eccentricities were set at values representative of their current mean values . for neptune , regardless of initial eccentricity , the eccentricity damping timescale @xmath272 was set to @xmath273 years . the integrations were performed using _ mercury _ s hybrid integrator , and run for @xmath274 ; at this point , the disk particles were down - selected to only those with semi - major axes in proximity to neptune s ( semi - major axis within one neptune hill radius ) . the integrations were then continued with only the giant planets and the down - selected disk particles until 100 myr total time had elapsed . for two small simulations , the integration was continued for 1 gyr . after the prime integration was complete , a short 100,000 year integration was performed where each particle s state vectors were saved very frequently . from this 100,000 year integration , the properties of the remaining objects were determined ; if they were librating around the l4 or l5 points of neptune with libration amplitudes less than @xmath275 and had eccentricities less than 0.12 , they were identified as stable neptune trojans . the number and properties of the objects identified as stable neptune trojans are compiled for each simulation and modeled with the same distributions used to model the observed sample in the preceding sections . table 3 outlines the initial conditions of each simulation , the number of trojans captured and the implied capture efficiency , and the properties of the inclination and libration amplitude distributions of the captured sample . capture efficiencies are inferred from the structure of the primordial disk imposed ; since neptune may only sweep up trojans from a small radial extent within the disk , these efficiencies could have strong spatial variance and simply using a different radial disk structure could produce substantially different capture efficiencies . these efficiencies ( @xmath276 ) , are often sufficiently high that trojan populations captured through the processes modeled here could be substantially reduced without being problematically low compared to the extant population size ( nesvorn & vokrouhlick 2009 ) . figure [ all_sim_results ] illustrates the initial conditions and capture results of all 25 simulations . in general , it is clear that for most simulations , particles prefer to become captured onto trojan orbits with inclinations similar to their primordial inclinations , and the mean primordial inclinations of particles that end up becoming neptune trojans generally closely match their mean inclinations as neptune trojans . this will be discussed further in the following section , and individual simulations will be highlighted . in all the simulations run , no captured trojan population was found to be significantly asymmetric between the l4 and l5 swarms . figure [ l4l5ratio ] illustrates the number captured into each cloud by each simulation , and no points fall outside the 95% confidence region . inclinations of captured trojan particles were found to change little for objects on initially inclined orbits . in general , most simulations resulted in the preferential capture of particles with higher initial inclinations , resulting in final widths higher than initial disk widths due in large part not to stirring or scattering of particles through the capture process , but rather due to this preference for capturing _ initially _ inclined objects . this produces a trend whereby the captured trojan width is nearly identical to the initial width of _ just those particles that would go in to become trojans _ ; see figure [ sigmain ] . this indicates that given the rates of neptune migration explored , the planetesimal disk must have had a significant component of high - inclination objects _ before _ neptune s arrival in order to produce the broad inclination widths that we see today ; indeed , initial widths of @xmath277 are required to produce widths consistent with the lower limit of the 95% confidence interval for the neptune trojan inclination width . , while circles indicate widths of brown s distributions fit to the initial inclinations of _ just those particles that would go on to become trojans , _ and vertical location indicates the final width of the captured trojans.,scaledwidth=45.0% ] ) and initial giant planet configurations , where only the migration timescale @xmath198 was varied from @xmath77 to @xmath129 years . points indicate the individual simulations and estimates on the uncertainty of captured trojan @xmath69 ; contours indicate acceptable regions spanned by linear models of @xmath69 vs. @xmath198 ( light gray indicates 68% confidence contour , dark gray indicates 95% confidence contour.),scaledwidth=45.0% ] in two integrations ( 15 and 20 ) , the captured trojan width is substantially higher than the initial disk width ; in these cases , neptune s initial eccentricity was set to 0.2 , and the uranus : neptune period ratio was allowed to sweep through a moderate range of values . other simulations with the same initial giant planet architecture , but colder initial disks ( @xmath278 ) did not produce any captured trojans and are not listed in table 3 . however , as in most other simulations , in these two integrations the captured trojans were not heated by the capture process ; on average , they ended up with nearly identical final inclinations and initial inclinations . this indicates that the capture process did not heat the trojans ; instead , the capture process _ preferred _ _ pre_-heated objects . as such , in order to produce large observed widths , a primordial disk must already be stirred to large widths to have sufficient high - inclination objects to populate the high inclination trojan orbits ; if the disk is too cold , and few objects are of sufficiently high inclination to be captured , the mean neptune trojan capture probability plummets . in both of these simulations , the lowest initial inclination of a captured object was @xmath279 . thus , while these integrations seem to suggest that very broad inclination distributions can be generated from initially modest disk widths , they _ still _ require that the initial disk have a significant population of high - inclination objects in order to populate the high - inclination trojan orbits . in several integrations with cold ( @xmath280 ) primordial disks ( eg . , 18 and 22 ) , several particles are visible in figure [ all_sim_results ] that attained captured inclinations substantially higher than their primordial values . however , the frequency of this occurrence in these simulations was low enough that the mean inclination was not substantially increased . longer migration timescales tended to produce larger captured inclination widths , as illustrated in figure [ wvt ] . again , this increase in width seems to be largely due to an increasing preference for capturing inclined objects , instead of from stirring during the capture process . the increase in captured trojan inclination width is relatively slow , increasing by only @xmath28120% over an order of magnitude change in migration timescale . recent arguments that the migration of neptune may have occurred on timescales much longer than those explored here ( morbidelli et al . 2014 ) may provide an alternative explanation . if the relatively slow increase in captured population width with increasing migration timescale changes slope beyond the regimes explored here , the long migration timescales suggested in morbidelli et al . ( 2014 ) could produce broad captured trojan inclination distributions from a cold initial disk . naive extrapolation from the simulations performed here suggest that capture efficiencies in such a scenario would be low , and may require problematically - massive initial disks if reduced by several orders ; however , as previously noted , the capture efficiencies determined by the simulations presented here could be very sensitive to the assumptions of the primordial disk s radial structure and extent . investing in simulations of capture simulations with much longer migration timescales than those considered here would be worthwhile . clones of two integrations ( 3 and 12 ) were extended beyond the nominal 100 myrs , and continued out to 1 gyr ( 4 and 13 ) to explore the long - term evolution of the captured populations . in both cases , the captured population decayed by factors of a few , and the nominal inclination distribution widths decreased marginally but the change was not statistically significant . in both cases , the nominal libration amplitudes decreased ; for integration 4 , the libration amplitude width decrease was statistically significant , dropping from @xmath282 to @xmath283 . for integration 13 , the nominal libration amplitude width decrease was smaller and not statistically significant given the captured trojan sample sizes . the neptune trojans have a high mean inclination ; this work has demonstrated that if the population s inclination distribution is well - modeled by a truncated brown s distribution like other resonant minor planet populations , then the width of that distribution must be greater than @xmath284 under conservative assumptions . by simulating the capture of neptune trojans from a dynamically - excited disk by a migrating , eccentric neptune , i have shown that the inclination widths generally do not change drastically between the initial disk and the captured trojan population ; what change there is does not come in large part from stirring during the capture process , but rather from preferential capture of inclined objects . this indicates that objects with excited orbits today were likely excited prior to neptune s arrival . a mechanism that might be responsible for this excitation is not yet clear . migration of planets driven by interactions with excited disks has not yet been explored in great detail . the evidence presented in this work indicates that if neptune migrated quickly , the disk neptune migrated into must have been heated prior to neptune s arrival ; further work investigating precisely how neptune s migration would have proceeded when driven by such a disk is merited . the computations in this paper were run on the odyssey cluster supported by the fas division of science , research computing group at harvard university . i thank matthew holman , rebekah dawson , and konstantin batygin for their input regarding computational techniques and processes involved in planet migration and resonant capture , and wesley fraser for useful discussions on statistical techniques for minor planet population synthesis . much of this manuscript was written in the _ 3 little figs _ cafe in somerville , massachusetts , and i thank the staff and owners for their coffee , lavender biscuits , and patience . bernstein , g. & kushalani , b. 2000 . aj 120 , 3323 . brown , m. 2001 . aj 121 , 2804 - 2814 . chambers , j. 1999 . mnras 203 , 793 - 799 . dawson , r. & murray - clay , r. 2012 . apj 750 , 43 . gladman , b. , and 10 co - authors . apj 144 , 23 . gulbis , a. , elliot , j. , adams , e. , benecchi , s. , buie , m. , trilling , d. , & wasserman , l. 2010 . aj 140 , 350 - 369 . horner , j. & lykawka , p. 2012 . mnras in press . jones , r. & 16 co - authors . 2006 . icarus 185 , 508 - 522 . lykawka , p. , horner , j. , jones , b. , & mukai , t. 2009 . mnras 298 , 1715 - 1729 . marin , j .- m . , pudlo , p. , robert , c. , & ryder , r. 2011 . arxiv:1101.0955v2 morbidelli , a. , gaspar , h. , nesvorny , d. 2014 . icarus 232 , 81 - 87 . nesvorn , d. & vokrouhlick , d. 2009 . aj 137 , 5003 - 5011 . parker , a. & 22 co - authors . aj 145 , 96 . petit . j .- m . and 15 co - authors . aj 142 , 131 . pritchard , j. , seielstad , m. , perez - lezaun , a. , & feldman , m. 1999 . molecular biology and evolution 16 , 1791 - 8 . sheppard , s. and trujillo , c. 2006 . science 313 , 511 . sheppard , s. and trujillo , c. 2010a . science 329 , 1304 . sheppard , s. and trujillo , c. 2010b . apjl 723 , l233 . wolff , s. , dawson , r. , & murray - clay , r. 2012 . apj 764 , 171 . zhou , l .- y . , dvorak , r. & sun , y .- s . mnras 398 , 1217 - 1227 . the pdf of the rayleigh distribution is given by @xmath285 if this distribution is allowed to be continuous over all positive reals , then the mean is given by @xmath286 . however , in the case where there is a finite upper limit @xmath168 applied , the mean is not as straightforward . the mean of such a truncated distribution with known pdf @xmath287 and cdf @xmath288 ( in the case where the pdf and cdf are defined for arguments greater than 0 ) is given by @xmath289 in the case of a rayleigh distribution , this evaluates to @xmath290 since as @xmath84 grows large , the truncated distribution approaches @xmath291 , the mean of this truncated rayleigh distribution has the asymptotic value of @xmath292 as @xmath76 . [ raymean ] is a transcendental equation , so in order to propose a @xmath70 as done in the text , eqn . [ raymean ] must be numerically solved for @xmath84 . in my implementation , this was accomplished with the _ scipy _ implementation of brent s method . as long as the truncation value @xmath168 is greater than @xmath84 , then the mode of the truncated rayleigh distribution remains @xmath84 . the median of the truncated rayleigh distribution is trivial to compute , but i include it for completeness here : @xmath293 which , as @xmath294 approaches the limit of the un - truncated distribution of @xmath295 . the brown s distribution ( brown 2001 ) is colloquially given by @xmath296 for this work , we will assume that it applies only to the range @xmath297 ( neglecting retrograde orbits ) . this means that , by nature , the brown s distribution does not have a simple pdf , as it is already truncated . additionally , since integrals of the form @xmath298 do not evaluate to a convenient form , determining the normalization of a pdf or cdf is not trivial . in order to determine the mean value , then , we must evaluate @xmath299 to simplify this prospect , we use the taylor expansion of @xmath300 to make these integrals more similar to those we computed for the rayleigh distribution . for truncations up to @xmath301 , the taylor series up to order 5 is sufficient for our purposes . @xmath302 where @xmath303 given that as @xmath84 grows large , the truncated distribution approaches @xmath304 , the mean of a truncated brown s distribution has the asymptotic value of @xmath305 as @xmath76 . for the approximation given by eqn . [ brownmeanapprox ] the astymptotic limit is @xmath306 . for truncations up to @xmath307 , these asymptotic values agree to within 0.05% . as with the truncated rayleigh distribution , this must be solved numerically for @xmath84 given a desired @xmath70 , and this was accomplished with the same numerical method described previously . given a proposed parametric model for orbit distributions , i generate a synthetic sample of neptune trojan orbits and observational circumstances in the following way . first , inclinations @xmath142 , libration amplitudes @xmath12 , and eccentricities @xmath13 are drawn from their respective proposed distributions ( semi - major axes are assumed to be identical ) . libration phases @xmath308 are uniformly sampled over @xmath309 , and the current resonant arguments are generated by assuming sinusoidal libration behavior : @xmath310 . since i am defining the distribution with respect to the centers of the l4 and l5 clouds , i set @xmath311 instead of @xmath10 , where @xmath312 is a small , positive angle derived from the empirical libration properties of the synthetic neptune trojans generated by the capture simulations in this work . this offset was found to depend on @xmath12 , @xmath13 , and ( weakly ) on @xmath142 . true majority of the variation in mean libration center was found to be well - represented by @xmath313 ; see fig . [ liboffset ] . longitudes of ascending node @xmath314 are uniformly sampled over @xmath309 . instead of proposing an argument of pericenter and mean anomaly , then solving kepler s equation , i use a small - eccentricity approximation to probabilistically determine an offset from the object s mean longitude caused by the object s eccentricity ( the epicyclic component of motion ) . for eccentricities @xmath315 , a sample of approximate offsets between true anomaly and mean anomaly can be drawn from @xmath316 where @xmath114 is a uniform random variate drawn over @xmath317 , and @xmath318 is in radians . the two other observational parameters needed are the longitudinal angle between the trojan and the mean trojan libration center , @xmath320 , and the absolute value of the heliocentric latitude @xmath321 . from the cartesian coordinates , now that @xmath142 , @xmath9 , and @xmath323 are available , they are passed through their respective coverage functions ; these functions represent the probability that a given object will be observed given its @xmath142 , @xmath9 , or @xmath323 . `` synthetic observed '' trojans are randomly sampled given the product of their coverage functions . the biased , `` synthetic observed '' sample of ( @xmath324 ) is returned to be compared directly to the real neptune trojan properties . libration centers found for synthetic neptune trojans . dashed line indicates trend of @xmath313 , the adopted model used to produce synthetic neptune trojan observation circumstances.,scaledwidth=50.0% ]

in an effort to characterize the mach numbers of ism magnetohydrodynamic ( mhd ) turbulence , we study the probability distribution functions ( pdfs ) of spatial increments of density , velocity , and magnetic field for fourteen ideal isothermal mhd simulations at resolution . in particular , we fit the pdfs using the tsallis function and study the dependency of fit parameters on the compressibility and magnetization of the gas . we find that the tsallis function fits pdfs of mhd turbulence well , with fit parameters showing sensitivities to the sonic and alfvn mach numbers . for 3d density , column density , and position - position - velocity ( ppv ) data we find that the amplitude and width of the pdfs shows a dependency on the sonic mach number . we also find the width of the pdf is sensitive to global alfvnic mach number especially in cases where the sonic number is high . these dependencies are also found for mock observational cases , where cloud - like boundary conditions , smoothing , and noise are introduced . the ability of tsallis statistics to characterize sonic and alfvnic mach numbers of simulated ism turbulence point to it being a useful tool in the analysis of the observed ism , especially when used simultaneously with other statistical techniques .

1103.3299

understanding the dynamics and evolution of the interstellar medium ( ism ) is critical to advancing our knowledge of many astrophysical phenomena spanning a wide range of scales such as star formation , cosmic ray physics , magnetic reconnection , galaxy evolution , and magnetic dynamo theory ( elmegreen & scalo 2004 ) . an essential component of the current paradigm of the ism is the ubiquitous existence of magnetohydrodynamic ( mhd ) turbulence ( see review by mckee & ostriker 2007 ) . turbulence and magnetic fields play a crucial roll in each of these processes and are some of the main drivers of ism evolution . evidence for the role of turbulence is seen in the `` big power law '' of the electron density fluctuations ( armstrong , rickett , & spangler 1995 ) , fractal structure in the molecular media ( elmegreen & falgarone 1996 , stutzki et al . 1998 ) , and intensity fluctuations contributed by both density and turbulent velocity in channel maps ( crovisier & dickey 1983 ; green 1993 ; deshpande , dwarakanath , & goss 2000 ; elmegreen , kim , & staveley - smith 2001 ) . despite the obvious importance of mhd turbulence to astrophysics , few methods exist to study it directly . due to advances in computational power and the general recognition of turbulence as an important ism process , major advances have been made in mhd turbulence theory and observation in the last ten years . however , the issue of turbulence and its effects on the ism ( and processes therein ) remains one of the most exciting and open problems in the field ( elmegreen & scalo 2004 ) . astrophysical turbulence is a complex nonlinear phenomena that can occur in a multiphase media with many energy injection sources on scales ranging from kpc down to sub - au . although limited in complexity , numerical simulations of turbulence provide one of the best avenues for researchers of the ism to understand the nature of magnetized turbulence . the combined efforts of predictive theory and numerical tests have greatly increased our knowledge of mhd turbulence , including its anisotropy , intermittency , and imbalanced nature ( see cho , lazarian , & vishniac 2002 , kowal & lazarian 2010 , beresnyak & lazarian 2010 ) . yet what about observationally driven studies of turbulence ? the most common observational techniques to study turbulence include scintillation studies , which are limited to fluctuations in only ionized media ( e.g. narayan & goodman 1989 ; spangler & gwinn 1990 ) , density fluctuations via column density maps , and radio spectroscopic observations via centroids of spectral lines ( falgarone et al . 1994 ; miesch & bally 1994 ; miesch & scalo 1995 ; lis et al . 1998 ; miesch , scalo , & bally 1999 ) . column densities are the most abundant and easily obtained observable data and have shown that density fluctuations can be a useful and straightforward way of gauging turbulence parameters ( monin & yaglom 1967 ; lithwick & goldreich 2001 ; cho & lazarian 2003 ) . position - position - velocity ( ppv ) spectroscopic data has the advantage over column density maps in that it contains information on the turbulent velocity field . however , this type of data provides contributions of _ both _ density and velocity fluctuations entangled together , and the process of separating the two has proven a challenging problem ( see lazarian 2006b ) . in addition , one must use caution when dealing with ppv data , as structures seen in velocity slices are not one - to - one with structures in three dimensional position space . one of the main approaches for characterizing ism turbulence is based on using statistical techniques and descriptions . the most common `` go - to '' tool for both observers and theorist alike is the spatial power spectrum . in fact , most of the attempts to relate observations to models has been by obtaining the spectral index ( i.e. the log - log slope of the power spectrum ) of column density and velocity . while obtaining the spectral index of column density is straightforward , more sophisticated techniques for obtaining the velocity spectral index from ppv data have been recently developed . these include : velocity channel analysis ( vca ) ( lazarian & pogosyan 2000 ; esquivel et al . 2003 ; lazarian & pogosyan 2004 ; lazarian et al . 2001 ; padoan et al . 2003 ; chepurnov & lazarian 2009 ) , the spectral correlation function ( scf ) ( rosolowsky et al . 1999 ; padoan , rosolowsky , & goodman 2001 ) , velocity coordinate spectrum ( vcs)(lazarian & pogosyan 2008 , 2006 ; chepurnov & lazarian 2006 , 2009 ; padoan et al . 2009 ) , and modified velocity centroids ( lazarian & esquivel 2003 ; esquivel & lazarian 2005 ; ossenkopf et al . 2006 ; esquivel et al . 2007 ) . for cases where turbulence is supersonic , the vca is most appropriate while centroids can be used in subsonic cases . although the power spectrum is useful for obtaining information about energy transfer over scales , it does not provide a full picture of turbulence , partially because it only contains information on fourier amplitudes . an example of this is illuminated in a study by chepurnov et al . ( 2008 ) , who showed that a substantially different distribution of density could have the same power spectrum . in light of this , many other techniques have been developed to study and parametrize observational magnetic turbulence . these include higher order spectrum , such as the bispectrum , higher order statistical moments , topological techniques ( such as genus ) , clump and hierarchical structure algorithms ( such as dendrograms ) , principle component analysis , and structure / correlation functions as tests of intermittency and anisotropy ( for examples of such studies see heyer & schloerb 1997 , burkhart et al . 2009 ; chepurnov & lazarian 2009 ; kowal , lazarian , & beresnyak 2007 ; goodman et al . 2009 ; burkhart et al . wavelets methods , and variations on them such as the @xmath1-variance method , have also been shown to be very useful in characterizing inhomogeneities in data ( see ossenkopf , krips , & stutzki 2008a , 2008b ) . in particular , many of the studies mentioned above focus on obtaining the parameters of turbulence from observations . these parameters include sonic and alfvnic mach numbers , injection scale , gas temperature , and reynolds number . in particular the sonic and alfvnic mach numbers provide much coveted information on the gas compressibility and magnetization . many of these techniques , geared towards obtaining the parameters of turbulence via density fluctuations studies , were successfully applied to observational data ( see burkhart et al . 2010 and chepurnov et al . 2008 , for examples ) . vca and vcs were also applied to galactic hi data and successfully recovered the spectrum of velocity ( see chepurnov et al . 2010 ) . in this vein , esquivel & lazarian ( 2010 ) , henceforth known as el10 , used the so - called tsallis statistic for studies of mhd turbulence . it is this statistic that is the focus of this work , and here we will further illuminate its uses . the tsallis distribution is a function that can be fit to incremental pdfs of turbulent density , magnetic field , and velocity . in astrophysical settings , tsallis statistics was originally used in the context of solar wind observations ( burlaga & vias 2004a ) . el10 applied this method to 3d mhd simulations with four varying values of sonic and alfvnic mach number at @xmath2 resolution . they explored density , magnetic field , velocity , and column densities , and found that the tsallis distribution is a very good fit to pdfs of increments of turbulence . they also found that the parameters of the fit that describe the width and tails of the pdfs showed dependency on the compressibility and magnetization of the simulation . the statistic is particularly useful in that it is scale independent and thus a comparison between the analysis of simulations and observations is not burdened with complicated scaling relations . this opens up the possibility of using the tsallis method on observed ism data in order to gain access to information on these parameters . while el10 was the first to implement this tool on simulations of ism mhd turbulence , they used low spatial resolution simulations , a small parameter regime , and did not explore the dependencies on the amplitude fit parameter . they also did not explore the use of tsallis statistics on spectroscopic data . in this paper , we will greatly extend their parameter range and resolution from 4 at @xmath2 isothermal mhd simulations to 14 at @xmath0 . in addition , we will more explicitly explore the ability of the tsallis function to describe data of an observable nature , such as smoothed synthetic column density maps and synthetic ppv data of varying velocity resolution . we also investigate the quality of our fits and subsequent fit parameters . the paper is organized as follows . in [ tsallis ] we describe the tsallis distribution which is fit to increments of mhd turbulence pdfs . in [ numericalsetup ] we discuss our numerical scheme and resulting simulations . in [ 3d ] we apply tsallis to non - observational 3d quantities ( density and directional components of magnetic field and velocity ) and test the accuracy of our fits . in [ observ ] we apply this tool to observational quantities such as column density and ppv data . in [ disc ] we discuss our results followed by the conclusions in [ conc ] . the tsallis distribution ( tsallis 1988 ) was originally derived as a means to extend traditional boltzmann - gibbs mechanics to fractal and multifractal systems . the complex dynamics of multifractal systems apply to many natural environments such as ism turbulence ( shivamoggi 1995 ) . it is therefore fitting to explore the extent to which tsallis statistics can be used to describe these systems and processes therein ( for further discussion see section [ relate ] ) . work of this nature was first carried out by burlaga and collaborators ( burlaga & vias 2004a , 2004b , 2005a , 2005b , 2006 ; burlaga , ness , & acuas 2006 , 2007 , 2009 ; burlaga , vias , & wang 2007 ) to describe the temporal variation in pdfs of magnetic field strength and velocity of solar wind measured by the _ voyager 1 _ & _ 2 _ spacecrafts . el10 used tsallis statistics to describe the spatial variation in pdfs of density , velocity , and magnetic field of mhd simulations similar to those used here . both efforts found that the tsallis distribution provided adequate fits to their pdfs and gave insight into statistics of turbulence . the tsallis function of an arbitrary incremental pdf @xmath3 has the form : @xmath4^{-1/(q-1)}\ ] ] the fit is described by the three dependent parameters @xmath5 , @xmath6 , and @xmath7 . the @xmath5 parameter describes the amplitude while @xmath7 is related to the width or dispersion of the distribution . parameter @xmath6 , referred to as the `` non - extensivity parameter '' or `` entropic index '' , describes the sharpness and tail size of the distribution . the arbitrary function used to describe density and the directional components of velocity and magnetic field in this application takes the form of our incremental pdf . it has the form @xmath8 , where @xmath9 ... @xmath10 refers to a spatial average . depending on the quantity in question , we set @xmath11 ; where `` @xmath12 '' is the lag or spatial scale . for a given lag this calculation is done for each pixel in the three cardinal directions . a normalized 100 bin histogram of these values results in our incremental pdf which is then fit with the tsallis function ( see figure [ fig:3dhistsall ] for an example pdf ) . the tsallis fit parameters are in many ways similar to statistical moments . moments , more specifically the third and fourth order moments , have been used to describe the density distributions and have shown sensitivities to simulation compressibility ( kowal , lazarian , & beresnyak 2007 ; burkhart et al . the first and second order moments simply correspond to the mean and variance of a distribution . skewness , or third order moment , describes the asymmetry of a distribution about its mode . skewness can have positive or negative values corresponding to right and left shifts of a distribution respectively . the fourth order moment , kurtosis , is a measure of a distribution s peaked or flatness compared to a gaussian distribution . like skewness , kurtosis can have positive or negative values corresponding to increased sharpness or flatness . in regards to the tsallis fitting parameters , the @xmath7 parameter is similar to the second order moment variance while @xmath6 is closely analogous to fourth order moment kurtosis . unlike higher order moments , however , the tsallis fitting parameters are dependent least - squares fit coefficients and are more sensitive to subtle changes in the pdf . we generate a database of 14 three dimensional numerical simulations ( 512@xmath13 resolution ) of isothermal compressible mhd turbulence by using the cho & lazarian ( 2002 ) code and varying the input values for the sonic and alfvnic mach number ( see table [ tab : table1 ] ) . the sonic mach number is defined as @xmath14 , where @xmath15 is the local velocity vector magnitude and @xmath16 is the sound speed . averaging is done over the entire simulation . similarly , the alfvnic mach number is @xmath17 , where @xmath18 is the alfvnic velocity , @xmath19 is the local magnetic field vector magnitude , and @xmath20 is density . below , we briefly outline the major points of the numerical setup ( for more details see cho & lazarian 2002 ) . the code is a second - order - accurate hybrid essentially non - oscillatory ( eno ) scheme which solves the ideal mhd equations in a periodic box : @xmath21 = { \bf f } , \\ \frac{\partial { \bf b}}{\partial t } - \nabla \times ( { \bf v } \times{\bf b } ) = 0,\end{aligned}\ ] ] with zero - divergence condition @xmath22 , and an isothermal equation of state @xmath23 , where @xmath24 is the gas pressure . on the right - hand side , the source term @xmath25 is a random large - scale driving force ] . boundary conditions are periodic . we drive turbulence solenoidally in fourier space at wave scale k equal to about 2.5 ( 2.5 times smaller than l , the size of the box ) . this defines the injection scale in our models and the driving is done in fourier space to minimize the influence of the driving force on the generation of density structures . the initial density and velocity fields are set to unity . we do not set the viscosity and diffusion explicitly in our models . the scale at which dissipation starts to act is defined by the numerical diffusivity of the scheme . the eno - type schemes are considered to be relatively low diffusion ( see liu & osher 1998 ; levy , puppo , & russo 1999 ) . the numerical diffusion depends not only on the adopted numerical scheme but also on the smoothness of the solution , so it changes locally in the system . in addition , it is also a time - varying quantity . all these problems make its estimation very difficult and incomparable between different applications . however , the dissipation scales can be estimated approximately from the velocity spectra . in the case of our models we estimate the dissipation scale at @xmath26 pixels . lllllc 1 & 0.1 & 10.0 & 2.0 & supersonic & super - alfvnic + 2 & 1.0 & 10.0 & 0.7 & supersonic & sub - alfvnic + 3 & 0.1 & 7.0 & 2.0 & supersonic & super - alfvnic + 4 & 1.0 & 7.0 & 0.7 & supersonic & sub - alfvnic + 5 & 0.1 & 6.0 & 2.0 & supersonic & super - alfvnic + 6 & 1.0 & 6.0 & 0.7 & supersonic & sub - alfvnic + 7 & 0.1 & 4.0 & 2.0 & supersonic & super - alfvnic + 8 & 1.0 & 4.0 & 0.7 & supersonic & sub - alfvnic + 9 & 0.1 & 3.0 & 2.0 & supersonic & super - alfvnic + 10 & 1.0 & 3.0 & 0.7 & supersonic & sub - alfvnic + 11 & 0.1 & 0.7 & 2.0 & subsonic & super - alfvnic + 12 & 1.0 & 0.7 & 0.7 & subsonic & sub - alfvnic + 13 & 0.1 & 0.1 & 2.0 & subsonic & super - alfvnic + 14 & 1.0 & 0.1 & 0.7 & subsonic & sub - alfvnic + [ tab : table1 ] as density fluctuations are generated by the interaction of mhd waves , the time @xmath27 is in units of the large eddy turnover time ( @xmath28 ) and the length in units of @xmath29 , the energy injection scale . the magnetic field consists of the uniform background field and a fluctuating field : @xmath30 . initially @xmath31 . we divided our models into two groups corresponding to sub - alfvnic ( @xmath32 ) and super - alfvnic ( @xmath33 ) turbulence . for each group we compute several models with different values of gas pressure ( see table [ tab : table1 ] ) . + the top of figure [ fig:3dpic ] presents a 3d density rendering of simulation # 2 ( table [ tab : table1 ] ) where @xmath34=10 and @xmath35=0.7 . the x and y axes are labeled on the figure with the mean magnetic field ( @xmath36*b*@xmath37 ) parallel to the x direction . the bottom displays , for the same simulation , a column density ( left ) and the same column density convolved with a radially decreasing gaussian function in order to create the effect of cloud - like boundaries . see section [ observ ] for descriptions of column density construction . for the first portion of our analysis we investigate tsallis fits of pdfs of 3d density and the three directional components of magnetic field and velocity . we fit the tsallis distribution to incremental pdfs ( see section [ tsallis ] ) using the levenberg - marquardt algorithm ( levenberg 1944 ; marquardt 1963 ) , for spatial separations ( lag ) 1 , 2 , 4 , 8 , 16 , 32 , 64 , and 128 pixels ( up to 1.5 times smaller then the injection scale ) . fits and pdfs are shown in figure [ fig:3dhistsall ] ( symbols are the data from the simulations , lines represent the tsallis fit ) . increasing lags are displayed vertically on the same logarithmic vertical scale . we only present magnetic and velocity fields directed along the mean magnetic field . el10 saw no strong variation with los orientation for magnetic and velocity fields , which we confirm . pdfs of perpendicular components are therefore omitted from the figure . figure [ fig:3dhistsall ] presents pdfs ( and their corresponding tsallis fits ) of 3d density and components of velocity and magnetic field parallel to the mean magnetic field ( red , green and blue lines respectively ) . panels show four different simulations @xmath34=10 @xmath35=0.7 , @xmath34=10 @xmath35=2.0 , @xmath34=0.7 @xmath35=0.7 , and @xmath34=0.7 @xmath35=2.0 ( left to right ) . visual analysis of these figures shows tightly correlated fits with only slight deviation near the tails of the pdfs ( y - axis is logarithmic ) . three outstanding trends can be seen across the simulations . first , is the increase in gaussianity as @xmath34 decreases ( turbulence becomes subsonic ) . for supersonic turbulence , density pdfs are highly kurtotic . this corresponds to a higher probability that @xmath38 due to shock filaments causing central spikes of in pdfs increasing their kurtosis ( agrees with trends seen in el10 and burlaga & vias 2004b ) . the second trend is the increase in gaussianity as the lag increases . at low lags density , magnetic field , and velocity have higher probabilities of being near the mean value ( here normalized to zero ) . this type of behavior is congruent with the results of falgarone et al.(1994 ) which analyzed the skewness and kurtosis of pdfs of varying scales . in the case of subsonic turbulence , the pdfs of density look very similar to pdfs of velocity and magnetic field for high lags . third , there is an increase in pdf kurtosis for density , velocity , and magnetic field for simulations of a high magnetic field ( see the first and second panel of figure [ fig:3dhistsall ] for small lags ) . magnetization plays an intimate role in the development of turbulence and density enhancements and this affect can be attributed to field freeze - in . in the following subsections we will further describe the 3d quantities and their fits individually . from top to bottom , figure [ fig:3dfinal ] displays fit parameters @xmath5 , @xmath6 , and @xmath7 ( plotted as @xmath39 ) versus the spatial lag for 3d density for all 14 simulations . the figure is separated on the left and right corresponding to sub- and super - alfvnic mach number simulations respectively . the top panels display parameter @xmath5 ( corresponding the amplitude ) and shows a strong sensitivity to the degree of sonic number . to emphasize this fact we break our simulations up into three categories ; highly supersonic ( @xmath34=10 , 7 , 6 ) , mildly supersonic ( @xmath34=4 , 3 ) , and subsonic ( @xmath34=0.7 , 0.1 ) . a solid line is plotted through each subgroup s average value at each lag . attention to the difference in scale between the right and left panel shows a sensitivity to @xmath35 as well . errors in the fit parameters are calculated from the standard deviation about each subgroup s mean ( i.e. highly supersonic , mildly supersonic , subsonic ) for each lag . error bars are omitted from figure [ fig:3dfinal ] for clarity but are displayed in figure [ fig:3d4 ] for @xmath39 . the @xmath5 parameter has a percent error generally @xmath36 25@xmath40 but reaches a maximum of 56@xmath40 for simulation @xmath34=3.0 @xmath35=2.0 ( square symbol ) at a lag of 1 pixel . the deviation from the mean mildly supersonic value is not large compared to the supersonic group but the lower numerical values produce a higher percent error . this error drops off significantly with increasing lag . this is to be expected for simulations since at low lags we are in the range of numerical dissipation . super - alfvnic errors all fall below 20@xmath40 . in the middle panels , @xmath6 ( related to the pdf s tails ) shows a slight sensitivity to the simulations compressibility and no magnetic sensitivity . for @xmath6 we break the simulations up into only supersonic ( @xmath34=10 , 7 , 6 , 4 , 3 ) and subsonic ( @xmath34=0.7 , 0.1 ) and plot a solid line through their average values . the minimal variation in @xmath6 between simulations results in low errors that are below a 20@xmath40 for each lag . parameter @xmath7 , or width , displayed in the bottom panels of figure [ fig:3dfinal ] is presented as the fit value @xmath39 as it appears in equation ( 1 ) for convenience and clarity of representation . as with @xmath5 , we plot solid lines of the average of highly super- , mildly super- , and subsonic simulations showing the same sensitivity to @xmath34 although to a lesser degree . both subsonic simulations lie along the lag ( x ) axis . the most outstanding trend seen in @xmath7 is its sensitivity to @xmath35 . taking note of the large difference in scales between sub- and super - alfvnic panels shows elevated values for simulations with high magnetic fields . figure [ fig:3d4 ] summarizes this sensitivity displaying the sub- and super - alfvnic versions of two highly supersonic ( @xmath34=10 , 7 ) , one mildly supersonic ( @xmath34=4 ) , and one subsonic ( @xmath34=0.7 ) simulation . for each @xmath41 , the sub - alfvnic ( @xmath42=1.0 ) simulation produces a @xmath39 value @xmath43 2 times its super - alfvnic counter part at lag = 1 pixel . errors for @xmath39 are the highest of the three parameters due to it large deviation between simulations . generally the errors are @xmath36 40@xmath40 but peak at 160@xmath40 for the @xmath34=3.0 @xmath35=2.0 simulation ( filled circle ) at 1 pixel lag . this is mainly due to its @xmath36 1 value . even with these significant errors , the differences between sub- and super - alfnic turbulence is apparent . aside from the values of @xmath5 , @xmath6 , and @xmath7 , the general shape of all three shows trends toward @xmath34 . both @xmath5 and @xmath39 show relatively consistent values over lag for subsonic simulations while supersonic simulations show steep decreasing slopes . the analysis of velocity is more directly related to investigations of turbulence than density and is particularly important in regards to solar wind measurements . tsallis provides excellent fits to our velocity pdfs and also shows dependencies on mach numbers , although the tsallis fit parameters show decreased sensitivities to @xmath44 and @xmath35 compared to the density analysis . the top panels of figure [ fig : bvx ] displays @xmath39 for sub- and super - alfvnic simulations on the left and right respectively for the component of velocity along the mean magnetic field ( x - direction ) . solid lines represent the average values of highly super- , mildly super- , and sub sonic simulations at each lag ( for top and bottom panels ) . a small sensitivity to @xmath34 can be seen with increased values for more turbulent simulations . the sensitivity to @xmath35 is most predominant parallel to @xmath36*b*@xmath37 ( which is the los shown in this figure ) where there is a near factor of 2 increase in @xmath39 . perpendicular to @xmath36*b*@xmath37 , the increase is @xmath45 . for sub - alfvnic simulations ( @xmath42=1 ) a @xmath4620% increase in @xmath39 is seen for velocity along @xmath36*b*@xmath37 while no preferred direction is seen for super - alfvnic simulations . we do not show figures for @xmath5 and @xmath6 as they displayed less sensitivity to mach numbers then did @xmath39 . generally for velocity , the @xmath5 parameter displays a very small sensitivity to compressibility and the values for all simulations span a limited range . compared to 3d density , velocity exhibits a two order of magnitude decrease in standard deviation for small lags . a slight increase in @xmath5 can be seen for velocity of highly magnetized simulations along the mean magnetic field ( x direction ) but no similar relationship is seen for velocity perpendicular to @xmath36*b*@xmath37 . fit parameter @xmath6 is even less descriptive , with no significant mach number dependencies . @xmath39 is the most sensitive to @xmath47 and @xmath35 maintaining the same trends seen in density but on a fraction of the scale . analysis of directional magnetic field strength is shown in the bottom of figure [ fig : bvx ] for @xmath39 . tsallis fits the pdfs of magnetic field well . parameter @xmath39 displays a sensitivity to the @xmath44 number of simulations of a given @xmath35 as seen in the bottom panels of figure [ fig : bvx ] for the magnetic field parallel to @xmath36*b*@xmath37 . in the presented figures @xmath39 has a similar value for high and low @xmath35 but along other lines of sight the relationship between @xmath39 values and increased magnetization generally does not hold . however , this may be useful for determining mean field direction in the ism . similar to velocity , we find that @xmath5 and @xmath6 are not as useful as @xmath7 in terms of describing mach numbers with component velocity and magnetic field , and hence we omit the figures . parameter @xmath5 shows a very small sensitivity to compressibility while @xmath6 shows no significant variation for any simulation . errors analysis for the fit parameters of velocity and magnetic field are carried out in the same manner as density ( calculating the standard deviation of each @xmath34 subgroup at each lag ) . velocity along the mean magnetic field has fit errors consistently @xmath36 18@xmath40 for all lags and simulations . magnetic field along the same los has similarly low errors ( @xmath36 20@xmath40 ) with the exclusion of the sub - alfvnic ( @xmath42=1 ) @xmath34=4 & 3 simulations ( triangles and squares respectively ) which reach 43@xmath40 and 109@xmath40 error respectively . one should keep in mind , however , that these measurements are done for one los as there are significant variations in the trends of velocity and magnetic field depending on the orientation . the behavior of tsallis fits to pdf increments of density , velocity , and magnetic field is in agreement with results found in el10 at higher resolution with a larger parameter range . our analysis of spatial variations provides insight into the underlying physics of mhd turbulence , where as the analysis of temporal variations of velocity and magnetic field in solar wind observations well described the multiphase structure of this phenomenon ( burlaga & vias 2004a , 2004b , 2005a , 2005b , 2006 ; burlaga , ness , & acuas 2006 , 2007 , 2009 ; burlaga , vias , & wang 2007 ) . in the latter a time scale , @xmath48 , ( opposed to spatial scale @xmath12 ) is used to describe incremental fluctuations by @xmath49 . in order to characterize the quality of our fits and reliability of results , we calculate the pearson s @xmath50 and coefficient of determination , @xmath51 , for our tsallis pdf fits presented in equations ( 5 ) and ( 6 ) below . @xmath52 in these equations , @xmath53 represents the observed pdf and @xmath54 is the tsallis fit . each measurement describes the accuracy of the fit . @xmath50 approaches zero for perfect fits while @xmath51 approaches one . in general , the tsallis function will fit pdfs of density , column density , magnetic field , and velocity with a @[email protected] and a @[email protected] resulting in tightly correlated fits for the 100 degrees of freedom ( number of bins in pdf histogram ) . figure [ fig : chir ] displays the @xmath50 and @xmath51 for the sub- and super - alfvnic counterparts of our most turbulent ( top ) and least turbulent 3d density simulations ( bottom ) . from these plots we find that the best fits are seen for super - alfvnic , subsonic simulations for lags greater than 16 pixels . due to the kurtotic nature of high turbulence , low lag pdfs are difficult for tsallis to fit at the central peak ( see figure [ fig:3dhistsall ] , right panel , bottom ) . in these extreme cases ( @xmath34=10 , @xmath35=0.7 ) values of @xmath50=0.67 and @xmath51=0.61 are obtained . the low lag regimes are in the dissipation range of the turbulence , and tsallis seems to have a difficult time fitting turbulence on scales sampling either the dissipation range or the injection scale . despite these difficulties at low ( or high ) lag , tsallis still can prove itself useful for characterizing turbulence . for example , if one is in the dissipation range of turbulence ( i.e. at scales similar to our lags less then 32 pixels ) , then these `` goodness of fit tests '' can also be used to determine what type of turbulence is present . subsonic and super - alfvnic turbulence both are better fit with the tsallis distribution at low lags then supersonic or sub - alfvnic turbulence . at higher lags the fit quality converges as we enter the inertial range of turbulence . these fit tests should not only be preformed to test fit quality but could also be used as an additional test of the mach number range in a given data set . an alternative source of potential error in our analysis is the quality of pdfs at large lags . not only are these lags on the upward scale of the inertial range , but also are subjected to degraded resolution . these resolution effects can be seen particularly in the case of the column density , which are 2d quantities discussed in the next section . the inspection of figure [ fig : cdxall ] shows fits becoming less tight at large lags . this manifests as `` jumps '' in the fits parameters which can be seen in the subsonic case in figure [ fig : cdx4 ] . this is the most extreme example of this trend but it is seen to some degree for every simulation at a large enough lag . in these cases the @xmath50 and @xmath51 remain constrained to the values stated above . we conclude that , for this analysis , our results are most stable for lags in the inertial range of turbulence and for lags at least 8 times smaller then the box size to provide enough sampling . this trend is consistent with the results seen using higher order moments to analyze the incremental pdfs . in section [ 3d ] we confirmed the results of el10 using higher resolution simulations and a substantially larger parameter range . in addition , we demonstrate the sensitivity the @xmath7 ( width ) fit parameter has toward the alfvnic mach number . however , observations of the ism do not provide direct 3d information of density . combining density and velocity along the line - of - sight ( los ) provides a 3d position - position - velocity ( ppv ) cube , however these types of data cubes can almost never be reliably interpreted as having a one - to - one correspondence with an actual 3d volume density . considering the difficulties of obtaining direct 3d ism information , a study of tsallis statistics on observables such as column density and ppv data is necessary . we create synthetic 2d column density maps perpendicular and parallel to the mean magnetic field for all 14 of our simulations by averaging the 3d density cubes along a given line of sight . we assume the emitting gas is optically thin and that the emissivity is linearly proportional to density ( such as in the case of hi ) . an example is presented in figure [ fig:3dpic ] ( bottom left ) . using the same method described in the section [ tsallis ] , pdfs are created using spatial lags ( increments ) in _ two _ directions which are then fit using the tsallis function . figure [ fig : cdxhists1 ] displays the distributions and fits for 4 simulated column densities created parallel to the mean magnetic field . the 4 simulations we show here are divided into panels : @xmath34=10 @xmath35=0.7 , @xmath34=10 @xmath35=2.0 , @xmath34=0.7 @xmath35=0.7 , and @xmath34=0.7 @xmath35=2.0 ( left to right ) . this is analogous to the arrangement in figure [ fig:3dhistsall ] . while there is an overall decrease in kurtosis of the column density pdfs compared to 3d density , the same three trends are still apparent . pdfs become more gaussian with decreased @xmath34 , gaussianity increases with lag for supersonic cases , and gaussianity increases with @xmath35 . a trend not seen in the 3d case is that subsonic pdfs become less smooth at high lags . this is due to the decrease in resolution going from 3d to 2d . el10 also observed this trend with their subsonic pdfs . figure [ fig : cdxall ] displays the fit parameters @xmath5 , @xmath6 , and @xmath39 from top to bottom for all 14 column density simulations created along the x direction ( parallel to @xmath36*b*@xmath37 ) . each parameter is divided into two panels with sub - alfvnic and super - alfvnic simulations on the left and right respectively . the @xmath5 parameter ( top panels ) is presented with three over - plotted lines , averaging the super- , mildly super- , and subsonic simulations . again , a strong sensitivity to @xmath34 is seen both in the value and the shape of the fit over spatial scale . the sensitivity to @xmath57 is less prevalent than in 3d density . errors for column density fit parameters are calculated in the same manner as 3d density by measuring the standard deviation of each @xmath34 subgroup at each lag . the @xmath5 parameter achieves @xmath36 23@xmath40 errors for each lag and simulation reaching the lowest errors ( @xmath36 5@xmath40 ) at lags between 16 and 64 pixels . as in density , error bars are excluded from figure [ fig : cdxall ] for clarity but displayed in figure [ fig : cdx4 ] for @xmath39 . parameter @xmath6 ( middle panels ) is presented with averaging super- and subsonic lines showing a slight sensitivity to @xmath34 and no coherent sensitivity to @xmath57 . errors for @xmath6 are generally less than 30@xmath40 but begin to reach percent errors greater than 100@xmath40 for subsonic simulations at lags greater than 34 pixels due to random variations . inspection of the right panel of figure [ fig : cdxhists1 ] shows that this is consistent with when the pdf start to become nonuniform . the @xmath7 parameter ( bottom panels ) , here plotted as @xmath39 , shows both a strong sensitivity to @xmath34 and @xmath35 . highly super- , mildly super- , and subsonic averaging lines are over - plotted which emphasize that values of @xmath39 are affected by @xmath41 . errors for @xmath39 are @xmath36 25@xmath40 for lags of 2 to 32 pixels . maximum errors reach values of 70@xmath40 at a 1 pixel lag and 60@xmath40 error at lags greater than 64 pixels . interestingly , as in el10 , subsonic column densities shows a more random behavior at larger lags due to low resolution and low density contrasts . figure [ fig : cdx4 ] summarizes @xmath7 s alfvnic sensitivity by displaying the sub- and super - alfvnic versions of two highly supersonic ( @xmath34=10 , 7 ) , one mildly supersonic ( @xmath34=4 ) , and one subsonic ( @xmath34=0.7 ) simulation . in each case , the simulation having the higher magnetic field results in a higher @xmath39 value . although the degree to which @xmath39 is elevated is less than for 3d density , a strong direct correlation remains . subsonic trends become less smooth at lag=32 pixels , which is @xmath58 6.4 times smaller then the injection scale . we may conclude that the column density has a stronger dependency on lag then in the 3d cases , especially for subsonic turbulence . we analyze fit parameters from column densities created along the y and z axes ( i.e. perpendicular to the mean magnetic field ) and while there were slight variations in the fits , sensitivities to alfvn and sonic mach numbers were still observed . no discernible trend to detect magnetic los orientation was seen . ( width ) for the sub- and super - alfvnic versions of the @xmath34=7 , 4 , and 0.7 simulations plotted against the smoothing degree in fwhm size . high and low magnetization are presented in red and blue respectively . supersonic simulations are more dramatically affected than subsonic as turbulent density enhancements are smoothed out . ] while the analysis simulated column density maps shows the strength of tsallis statistics in describing turbulent characteristics , analogous observational column density data do not have pencil thin beam resolution . to address this issue , we experiment with smoothing our column densities to different degrees to determine the role degradation of resolution plays in distributions and fit sensitivities . using a gaussian smoothing kernel , we degrade our column density simulations with a full - width - half - maximum ( fwhm ) of 3 , 5 , and 10 pixels . in an effort to characterize the effects of smoothing we increase our lag resolution by creating the same number of distributions over a smaller range of lags . figure [ fig : lag1 ] presents the the original and smoothed fit parameter @xmath39 for sub- and super - alfvnic simulations versions of simulations @xmath41=10 and 0.7 from top left to bottom right respectively . while we investigated the effects of smoothing on all three fit parameters , we only present @xmath39 as the effects on @xmath6 and @xmath5 are similar . in order to characterize how the slope of the tsallis parameters vs. lag change with increasing smoothing a least squares linear fit is applied to each scenario with the slopes displayed in the respective legends . the red stars denote the non - smoothed case for comparison . from this figure it is clear that increased smoothing lowers the values of @xmath39 for each lag and its overall slope . these effects are present across all simulations . while the supersonic ( top row ) simulations show smooth monotonic trends for both smoothed and pencil beam data , subsonic simulations ( bottom row ) shows very bumpy trends for the case where no smoothing is introduced ( red stars and line ) especially as lag increases . however , as we introduce smoothing , the trends become highly monotonic , even for the case of fwhm=3 pixels . applying tsallis fits to incremental pdfs while varying the level of smoothing will act as an additional tool to characterize turbulent parameters through the overall change in slope . figure [ fig : mvsmooth ] provides a summary of the effect smoothing has on the slope of @xmath39 by plotting it against the smoothing degree . simulations of the same sonic number are shown with the same symbol while the color corresponds to the previously used color scheme ( red being @xmath42=1 , sub - alfvnic and blue being @xmath42=0.1 , super - alfvnic ) . we see that increased smoothing flattens the @xmath39-lag slope . a sensitivity to the @xmath35 is still present in both slope and value in every smoothing cases ( note scale between sub- and super - alfvnic simulations in figure [ fig : lag1 ] ) . high sonic mach number simulations show slopes that are the most altered as smoothing is increased . this is due to shock density enhancements being smoothed out . however , the supersonic cases show larger variation with differing @xmath35 then their subsonic counterparts . this leads us to believe that one can more easily tell the magnetization strength of supersonic gas from tsallis fits . increasing the smoothing does not affect the subsonic cases , as these simulations already have gaussian distributions and have no peak density enhancements from shocks to smooth out . parameter @xmath5 was affected by smoothing in the exact same way @xmath39 . values and slopes decreased with increased smoothing . plots are excluded for brevity but both @xmath5 and @xmath7 could be explored with increasing smoothing to put estimates on turbulence . the @xmath6 parameter saw little deviations due to smoothing and a slightly weakened @xmath34 sensitivity . fit parameter errors for our smoothing discussing can be considered to decrease with smoothing from their unsmoothed values ( quoted in the previous section ) as differences in @xmath34 sub groups of simulations become washed out in parameter space . while analysis using tsallis statistics is very adept at providing turbulence parameters when smoothing is applied to column density maps , more pressing maybe the issue of noise . one may not expect noise to affect results at larger lags however , noise may make it challenging to distinguish trends on smaller scales where the difference between our simulations is prominent . it is our intent in this subsection to test the effectiveness of this statistical tool with the addition of noise , determine our confidence range between trends with noise and with no noise , and explore which lags are most affected by the addition of noise . in order to achieve this end we add random gaussian noise to our column density maps . we do this by setting a given average signal - to - noise ( snr ) ratio and scaling the power of the noise and signal to match this snr . we look at column density integrated parallel to the mean magnetic field with average snr of 400 , 200 , 100 , 50 , 20 and unity . the addition of noise on the column density pdfs has a general smoothing effect toward gaussianity ( pdfs not shown ) . this is to be expected as the noise added to the column density is gaussian white noise . figures [ fig : noisea ] and [ fig : noisew ] illustrate the effects of noise for the parameters @xmath5 and @xmath39 respectively for four simulations with given sonic mach numbers of @xmath34= 7.0 and 0.7 and alfvnic number @xmath35= 2.0 and 0.7 . from these figures it is clear that the smaller spacial lags ( lags = 1 - 32 pixels ) are highly affected by the addition of noise . the noise injected simulations ( lines with colors and symbols ) do not converge to their no - noise counter part ( shown as a solid black line ) until lag @xmath4620 pixels . at around lag 20 , simulations with snr greater than 50 show similar shaped curves to the no - noise case . as the snr decreases , the values of @xmath7 and @xmath5 both decrease for the supersonic case . the subsonic case shows a more pronounced non - monotonic behavior . in all cases above snr of 20 , the sub - alfvnic supersonic turbulence shows heightened values of @xmath5 and @xmath39 . the sonic number between these two simulations can be determined within 1.5 @xmath59 confidence at snr 20 and above . at snr of unity it is impossible to distinguish between either sonic or alfvnic numbers , which is to be expected . a snr greater than 100 is sufficient for showing differences in @xmath39 and @xmath5 between sub- and super - alfvnic simulations , with sub - alfvnic producing larger values . in the case of supersonic turbulence a snr of 20 is generally sufficient to discern between sub- and super - alfvnic gas using tsallis fit parameters with 1.5@xmath59 confidence for all lags . while this is promising that all pixel scales show statistically significant differences , the shape of the @xmath39 or @xmath5 curve vs. lag converges to non - noise levels only for lag @xmath4620 pixels and therefore , small lags may not be as reliable as larger lags in the presence of low snr . subsonic cases can distinguish alfvnic number with confident in the 1 @xmath59 range at snr 20 and greater for lags larger then 32 pixels . studies of the ism frequently deal with both clouds - like objects as well as diffuse gas . while our simulations are most directly applicable to diffuse ism , we can also study the effects the tsallis fits will have on ism gas with boundaries . an example of such a situation would be molecular clouds which have characteristic radially decreasing density values . to mimic this , we convolve our 3d density simulations with a spherical function which maintains the value inside a given radius r from the center and creates a gaussian decay outside this radius . for our simulations we choose an r of 205 , 90 , and 10 pixels corresponding to 2.5 , 6 , and 51 times smaller than box size respectively . the cloud convolved 3d simulations are then averaged along a los creating a column density map as described in section [ coldn ] . for reference a column density map with a radial decreasing gaussian boundary starting at r=90 parallel to the mean magnetic field is presented in figure [ fig:3dpic ] ( bottom right ) for simulation # 2 ( @xmath34=10 , @xmath35=0.7 ) . in general , a radially decreasing cloud boundary creates an increase in the kurtosis of the pdfs as compared with unbounded column density ( not shown ) . this change comes from the increase in zero point values of @xmath60 as more contrast is created . at a radius of 205 pixels the pdfs are still fairly uniform and only see slight variations from gaussianity for subsonic simulations and fits remain similar to those seen in figure [ fig : cdxhists1 ] . figure [ fig : cloudall ] presents the @xmath5 ( top ) , @xmath6 ( middle ) and @xmath39 ( bottom ) fit parameters for the r=205 bounded simulations parallel to the mean magnetic field . as with 3d density and unbounded column density , sub- and super - alfvnic simulations are presented on the left and right panels respectively . three averaging lines are plotted through the highly super- , mildly super- , and subsonic simulations for @xmath5 and @xmath7 , while @xmath6 has only super- and subsonic averaging lines . sensitivities to @xmath41 are still apparent throughout the @xmath5 and @xmath39 fits values and slopes but emphasized more in sub - alfvnic simulations . @xmath35 remains strongly tied to @xmath7 , as presented in figure [ fig : cloud4 ] . in each scenario , the simulation containing a stronger magnetic field results in a @xmath432 times higher value of @xmath39 . the @xmath6 parameter shows little effect from a radially decreasing boundary condition and still is only mildly affected by simulation compressibility . fit parameter errors of cloud bounded column densities shows similar trends to those seen in full resolution 2d density maps . parameter @xmath5 maintains the lowest error from lag 2 to 32 pixels where errors remain @xmath36 16@xmath40 . outside this range , errors increase and reach a maximum error of 46@xmath40 at the 1 pixel lag . @xmath6 achieves errors below 36@xmath40 for most simulation lags excluding subsonic simulations and large lags . in this regime error reaches 90@xmath40 . errors of the @xmath39 parameter display similar trend as @xmath6 where errors below 50@xmath40 are produced for mid range lags ( 4 - 32 pixels ) and highly and mildly supersonic simulations . errors reach @xmath37 150@xmath40 at small lags and subsonic simulations at high lags . this further confirms the trend that decreasing resolution has on the quality of fits and fit parameter values . even as errors increase , the @xmath39 parameter is still able to distinguish @xmath35 within the error bars . similar to the incremental column density pdfs , our largest clouds proved well described by tsallis statistics . however , the decrease in cloud size greatly affects the pdfs and the ability of the tsallis function to fit them . for example , in the case of our smallest cloud ( r=50 pixels ) , the pdf and fit for simulation # 2 ( @xmath41=10 , @xmath35=0.7 ) at even our largest lag ( 128 pixels , not shown ) , is extremely kurtotic and skewed to negative values . this is due to radially decreasing density , the value of @xmath61 is in general @xmath37 0 for three out of the four directional calculations ( + x , -x , + y , -y ) . tsallis is a symmetric function about @xmath60 and this skewness of the pdf prevents tight fits . while quality of the fit is no longer intact , the relative amplitudes and widths change with lag , @xmath34 , and @xmath35 . cloud convolved column densities with los perpendicular to the mean magnetic field were also analyzed and showed the same trends and no discernible correlation with the los orientation . the application of tsallis statistics has proven itself to be a useful tool for exploring a wide parameter range of sonic and alfvnic mach numbers in density and column density data . however , while density fluctuations are indeed useful for characterizing turbulence , more appropriate is the use of velocity information . we test the usefulness of tsallis pdf fits on position - position - velocity ( ppv ) data in order to see if additional information is provided by including the velocity axis . we create synthetic ppv cubes of our full simulation range using density and velocity with los perpendicular to mean magnetic field . in order to test a variety of velocity resolutions we create ppv cubes with velocity widths of 0.15 ( @xmath4630 channels ) , 0.07 ( @xmath4660 channels ) , and 0.007 ( @xmath46600 channels ) . units remain scale free . figure [ fig : ppv ] presents @xmath39 vs. lag for the sub- and super - alfvnic versions of the @xmath34=10 , 4 , and 0.7 simulations from top to bottom respectively . each panel has the three velocity resolutions over - plotted for comparison . low , middle , and high resolutions are represented by the circle , diamond , and triangle respectively . within each resolution , red symbols correspond to sub - alfvnic ( @xmath42=1.0 ) while blue corresponds to super - alfvnic ( @xmath42=0.1 ) . all three panels display a common trend of decreasing @xmath39 value as the velocity resolution increases ( width of the velocity channels decreases ) . also , it is clear that the @xmath7 parameter is sensitive to the @xmath35 number for ppv simulations and produces elevated values for simulations with high magnetization . the general shape of @xmath39 over increasing lags is distinctly different for ppv data than in density , velocity , or magnetic field . the pdfs , and therefore fit parameters are more consistent with lag variations and produce smoother trends . this alternative trend is only seen in supersonic ppv data while values of @xmath39 for subsonic ppv data more closely resemble the trends seen in density analysis . parameter @xmath5 shows minimal variations to @xmath35 or @xmath34 for ppv simulations . sensitivities to @xmath34 are seen on a scale too small to be used observationally and there is no coherent relationship to @xmath35 . similar to previous discussions @xmath6 only provides limited information on compressibility and none on @xmath35 . it should be noted that during the distribution creation process , especially for high velocity resolutions , velocity channels with little emission were excluded . in addition , a majority of the small lag pdfs had a single multiple order of magnitude jump at the center of the distribution ( where @xmath60 ) . this point was excluded from the fitting process . this is simply due to the way the emission is spread over the channels in ppv with a maxwellian velocity distribution . most gas is near the mean velocity and a gaussian decrease in emission is seen as one goes to the extreme minimum and maximum channel velocities . analysis of density distributions along the velocity axis for individual pixels were also analyzed . an attempt to fit these distributions with the tsallis function proved ineffective . we find the distribution of density emission along velocity channels , especially for high turbulence , are highly skewed and irregular which is consistent with the findings of falgarone et al . as tsallis is a symmetric function it is not able to provide tight fits to these highly irregular and varying distributions . we studied tsallis statistics as a way to characterize mhd ism turbulence . while we do not address the theoretical justification of the particular form of the fits predicted by the equation ( 1 ) , our study shows that this fit provides a very good correspondence with experimentally obtained distributions . we showed that the parameters that enter the tsallis expression are dependent on the sonic mach ( @xmath34 ) and alfvnic mach ( @xmath35 ) numbers , confirming and extending the earlier claims of esquivel & lazarian ( 2010 ) . in addition , we explored the differences in the dependencies of the fitting parameters on both @xmath35 , @xmath34 and on the resolution . this allows one to find @xmath34 and @xmath35 independently . we also applied tsallis statistics to ppv data , which , as far as we know , is the first time such a study has been undertaken . another important point that we address here is the influence of telescope resolution in the form of smoothing . this issue was largely omitted in our earlier publications , but our results show that the influence of the telescope beam may be important for observational studies . not only does smoothing affect the statistics , but it could actually be a helpful addition to the statistical procedure to study mach numbers . indeed , telescope measurements introduce their own smoothing , which depends on the telescope resolution . as the injection scale of the turbulence is frequently not known , the given measurements of the parameters of the tsallis statistics may not be straightforward to identify with the particular simulation . in this situation , additional smoothing can help to see to what parameters of turbulence the observed data corresponds to . burkhart et al . ( 2010 ) studied the small magellanic cloud ( smc ) to characterize the properties of turbulence in its individual parts . the same approach is applicable with tsallis statistics . it is convenient to note , that one can gain information on regions turbulence or magnetization without knowing the precise resolution of the telescope . our motivation for applying boundaries to our simulations was fueled by making tsallis applicable to more compact objects , such as those that may be found in molecular clouds . our results show clear tendencies to determine mach number information from the tsallis fitting parameters . in addition , our study shows that the use of the full ppv data can bring additional advantages . for instance , the evolution of tsallis parameters with is clearly different for subsonic and supersonic cases ( see figure [ fig : ppv ] ) in ppv data . we feel that the tsallis fit provides useful way of studying turbulence , which should be implemented along with other techniques used previously ( see kowal , lazarian , & beresnyak 2007 ; burkhart et al . 2009 , 2010 ) . we advocate an approach which combines the use of different techniques . for instance , the effect of the telescope beam smoothing depends on the ratio of the angular size of the turbulence energy scale in the object under study and the telescope resolution . the former can be obtained by studying the power spectra of turbulence . while we are still searching for the basic approaches to study turbulence observationally , the successes in theoretical and numerical studies provide us with a reliable guidance for obtaining statistics of turbulence from observations . the development of the vca and vcs techniques ( see lazarian & pogosyan 2000 , 2004 , 2006 , 2008 , chepurnov & lazarian 2008 ) for instance , quantified how velocity and density create structures observed in the ppv . results on the anisotropies of observational data statistics in the plane of the sky in lazarian , pogosyan & esquivel ( 2002 ) and subsequent publications ( esquivel & lazarian 2005 ) reveals strong dependencies on media magnetization . kurtosis and skewness measures of the density ( kowal , lazarian , & beresnyak 2007 ; burkhart et al . 2009 ) provide good insight mostly into the intensity of turbulence , i.e. its @xmath34 . genus ( see lazarian 1999 , kim & park 2007 , chepurnov et al . 2008 ) provides a measure of topology , while bispectrum ( see burkhart et al . 2009 ) provides insight into mode correlation and phase information of the turbulence . additional information can be obtained by employing other tools , e.g. dendrograms ( rosolowsky et al . 2008 , goodman et al . 2009 , burkhart et al . 2011 ) . at the same time the synergy of the simultaneous use of different techniques is still to be explored . in this paper we fit the tsallis distribution function to pdfs of turbulence with little or no theoretical justification . indeed , while the history of turbulence studies can be traced back to kolmogorov ( 1941 ) , turbulence studies have since taken many different directions . in this paper , we are particularly interested in characterizing magnetized turbulence in the ism . the tsallis distribution formalization comes from a different background , specifically the issue of quantities which are invariant in the navier - stokes equations at high reynolds numbers and the assumption that the singularities due to the invariance distribute themselves multifractally in physical space . much work has been done to address both the theoretical frame work and application of the tsallis statistics to numerical and laboratory turbulence pdfs ( see tsallis 1999 and arimitsu & arimitsu ( 2000a , 2000b , 2001 , 2002 , 2003 ) ) . for example , in arimitus & arimitus 2003 , the authors successfully fit the tsallis distribution to pdfs of turbulence in three different laboratory experiments . burlaga and collaborators successfully fit tsallis to solar wind data ( burlaga , vias , & wang 2009 ) . in this paper , we successfully fit tsallis to our numerical magnetized ism turbulence , even in the case of observable quantities such as column density and ppv emission cubes . in the case of the laboratory , space observations , and numerical experiments ; the geometry of the system , the reynolds numbers , and the type of turbulence in question , are all very different . never - the - less , in all cases the tsallis distribution provided good fits to the data . motivated by these past studies , we applied the tsallis function in an _ empirical _ way to determine the parameters of turbulence , and we direct the curious reader to the literature mentioned above for a more in depth treatment of the theoretical foundations of the tsallis function . also , our paper is motivated towards making tsallis statistics useful for observational studies of turbulence , and in this case the discussion of multifractal theory is beyond our current scope . we do not attempt to justify or even test these ideas , but investigate to what degree the fit provided by the two parameter tsallis formula corresponds to our numerical data . while there is no doubt that substantial progress has been made in last decade in numerical simulations of turbulence , the community still has a long way to go in terms of achieving the resolution required to fully capture ism physics . one issue in particular is the discrepancy between astrophysical values of reynolds ( re ) and magnetic reynolds ( rm ) and the numerical values . as a rule , astrophysical environments are turbulent and astrophysical turbulence is characterized by enormously large re and rm . for instance , the rm number , which characterizes the degree of frozenness of magnetic field within eddies , may differ in astrophysical environments and numerical simulations by a factor larger than @xmath62 . with these sorts of discrepancies in mind ( which exist for all numerical codes that simulate turbulence ) , how are we able to justify our results ? there are several avenues to justify the independence of our results from re and rm values . the particular numerical code used here provides enough resolution to allow us to see the inertial range of turbulence . if turbulence is evolving along the inertial range it is self - similar , and the change in re and rm does not change the structure of the large scale motions which we study here . indeed , according to the lazarian & vishniac ( 1999 ) model of turbulent reconnection ( tested in kowal et al . 2009 ) , magnetic reconnection in a turbulent media depends on the large scale field wondering determined by large scale motions . thus we do not expect the unresolved small scales to affect the physics of our simulations at large scales that we study . keeping all of this in mind , we also tested our tsallis fits at resolutions of @xmath2 and @xmath0 and found no deviation in how well tsallis was able to both fit the pdfs and describe the turbulence in question . in this paper we applied the tsallis formalism to simulated diffuse ism isothermal ideal mhd turbulence using incremental pdfs of density , velocity , and magnetic field . this method was also applied to simulated column densities with measures taken to duplicate observed measurements such as smoothing , noise , and radial decreasing cloud boundaries . the use of tsallis statistics on ppv data was also explored . a summary of our findings is as follows : * the tsallis function is capable of well describing incremental pdfs of a wide range of simulated mhd density , velocity , and magnetic field with its three fit parameters @xmath5 ( amplitude ) , @xmath7 ( width ) , and @xmath6 ( related to the pdf s tails ) . * for 3d density , fit parameters @xmath5 , @xmath6 , and @xmath39 are sensitive to @xmath34 . @xmath5 and @xmath39 show sensitivities to @xmath35 with both displaying greater values for sub - alfvnic simulations . * magnetic field and velocity show sensitivities to @xmath34 and @xmath35 but these are not as strong as density . * pdfs of column density are equally well described by the tsallis distribution . fit parameters @xmath5 , @xmath6 , and @xmath39 remain sensitive to @xmath34 . @xmath5 looses some of its magnetic sensitivity but @xmath39 consistently produces elevated values for sub - alfvnic simulations . * the affect of degrading resolution lowers both @xmath5 and @xmath39 vs. lag monotonically . supersonic simulations are most affected by smoothing . @xmath5 and @xmath39 remain sensitive to @xmath34 and @xmath35 for mild smoothing . @xmath6 looses @xmath34 sensitivity with smoothing . * the addition of smoothing at varying degrees acts as an additional method to constrain sonic mach number through the analysis of the slope of parameters @xmath5 or @xmath39 over lag . * noise has the largest affect on fit parameters . small scale variations strongly alter the pdfs of low lags and the shape of the fit distributions . sensitivities to @xmath35 and @xmath34 are still seen on smaller scales as the spatial lag increases . * radially deceasing cloud boundaries have little affect as long as a there are enough points inside providing good signal . as clouds become smaller , pdfs become more kurtotic and skewed preventing meaningful fits . * ppv data was fit with tsallis distribution excluding the zero point and parameter @xmath39 showed sensitivities to @xmath34 and @xmath35 . * tsallis statistics of incremental pdfs is a successful tool in describing a wide range of high resolution mhd simulations and their observational counter parts . along with its sensitivities to turbulence and magnetic fields , the tsallis distribution is highly complimentary to power spectrum and other ism statistical tools . lazarian , a. 1999 , plasma turbulence and energetic particles in astrophysics , proceedings of the international conference , cracow ( poland ) , 5 - 10 september 1999 , eds . : michal ostrowski , reinhard schlickeiser , obserwatorium astronomiczne , uniwersytet jagiellski , krakw 1999 , p. 28 , 28 authors wish to thank a. esquivel for use of the tsallis code and productive discussion and the anonymous referee for thoughtful comments . is supported by the nsf funded research experience for undergraduates ( reu ) program through nsf award ast-1004881 . b.b . acknowledges the nsf graduate research fellowship and the nasa wisconsin space grant institution . thanks both nsf ast 0808118 and the center for magnetic self - organization in astrophysical and laboratory plasmas .

we study stochastic particle systems that conserve the particle density and exhibit a condensation transition due to particle interactions . we restrict our analysis to spatially homogeneous systems on finite lattices with stationary product measures , which includes previously studied zero - range or misanthrope processes . all known examples of such condensing processes are non - monotone , i.e. the dynamics do not preserve a partial ordering of the state space and the canonical measures ( with a fixed number of particles ) are not monotonically ordered . for our main result we prove that condensing homogeneous particle systems with finite critical density are necessarily non - monotone . on finite lattices condensation can occur even when the critical density is infinite , in this case we give an example of a condensing process that numerical evidence suggests is monotone , and give a partial proof of its monotonicity .

1505.02049

we consider stochastic particle systems which are probabilistic models describing transport of a conserved quantity on discrete geometries or lattices . many well known examples are introduced in @xcite , including zero - range processes and exclusion processes , which are both special cases of the more general family of misanthrope processes introduced in @xcite . we focus on spatially homogeneous models with stationary product measures and without exclusion restriction , which can exhibit a condensation transition that has recently been studied intensively . a condensation transition occurs when the particle density exceeds a critical value and the system phase separates into a fluid phase and a condensate . the fluid phase is distributed according to the maximal invariant measure at the critical density , and the excess mass concentrates on a single lattice site , called the condensate . most results on condensation so far focus on zero - range or more general misanthrope processes in thermodynamic limits , where the lattice size and the number of particles diverge simultaneously . initial results are contained in @xcite , and for summaries of recent progress in the probability and theoretical physics literature see e.g. @xcite . condensation has also been shown to occur for processes on finite lattices in the limit of infinite density , where the tails of the single site marginals of the stationary product measures behave like a power law @xcite . in general , condensation results from a heavy tail of the maximal invariant measure @xcite , and so far most studies focus on power law and stretched exponential tails @xcite . as a first result , we generalize the work in @xcite and provide a characterization of condensation on finite lattices in terms of the class of sub - exponential tails that has been well studied in the probabilistic literature @xcite . this characterization holds for a particular definition of condensation given in section [ sub sec : ips ] , which was also used in @xcite . our main result is that all spatially homogeneous processes with stationary product measures that exhibit condensation on finite lattices with a finite critical density are necessarily non - monotone . monotone ( attractive ) particle systems preserve the partial order on the state space in time , which enables the use of powerful coupling techniques to derive rigorous results on large scale dynamic properties such as hydrodynamic limits ( see @xcite and references therein ) . these techniques have also been used to study the dynamics of condensation in attractive zero - range processes with spatially inhomogeneous rates @xcite , and more recently @xcite . as we discuss in appendix [ sec : statmech ] , non - monotonicity in homogeneous systems with finite critical density can be related , on a heuristic level , to convexity properties of the canonical entropy . for condensing systems with zero - range dynamics , it has been shown that this is related to the presence of metastable states , resulting in the non - monotone behaviour of the canonical stationary current / diffusivity @xcite . this corresponds to a first order correction of a hydrodynamic limit leading to an ill - posed equation with negative diffusivity in the case of reversible dynamics . heuristically , this is of course consistent with the concentration of mass in a small , vanishing volume fraction , but poses great technical difficulties to any rigorous proof of hydrodynamic limits for such particle systems . first results in this direction only hold for sub - critical systems under restrictive conditions @xcite , and due to a lack of monotonicity there are no results for non - reversible dynamics . condensing monotone particle systems would , therefore , provide interesting examples of homogeneous systems for which coupling techniques could be used to derive stronger results on hydrodynamic limits . however , our result implies that this is not possible for condensing models with stationary product measures and a finite critical density on finite lattices . in the thermodynamic limit condensation has been defined through the equivalence of ensembles , which can be established in generality for a class of long - tailed distributions with a finite critical density @xcite . this class has also been studied before @xcite and includes the class of sub - exponential distributions , for which our results apply also in the thermodynamic limit . a detailed discussion of their connections and the resulting differences between condensation on finite lattices and in the thermodynamic limit is given in sections [ sec : subexp ] and [ tlimit ] . we remark that for systems where the dynamics is directly defined on infinite lattices there are no rigorous results or characterizations of condensation to our knowledge , and we do not discuss this case here . for systems with infinite critical density condensation can still occur on finite lattices , and since non - monotonicity typically occurs above the critical density , such processes can also be monotone . when the tail of the stationary measure is a power law and decays faster than @xmath0 with the occupation number @xmath1 , we prove that the process is still non - monotone . in section [ examples ] we present preliminary results for tails that decay slower than @xmath0 , which strongly suggests that a monotone and condensing particle system exists ( see @xcite for further discussion ) . the paper is organised as follows . in section [ sec : notation ] we introduce the background used to study condensation and monotonicity in particle systems , and state our main results . in section [ sec : proof ] we prove our main theorem by induction over the size of the lattice , showing that the family of canonical stationary measures is necessarily not monotonically ordered in the number of particles . in section [ sec : char cond ] we discuss the differences between condensation on fixed lattices and in the thermodynamic limit , and prove equivalence of condensation on finite lattices with the tail of the maximal invariant product measure being sub - exponential . in section [ examples ] we review examples of homogeneous processes that have been shown to exhibit condensation , and present some explicit computations for misanthrope processes and processes with power law tails . we consider stochastic particle systems on fixed finite lattices @xmath2 , which are continuous - time markov chains on the countable state space @xmath3 . for a given configuration @xmath4 the local occupation , @xmath5 for @xmath6 , is a priori unbounded . the jump rates from configuration @xmath7 to @xmath8 are denoted @xmath9 , and the dynamics of the process is defined by the generator @xmath10 for all continuous functions @xmath11 . we assume the process conserves the total number of particles @xmath12 and conditioned on @xmath13 , the process is assumed to be irreducible , so that @xmath14 is the only conserved quantity . the process is therefore ergodic on the finite state space @xmath15 for all fixed @xmath16 . on @xmath17 the process has a unique stationary distribution @xmath18 , and the family @xmath19 is called the canonical ensemble . we focus on systems for which the stationary distributions are spatially homogeneous , the marginal distributions @xmath20 $ ] are identical for all @xmath21 . this typically results from translation invariant dynamics on translation invariant lattices with periodic boundary conditions , but the actual details of the dynamics are not needed for our results . for these systems we define condensation in terms of the maximum occupation number @xmath22 [ condensation ] a stochastic particle system with canonical measures @xmath18 on @xmath17 with @xmath23 exhibits * condensation * if @xmath24 = 1 \ . \label{eqn : cond1}\ ] ] this condition implies in particular the existence of all limits involved . the interpretation of ( [ eqn : cond1 ] ) is that due to the sub - exponential tails of the measure , in the limit @xmath25 all but finitely many particles concentrate in a single lattice site and that the distribution of particles outside the maximum is non - degenerate . as we will see in proposition [ prop1 ] below , the latter is in fact given by the maximal invariant measure when the system exhibits product stationary measures . there are of course other possible definitions of condensation which are less restrictive or more appropriate in other situations . for inhomogeneous systems or thermodynamic limits ( with @xmath26 ) the condensed phase can be localized in particular sites or have a more complicated spatial structure ( see e.g. @xcite and for recent summaries @xcite ) . for our case of spatially homogeneous systems on finite lattices , @xmath27 with fixed @xmath28 , the above is the most convenient definition and has been used in previously studied examples @xcite . a more detailed discussion is provided in section [ sec : char cond ] . + we use the natural partial order on the state space @xmath29 given by @xmath30 if and only if @xmath31 for all @xmath6 . a function @xmath32 is said to be increasing if @xmath33 implies @xmath34 . two measures @xmath35 on @xmath29 are stochastically ordered with @xmath36 , if for all increasing functions @xmath32 we have @xmath37 , where @xmath38 denotes the expectation of @xmath39 with respect to @xmath40 . a stochastic particle system on @xmath29 with generator @xmath41 and semi - group @xmath42 is called monotone ( attractive ) if it preserves stochastic order in time , i.e. @xmath43 coupling techniques for monotone processes are an important tool to derive rigorous results on the large scale dynamics of such systems such as hydrodynamic limits . there are sufficient conditions on the jump rates ( [ gene ] ) to ensure monotonicity for a large class of processes ( see e.g. @xcite for more details ) , however for our results we only need a simple consequence for the stationary measures of the process . [ order ] if the stochastic particle system as defined in section [ sub sec : ips ] is monotone , then the canonical distributions @xmath18 are ordered in @xmath44 , i.e. @xmath45 the proof is completely standard but short , so we include it for completeness . fix a monotone process as defined in section [ sub sec : ips ] . consider two initial distributions @xmath40 and @xmath46 , concentrating on @xmath17 and @xmath47 respectively , given by @xmath48 = { \mathds{1}}(\eta_{1 } = n ) \quad \textrm{and}\quad \mu'[\xi ] = { \mathds{1}}(\xi_1 = n+1)\ , \ ] ] for @xmath49 and @xmath50 . clearly @xmath51 , and so by monotonicity of the process this implies @xmath52 for all @xmath53 . furthermore , by ergodicity we have @xmath54 all rigorous results on condensing particle systems so far have been achieved for processes which exhibit stationary product measures , for which the measures @xmath18 take a simple factorized form . these can then be expressed in terms of un - normalized single - site weights @xmath55 , @xmath56 . due to conservation of @xmath14 such processes exhibit a whole family of stationary homogeneous product measures @xmath57=\prod_{x\in\lambda } \nu_\phi [ \eta_x ] \quad\mbox{with marginals}\quad \nu_\phi [ \eta_x ] = \frac{w(\eta_x ) } { z(\phi ) } \,\phi^{\eta_x}\ . \label{eq : pm}\ ] ] the measures are defined whenever the normalization @xmath58 is finite . this is the case for all fugacity parameters @xmath59 where @xmath60 or @xmath61 $ ] , and @xmath62{w(n)}\big)^{-1}\ ] ] is the radius of convergence of ( [ zphi ] ) . the family @xmath63 is also called the grand - canonical ensemble and @xmath64 the ( grand - canonical ) partition function . since the process is irreducible on @xmath17 for all @xmath65 we have @xmath66 for all @xmath67 . the canonical distribution can be written as @xmath68=\nu^{l}_\phi [ \eta |s_l = n]\quad\mbox{for all } \phi\in d_\phi \ , \ ] ] which is independent of the choice of @xmath69 . equivalently @xmath70 = \frac{1}{z_{l , n } } \prod_{x\in\lambda } w(\eta_{x})\quad\mbox{where}\quad z_{l , n } = \sum_{\eta\in \omega_{l , n } } \prod_{x\in\lambda } w(\eta_{x } ) \label{eqn : prod measure}\end{aligned}\ ] ] is the ( canonical ) partition function . note that throughout the paper we characterize all measures by their mass functions since we work only on a countable state space @xmath29 and the measures @xmath18 concentrate on finite state spaces @xmath17 , which is the framework we rely on in this paper . + our results hold for systems with general stationary weights , @xmath66 for each @xmath71 , subject to the regularity assumption that @xmath72 \label{regu}\ ] ] exists , which is then necessarily equal to @xmath73 . if @xmath74 then weights that satisfy are sometimes called long - tailed @xcite , which is discussed in more detail in section [ tlimit ] . for processes with such stationary product measures there is a simple equivalent characterization of condensation which we prove in section [ proof equiv ] . [ prop1 ] consider a stochastic particle system as defined in section [ sub sec : ips ] with stationary product measures as defined in section [ sec : mon ] satisfying regularity assumption . then the process exhibits condensation according to definition [ condensation ] if and only if @xmath75 , @xmath76 $ ] , and @xmath77}{\nu_{\phi_c } [ \eta_1 = n ] } = \lim_{n\to\infty } \frac{z_{2,n}}{w(n)z(\phi_c ) } \in ( 0,\infty ) \quad\mbox{exists}\ . \label{paff}\ ] ] in this case , the distribution of particles outside of the maximum converges weakly ( equivalently in total variation ) to the critical product measure @xmath78 , _ i.e. _ for fixed @xmath79 we have @xmath80 \to \prod_{i=1}^{l-1}\nu_{\phi_{c}}[\eta_{i } = n_i ] \quad \textrm{as}\quad n \to \infty \ .\end{aligned}\ ] ] note that for @xmath81 we may rescale the exponential part of the weights to get @xmath82 and we can further multiply with a constant , so that in the following we can assume without loss of generality that @xmath83 the condition ( [ paff ] ) can also be written as @xmath84 where @xmath85 is the convolution product . this is a standard characterization to define the class of sub - exponential distributions ( see e.g. @xcite ) . sub - exponentiality implies that a large sum of two random variables is typically realized by one of the variables taking a large value ( see section [ sec : subexp ] for more details ) , which is of course reminiscent for the concept of condensation . implications and simpler necessary conditions on @xmath86 which imply ( [ paff2 ] ) have been studied in detail , and we provide a short discussion in section [ sec : char cond ] . as a consequence of proposition [ prop1 ] , condensation is only a property of the tail of @xmath87 , therefore if a process with stationary product measures condenses in the sense of definition [ condensation ] for some @xmath23 , it condenses for all @xmath88 . proposition [ prop1 ] provides a generalization of previous results on condensation on finite lattices @xcite and is used in the proof of our main result , which is the following . [ mainres ] consider a spatially homogeneous stochastic particle system as defined in section [ sub sec : ips ] which exhibits condensation in the sense of definition [ condensation ] , has stationary product measures that satisfy , and has finite critical density @xmath89 then the process is necessarily not monotone . + the same is true if ( [ fmean ] ) is replaced by the assumption that we use the notation @xmath90 if @xmath91 as @xmath92 . ] @xmath93 with @xmath94 $ ] . the class of distributions which fulfil ( called sub - expontial ) , and therefore exhibit condensation on finite lattices , is large ( see e.g. ( * ? ? ? * table 3.7 ) ) , and includes in particular * power law tails @xmath95 where @xmath96 , * log - normal distribution @xmath97 where @xmath98 and @xmath99 , which always has finite mean , * stretched exponential tails @xmath100 for @xmath101 , @xmath102 , * almost exponential tails @xmath103 for @xmath104 . for the last two examples all polynomial moments are finite . this covers all previously studied models on condensation in zero - range processes @xcite . as we will discuss in section [ tlimit ] all these examples also exhibit condensation in the thermodynamic limit , whenever they have finite first moment . it can also be shown that the limit in ( [ paff2 ] ) is necessarily equal to @xmath105 and that in fact @xmath106 for any fixed @xmath23 ( see @xcite and proposition [ 2tol ] ) . since we consider a fixed lattice @xmath27 , @xmath107 is not a necessary condition for condensation as opposed to systems in the thermodynamic limit . even if the distribution of particles outside the maximum has infinite mean , condensation in the sense of definition [ condensation ] can occur . however , if @xmath108 ( e.g. for power law tails with @xmath109 ) , the distribution of particles outside the maximum can not be normalized , condition fails , and there is no condensation in the sense of our definition . 0.49 0.5 0.495 [ back : almostexp ] we will prove non - monotonicity in the next section by showing that expectations for a particular monotone decreasing observable @xmath110 under @xmath18 are not decreasing in @xmath44 . the chosen function is related to ( but not equal to ) the number of particles outside the maximum ( condensate ) , which has been shown previously to exhibit non - monotone behaviour for a class of condensing zero - range processes in the thermodynamic limit @xcite . when the number of particles @xmath111 just exceeds the critical value , typical configurations still appear homogeneous with a maximum occupation number of we use the notation @xmath112 if @xmath113 as @xmath114 . ] @xmath115 . only when the number of particles is increased further the system switches to a condensed state with a maximum that contains a non - zero fraction of all particles . we present numerical evidence of this non - monotone switching behaviour for the background density @xmath116 in figure [ backnumerics ] . this is a finite size effect which disappears in the limit @xmath117 , and for specific models it has been shown to be related to the existence of super - critical homogeneous metastable states @xcite . for large @xmath28 , the switching to condensed states occurs abruptly over a relatively small range of values for @xmath44 . since the @xmath18 are conditional product measures the correlations in the system are very weak , which causes metastable hysteresis effects and non - monotonicity of the canonical ensemble around the critical point . metastable hysteresis has been established in @xcite for zero - range processes . our result implies that this behaviour is generic for all condensing systems with finite critical density . we also give a heuristic discussion of the connection to convexity properties of the entropy of the system in appendix [ sec : statmech ] . there are several examples of homogeneous , condensing , monotone particle systems with finite critical density which have been studied on a heuristic level and which we summarize in section [ examples ] . their stationary measures are not of product form and no explicit formulas are known , so these systems are therefore hard to analyse rigorously . for systems with non - product stationary measures , upward fluctuations in the density which are homogeneously distributed may be suppressed strongly enough , so that the metastable states do not exist . such models may then also be monotone , and examples are given in section [ homnonpro ] . we excluded the case @xmath118 in the presentation in section [ sec : resu ] for notational convenience , but it is easy to see that our results also hold in this case . with the convention @xmath119 we have @xmath120 and @xmath121 , and then existence of the limit @xmath122 is equivalent to @xmath123=2\frac{w(n)w(0)}{z_{2,n } } \to \frac{2w(0)}{2z(0)}=1\ , \textrm { as } n \to \infty , \ ] ] i.e. condensation of all @xmath44 particles on a single site . this can easily be extended to all @xmath23 with proposition [ 2tol ] . considering only events with all @xmath44 particles on one site , or @xmath124 particles on one site and @xmath125 particle elsewhere , we have convergence from above @xmath126 this implies the non - monotonicity of @xmath18 as discussed in section [ sec : proof ] . we assume that the process exhibits condensation in the sense of definition [ condensation ] and has stationary product measures , so the canonical measures @xmath127 are of the form . furthermore , we assume the weights satisfy the regularity assumption , and without loss of generality @xmath128 , see . we show that the family of canonical measures is not stochastically ordered in @xmath44 , which implies non - monotonicity of the process by lemma [ order ] . to achieve this , we use the test function @xmath129 which indicates the event where all particles concentrate in the maximum at site @xmath28 . [ statement : monotonicity ] the function @xmath110 defined in ( [ eqn : dec observable ] ) is monotonically decreasing , which implies that @xmath130 whenever the family of canonical measures @xmath18 is stochastically ordered in @xmath44 . fix configurations @xmath131 such that @xmath30 . if @xmath132 then @xmath133 has at least one particle outside of site @xmath28 , therefore so does @xmath134 which implies @xmath135 . if @xmath136 then necessarily @xmath137 since @xmath138 . therefore @xmath39 is a decreasing function . using and the convention ( [ wlog ] ) , we find that the canonical expectation of the function is given by @xmath139 so if the canonical measures are monotone in @xmath44 , monotonicity of @xmath39 implies ( [ eq : zmono ] ) . 0.49 0.5 by proposition [ prop1 ] we know that for condensing systems the ratio @xmath140 converges . this implies the sequence @xmath141 in lemma [ statement : monotonicity ] converges , as summarised in the following proposition using our notation ( see ( * ? ? ? * theorem 1 and lemma 5 ) ) . [ 2tol ] consider conditional product measures ( [ eqn : prod measure ] ) with weights @xmath66 for all @xmath71 , which satisfy * @xmath142 as @xmath114 , the regularity assumption ( [ wlog ] ) , + * @xmath143 , + * @xmath144 as @xmath145 . then @xmath146 and furthermore , @xmath147 note that the limit in states that the probability of observing a large total number of particles under the critical product measure is asymptotically equivalent to the probability of observing a large number of particles on any one of the @xmath28 sites , precisely @xmath148}{l\,\nu_{\phi_c}[\h_1 = n ] } = 1\ , . \end{aligned}\ ] ] this is further equivalent to the canonical probability of the maximum containing the total mass converges to the critical probability that @xmath149 sites are empty , i.e. @xmath150 \to \nu_{\phi_c}^{l-1 } [ \h \equiv 0 ] $ ] . to complete the proof we show that a subsequence of @xmath141 converges from above , which contradicts the assumption of monotonicity by lemma [ statement : monotonicity ] . we present a numerical illustration for the monotonicity properties of the function @xmath151 in figure [ testnumerics ] , which is normalized such that @xmath152 as @xmath25 . the proof of the following lemma represents the most significant part of the proof of theorem [ mainres ] and is given in section [ sec : finite - mean ] for the case of finite mean and in section [ sec : power - law ] for the power law case . [ res2 ] under the conditions of theorem [ mainres ] , and assuming without loss of generality @xmath153 , for each @xmath88 there exists a @xmath154 and a sequence @xmath155 with @xmath156 as @xmath157 such that @xmath158 therefore , we know that there exists some @xmath159 such that @xmath160 , which by lemma [ statement : monotonicity ] implies that the canonical measures are not stochastically ordered in @xmath44 , and thus the process can not be monotone by lemma [ order ] . this completes the proof of theorem [ mainres ] . in order to prove that @xmath161 converges from above for some non - decreasing sequence @xmath162 we first specify a sequence on which we can bound the ratio @xmath163 below . [ statement : ratio upper bound ] for weights @xmath164 with finite and non - zero first moment , _ i.e. _ @xmath165 , there exists a sequence @xmath166 with @xmath167 as @xmath168 such that for all @xmath169 @xmath170 for each @xmath171 , define @xmath172 as follows @xmath173 by definition @xmath162 is a non - decreasing sequence . assume for contradiction that @xmath162 is bounded above , then for all @xmath174 we would have @xmath175 for some @xmath176 , and therefore @xmath177 contradicting the assumption of finite first moment . for @xmath178 we have @xmath179 [ statement : 2 site limit above ] for weights @xmath164 with finite first moment , there exists a subsequence @xmath180 of the sequence defined in claim [ statement : ratio upper bound ] such that @xmath181 for all @xmath182 and @xmath183 sufficiently large . by neglecting at most a single term in the sum defining @xmath184 , the ratio @xmath185 can be bounded below as follows , @xmath186 we define @xmath187 to be the largest index where the ratio @xmath188 is minimized . in particular @xmath189 by definition @xmath190 , and so @xmath191 . there exists a subsequence @xmath192 such that @xmath193 , with a slight abuse of notation we denote the subsequences @xmath194 and @xmath195 simply by @xmath196 and @xmath197 . suppose @xmath198 , by claim [ statement : ratio upper bound ] we have @xmath199 which together with contradicts proposition [ 2tol ] , therefore @xmath200 and @xmath201 for all @xmath183 . applying claim [ statement : ratio upper bound ] we then have @xmath202 subtracting @xmath203 we get @xmath204 neglecting the final term in we have @xmath205 as @xmath206 , where @xmath207 is the critical density defined in . together with this completes the proof of claim [ statement : 2 site limit above ] . to complete the proof of lemma [ res2 ] we proceed by induction on the system size , @xmath28 . we make the following inductive hypothesis ; 1 . [ h1 ] there exists a sequence @xmath208 such that @xmath209 for all @xmath210 and @xmath211 sufficiently large . the case @xmath212 is given by claim [ statement : 2 site limit above ] . analogously to the proof of claim [ statement : 2 site limit above ] we define @xmath213 by the same argument as in the proof of claim [ statement : 2 site limit above ] there exists a subsequence @xmath192 such that @xmath214 , again we denote the respective subsequences by @xmath197 and @xmath196 . for @xmath183 sufficiently large , we have @xmath215 where the final inequality follows from the inductive hypothesis ( h ) and claim [ statement : ratio upper bound ] . subtracting @xmath216 we get @xmath217 now , following the proof of claim [ statement : 2 site limit above ] , multiply ( [ arghh ] ) by @xmath218 and neglect the second term on the second line . then the first term vanishes , since @xmath219 in terms of the normalized grand - canonical measure @xmath220 $ ] , so we have @xmath221 where @xmath207 is the critical density as defined in ( [ fmean ] ) . this implies that the first term in the second line of ( [ arghh ] ) , after multiplication with @xmath196 , converges to a strictly positive constant . finally , the third line in ( [ arghh ] ) converges to zero after multiplying by @xmath196 since we have @xmath222 , which implies @xmath223 as @xmath206 , by . using the definition of @xmath197 in ( [ kmdef ] ) , this implies that there exists a constant @xmath102 such that for all @xmath183 large enough @xmath224 so ( h ) holds for @xmath225 , completing the induction . this concludes the proof of lemma [ res2 ] for the case where the critical measure has finite mean . we consider stationary weights of the form @xmath226 with @xmath227 , @xmath228 , and @xmath229 . we prove non - monotonicity of @xmath141 for @xmath230 and @xmath231 for all @xmath71 via an exact computation . the case @xmath232 can be done completely analogously but involves different expressions with logarithms in the resulting limits , and is presented in appendix [ bis2 ] . the proof remains valid for general converging @xmath233 with only minor differences , which we explain in a remark at the end of this section . convergence of @xmath234 from above or below for the exact power law depends on the parameter @xmath229 , as summarized in the next result . [ lemma : infinite mean ] for stationary weights of the form @xmath235 and @xmath227 with @xmath229 @xmath236 where @xmath237 for @xmath238 we have @xmath239 which has the same sign as @xmath240 . this result implies that whenever @xmath235 for @xmath241 and @xmath227 with @xmath230 lemma [ res2 ] holds with @xmath242 . this completes the proof of lemma [ res2 ] in the case @xmath243 . to prove this result we make use of the full taylor series of @xmath244 at @xmath245 and integral approximations to compute the asymptotic behaviour of summations . to simplify notation we assume that @xmath44 is even . for odd @xmath44 there is no term with multiplicity one and there exists an obvious modification . first note that @xmath86 fulfils the regularity assumption and @xmath246 as @xmath247 for all @xmath88 @xcite , so by proposition [ prop1 ] a process with stationary measures @xmath18 will exhibit condensation . for @xmath212 we subtract @xmath203 from @xmath140 to get @xmath248 substituting the taylor expansion of @xmath244 we find @xmath249 in the last line the @xmath250 term was combined with the second term , and we adopt the usual convention that empty products are equal to one . both summations in @xmath1 are over continuous and monotone functions @xmath251 , therefore we can use the usual integral approximation for decreasing ( increasing ) functions @xmath252 for all @xmath253 and @xmath254 . multiplying by @xmath255 we find the limit as @xmath145 of ( [ powerlaw difference ] ) to be @xmath256 it is shown in appendix [ appendixsign ] that this is positive ( and finite ) in the region @xmath230 and negative ( and finite ) in the region @xmath257 , completing the proof of lemma [ lemma : infinite mean ] for @xmath258 . the result holds for general system size , @xmath88 , and is proved by induction . the inductive hypothesis states @xmath259 similar to the case @xmath212 we write @xmath260 we first establish the limit of the function @xmath261 in equation . the inductive hypothesis can be written as @xmath262 which implies @xmath261 can be written as @xmath263 - l z(1)^{l } \right ) \ .\end{aligned}\ ] ] rearranging terms and noting that @xmath264 we then have @xmath265 after taylor expanding @xmath266 appearing in the first line above , it is easy to see that the limit of the first line is given by @xmath267 as @xmath247 . using the @xmath212 result to calculate the limit of the second line we find @xmath268 to identify the limit of @xmath269 in , we again make use of the taylor expansion of @xmath244 similarly to the two site case and we write @xmath270 changing the order of summations , separating the @xmath271 term and using @xmath222 we have @xmath272 for all @xmath273 and @xmath229 we have @xmath274 as @xmath145 , which implies that for any fixed @xmath275 we have @xmath276 . therefore , the following limits are equal @xmath277 using the inductive hypothesis we have @xmath278 is given by @xmath279 now applying the @xmath212 result it is possible to show that @xmath280 where the limit of the first line of was @xmath281 by the additional factor @xmath282 appearing in the summations . combining and we have @xmath283 concluding the induction so the result holds for all @xmath88 . from the recursion it is obvious that @xmath284 will have the same sign as @xmath240 , completing the proof of lemma [ lemma : infinite mean ] . a slightly modified version of lemma [ lemma : infinite mean ] also holds if the stationary weights are of the form @xmath285 where @xmath286 . the limit in only depends on the tail behaviour of the weights and is now given by @xmath287 . briefly , this can be seen as follows , becomes @xmath288 taylor expanding @xmath244 and rearranging to find terms of the form @xmath289 and using the same argument to calculate the limit of @xmath269 we have @xmath290 for all @xmath273 and any @xmath275 , and the result follows . similar modifications are required in the inductive step and the new limit in is given by @xmath291 for all @xmath88 . this does not change the sign of the limit in and therefore lemma [ res2 ] still holds . condensation arises in spatially homogeneous systems with stationary product measures due to the sub - exponential tail of the stationary weights @xmath87 , which has been studied extensively in previous work . in this section we review relevant results on heavy - tailed distributions and discuss the links between condensation on finite fixed lattices and in the thermodynamic limit before we give the proof of proposition [ prop1 ] in section [ proof equiv ] . sub - exponential distributions are a special class of heavy - tailed distributions , the following characterization was introduced in @xcite with applications to branching random walks , and has been studied systematically in later work ( see e.g. @xcite ) , for a review see for example @xcite or @xcite . a non - negative random variable @xmath292 with distribution function @xmath293 $ ] is called heavy - tailed if @xmath294 , @xmath295 for all @xmath296 , and @xmath297 it is called sub - exponential if @xmath294 , @xmath295 for all @xmath296 , and @xmath298 here @xmath299 $ ] denotes the convolution product , the distribution function of the sum of two independent copies @xmath300 and @xmath301 . it has been shown @xcite that is equivalent to either of the following conditions , @xmath302}{\mathbb{p}\big[\max\{x_{i } : i \in \{1 , \ldots , l \ } \}>x\big ] } = 1 \textrm{\quad for all \quad } l { \;\geqslant\;}2 \ . & \end{aligned}\ ] ] the second characterization shows that a large sum of independent sub - exponential random variables @xmath303 is typically realized by one of them taking a large value , which is of course reminiscent of the condensation phenomenon . it was further shown in @xcite that sub - exponential distributions also have the following properties , @xmath304 most results in the literature are formulated in terms of distribution functions and tails and apply to discrete as well as continuous random variables . @xcite provides a valuable connection to discrete random variables in terms of their mass functions @xmath86 , @xmath305 . + assume the following properties for a sequence @xmath306 , * @xmath307 as @xmath114 , * @xmath308(normalizability ) , * @xmath309 exists . then ( * ? ? ? * theorem 1 ) asserts that @xmath310 and @xmath311 is the mass function of a discrete , sub - exponential distribution . the implication @xmath312 is given in ( * ? ? ? * lemma 5 ) . sufficient ( but not necessary ) conditions for assumption ( c ) to hold are given in ( * ? ? ? * remark 1 ) . + provided @xmath313 , then ( c ) holds if either of the following conditions are met : * @xmath314 for some constant @xmath315 , or * @xmath316 where @xmath317 is a smooth function on @xmath318 with @xmath319 and @xmath320 as @xmath321 , and @xmath322 . + case ( i ) includes distributions with power law tails , @xmath95 with @xmath96 . the stretched exponential with @xmath323 , @xmath324 , and the almost exponential with @xmath325 , @xmath326 , are covered by case ( ii ) . the class of sub - exponential distributions includes many more known examples than the list given in section [ sec : discussion ] ( see e.g. ( * ? ? ? * table 3.7 ) ) . analogous to the characterisation of sub - exponential distributions , given by , for discrete distributions the existence of the limit @xmath327 is equivalent to the existence of the following condition @xmath328}{\mathbb{p}\big[\max\{x_{1},x_{2}\ } = n \big ] } \to 1 \textrm { as } n \to \infty \ .\ ] ] this holds , since we have the following equality of ratios @xmath329}{\mathbb{p}\big[\max\{x_{1},x_{2}\ } = n\big ] } = \frac{z_{2,n}}{2w(n)\sum_{n=0}^{n}{w(n ) } } = \frac{(w*w)(n)}{2w(n)\sum_{n=0}^{n}{w(n ) } } \ .\ ] ] specific properties of power law tails @xmath86 are used in @xcite to show condensation for finite systems in the sense of definition [ condensation ] . in proposition [ prop1 ] , proved in section [ proof equiv ] , we extend this result to stationary product measures with general sub - exponential tails . in this context , condensation is basically characterized by the property which assures emergence of a large maximum when the sum of independent variables is conditioned on a large sum . as summarized in the introduction , condensation in stochastic particle systems has mostly been studied in the thermodynamic limit with particle density @xmath330 , where @xmath331 such that @xmath332 . in that case conditions on the sum of @xmath28 independent random variables become large deviation events , which have been studied in detail in @xcite . + in the thermodynamic limit , a definition of condensation is more delicate and the approach presented in @xcite follows the classical paradigm for phase transitions in statistical mechanics via the equivalence of ensembles ( see e.g. @xcite for more details ) . a system with stationary product measures exhibits condensation if the critical density is finite , _ i.e. _ @xmath333 and the canonical measures @xmath18 are equivalent to the critical product measure @xmath334 in the limit @xmath331 such that @xmath332 for all super - critical densities @xmath335 . the interpretation is again that the bulk of the system ( any finite set of sites ) is distributed as the critical product measure in the limit . it has been shown in @xcite ( see also @xcite for a more complete presentation ) that the regularity condition and @xmath107 imply the equivalence of ensembles , which has therefore been used as a definition of condensation in ( * ? ? ? * definition 2.1 ) . therefore , any process that condenses for fixed @xmath336 with @xmath337 in the sense of definition [ condensation ] also condense in the thermodynamic limit . this includes all previously studied examples @xcite , however there exists distributions that satisfy with @xmath337 but do not satisfy the conditions of proposition [ prop1 ] and do not condense for fixed @xmath336 . this is illustrated by an example given below . it is also discussed in ( * ? ? ? * section 3.2 ) that assumption is not necessary to show equivalence , but weaker conditions are of a special , less general nature and are not discussed here . note also that equivalence of ensembles does not imply that the condensate concentrates on a single lattice site , the latter has been shown so far only for stretched exponential and power - law tails with @xmath107 in @xcite . both definitions involve only the sequence @xmath18 of canonical measures and not on the dynamics of the underlying process . since the canonical measures are fully characterised by the weights @xmath86 condensation can be viewed as a feature of the tails of the weights @xmath86 . the condensation phenomena can also be studied for continuous random variables on the local state space @xmath338 , see for example @xcite . the following continuous example , taken from @xcite is shown to satisfy but is not sub - exponential . we show that the distribution has a finite mean and therefore exhibits condensation in the thermodynamic limit but not on a finite lattice in the sense of definition [ condensation ] . for a real - valued random variable @xmath292 with distribution function @xmath339 $ ] , assume @xmath340 . let @xmath341 to be an increasing sequence with @xmath342 and @xmath343 be a continuous and piecewise linear function such that @xmath344 and @xmath345 for @xmath346 . the sequence @xmath341 is defined iteratively as follows @xmath347 and @xmath348 for @xmath349 . the mean is finite since @xmath350 where the final step uses the relation @xmath351 to bound the series from above . for all distributions satisfying which are not sub - exponential @xmath141 does not have a limit in @xmath352 as @xmath145 and with proposition [ prop1 ] there is no condensation on finite lattices according to definition [ condensation ] . for a discretized version of the example given above with weights @xmath353 we have @xmath354 for @xmath355 as @xmath114 @xcite . for this example , following the proof of proposition [ prop1 ] this implies that @xmath356\to 0 $ ] for @xmath355 as @xmath145 and all @xmath357 . therefore , the @xmath212 bulk occupation number @xmath358 diverges in distribution as @xmath145 by receiving a diverging excess mass from the condensate due to the light tail of @xmath86 . it can be shown that these weights satisfy and therefore exhibit condensation in the thermodynamic limit where the excess mass can be distributed on a diverging number of sites . for a process that exhibits condensation in the thermodynamic limit with a sub - exponential critical product measure , proposition [ prop1 ] implies that condensation occurs also on finite lattices with @xmath359 . theorem [ mainres ] then implies that this process is necessarily non - monotone for all fixed system sizes @xmath28 . however , monotonicity for condensing processes with long - tailed but not sub - exponential stationary measures remains open . let us first assume that the process exhibits condensation according to definition [ condensation ] and has canonical distributions of the form ( [ eqn : prod measure ] ) where the weights fulfil ( [ regu ] ) , i.e. @xmath360 $ ] as @xmath92 . in this part of the proof we establish that ; 1 . @xmath74 , 2 . @xmath361 has a limit as @xmath145 , 3 . @xmath362 , which also implies @xmath363 as @xmath145 , and 4 . convergence of @xmath363 for some @xmath88 implies convergence for @xmath212 and therefore holds . step ( 1 ) , show @xmath75 . assume first that @xmath364 as @xmath114 . for all @xmath365 and @xmath366 we have @xmath367 & = \frac{l}{z_{l , n}}\sum_{n=0}^{k}z_{l-1,n}w(n - n ) \\ & { \;\leqslant\;}l\frac{k+1}{z_{l , n}}\max_{0{\;\leqslant\;}n { \;\leqslant\;}k}\left(z_{l-1,n}\right)\max_{0{\;\leqslant\;}n { \;\leqslant\;}k } \left(w(n - n)\right ) \ .\end{aligned}\ ] ] let @xmath368 . the partition function @xmath369 is trivially bounded below by the event that site 1 takes @xmath370 particles and the second site takes the remaining @xmath371 particles , @xmath372 therefore @xmath373 { \;\leqslant\;}\frac{l}{w(0)^{l-2}}\frac{k+1}{w(n^{\star}+1 ) } \frac{w(n - n^{\star})}{w(n - n^{\star}-1)}\max_{0{\;\leqslant\;}n { \;\leqslant\;}k}\left(z_{l-1,n}\right)\to 0 \end{aligned}\ ] ] as @xmath145 , which implies condensation can not occur in the sense of definition [ condensation ] contradicting the initial assumption , therefore @xmath74 . step ( 2 ) , prove @xmath141 converges as @xmath145 . by definition [ condensation ] the limit @xmath374\ , \ ] ] exists and @xmath375 for @xmath376 sufficiently large . for @xmath377 we have @xmath378 = l\frac{w(n)}{z_{l , n}}\sum_{n=0}^{k}z_{l-1,n}\frac{w(n - n)}{w(n ) } \ .\ ] ] since @xmath379 , @xmath376 is fixed , and @xmath375 , implies the convergence of @xmath141 as @xmath145 . step ( 3 ) , prove @xmath362 . by we have @xmath380 as @xmath381 , taking the limit as @xmath145 of this implies @xmath382 since we also have @xmath383 , this implies @xmath143 . using @xmath380 , then also implies @xmath384 as @xmath145 . step ( 4 ) . we have seen above that condensation implies @xmath74 , @xmath385 , and @xmath384 as @xmath145 , then ( * ? ? ? * theorem 2.10 ) implies @xmath386 completing this part of the proof . now , let us consider a stochastic particle system with canonical distributions of the form ( [ eqn : prod measure ] ) which fulfil ( [ wlog ] ) and ( [ paff2 ] ) with @xmath82 and @xmath387 . we keep the notation for @xmath82 general in the following to clarify the argument . it is immediate from proposition [ 2tol ] , and remembering that we set @xmath388 , that @xmath389=lw(n)/z_{l , n } \to z(\phi_c ) ^{-(l-1 ) } > 0\ .\ ] ] then we have for all fixed @xmath376 and @xmath366 @xmath390&=l\sum_{n=0}^k \frac{w(n - n)z_{l-1,n}}{z_{l , n } } = \sum_{n=0}^k z_{l-1,n}\frac{w(n - n)}{w(n)}\frac{lw(n)}{z_{l , n}}\nonumber\\ & \to \sum_{n=0}^k \frac{z_{l-1,n } \phi_c^n } { z(\phi_c ) ^{l-1 } } = \nu_{\phi_c } ( \eta_1 + \ldots + \eta_{l-1 } { \;\leqslant\;}k)\end{aligned}\ ] ] as @xmath25 . since @xmath391 is a non - degenerate probability distribution , this implies that @xmath392 as @xmath393 , which is ( [ eqn : cond1 ] ) . to compute the distribution outside the maximum we get for fixed @xmath394 and large enough @xmath44 @xmath395 = \frac{w(n_1 ) \cdots w(n_{l-1})w(n - n_{1 } { -}\ldots { -}n_{l-1})}{\pi_{l , n } [ m_l { = } \eta_l]\ z_{l , n } } } \nonumber\\ & & = \frac{1}{l\pi_{l , n } [ m_l = \eta_l]}w(n_{1 } ) \cdots w(n_{l-1})\frac{w(n - n_{1 } -\ldots -n_{l-1})}{w(n)}\frac{lw(n)}{z_{l , n}}\nonumber\\ & & \to w(n_1 ) \cdots w(n_{l-1})\phi_c^{n_{1 } + \ldots + n_{l-1 } } /z(\phi_c ) ^{l-1}\ , \end{aligned}\ ] ] as @xmath25 . here we have used that spatial homogeneity of the measure and asymptotic uniqueness of the maximum according to imply @xmath396\to 1/l$ ] . this completes the proof of proposition [ prop1 ] . in this section we review several stochastic particle systems that exhibit condensation . by theorem [ mainres ] , if these processes are homogeneous and monotone with a finite critical density they do not have stationary product measures . to prove monotonicity for the examples mentioned below it is sufficient to construct a basic coupling of the stochastic process which preserves the partial order and particles jump together with maximal rate . for a definition of a coupling see @xcite and for the statement of strassen s theorem linking stochastic monotonicity and the coupling technique see @xcite . the steps to construct a basic coupling are outlined in @xcite . + condensation in homogeneous particle systems has mostly been studied in the framework of misanthrope processes @xcite . at most one particle is allowed to jump at a time and the rate that this occurs depends on the number of particles in the exit and entry sites . the misanthrope process is a stochastic particle system on the state space @xmath397 defined by the generator @xmath398 here @xmath399 denotes the configuration after a single particle has jumped from site @xmath400 to site @xmath401 , which occurs with rate @xmath402 . the purely spatial part of the jump rates , @xmath403 , are transition probabilities of a random walk on @xmath27 . such models are usually studied in a translation invariant setting with periodic boundary conditions , typical choices are symmetric , totally asymmetric or fully connected jump rates with @xmath404 , @xmath405 , or @xmath406 , respectively . misanthrope processes include many well - known examples of interacting particle systems , such as zero - range processes @xcite , the inclusion process @xcite , and the explosive condensation model @xcite . it is known @xcite that misanthrope processes with translation invariant dynamics @xmath407 exhibit stationary product measures if and only if the rates fulfil @xmath408 and , in addition , either @xmath409 the corresponding stationary weights satisfy @xmath410 misanthrope processes are monotone ( attractive ) @xcite if and only if the jump rates satisfy @xmath411 in theorem [ mainres ] we have proved that processes that exhibit stationary product measures and condensation with finite mean or power law tails , @xmath95 , with @xmath412 $ ] are necessarily not monotone . for power law tails with @xmath413 $ ] convergence of @xmath141 is from below and our method does not disprove monotonicity of the measures @xmath18 or monotonicity of the underlying process . using the specific form of the stationary measures , it is clear that possible examples of monotone processes with stationary product measures of this form can not be of misanthrope type . [ lemma : mismonotone ] a misanthrope process defined by the generator , that has stationary product measures and exhibits condensation is not monotone . gives necessary conditions for the monotonicity of the misanthrope process and implies with that @xmath414 is non - decreasing . this implies that the ratio converges to @xmath415 $ ] , which is the regularity assumption . assuming the process condenses in the sense of definition [ condensation ] , then proposition [ prop1 ] implies @xmath74 . now we have @xmath416 for all @xmath71 . therefore , @xmath417 which implies @xmath418 we conclude that the critical partition function diverges and the critical measure @xmath391 does not exist , which is a necessary condition for condensation . therefore condensation does not occur in misanthrope processes with stationary product measures . in @xcite generalised misanthrope processes have been introduced where more than one particle is allowed to jump simultaneously . they are defined via transitions @xmath419 for @xmath420 at rate @xmath421 and conditions on the jump rates for monotonicity are characterized . this class provides candidates for possible monotone , condensing processes with product measures as we discuss in the next subsection . the generalised zero - range process ( gzrp ) @xcite is a stochastic particle system on the state space @xmath422 defined by the generator @xmath423 here @xmath424 is the configuration after @xmath425 particles have jumped from @xmath400 to @xmath426 . the jump rates @xmath427 satisfy @xmath428 if @xmath429 , and we use the convention that empty summations are zero . we consider translation invariant @xmath430 on a finite lattice @xmath431 and note that the process preserves particle number @xmath432 . it is known @xcite that these processes exhibit stationary product measures if and only if the jump rates have the explicit form @xmath433 where @xmath434 are arbitrary non - negative functions with @xmath435 strictly positive . the stationary weights are then given by @xmath436 . monotonicity of the gzrp can be characterized in terms of @xmath437 the gzrp is monotone if and only if @xmath438 we note these conditions arise from a special case of the results in ( * ? ? ? * theorem 2.11 ) on generalised misanthrope models , since @xmath427 depends only on the occupation of the exit site and not the entry site . in this class , which is also discussed in detail in @xcite , condensing processes which are monotone , homogeneous , and have stationary product measures with a power tail @xmath95 with @xmath413 $ ] are conjectured to exist . as an example , consider the gzrp with rates given by @xmath439 since @xmath427 is of the form the process exhibits stationary product measures with weights of the form @xmath440 for all @xmath96 and @xmath88 the ratio @xmath441 converges to @xmath442 as @xmath145 @xcite so by proposition [ prop1 ] the process exhibits condensation . to prove the process is monotone we must show the rates satisfy the conditions given in equation . we first prove @xmath443 for all @xmath444 and @xmath445 . since @xmath446 for all @xmath445 we can drop the @xmath447 term from the definition of @xmath448 . we have @xmath449 \ .\end{aligned}\ ] ] since @xmath244 is increasing in @xmath400 for @xmath450 and @xmath451 we have @xmath452 we also need to show @xmath453 for all @xmath454 and @xmath241 . taking discrete derivatives in @xmath425 @xmath455 so @xmath456 is an increasing function in @xmath425 . therefore , @xmath457 for all @xmath458 , and it suffices to show @xmath459 we present numerical evidence in figure [ gzrpnumerics ] which corroborates our claim that the process with rates ( [ gzrpex ] ) is indeed monotone for @xmath460 $ ] and is not for @xmath461 . the chipping model is a stochastic particle system on the state space @xmath462 , introduced in @xcite . the dynamics are defined by the generator @xmath463 here @xmath464 denotes the configuration after all the particles at site @xmath400 have jumped to site @xmath401 , which occurs at rate @xmath125 , and single particles jump at rate @xmath465 . the spatial part @xmath430 is again spatially homogeneous as described in section [ sec : mis ] . it is easy to see that a basic coupling will preserve the partial order on the state space @xmath29 as defined in section [ sec : mon ] . therefore , by strassen s theorem @xcite , the chipping model is a monotone process and lemma [ order ] implies that conditional stationary measures of the process are ordered in @xmath44 . the condensation transition in the chipping model was established on a heuristic level in @xcite . we have defined the critical density @xmath466 only for systems with product stationary measures ( see ) . in general , the critical density on a fixed system of size @xmath23 , with unique invariant measures @xmath467 , can be defined as the background density of bulk sites @xmath468 notice if @xmath467 are conditional product measures ( see ) then @xmath469 is consistent with and in particular independent of @xmath28 , which follows from proposition [ prop1 ] ( more explicitly ) . for the chipping model in the case @xmath212 , the process reduces to a @xmath125-dimensional process on @xmath470 and the measure @xmath471 and @xmath472 can be computed explicitly to find @xmath473 in @xcite the critical density in the thermodynamic limit is defined as @xmath474 inspired by the fact that in case of condensation the second moment is either dominated by the condensate and scales like the system size @xmath28 , or it diverges since the maximal invariant measure does not have finite second moment . it is shown by heuristic computations in a mean - field limit that @xmath475 this suggests that the critical density can depend on the system size @xmath28 for distributions with non - product stationary measures . the @xmath476 scaling of the critical density can be intuitively understood in the two site chipping model with @xmath44 particles . this process can be interpreted as a symmetric random walk on the state space @xmath477 with jumps @xmath478 at rate @xmath87 and random jumps to either boundary ( resetting , @xmath479 or @xmath44 ) at rate @xmath125 . after a reset the particle diffuses at rate @xmath87 and reaches a typical distance @xmath476 from the boundary until the next reset . so this model is a monotone and spatially homogeneous process that heuristically exhibits a condensation transition with finite ( size dependent ) critical density , but it does not exhibit stationary product measures . condensation is also observed in models where chipping is absent ( @xmath480 ) and the dynamics result in a single block of particles jumping on the lattice @xmath481 corresponding to the critical density @xmath121 . condensation and non - monotonicity are also related to convexity properties of the entropy , which we briefly describe in the following in a non self - contained and non - rigorous discussion that is aimed at readers with a background in statistical mechanics . in the thermodynamic limit the canonical entropy is defined as @xmath482 for the processes we consider , equivalence of canonical and grand - canonical ensembles has been established in @xcite for condensing or non - condensing systems , so @xmath483 is given by the ( logarithmic ) legendre transform of the pressure @xmath484 this takes a particularly simple form since the grand - canonical measures are factorisable , and is a strictly convex function for @xmath485 . general results then imply that @xmath486 also has to be strictly convex below the critical density @xmath207 ( see e.g. @xcite ) , which holds for non - condensing systems and condensing systems with @xmath487 . for condensing systems with finite critical density @xmath483 is linear for @xmath488 , consistent with phase separation phenomena , where in this case the condensed phase formally exhibits density @xmath489 ( see e.g. @xcite for a general discussion ) . it is not possible to derive general results for finite @xmath28 and @xmath44 , but if we assume that the ratio of weights @xmath490 is monotone increasing in @xmath1 , we can show that a monotone order of @xmath18 implies that @xmath491 is necessarily convex . note that with ( [ regu ] ) our assumption implies that @xmath86 has exponential tails with @xmath81 or decays super - exponentially with @xmath492 , and in both cases the system does not exhibit condensation . we can define @xmath493 so that @xmath494 is a monotone increasing test function on @xmath495 . it is easy to see that for its canonical expectation we have for all @xmath23 and @xmath496 @xmath497 therefore , monotonicity of the canonical measures implies that the ratio of partition functions ( [ eq : exe ] ) is increasing and the discrete derivative of @xmath498 in @xmath44 is decreasing . we expect that in the limit @xmath499 the monotonicity assumption on @xmath490 is not necessary , and @xmath500 is convex in @xmath44 for all non - condensing systems , consistent with strict convexity of @xmath483 . for condensing systems the weights @xmath87 decay sub - exponentially , and if @xmath490 is monotone then it has to be decreasing in @xmath1 . therefore the choice @xmath493 implies @xmath501 is not a monotone test function , and the above general arguments can not be used to relate non - convexity of @xmath500 to the absence of a monotone order in @xmath18 . for particular condensing systems , however , it has been shown that @xmath500 is typically convex for small @xmath502 and concave for larger @xmath111 @xcite . these results focus on power law and stretched exponential tails for @xmath86 , and have been derived for zero - range processes where the ratio @xmath503 is equal to the canonical current . non - monotone behaviour around the critical density therefore has implications for finite - size corrections and derivations of hydrodynamic limits as mentioned in the introduction . consider stationary weights of the form @xmath504 with @xmath227 , we prove the non - monotonicity of @xmath141 in a similar fashion to the proof of lemma [ lemma : infinite mean ] summarised in the following lemma . [ lemmabis2 ] for stationary weights of the form @xmath504 with @xmath227 we have @xmath505 for @xmath238 we have @xmath506 which is positive for all @xmath88 since @xmath507 . first consider the case @xmath258 . as in the proof of lemma [ lemma : infinite mean ] we will utilise the full taylor expansion of @xmath508 , integral bounds on monotone series , and assume @xmath44 is even , for @xmath44 odd there exists obvious modifications to the proof . we have @xmath509 where the terms @xmath510 in the above summations cancel . substituting the taylor expansion of @xmath508 we find @xmath511 now we are in a position to apply the integral bounds , first noting that @xmath512 is decreasing for @xmath513 , constant and equal to @xmath125 for @xmath514 , and increasing for @xmath515 . multiplying both sides of and applying the integral bounds it is easy to show @xmath516 now consider the case @xmath517 and make the following inductive hypothesis @xmath518 as in the proof of lemma [ lemma : infinite mean ] write @xmath519 we fist establish the limit of @xmath520 in . the inductive hypothesis can be rewritten as @xmath521 similar to the proof of lemma [ lemma : infinite mean ] @xmath520 can be written in the form @xmath522 since @xmath523 is decreasing for @xmath524 we can find upper and lower bounds of the first term , by pulling out the logarithm , of the form @xmath525 applying the same steps exactly as they appear in the proof of lemma [ lemma : infinite mean ] to the upper and lower bounds we have @xmath526 to identify the limit of @xmath527 in we again follow the steps given in the proof of lemma [ lemma : infinite mean ] , which implies @xmath528 combining this with we have @xmath529 from the recursion it is obvious that @xmath530 will have the same sign as @xmath531 , completing the proof of lemma [ lemmabis2 ] . in this section , we compute the sign of @xmath240 for @xmath229 , where @xmath532 recall the definition of the pochhammer symbol @xmath533 and the hypergeometric function @xmath534 we now show @xmath535 which in particular implies @xmath536 by evaluating the hypergeometric formula . factorising the term @xmath537 from @xmath240 and rearranging terms inside the summation we have @xmath538 now use the following identities to simplify the terms inside the summation @xmath539 which gives the required result . to complete the proof we use the following two relations for hypergeometric functions , euler s transform @xcite @xmath540 and gauss s second summation theorem @xcite @xmath541 therefore , @xmath542 to calculate the sign of @xmath240 we first note that the gamma function @xmath543 is positive for all @xmath296 and negative in the region @xmath544 , which implies @xmath545 this work was supported by the engineering and physical sciences research council ( epsrc ) , grant no . p.c . acknowledges fellowship funding from the university of warwick , institute of advanced study . the authors are grateful to ellen saada and thierry gobron for useful discussions and comments on the manuscript .

hd dominates the cooling of primordial clouds with enhanced ionization , e.g. shock - heated clouds in structure formation or supernova remnants , relic hii regions of pop iii stars , and clouds with cosmic - ray ( cr ) irradiation . there , the temperature decreases to several 10 k and the characteristic stellar mass decreases to , in contrast with first stars formed from undisturbed pristine clouds ( ) . however , without cr irradiation , even weak far ultra - violet ( fuv ) irradiation suppresses hd formation / cooling . here , we examine conditions for hd cooling in primordial clouds including both fuv and cr feedback . at the beginning of collapse , the shock - compressed gas cools with its density increasing , while the relic hii region gas cools at a constant density . moreover , shocks tend to occur in denser environments than hii regions . owing to the higher column density and the more effective shielding , the critical fuv intensity for hd cooling in a shock - compressed gas becomes times higher than in relic hii regions . consequently , in the shock - compressed gas , the critical fuv intensity exceeds the background level for most of the redshift we consider ( ) , while in relic hii regions , hd cooling becomes effective after the cr intensity increases enough at . our result suggests that less massive ( ) pop iii stars may be more common than previously considered and could be the dominant population of pop iii stars . stars : formation - stars : population iii - dark ages , reionization , first stars - early universe .

1405.6901

the first generation of stars are believed to have had a significant impact on the cosmic evolution through their radiative , mechanical and chemical feedback ( e.g. , * ? ? ? stellar extreme - ultraviolet ( euv ) photons ( @xmath5 ev ) ionize the circumstellar gas to form an hii region , thereby initiating the process of the reionization of the intergalactic medium ( igm ) . even the single supernova ( sn ) of a massive ( @xmath6 ) first - star is able to totally evacuate the gas and suppresses subsequent star formation in the same minihalo @xcite although some of the materials blown out by the sn of a less - massive ( @xmath7 ) progenitor may fall back again and commence another episode of star formation @xcite . the ejected gas by a sn contributes to the metal enrichment of the igm . the extent of such feedback depends crucially on the initial mass function ( imf ) , as well as the star formation rate ( sfr ) , of first stars . stars formed from the metal - free primordial gas , or population iii ( pop iii ) stars , have been classified into two sub - populations depending on whether the gas is still in an undisturbed pristine state with a low electron fraction @xmath8 ( pop iii.1 ) , or the gas has already been affected by the radiative or kinematic feedback from pre - existing stars ( pop iii.2 ) , following the nomenclature proposed by oshea et al . despite with the same primordial composition , the thermal evolution of the gas is fairly different in those cases , resulting in different characteristic stellar masses . in the standard @xmath9cdm cosmology , pop iii.1 stars are predicted to be formed in minihalos of @xmath10 @xcite . during its protostellar collapse , cooling is entirely by @xmath11 , which is produced by the following electron - catalyzed reaction ( so - called the @xmath12 channel ; * ? ? ? * ; * ? ? ? * ) : @xmath13 @xmath14 with the high value of the minimum temperature @xmath15 attained at the density @xmath16 , the fragmentation mass is as large as @xmath17 @xcite , which is set by the jeans mass at this epoch . in this case , the final mass of formed stars is set by the radiative feedback from the growing protostar , and its value is a few 10 - a few 100 @xmath18 , much higher than the present - day counterpart @xcite . on the other hand , pop iii.2 stars are formed from the metal - free gas affected either by the radiative or kinematic feedback from earlier generation of stars , typically in atomic - cooling halos with virial temperatures @xmath19 @xcite . an example of pop iii.2 stars is those formed in the relic hii regions of defunct pop iii stars @xcite . in this case , a larger amount of @xmath11 is produced ( @xmath20 ) via the @xmath12 channel ( eqs . [ eq : h-1 ] and [ eq : h-2 ] ) than in the pop iii.1 case ( @xmath8 ) ( e.g. , * ? ? ? once the temperature plummets below @xmath21 by this @xmath11 cooling , deuterium is rapidly converted into hd via the exothermic reaction , @xmath22 this cooling by hd lowers the temperature further to @xmath23 ( e.g. , * ? ? ? the characteristic mass of stars formed in such an environment is predicted to be @xmath24 @xcite . that is , pop iii.2 stars may be typically less massive than preceding pop iii.1 stars although they are more massive than ordinary present - day stars . with high effective temperatures ( @xmath25 ) @xcite , massive pop iii stars emit a copious amount of far - uv ( fuv ) photons in the lyman - werner ( lw ) bands ( @xmath26 ev ) . the fuv radiation field photodissociates @xmath11 , thereby prohibiting efficient cooling in the primordial gas . even a single massive pop iii star emits enough fuv photons to suppress the subsequent star formation in the same halo @xcite . in addition , the gas in a minihalo is fairly vulnerable to the external irradiation of fuv photons and thus the number of pop iii.1 stars formed can be severely regulated by the fuv background ( e.g. , * ? ? ? on the other hand , pop iii.2 star formation in more massive halos still goes on by atomic cooling even under the fuv background and is not strongly affected . ( haiman et al . 1997 ; * ? ? ? this indicates the possibility that pop iii.2 stars can be the major population of primordial stars and significant sources of radiative and chemical feedback to subsequent star formation @xcite . moreover , the pop iii.2 star - formation epoch may have lasted until redshifts as low as @xmath27 @xcite . if so , gamma - ray bursts ( grbs ) or sne of pop iii.2 stars are detectable with future facilities , which allow us to peer into the nature of primordial stars @xcite . later , however , it was found that hd formation is strongly suppressed in the presence of even a weak fuv radiation field ( yoshida et al . 2007b ; * ? ? ? * ) since the slight photodissociation of @xmath11 causes the significant reduction of the hd fraction . without hd cooling , the thermal evolution of a pop iii.2 star - forming gas becomes identical to the pop iii.1 case , resulting in pop iii.2 stars having similar masses to pop iii.1 stars . however , this is not yet the conclusive result since the previous studies have not properly taken account of the following effects , which favors hd formation / cooling . in the galaxy formation epoch , shocks occur ubiquitously , for example , associating with the virialization of halos or sn explosions . in a shocked gas , the enhanced ionization degree enables efficient h@xmath28/hd formation and cooling @xcite . for a strong shock , the post - shock gas is compressed isobarically with its cooling , i.e. , the temperature decreases with increasing the density ( e.g. , * ? ? ? on the other hand , for the free - falling cloud in a relic hii region , rapid cooling makes the temperature to plummet almost isochorically , i.e. , the temperature decreases at a constant density ( yoshida et al . 2007b ; * ? ? * ; * ? ? ? * ) . with the higher column density , the fuv radiation field is more effectively shielded for the post - shock flow than in the relic hii region , resulting in the lower fuv flux in the former case . nonetheless , in the previous studies ( e.g. , * ? ? ? * ) , the post - shock flows are treated without the proper account of the isobaric nature of evolution . associated with star - formation activities , not only the fuv background , but also the cosmic ray ( cr ) background is expected to be generated by way of the cr acceleration in sn remnants . cr irradiation promotes @xmath11 and hd formation / cooling by enhancing the ionization degree @xcite . thus , cr irradiation is able to counteract with the negative feedback of fuv radiation . however , only either fuv or cr irradiation has been considered in the previous studies @xcite and their combined effects have not been fully understood . in this paper , we study the conditions for hd cooling to become important in pop iii.2 star formation under the presence of both fuv and cr backgrounds . we calculate the temperature and chemical evolution of ( i ) a cloud compressed by plane - parallel steady shocks @xcite , and of ( ii ) a free - falling cloud in a relic hii region . we find that the critical value for fuv intensity above which hd cooling is quenched is about an order of magnitude higher in the former than in the latter case . also the critical fuv intensity increases with increasing cr intensity . we also estimate the background fuv and cr intensities from theoretically predicted values for the sfr in the high - redshift universe . by comparing the background fuv intensity with the critical value , we conclude that the hd mode of pop iii.2 star formation is fairly common at the galaxy formation epoch . the rest of this paper is organized as follows . in [ sec : model ] , we describe the numerical model and initial conditions for our calculation . in [ sec : result ] , we present the results for the free - falling clouds in relic hii regions and for the clouds compressed either by the virialization shock or by the sn blast wave . we also present the critical fuv intensity for hd cooling as a function of the cr intensity . in [ sec : estimate ] , we estimate the background fuv and cr intensities in high - redshift universe , and discuss whether the hd cooling condition is satisfied at these epochs . after discussing the implications and uncertainties of our study in [ sec : discussion ] , we summarize our findings in [ sec : conclusion ] . in this section , we present our model for calculating the thermal and chemical evolution of primordial clouds . here , we first describe the method of calculation for the free - falling cloud in a relic hii region and then for the shock - compressed gas in structure formation or an sn explosion . we use the one - zone model , where we focus on the evolution of the central core for the free - falling cloud and of the coolest gas layer for the shock - compressed gas before the self - gravity becomes effective by using the plane - parallel and steady shock approximation . when a pop iii star dies and collapses directly to a black hole ( bh ) , the gas in the relic hii region surrounding the star begins to recombine and cool to form a subsequent star @xcite . here , we assume that the formed bh has no influence on the evolution of the gas in the relic hii region for simplicity . at first , the baryon mass in the relic hii region is too small for the gas to contract by self - gravity , but the gas loses pressure - support by efficient radiative cooling and contracts being attracted by the gravitational potential of the dm halo . after the temperature decreases so that the jeans mass becomes lower than the hii region mass , the gas begins to contract in the almost free - fall way by self - gravity . owing to its high pressure , the gas in the hii region is accelerated monotonically with radius up to a supersonic velocity and expands into the interstellar medium ( whalen et al . 2004 ; kitayama et al . . however , our model does not take into account the initial velocity profile within the hii region for simplicity . the following prescription is the same as the previous studies ( johnson & bromm 2006 ; yoshida , omukai & hernquist 2007 ; wolcott - green & haiman 2011 ) . an isolated cloud collapsing at the rate close to free - fall by its self - gravity develops a core - envelope structure @xcite , consisting of a nearly jeans - scale core with the constant density and an envelope where the density decreases radially as @xmath29 . in our one - zone model , we follow the evolution of the physical quantities at the core center @xcite . since collapse proceeds roughly at the free - fall rate , the density evolves as @xmath30 where @xmath31 is the free - fall time and @xmath32 the gravitational constant . temperature evolution follows the energy equation : @xmath33 where @xmath34 is the internal energy per unit mass , @xmath35 the pressure , @xmath36 the net cooling rate per unit volume . to supplement the equations above , the ideal - gas equation of state @xmath37 is used , where @xmath38 is the boltzmann constant , @xmath39 the proton mass and @xmath40 the mean molecular weight . using the ratio of specific heat @xmath41 , specific energy is related to pressure as @xmath42 . throughout this paper , we take @xmath43 and @xmath44 neglecting molecular contribution . considering that the size of the contracting core is half the jeans length @xmath45 , we calculate the column density of the core as @xmath46 @xcite . in this case , the initial size of this region is estimated as @xmath47 kpc , which is an overestimation compared to the typical size of an hii region @xmath48 pc ( whalen et al . 2004 ; kitayama et al . . however , at the beginning of collapse , owing to the low gas density and h@xmath28 fraction , the h@xmath28 column density will be too low for effectively shielding the gas from the fuv filed , and the above overestimation will not matter for estimating the shielding factor . in our model of free - fall collapse , the collapse equation and the energy equation ( eqs . [ eq : drho_dt ] , [ eq : energy ] ) are decoupled and feedback from thermal pressure is not included for simplicity . in the runaway - collapse phase , an isothermal cloud collapses approximately at a free - fall rate @xcite . although the pressure gradient is not negligible , the deviation of the collapse time from free fall is only by a factor of 1.58 @xcite . for a non - isothermal case , the deviation is represented as a monotonically increasing function of the effective adiabatic index @xmath41 of the cloud in equations 7 - 9 of @xcite . for example , the collapse time becomes @xmath49 times longer than the free - fall time for @xmath50 . if we use this generalized collapse time for the collapse equation , the collapse equation 4 is coupled with the energy equation 5 . however , we find that the effective adiabatic index becomes smaller than 1.2 for most of the evolutionary trajectory , and the deviation from free - fall may be at most by a factor of a few . thus , feedback from thermal pressure may change our results little . here , we consider that an overdense region collapsing within an hii region is large enough . then , since the overdense region is strongly bounded by the gravitational field of dm and gas , it contracts by the gravitational force which exceeds the pressure force from its less - dense surroundings . therefore , the compression of the overdense region by external pressure may be neglected , and it will collapse roughly at a free - fall rate . although our formulation does not include dm gravity , it becomes significant at the beginning of gas evolution . then , we recalculate the evolution of the gas in relic hii regions taking account of dm gravity by modifying @xmath51 in @xmath52 of eq . to @xmath53 , but we find little change in our results . here , @xmath54 is the mean dm mass density of a virialized halo at @xmath55 . after the passage of a blast wave , which is generated in structure formation or an sn explosion , the post - shock gas has a very high temperature due to shock heating and it cools and contracts with the radiative cooling time . at first , self - gravity is not effective in the shocked gas layer . we focus on the evolution of the most cooled gas layer , assuming that the post - shock gas layer is plane - parallel and the flow is steady @xcite . the initial condition of a shock - compressed gas is characterized by two parameters : the pre - shock density @xmath56 and the velocity of the shock front @xmath57 . in the strong shock limit , the physical quantities just behind the shock front are given by the rankine - hugoniot relation : @xmath58 @xmath59 @xmath60 @xmath61 where the subscripts 0 and 1 refer to the pre- and post - shock quantities , respectively . behind the shock front , the flow evolution is described by @xmath62 @xmath63 along with the energy equation ( [ eq : energy ] ) . the column density of the post - shock gas can be estimated from @xmath64 where @xmath65 is the number density in the pre - shock gas and @xmath66 is the time since the shock occurs . rapid cooling in the post - shock layer leads to the formation of a dense gas shell , which eventually becomes gravitationally unstable and fragments . for the sn shock , we assume that fragmentation occurs when the column density of the post - shock layer @xmath67 exceeds that in the jeans length @xmath68 @xcite : @xmath69 for the structure - formation shock , the mass of the post - shock layer is doubled compared to the sn case , since in one dimension , two flows collide to form the shocked region and two shock fronts propagate in opposite directions to each other . in this case , fragmentation occurs under the condition : @xmath70 in addition to them , for fragmentation , the growth time of a perturbation , which is approximately given by the local free - fall time @xmath52 , must be shorter than the contraction time of the layer , i.e. , the cooling time @xmath71 ( yamada & nishi 1998 ) . we have confirmed this condition is always satisfied at the fragmentation epoch determined by the column - density conditions given above . after fragmentation , each fragment or clump begins to contract by its self - gravity . we also follow its evolution by the same one - zone model as in sec . [ subsec : free - fall_form ] . the net cooling rate @xmath36 in eq . consists of the following terms : @xmath72 where @xmath73 , @xmath74 and @xmath75 are the radiative cooling rates by h lyman-@xmath76 line emission , @xmath11-line emission and hd - line emission , respectively , and @xmath77 is the cr heating rate . we adopt the cooling rates @xmath73 from @xcite , @xmath74 from @xcite , and @xmath75 from @xcite . since each cr ionization associates with heating of @xmath78 @xcite , @xmath79 where @xmath80 is the hydrogen ionization rate by crs , and @xmath81 is the number density of hydrogen atoms . at densities we consider ( @xmath82 ) , the contribution from chemical cooling and heating is small and is neglected in this paper @xcite . our chemical network includes following 11 species : h , @xmath11 , @xmath34 , @xmath83 , @xmath84 , @xmath12 , d , hd , @xmath85 , @xmath86 and @xmath87 . the deuterium fraction is set to @xmath88 , and the he chemistry is neglected since it is thermally inert in the temperature range we consider . chemical reactions include 18 hydrogen collisional reactions in table 1 of @xcite , 18 deuterium collisional reactions in table 1 of @xcite , photo - dissociation of h@xmath28 , hd , and h@xmath89 , and cr ionization . for collisional reactions , we use the updated rate coefficients by @xcite . the photo - dissociation rate coefficients are taken from @xcite and @xcite , and the @xmath11 and hd shielding factors are from @xcite . for the shielding factors , we take account of the doppler shift due to the bulk motion of the cloud as in @xcite for the free - falling case and in @xcite for the shock - compressed case . the stellar radiation field is assumed to have the diluted black - body spectrum with the temperature of @xmath90 k , resembling that from a pop ii star cluster , and its intensity is parameterized by @xmath91 , where @xmath92 is the intensity at the lyman limit . the cr ionization rate of neutral hydrogen @xmath80 is treated as a free parameter , since the cr intensity at the high - redshift universe is highly uncertain . however , the cr intensity may be related with the star - forming activity , and the sfr at @xmath93 is theoretically estimated to be no more than the present - day value ( e.g. , tornatore et al . 2007 ; trenti & stiavelli 2009 ; johnson et al . 2013 ) . we thus consider the cr intensity in the range of @xmath94 , which are smaller than the galactic value @xmath95 @xcite . in the relic hii region of a defunct pop iii star , a primordial cloud begins to collapse from a highly ionized state . the temperature and density in the relic hii region are almost uniform with @xmath96 and @xmath97 ( kitayama et al . 2004 ; whalen et al . 2004 ; abel , wise & bryan 2007 ; yoshida et al . we thus adopt @xmath98 and @xmath99 as the fiducial initial values , and also study the cases with higher and lower densities @xmath100 to see the dependence on the initial density . we set the initial chemical abundances as @xmath101 , @xmath102 , @xmath103 , and @xmath104 for other species , where @xmath105 is the fractional abundance of the @xmath106-th species to the hydrogen nuclei . we consider two kinds of shocks which occur in the virialization of halos and sn explosions . here , we need to specify the shock velocity @xmath57 , pre - shock density @xmath65 , and initial chemical abundance @xmath107 for each case . for the virialization shock , the mean gas number density within the virialized halo is estimated as @xcite @xmath108 however , according to the recent studies on first galaxy formation , the shock front does not sustain at the virial radius and its location shrinks inward owing to efficient ly-@xmath76 emission @xcite . protogalaxies accrete gas via cold flows in dense filaments , with the infall velocity of @xmath109 . such cold flows penetrate deep inside the halo and collide each other to form virialization shocks with pre - shock densities @xmath110 , much higher than in eq . . considering this effect , we adopt @xmath111 as the fiducial values for the pre - shock gas . the shocked gas behind a sn blast wave cools and is compressed significantly after the sn remnant enters the pressure - driven expansion phase @xcite . the shock velocity early in this phase is @xmath112 ( machida et al . 2005 ; nagakura et al . 2009 ; chiaki et al . the pre - shock density is comparable to the density in relic hii regions @xmath97 , since core - collapse sne explode within hii regions which their progenitors produce . hence , we adopt @xmath113 as the fiducial values for the sn shock . @xcite and @xcite studied the chemical composition in the primordial gas just behind a shock front . we adopt their results as the initial composition in the shock - compressed gas , which are summarized in table [ tab : shock_comp ] for the virialization and sn shocks . the fractions of species not presented in table [ tab : shock_comp ] are set to zero , except for the electron fraction @xmath103 . .initial chemical composition in the shock - compressed gas [ cols="^,^,^,^,^",options="header " , ] [ tab : shock_comp ] here , we present the results for the free - falling cloud in a relic hii region . we also find the critical fuv intensity above which hd cooling is suppressed for the different values of the cr intensity . first , we see the results without cr irradiation to extract the fuv effect alone . fig . [ fig : h2_nt_cr0 ] shows the temperature evolution , and fig . [ fig : h2_nt_cr0_chem ] shows the fractions of ( a ) @xmath11 , ( b ) electron , and ( c ) hd , for five different fuv intensities : @xmath114 , and @xmath115 , both as a function of the number density @xmath116 . the temperature first drops isochorically to @xmath117 via efficient ly-@xmath76 cooling in all the cases in fig . [ fig : h2_nt_cr0 ] , and the subsequent behavior depends on the fuv intensities . the same results are obtained in johnson & bromm ( 2006 ) , yoshida , omukai & hernquist ( 2007 ) , and wolcott - green & haiman ( 2011 ) . in the absence of fuv irradiation ( @xmath118 ) , abundant @xmath11 with the fraction @xmath20 forms via the h@xmath119 channel ( eqs . [ eq : h-1 ] and [ eq : h-2 ] ) owing to the high initial ionization degree ( solid line in fig . [ fig : h2_nt_cr0_chem]a ) . the cooling by this enhanced h@xmath28 makes the temperature to fall below @xmath120 , the minimum value attainable in the primordial pristine gas , where the ionization degree is only @xmath8 @xcite . once the temperature reaches @xmath121 , the exothermic reaction ( eq . [ eq : hd ] ) converts most of the deuterium into hd ( solid line in fig . [ fig : h2_nt_cr0_chem]c ) . hd cooling further drops the temperature to the minimum value @xmath122 at density @xmath123 , which is the critical density for hd to reach local thermodynamic equilibrium . toward higher density , the temperature increases gradually by compressional heating . at the temperature minimum , the parent cloud is considered to fragment into clumps with the jeans mass of @xmath124 ( yoshida et al . 2007b ; * ? ? ? we call hereafter this mode of star formation as _ hd - pop iii star formation _ , in contrast to _ h@xmath28-pop iii star formation _ , where h@xmath28 is the sole coolant during proto - stellar collapse . with fuv irradiation , the temperature becomes higher owing to the photodissociation of the coolants , h@xmath28 and hd ( fig . [ fig : h2_nt_cr0_chem]a , c ) . with @xmath125 as low as 0.01 , the evolutionary track is similar to that of @xmath118 , although with slightly higher temperature . with intensity @xmath126 , the @xmath11 fraction is reduced to @xmath8 , and the temperature does not reach the regime of efficient hd formation / cooling ( short - dashed line in fig . [ fig : h2_nt_cr0_chem]a , c ) . with the even higher fuv intensity @xmath127 , photodissociation strongly reduces the h@xmath28 fraction at low densities ( dotted and dot - dashed lines in fig . [ fig : h2_nt_cr0_chem]a ) . the clouds contract isothermally at @xmath117 by ly-@xmath76 cooling until @xmath128 for @xmath129 ( @xmath130 for @xmath131 , respectively ) , where the h@xmath28 self - shielding against fuv radiation becomes effective . thereafter , the temperature drops first almost vertically to @xmath132 k and then gradually to @xmath133 by @xmath11 cooling . in the cases with @xmath134 , without hd cooling , the gas cools only to the minimum temperature @xmath135 by h@xmath28 cooling at @xmath136 as in the case of pop iii.1 star formation . namely , star formation proceeds in the h@xmath28-pop iii mode . in this mode , the clouds fragment into rather massive clumps of @xmath137 . note that , in the high density regime ( @xmath138 ) , all the evolutionary tracks converge to either of the two tracks , one by hd cooling for @xmath139 or the other by @xmath11 cooling for @xmath134 . . different lines correspond to the different fuv intensities : @xmath140 , and @xmath115 from bottom to top . the thin dashed lines indicate the constant jeans masses . ] , ( b ) @xmath34 , and ( c ) hd in the same clouds as in fig . [ fig : h2_nt_cr0 ] . in each panel , different lines correspond to the different fuv intensities : @xmath141 ( solid ) , @xmath142 ( long - dashed ) , @xmath143 ( short - dashed ) , @xmath144 ( dotted ) , and @xmath115 ( dot - dashed ) . ] next , we see the results with cr irradiation on cloud evolution , as shown in figs . [ fig : h2_nt_j01 ] and [ fig : h2_nt_j01_chem ] , for the cases with fuv intensity @xmath126 . the different lines in the figures show different values of the cr intensity : @xmath145 ( i.e. , @xmath146 ) , @xmath147 and @xmath148 . recall that , with @xmath149 , hd cooling is suppressed without cr irradiation . as seen in fig . [ fig : h2_nt_j01 ] , with higher @xmath80 , the temperature becomes lower . this is because the elevated ionization degree by cr ionization causes more efficient h@xmath28 formation and cooling ( see fig . [ fig : h2_nt_j01_chem ] ) , which outweighs the effect of cr ionization heating . for @xmath150 , the temperature falls below 150 k , where the hd fraction increases significantly , owing to the enhanced @xmath11 cooling ( solid and long - dashed lines in fig . [ fig : h2_nt_j01_chem]c ) . by hd cooling , the temperature further drops to @xmath122 , and hd - pop iii star formation is realized in these cases . in this way , cr ionization can compensate the negative feedback of fuv photodissociation . , and the initial density to @xmath99 . different lines correspond to the different cr intensities : @xmath151 , and @xmath148 . the thin dashed lines indicate the constant jeans masses . ] , ( b ) @xmath34 , and ( c ) hd fractions for the same clouds as in fig . [ fig : h2_nt_j01 ] . in each panel , the different lines correspond to the different values of cr intensity : @xmath152 ( dot - dashed ) , @xmath153 ( dotted ) , @xmath154 ( short - dashed ) , @xmath155 ( long - dashed ) , and @xmath148 ( solid ) . ] since in our model for the free - falling cloud , the collapse equation ( eq . [ eq : drho_dt ] ) is decoupled from the energy equation ( eq . [ eq : energy ] ) , the collapse time becomes the same for the same initial density and is obtained as @xmath156 myr for @xmath157 . the hubble time is estimated as @xmath158 and this is roughly equal to the galaxy merger time . we find that the collapse timescale estimated above is smaller than the hubble time for @xmath159 . moreover , in the early universe we consider , the halo does not have an angular momentum large enough to form a rotation - supported disk which is comparable to the protogalaxy scale ( abel et al . thus , the gas cloud in a relic hii region may remain isolated without being affected by the protogalaxy rotation and merger , and star formation there proceeds within the local hubble time . we have learned that with the fuv intensity exceeding a critical value @xmath160 , hd cooling is totally suppressed even in initially ionized clouds . in addition , the value of the critical fuv intensity is elevated under cr irradiation . in fig . [ fig : h2_zeta_jcrit ] , the critical fuv intensity @xmath160 is plotted as a function of @xmath80 for initial densities @xmath161 ( dashed ) , 0.3 ( solid ) and 3 ( dot - dashed)@xmath162 . in finding the value of @xmath160 , we judge hd cooling is effective if the following two conditions are satisfied : ( i ) the minimum temperature at @xmath163 is less than 100 k , and ( ii ) the hd - cooling rate exceeds a half of the total ( i.e. the sum of compressional and cr ) heating rate at some moment . these two conditions are imposed to discriminate whether the high - density temperature tracks are either those by h@xmath28 cooling or by hd cooling . in the fiducial case of @xmath164 ( solid line ) , the critical intensity is @xmath165 for the negligible cr intensity ( @xmath166 ) , in accord with previous studies ( yoshida et al . 2007b ; * ? ? ? the critical intensity @xmath160 increases with the cr strength @xmath80 , since strong enough cr irradiation enables efficient hd cooling even under an intense fuv field . in the weak cr regime @xmath167 , the electrons present initially , rather than those produced by cr ionization , work as catalysts for the h@xmath28 formation reaction . abundant h@xmath28 is formed if the self - shielding against the fuv field becomes effective before the ionization degree is reduced significantly by recombination . since the self - shielding is more effective for larger column densities and thus for larger @xmath168 , more h@xmath28 is formed under the same fuv intensity . this results in higher @xmath160 for higher @xmath168 . on the other hand , in the strong cr regime @xmath80 @xmath169 , cr ionization provides the necessary electrons for abundant h@xmath28 formation . moreover , with a fuv field as high as @xmath160 , the shielding against the fuv field becomes effective and @xmath11 formation occurs only after the density is significantly increased from the initial value . therefore , the evolutionary track and critical fuv intensity @xmath160 do not depend on the initial density . for hd cooling as a function of the cr intensity @xmath80 . each curve corresponds to the different initial densities : @xmath161 ( dashed ) , 0.3 ( solid ) and 3 ( dot - dashed)@xmath162 . below @xmath160 , hd dominates the cooling in the course of cloud collapse . the shaded region shows the background fuv and cr intensities estimated from eqs . and with @xmath3 and @xmath170 . the blue thin - solid line shows the evolution of the background intensities evaluated using the pop ii sfr @xmath171 of johnson et al . the blue points refer to values at @xmath172 and @xmath173 from left to right . ] here , we present the results for a shock - compressed gas in the two cases of the shock velocity : ( i ) @xmath174 ( virialization shock ) and ( ii ) @xmath175 ( sn shock ) , respectively . first , to see the fuv effect on the gas compressed by the virialization shock ( @xmath176 ) , the results are shown for the cases without cr irradiation in figs . [ fig : shock20_nt_cr0 ] and [ fig : shock20_nt_cr0_chem ] . just behind the shock front , the gas is heated to @xmath19 and then cools immediately to @xmath117 by ly-@xmath76 cooling . with the negligible or weak fuv field ( @xmath177 ) , h@xmath28 forms abundantly owing to the high ionization degree ( @xmath178 ) in the post - shock gas ( solid and long - dashed lines in fig . [ fig : shock20_nt_cr0_chem]a , b ) . as in the relic hii region , the gas enters the hd formation regime ( @xmath179 k ) by h@xmath28 cooling ( solid and long - dashed lines in fig . [ fig : shock20_nt_cr0_chem]c ) , and the temperature further decreases to @xmath23 by hd cooling . as long as the cooling time is shorter than the free - fall time , the post - shock gas is compressed almost isobarically ( e.g. , * ? ? ? * ) . however , below @xmath122 , hd cooling is not efficient anymore , and the cooling time exceeds the free - fall time . when the sufficient mass for gravitational instability accumulates in the dense layer , it fragments into clumps of approximately the jeans mass @xmath1 , which then continue collapsing by hd cooling in the free - fall manner . this late - time temperature evolution is along the hd - cooling track , like the free - falling clouds in relic hii regions ( @xmath141 in fig . [ fig : h2_nt_cr0 ] ) . for @xmath126 , the isobaric contraction proceeds in the same way until @xmath23 as with the weaker fuv case . however , during the dense gas layer stays at @xmath23 until the jeans instability sets in , the hd fraction is significantly reduced by electron recombination and h@xmath28/hd photodissociation ( short - dashed lines in fig . [ fig : shock20_nt_cr0_chem]c ) . as a result , hd cooling does not become important in the subsequent free - fall phase and temperature evolution eventually converges to the h@xmath28-cooling track . for @xmath127 , strong fuv radiation completely quenches hd cooling and massive clumps of @xmath180 will be formed . especially , in the case of @xmath181 , fragmentation takes place early at @xmath182 . the subsequent free - fall evolution is similar to that of the free - falling clouds in relic hii regions with @xmath127 ( fig . [ fig : h2_nt_cr0 ] ) . and @xmath174 . different lines correspond to the different fuv intensities : @xmath140 , and @xmath115 . ] , ( b ) @xmath34 , and ( c ) hd in the same clouds as in fig . [ fig : shock20_nt_cr0 ] . different lines correspond to the different fuv intensities : @xmath141 ( solid ) , @xmath142 ( long - dashed ) , @xmath143 ( short - dashed ) , @xmath144 ( dotted ) , and @xmath115 ( dot - dashed ) . ] next , we see the cr effect for the case with fuv intensity @xmath183 . the results are shown in figs . [ fig : shock20_nt_j1 ] and [ fig : shock20_nt_j1_chem ] . recall that , with @xmath184 , hd cooling is suppressed in the absence of cr irradiation . as in the case of the relic hii region , more @xmath11 is formed by cr ionization ( fig . [ fig : shock20_nt_j1_chem]a , b ) and the temperature becomes lower by the enhanced h@xmath28 cooling . with the cr strength exceeding @xmath185 , most deuterium is converted into hd ( long - dashed and solid lines in fig . [ fig : shock20_nt_j1_chem]c ) , and temperature falls below @xmath186 k. thermal evolution finally converges to the hd - cooling track in the free - falling phase . . the initial density and velocity are the fiducial values : @xmath187 and @xmath174 . the different lines correspond to the different cr intensities : @xmath188 , and @xmath148 . ] , ( b ) @xmath34 , and ( c ) hd in the same clouds as in fig . [ fig : shock20_nt_j1 ] . the different lines correspond to the different cr intensities : @xmath152 ( dot - dashed ) , @xmath154 ( short - dashed ) , @xmath155 ( long - dashed ) , and @xmath148 ( solid ) . ] in our model for the shock - compressed gas , the contraction equations ( eqs . [ eq : rho_v ] , [ eq : rho_v^2 ] ) are coupled with the energy equation ( eq . [ eq : energy ] ) before the self gravity becomes effective , so that the timescale of the isobaric contraction phase differs among the lines . however , the evolution time after self - gravity becomes effective is almost the same and the time to reach @xmath189 is @xmath190 myr for the structure - formation shock , which is by an order of magnitude smaller than the local hubble time @xmath158 for @xmath159 . thus , the shock - compressed gas in structure formation may be unaffected by the protogalaxy rotation and merger , and star formation there proceeds within the local hubble time . in fig . [ fig : shock100_nt_cr0 ] , we show the fuv effect on the shock - compressed gas in an sn explosion ( @xmath191 ) for cases without cr irradiation . we find that , despite the different initial conditions , the evolutionary tracks are similar to those in the virialization shock with the same fuv intensity ( see fig . [ fig : shock20_nt_cr0 ] ) . this is because , although the initial density @xmath192 is ten times lower in the sn case , the post - shock temperature is 25 times higher and thus the pressure in the isobaric evolution phase differs only by a factor of 2.5 ( see fig . [ fig : shock20_nt_cr0 ] ) . we also find that the chemical compositions vary in a similar way as in the case of virialization shock ( see fig . [ fig : shock20_nt_cr0_chem ] ) , except that the ionization degree reaches @xmath47 due to the very high temperature @xmath25 behind the sn shock front . and @xmath193 . the different lines correspond to the different fuv intensities : @xmath140 , and @xmath115 . ] the same trend holds also in the cases with cr irradiation , and the evolutionary tracks are similar to those in the virialization shock with the same cr intensity ( see figs . [ fig : shock20_nt_j1 ] and [ fig : shock100_nt_j1 ] ) . . the fuv intensity is fixed to @xmath184 . the different lines correspond to the different cr intensities : @xmath188 , and @xmath148 . ] for the sn - shock case , the evolution time after self - gravity becomes effective is almost the same and the time to reach @xmath189 is calculated as @xmath194 myr , which is much longer than the lifetime of an sn remnant @xmath47 myr . thus , the sn shock will dissolve into the ism before it collects enough materials to trigger fragmentation and star formation in the post - shock region . this result is consistent with chiaki et al . ( 2013 ) , where they find that the sn - shock - compressed gas can collapse within the lifetime of the sn remnant for the ism density @xmath195 and @xmath196 erg . however , we discuss the evolution of the sn - shock - compressed gas for comparison with that of structure formation . as in [ subsec : h2_zeta_jcrit ] , the critical fuv intensity for hd cooling @xmath160 is calculated for a shock - compressed gas and is presented in fig . [ fig : sh_zeta_jcrit ] as a function of the cr intensity @xmath80 . the solid and dashed lines are for the virialization shock in the fiducial case @xmath111 and in the high density case @xmath197 , respectively , and the dot - dashed line indicates the sn shock case @xmath113 . from fig . [ fig : sh_zeta_jcrit ] , we see that for @xmath198 , the critical value @xmath160 becomes higher for a higher initial density @xmath168 or higher shock velocity @xmath57 . the reason is the same as in the case of a relic hii region : in the weak cr regime , abundant h@xmath28/hd formation is caused by the high post - shock ionization degree , and thus the prompt fuv shielding before significant recombination is needed . this can be realized more easily with a higher initial density @xmath168 or higher shock velocity @xmath57 , since the column density at @xmath199 k , below which recombination proceeds significantly , is higher in these cases . on the other hand , for @xmath200 , the critical value @xmath160 is solely determined by the cr strength @xmath80 and becomes independent of the initial condition @xmath201 . again the reason is the same as in the relic - hii - region case : the electrons from cr ionization , rather than those present initially , now work as catalysts for h@xmath28 formation and thus the h@xmath28 formation rate does not depend on the initial parameters . the comparison between figs . [ fig : h2_zeta_jcrit ] and [ fig : sh_zeta_jcrit ] tells us that in the weak cr regime , @xmath202 , the critical intensity is @xmath2 times higher for the shock - compressed gas ( @xmath203 ) than in the relic hii region ( @xmath204 ) . in this case , for abundant h@xmath28/hd formation , h@xmath28 should be promptly shielded against the fuv field before the electrons present initially recombine significantly . this condition can be more easily satisfied in the shock - compressed gas than in the relic hii region . the reason is that the shock - compressed gas evolves isobarically ( see fig . [ fig : shock20_nt_cr0 ] ) rather than isochorically as in the relic hii region ( see fig . [ fig : h2_nt_cr0 ] ) , which leads to the larger column density in the shock - compressed gas . in addition , shocks tend to occur in denser environments ( e.g. , cold - accretion flows ) than relic hii regions . consequently , in the shock - compressed gas , h@xmath28 is shielded from the fuv field more effectively , and the critical fuv intensity becomes higher than that in the relic hii region . for hd cooling as a function of the cr intensity @xmath80 in shock - compressed clouds ( see also fig . [ fig : h2_zeta_jcrit ] ) . the solid and dashed lines correspond to the virialization shock with the fiducial case , @xmath111 , and the high density case , @xmath197 , respectively . the dot - dashed line shows the sn - shock case , @xmath113 . ] we have seen that pop iii.2 star formation proceeds through the hd - cooling track , i.e. , in the hd - pop iii star formation mode , under a weak fuv field or strong enough cr irradiation . in this section , we estimate the intensities of the fuv and cr backgrounds as a function of redshift . then , by comparing them with the critical value @xmath160 obtained in [ sec : result ] , we discuss whether hd - pop iii star formation is realized at the epoch of galaxy formation ( @xmath205 ) . at this epoch , metal enrichment in an average galaxy may have proceeded to the level for pop ii star formation to occur ( e.g. , wise et al . 2012 ; johnson et al . we thus consider massive pop ii stars and their sn remnants as the plausible sources of the fuv and cr backgrounds , respectively . then , the intensity of the fuv background is estimated as @xcite @xmath206 where @xmath207 is the planck constant , @xmath208 the speed of light , @xmath209 the average frequency of the lw band ( @xmath210 ev ) , @xmath211 the lw band width ( @xmath212 ev ) , @xmath213 the number of fuv photons emitted per stellar baryon , and @xmath214 the stellar mass density at @xmath215 . johnson et al . ( 2013 ) evaluated @xmath214 by the relation @xmath216 , where @xmath171 is the pop ii sfr density , and @xmath217 the lifetime of a massive star . then , eq . can be rewritten as @xmath218 following @xcite , the energy density of the background crs @xmath219 is estimated from @xmath220 @xmath221 where @xmath222 is the sn energy , @xmath223 the efficiency of @xmath224 converted to cr acceleration , @xmath225 the number of sne per unit mass of stars formed , and @xmath226 the hubble time at @xmath215 . @xmath225 is obtained as @xmath227 assuming that massive stars with @xmath228 explode as sne for the salpeter imf . the cr ionization rate of a hydrogen atom @xmath80 is related with @xmath229 as @xcite : @xmath230 combining eqs . and , @xmath231 the high-@xmath215 sfr density has been evaluated theoretically by some authors ( e.g. , tornatore et al . 2007 ; trenti & stiavelli 2009 ; johnson et al . 2013 ) . according to johnson et al . ( 2013 ) , @xmath171 increases almost monotonically from @xmath232 to @xmath233 as @xmath215 decreases from @xmath234 to @xmath173 . the blue thin - solid lines in figs . [ fig : h2_zeta_jcrit ] and [ fig : sh_zeta_jcrit ] show the evolution of @xmath235 and @xmath236 evaluated from eqs . , and @xmath171 of johnson et al . the background intensities move from left to right along the line , and the blue points indicate the values at @xmath172 and @xmath173 , respectively . although the high-@xmath215 sfr density is still uncertain , most studies concluded @xmath237 for @xmath238 ( e.g. , tornatore et al . 2007 ; trenti & stiavelli 2009 ; johnson et al . 2013 ) . the shaded regions in figs . [ fig : h2_zeta_jcrit ] and [ fig : sh_zeta_jcrit ] indicate the range of @xmath235 and @xmath236 corresponding to this parameter range . first , we examine the case of free - falling clouds in relic hii regions ( fig . [ fig : h2_zeta_jcrit ] ) . for initial densities lower than the fiducial case @xmath164 ( black solid line ) , the critical fuv intensity for hd cooling @xmath160 is below the background level @xmath239 at @xmath240 , and hd cooling is suppressed in pop iii.2 star formation . as the background cr intensity @xmath80 increases with cosmic star formation , the critical intensity @xmath160 rises and eventually exceeds the background level @xmath239 at @xmath241 . below this redshift , hd cooling becomes efficient and hd - pop iii star formation occurs if some primordial environments still survive . although in a very dense hii region ( dot - dashed line ) , the critical intensity @xmath160 is above the background level @xmath239 even at @xmath240 , such dense conditions appear to be hardly realized ( kitayama et al . 2004 ; whalen et al . 2004 , yoshida et al . 2007a , abel2007 ) . we conclude that , in relic hii regions , hd - pop iii star formation is possible only at @xmath241 , where the intensity of the cr background is modest . on the other hand , for the shock - compressed gas ( fig . [ fig : sh_zeta_jcrit ] ) , the critical value @xmath160 exceeds the background level @xmath239 for most of the redshift we consider ( @xmath3 ) , and thus hd - pop iii star formation occurs . note that the pre - shock densities we adopt in fig . [ fig : sh_zeta_jcrit ] are rather low , and our results may give the lower bound of the critical fuv intensity @xmath160 . in fact , in the mass assembly process of first galaxies , cold - streaming shocks occur at the regions as dense as @xmath242 ( wise & abel 2007 ; wise et al . in such a dense shock environment , hd - pop iii star formation would be more common than the cases we have studied . in conclusion , the hd - pop iii mode is the main mode of pop iii.2 star formation in the cloud compressed either by a virialization or sn shock . so far , we have considered that the background fuv photons are uniformly distributed in the universe . in reality , they have some spatial fluctuations caused by local stellar sources , like star - forming galaxies ( dijkstra et al . 2008 ; ahn et al . 2009 ; agarwal et al . 2012 ; johnson et al . 2013 ) . in the primordial clouds irradiated with the strong fuv field as @xmath243 , hd formation is suppressed even in the shock - compressed gas unless the initial density is high ( @xmath244 ) or the cr intensity is strong ( @xmath245 : see fig . however , according to johnson et al . ( 2013 ) , only @xmath246 percent of the dense primordial cloud with @xmath247 is exposed to such a strong fuv field as @xmath243 . thus , our analysis with the uniform background fuv field is enough to discuss the major character of primordial star formation and the spatial fluctuations of it do not affect our main conclusions . we also assume that crs accelerated in sn remnants escape freely from their host galaxies and fill the universe homogeneously . however , in reality , they may be trapped in the host galaxy by the interaction with the magnetic field , and propagate diffusively through the ism of the galaxy @xcite . on the way of propagation , crs also lose their kinetic energy by ionizing the neutral ism . for the efficient cr escape into the igm , the cr escape time by diffusion @xmath248 should be shorter than both the ionization loss time @xmath249 and the age of the system @xmath250 , which is roughly equal to the age of the universe . here , @xmath248 and @xmath249 depends on the kinetic energy of a cr particle @xmath251 and the redshift @xmath215 . @xcite estimated the above timescales for crs with @xmath252 mev and discussed the cr escape efficiency during the cosmic structure formation . here , we extend their arguments by examining the @xmath251 dependence of the timescales . following @xcite , we assume that the energy spectrum of the cr number density has a power - law shape @xmath253 with @xmath254 ev . we find that the inequality @xmath255 holds for all the cr energy and redshift we consider ( @xmath3 ) . on the other hand , the inequality @xmath256 holds only for high energy crs with @xmath257 ev at @xmath3 . thus , the cr background is composed of crs with @xmath257 ev and its intensity varies with the spectrum index @xmath258 . in the standard fermi acceleration theory , the spectrum index is given as @xmath259 ( e.g. , * ? ? ? * ) , and the background intensity is reduced only by 10 % due to the confinement of crs with @xmath260 ev . on the other hand , if we adopt a steeper slope as @xmath261 in @xcite , it is reduced by an order of magnitude . even in this case , if a star - forming pristine gas resides in the same galaxy hosting sne , it is exposed to the cr intensity in the original or even the enhanced level , because of the efficient confinement within the galaxy . for more quantitive and definitive discussions , a detailed study of cr propagation in the ism of high-@xmath215 galaxies is needed and is beyond the scope of this paper . at the epoch of galaxy formation , high - mass x - ray binaries and miniquasars may be the plausible sources of high energy photons extending to the x - ray band @xcite . since the photoionization cross section of neutral atoms generally decreases with frequency in the power - law way , the igm becomes optically thin for x - ray photons with @xmath262 kev and the x - ray background develops at this epoch ( haiman et al . 2000 ) . the injection of x - ray photons enhances the cooling efficiency of primordial clouds by indirectly promoting h@xmath28 and hd formation through the photoionization of neutral atoms ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? nonetheless , we did not include this x - ray feedback , and here we briefly discuss how it changes our results . following @xcite , we assume that the x - ray background has a power - law spectrum @xmath263 where @xmath264 and @xmath265 is the intensity at 1 kev . then , using eqs . ( 17 - 19 ) in @xcite , the ionization rate of h is calculated as @xmath266 for the hydrogen column density @xmath267 . since the background field derives from the stellar activity , the x - ray intensity is related with the cr energy density as @xmath268 where we suppose that cr sources are pop ii clusters with the salpeter imf and the mass range @xmath269 @xcite erg as the energy of a pop ii sn instead of @xmath270 erg in @xcite . ] . substituting eq . into eq . and eq . into eq . , the x - ray ionization rate is related with the cr ionization rate as @xmath271 thus , we find that , at each redshift , x - ray feedback changes the ionization rate in pristine clouds not by an order of magnitude , but by only up to a factor of two . note that this is the upper limit of the x - ray effect since x - ray is more effectively shielded than crs . therefore , our main conclusion still remains valid even if we consider x - ray feedback . as we have seen above , pop iii star formation at the epoch of galaxy formation is affected significantly by the injection of crs . nonetheless , some previous studies calculated the pop iii sfr by considering the regulation process of star formation from fuv feedback alone , and the positive feedback from crs is neglected ( e.g. , tornatore et al . 2007 ; trenti & stiavelli 2009 ; de souza et al . 2011 ; johnson et al . 2013 ) . future studies should take all these feedback into consideration . we must admit that our treatment of the plane - parallel and steady shock is too simplistic to capture possible complex phenomena induced by e.g. , the turbulence , rotation , thermal instability , and magnetic field . for example , @xcite studied the effect of magnetic fields on the thermal evolution of a shock - compressed gas , using the two - dimensional mhd simulation . they found that after the short isobaric contraction , the magnetic pressure is amplified to be comparable to the ram pressure of the pre - shock materials , and thereafter the shock - compressed gas cools almost isochorically . this implies that the shielding of h@xmath28 against a fuv field becomes less effective and the resultant h@xmath28 fraction is lowered in the presence of magnetic fields . in this case , we expect that more massive clumps may form after the gas becomes gravitationally unstable and fragments , compared to the isobaric contraction case without magnetic fields . however , their calculation did not follow the evolution until the self - gravity of the shocked layer becomes important . they did not consider the gas in the primordial composition as well . thus , the above effects on our results should be investigated through the 3d mhd simulations in the future . another example is that our assumption of the time - independent shock velocity becomes invalid at least in the sn - shock , according to the one - dimensional numerical calculations @xcite . they showed that the sn - shock velocity becomes lower than @xmath175 at @xmath272 yr after the explosion . moreover , at @xmath273 yr , the ram pressure of the ambient medium decreases below the pressure in the shell so that the shell re - expands and decreases its temperature and density in an almost adiabatic manner . although in this case , h@xmath28 is produced efficiently to @xmath20 and the temperature decreases to @xmath274 k by h@xmath28 cooling , hd is produced only to @xmath275 , which is @xmath2 times less than our results , in the adiabatic expansion phase owing to the low gas density . in this case , the temperature decrease below @xmath276 k is not caused by hd cooling but by the adiabatic expansion of the shell @xcite . however , if the sn shock can collect enough materials to trigger gas fragmentation and contraction by self - gravity before dissolving into the ism , hd formation becomes effective and hd cooling becomes dominated in the free - falling phase , as we can see from fig . 6 in @xcite and fig . 6 in @xcite thus , whether this realistic evolution of the sn shock front significantly changes our results is not clear , and more detailed numerical calculations are needed for more concrete discussion . hd molecules form abundantly in the primordial gas with enhanced initial ionization and play a dominant role in gas cooling . at the epoch of galaxy formation ( @xmath205 ) , shocks occur ubiquitously associated with the mass assembly as well as the sn explosions . in such a shock - compressed gas , the ionization degree jumps up owing to collisional ionization . other examples include the cloud in the relic hii region of a defunct pop iii star and that with modest cr irradiation . however , even a low level of the fuv background suppresses hd formation / cooling via @xmath11 photodissociation in cases without cr irradiation . in this paper , we have examined the conditions for efficient hd cooling in primordial star formation by calculating the thermal and chemical evolution of a gas cloud under both fuv and cr irradiation . we have obtained the critical fuv intensity @xmath160 for hd cooling as a function of the cr intensity @xmath80 , and compared it with the estimated background level @xmath239 . we have considered both cases of the free - falling cloud in a relic hii region and a shock - compressed gas . at @xmath277 , the background cr intensity is estimated as @xmath278 of the galactic value . in this case , the critical fuv intensity @xmath160 in relic hii regions is below the background level @xmath239 and hd cooling is suppressed there . however , as the background cr intensity increases with the cosmic star formation , the critical intensity @xmath160 is also elevated and eventually exceeds the background level @xmath239 at @xmath4 . below this redshift , hd cooling becomes efficient and hd - pop iii star formation proceeds if some primordial environments still survive . on the other hand , the critical fuv intensity @xmath160 for the shock - compressed gas is @xmath2 times higher than that for the relic hii region even in the weak cr regime as @xmath278 of the galactic value . this is because the shock - compressed gas evolves isobarically ( i.e. , gas cools with its density increasing ) , while the gas in a relic hii region evolves isochorically ( i.e. , the gas cools at a constant density ) due to the initial rapid cooling . moreover , shocks tend to occur in denser environments ( e.g. , cold - accretion flows ) than relic hii regions . as a result , the shock - compressed gas is shielded from the fuv field more effectively , and has higher critical intensity . thus , for the shock - compressed gas , the critical value @xmath160 exceeds the background level @xmath239 and hd - pop iii star formation proceeds for most of the redshift we consider ( @xmath3 ) . our result suggests that hd - pop iii stars can be more common than previously considered and could be even the dominant population of pop iii stars . iii stars are born in clumps as massive as @xmath279 , but the final stellar mass is set to @xmath1 by the stellar radiative feedback onto the accretion flow @xcite . on the other hand , hd - pop iii stars form in clumps of @xmath1 , and the final mass is typically set to a few @xmath280 by the stellar feedback @xcite . these previous results may suggest that the star formation efficiency , i.e. , the mass ratio of formed stars to the parent clouds , of hd - pop iii stars is several times higher than that of h2-pop iii stars . they also imply the possibility that hd - pop iii stars could be the majority of pop iii.2 stars . then , hd - pop iii stars of a few 10 @xmath281 can be important sources of radiative feedback at the epoch of galaxy formation . moreover , according to the observations of the abundance pattens in the metal - poor halo stars in the galaxy , the imprints of a pair - instability sn have not been discovered yet @xcite . since stars with @xmath282 end their lives as core - collapse sne @xcite , hd - pop iii stars might be the major contributor to the metal enrichment in the ism / igm at that epoch , if there were no pisne in the universe . recent discoveries of the almost pristine gas with metallicity @xmath283 in the damped ly @xmath76 systems at @xmath284 and @xmath285 @xcite imply that metal enrichment in the igm has proceeded quite inhomogeneously . theoretical models also show the inhomogeneous nature of metal enrichment in the igm . they predict that pop iii star formation can continue up to @xmath27 or even lower , and evaluate the ratio of the pop iii sfr to the total sfr as @xmath286 ( @xmath8 ) at @xmath287 ( @xmath288 ) ( tornatore et al . 2007 ; johnson et al . 2013 ) . this indicates that pop iii stars can form to some extent even at @xmath241 , and our study show that they are typically formed as hd - pop iii stars with a few @xmath280 . grbs and sne from these pop iii progenitors at @xmath4 will be detected with the current or future facilities and may provide us with information on the mass , imf and sfr of these stars ( nakauchi et al . 2012 ; whalen et al . 2013 ; tanaka et al . 2013 ) . moreover , if the hd absorption lines are detected in the spectrum of a grb afterglow which lacks metal absorption , hd is confirmed as the dominant coolant in that primordial gas @xcite . we thank the anonymous referee for helpful comments and improving the quality of this paper . we also thank k. kashiyama , t. nakamura y. suwa , and k. tanaka for fruitful discussions and comments . this work is supported in part by the grant - in - aid from the ministry of education , culture , sports , science and technology ( mext ) of japan ( 23 - 838 ki ; 25287040 ko ) . haiman , z. , rees , m. j. , & loeb , a. 1996 , apj , 467 , 522 haiman , z. , thoul , a. a. , & loeb , a. 1996 , apj , 464 , 523 haiman , z. , rees , m. j. , & loeb , a. 1997 , apj , 476 , 458 haiman , z. , abel , t. , & rees , m. j. 2000 , apj , 534 , 11 hayakawa , s. , nishimura , s. , & takayanagi , t. 1961 , pasj , 13 , 184

a new moving group comprising at least four blue horizontal branch ( bhb ) stars is identified at ) . the horizontal branch at magnitude implies a distance of 50 kpc from the sun . the heliocentric radial velocity is km s , corresponding to km s ; the dispersion in line - of - sight velocity is consistent with the instrumental errors for these stars . the mean metallicity of the moving group is [ fe / h] , which is significantly more metal poor than the stellar spheroid . we estimate that the bhb stars in the outer halo have a mean metallicity of [ fe / h]= , with a wide scatter and a distribution that does not change much as a function of distance from the sun . we explore the systematics of sdss dr7 surface gravity metallicity determinations for faint bhb stars , and present a technique for estimating the significance of clumps discovered in multidimensional data . this moving group can not be distinguished in density , and highlights the need to collect many more spectra of galactic stars to unravel the merger history of the galaxy . [ firstpage ] galaxy : structure galaxy : halo methods : statistical .

1002.3607

in recent years , stellar photometry primarily from the sloan digital sky survey ( sdss ) has allowed astronomers to discover previously unknown spatial substructure in the galactic spheroid . other surveys such as 2mass and quest have also been influential . highlights include the discovery of new tidal streams @xcite ; dwarf galaxies , star clusters , and transitional objects @xcite ; and previously unidentified spatial structure of unknown or controversial identity @xcite . the discovery of spatial substructure immediately suggests the need for spectroscopic follow - up to determine the orbital characteristics of tidal debris , the masses of bound structures , and the character of structures of unknown identity . it has long been known that a spheroid that is formed through accretion will retain the record of its merger history much longer in the velocities of its component stars than it will in their spatial distribution @xcite . for this reason early searches concentrated on velocity to find moving groups " in the spheroid @xcite rather than on the search for spatial substructure . groups of spheroid stars that have similar spatial positions and velocities ( moving groups ) are generally equated with tidal disruption of star clusters or dwarf galaxies , unlike local disc moving groups " that were once hypothesized to be the result of disruption of star clusters @xcite but are now thought to be the result of orbital resonances . recently , searches for spheroid velocity substructure in sdss data have concentrated on local ( up to 17.5 kpc from the sun ) metal - poor main sequence stars @xcite , finding more than a dozen groups of stars with coherent velocities . @xcite estimate that there are @xmath7 cold substructures in the milky way s stellar halo . blue horizontal branch ( bhb ) stars are particularly important for studies of spheroid substruscture ( e.g. clewey & kinman 2006 ) because they can be seen to large distances in the sdss and because a large fraction of the sdss stellar spectra are bhbs . in the future we need more complete spectroscopic and proper motion surveys designed to discover velocity substructure in the milky way s spheroid . in this paper we present evidence for a tidal moving group of bhb stars , discovered in the sdss and sloan extension for galactic understanding and exploration ( segue ; yanny et al . 2009 ) spectroscopic survey . additional evidence that the stars are part of a coherent structure comes from the unusually low metallicity of the stars in the moving group . we are unable to isolate this moving group in density substructure , highlighting the power of velocity information to identify low density contrast substructure in the spheroid . historically many of the spheroid substructures have been discovered by eye in incomplete datasets , rather than by a mechanical analysis of data with a well - understood background population . as the size of the substructures decreases , it becomes more critical to determine whether the observed structure could be a random fluctuation of the background . in this paper , we present a method ( appendix a ) for estimating the probability that a random fluctuation could produce an observed lump . " as we work towards the fainter limits of the sdss data , it becomes more important to understand how stellar parameters derived from spectra are affected by a low signal - to - noise ( s / n ) ratio . the stars in the newly detected halo substructure have s / n ratio between 7 and 10 , as measured from the ratio of fluctuations to continuum on the blue side of the spectrum . in the past we have successfully used spectra with s / n as low as 5 to measure radial velocities of f turnoff stars in the virgo overdensity @xcite . in this paper , we show that although higher s / n is preferable , some information about metallicity and surface gravity can be gained for bhb stars with s / n ratios less than 10 . since the strongest evidence that these bhbs form a coherent group is that they have unusually low metallicity , even for the spheroid , we also explore the segue metallicity determinations for bhb stars in the outer spheroid . a great variety of previous authors have measured a large metallicity dispersion for spheroid stars , with average metallicities somewhere in the @xmath8 < -1.7 $ ] range . these investigations analysed globular clusters @xcite , rr lyraes @xcite , k giants @xcite , and dwarfs @xcite . @xcite found an average [ fe / h ] of @xmath9 for globular clusters in the outer halo , and an average metallicity of @xmath10 for field stars in the outer halo , again with a large metallicity dispersion that does not depend on distance . the besanon model of the milky way adopted an average [ fe / h ] of @xmath11 , with @xmath12 , for the stellar halo . recently , @xcite analysed the full space motions of @xmath13 stars within 4 kpc from the sun , and found evidence that the ones that are kinematic members of the outer halo have an average metallicity of [ fe / h]=@xmath14 . our analysis of the sdss bhb stars shows that the metallicity of bhb stars in the galactic spheroid does not change from @xmath15 ( 6 kpc from the sun ) to @xmath16 ( 55 kpc from the sun ) at high galactic latitudes . the mean measured metallicity of these stars is [ fe / h]=@xmath17 , but analyses of globular clusters in the sample show that there could be a systematic shift in the bhb measured metallicities of a few tenths of a magnitude , and in fact [ fe / h]=@xmath6 is our best guess for the proper calibrated value . in order to search the stellar halo for co - moving groups of stars , we selected bhb stars from the sixth data release ( dr6 ; adelman - mccarthy et al . 2008 ) of the sdss . this release contains both legacy survey and segue photometric data from 9583 square degrees of sky . more technical information for the sdss survey can be found in @xcite . we selected 8753 spectra with the colors of a stars from the sdss dr6 star database using the color cuts @xmath18 and @xmath19 , within the magnitude range @xmath20 , where the @xmath21 subscript indicates that the magnitudes have been corrected for reddening and extinction using the @xcite reddening map . within this color box , the stars that are bluer in @xmath22 tend to be high surface gravity blue straggler ( bs ) or main sequence stars , and those that are redder in @xmath22 tend to be low surface gravity blue horizontal branch ( bhb ) stars . using @xmath22 , @xmath23 color selection parameters as described in figure 10 of @xcite , we divided the sample into @xmath24 candidate bhb stars and @xmath25 candidate bs stars with spectra . we separated the stars in surface gravity using a photometric technique , because in 2007 when we were selecting the stars , we were concerned that the s / n of the spectra would degrade more quickly than the s / n of the photometry at the faint end of our sample . the s / n of the spectroscopy increases over the whole magnitude range from @xmath26 , while the photometric accuracy degrades only for @xmath27 . the photometric degradation affects halo bhb stars fainter than about @xmath28 . with more recent reduction software ( see below ) , we have found similar sensitivity to surface gravity in photometric and spectroscopic indicators for faint sdss spectra , but the spectroscopic separation is much better than photometric separation for bright , high s / n spectra . the photometric selection technique does not introduce gradients in the selection effect with apparent magnitude , and can be applied identically for stars with and without measured spectra . at bright magnitudes , the spectral s / n is high enough that the dr6 surface gravities are reliable . the sdss spectra are r=1800 , and the s / n of a star spectra varies from 50 at @xmath29 to 7 at @xmath30 . we tested the efficiency and completeness of the photometric selection technique using 3996 sdss spectra of stars with @xmath31 and colors of a - type stars . of the 1948 that had surface gravities consistent with bhb stars , 84% were classified as bhbs by the photometric technique . of the 2048 spectra with high surface gravity ( bs stars ) , 35% were classified as bhbs in the photometric technique . these numbers for completeness and contamination are similar to estimates that we made from examination of figure 13 from @xcite , and apply only to stars with @xmath32 fainter than this , the photometric errors in @xmath33 will cause increased mixing of the populations . we searched our bhb catalog for stars that clustered in velocity , apparent magnitude , and galactic coordinates . our search was less systematic than that used by @xcite , but allows us to find moving groups that are more extended , such as tidal streams . because they first searched for pairs of co - moving stars that were within @xmath34 kpc of each other , they reduced their sensitivity to tidal debris streams that could be spread over tens of kpc . we first divided the catalog into sky areas in @xmath35 that were @xmath36 by @xmath37 in the respective coordinates , and made plots of @xmath38 vs. @xmath39 magnitude . we used @xmath40 so that each partition of the data would cover an equal area of the sky . here , @xmath38 refers to the line - of - sight velocity transformed to the galactic standard of rest using the solar motion of @xmath41 km s@xmath3 @xcite . by examining the line - of - sight velocity as a function of magnitude , we found a number of co - moving groups of stars that were associated with known globular clusters , dwarf galaxies , and tidal streams . halo substructures that had been previously identified were not studied further . we concentrated only on the most significant of the remaining bhb association ; others were left for future studies . figure 1 shows the @xmath38 vs. @xmath39 plot for the most significant co - moving group of bhb stars , along with the distribution of these stars in @xmath42 . although there is a second cluster of stars at @xmath43 in the top left plot , we show in the lower left plot that these stars are dispersed in position on the sky so we did not explore that association further . the moving group we describe here has bhb stars at @xmath1 , @xmath44 ; and @xmath4 km s@xmath3 , @xmath45 km s@xmath3 . the dispersion is consistent with the instrumental error in the velocity of 19@xmath46 magnitude bhb stars @xcite . these stars have an average heliocentric radial velocity of @xmath47 km s@xmath3 , which rules out any association of these stars with the disc population . in the direction @xmath48 , the line - of - sight , heliocentric velocities expected for the thin disc , thick disc , and spheroid components are @xmath49 , @xmath50 , and @xmath51 km s@xmath3 , respectively . the heliocentric radial velocity of our moving group is not far from that of the average spheroid star in that direction ( @xmath52 for the spheroid and @xmath4 km s@xmath3 for the moving group ) , but the dispersion rules out an association with the general spheroid population . the lower right panel of figure 1 shows that the association of horizontal branch stars could not be detected spatially ; velocity information was required to identify the moving group . in the same area , at @xmath53 , another cluster identified from sdss bhb stars was found by @xcite . although this cluster lies at roughly the same ( l , b ) angular position as the cluster identified in this paper , the estimated distance to the clewley et al . clump is 7.8 kpc , which is much closer to the sun than our moving group , with a horizontal branch brighter than @xmath29 . the parameters of the seven bhb stars in the moving group are presented in table 1 . the new moving group found in this study is located at @xmath54 . the stars have apparent magnitude in the @xmath55 filter of @xmath56 and velocities located within @xmath57 km s@xmath3 , with an average velocity of @xmath58 km s@xmath3 . a @xmath39 apparent magnitude of @xmath59 for bhb stars , with an estimated absolute magnitude of @xmath60 @xcite corresponds to a distance of @xmath61 kpc from the sun . the magnitude dispersion of the seven stars is @xmath62 . note that the dispersion in apparent magnitude for these stars is @xmath44 , and drops to @xmath63 if the two outliers with larger magnitudes ( one of which has a high estimated surface gravity ) are excluded . the magnitudes are consistent with a structure that has zero depth , but the depth could be as large as @xmath64 kpc . [ cols= " < , > , > , > , > , > , > , > , > , > , > , > " , ] @xmath65 highly likely member of moving group the sdss dr7 sspp pipeline @xcite generates ten different measures of the surface gravity for each stellar spectrum , and additionally computes an adopted " @xmath66 , which is determined from an analysis of the ten different methods . for stars with s / n less than 10 , the adopted @xmath66 is set to an error code . we analyzed histograms of the distribution in @xmath66 of bright and faint a stars in our sample , looking for a method that showed two peaks : one at low surface gravity ( bhb stars ) , and one at high surface gravity ( bs or main sequence stars ) . several of the methods appeared to separate bright a stars by surface gravity , and none appeared to separate the stars with s / n@xmath67 very well . we selected the logg9 parameter , which is described in general terms in @xcite , in which it is referred to as the k24 grid " . the original reference to the method is @xcite , where the method is described in great detail . the method finds the best match between the sdss spectrum and a set of model atmospheres from @xcite , using the @xmath68 color index and the normalized spectral fluxes in the wavelength range @xmath69 , using a resolving power of @xmath70 . the surface gravities using this dr7 logg9 estimator are tabulated in table 1 . six of the seven candidate members of the moving group have surface gravities less than @xmath71 , as expected for bhb stars . subsequent to our search for spheroid structure using bhb stars , @xcite published a list of sdss dr6 a stars along with measurements of @xmath72 and @xmath73 measurements . these are classical indices used by @xcite to classify bhb stars from the @xmath74 line width and depth . using this method for determining the surface gravities of a stars , only 2 of our candidate a stars appear to be low surface gravity . there are two effects that contribute to the small fraction of confirmed bhbs : ( 1 ) the performance of the @xmath72 indicator is degraded at low signal - to - noise , and ( 2 ) we disagree with the published @xmath72 values for two of the stars . in the top panels of figure 2 , we show the performance of the @xmath72 , @xmath73 indicators for the set of @xmath75 stars from @xcite , and for the subset of 1176 of these stars for which the s / n of the spectrum is between 7 and 10 . while the bright , high signal - to - noise spectra that were used by @xcite are separable , the lower signal - to - noise spectra do not have two clear peaks in the diagram . we show histograms of @xmath72 for all of the stars between the two vertical lines ( @xmath76 ) in the top two panels of figure 3 . for all of the stars ( most of which are high signal - to - noise ) , we see two peaks in the @xmath72 distribution . the low s / n sample shows no separation . we compare the separation using the properties of one line of the spectrum ( @xmath77 vs. @xmath73 ) with the sspp logg9 surface gravities using photometry and the continuum levels of the spectra in the bottom panels of figures 2 and 3 . figure 2 shows that the logg9 indicator works about as well as the @xmath72 indicator for bright stars . note that in the lower right panel , for stars with marginal s / n , the distribution of stars is much wider in the vertical direction , and the width in the horizontal axis is wider but not dramatically so . the lower panels of figure 3 show histograms of logg9 for all of the stars between the two horizontal lines ( @xmath78 ) . the brighter star data on the left shows two separate peaks for low and high surface gravity stars , while the marginal s / n data in the right panel does not show that separation however , we can see from the correlation between @xmath72 and logg9 in the lower right panel of figure 2 that there is some information on the surface gravity of even these lower s / n spectra . because we were surprised that the @xmath72 numbers could be so high , even at low signal - to - noise , for stars that we expected were bhbs in a moving group , we independently computed new values for the width of the @xmath74 at 80% of the continuum for the eight stars in table 1 . for six of them , we measured widths that were reasonably consistent with the published values . however , we measured a width of 23 for 1335 - 52824 - 167 instead of 40 , and a width of 23 for 1335 - 52824 - 008 instead of 33 . with these corrections , six of the eight stars in table 1 have @xmath72 widths less than 30 , including all four of the stars that we will later show are highly likely members of the moving group . given that we were searching through the data by eye and looking for clusters , it is important to verify that the clustering of stars is not a chance coincidence . in this section we will estimate an upper bound for the probability that this clump could be a random fluctuation in the number of stars in a a small region of angle in the sky , magnitude , and line - of - sight velocity . most of the sky that we searched for clumps was at high galactic latitude ; at low galactic latitude the stellar population is different , the density of stars is higher , and the sky coverage is less uniform . so for statistical purposes we limited our sample to @xmath79 . there are 2893 stars in our sample with @xmath79 . the seven stars were discovered in a 19 square degree area of the sky ( compared with 8800 square degrees above @xmath79 ) , with the requirement that this sky area be rectangular in galactic latitude and longitude ( @xmath80 ) . in comparison with the entire high latitude sample of bhb candidates , the seven stars in the clump span about 1/46@xmath46 of the entire longitude range of the input data , and 1/10@xmath46 of the latitude range of the data in an equal - area plot . from figure 1 , we find that the range of velocities in the clump includes @xmath81 of the sample stars , and the range of magnitudes in the clump is the same as @xmath82 of those in the high latitude sample . appendix a presents a statistical method to determine the significance of a clump " of data points in a @xmath83-dimensional search space . in this case we have four dimensions : galactic longitude , galactic latitude , radial velocity , and apparent magnitude . the statistical method measures the probability that random data will produce a clump of seven or more points within a four - dimensional box of the measured size . by using percentiles , we account for the non - uniformity of the data in the radial velocity and apparent magnitude dimensions . we also adjust the statistics , as explained in appendix a.4.2 , for the periodic boundary condition in galactic longitude . what is important for the calculation is that each of the dimensions be independent ( separable , in the terms of appendix a ) . in our case , the variables are not quite independent , especially galactic coordinates . because the density of a stars with sdss spectra varies as a function of position in the sky , the relative percentiles in @xmath42 would be different if a clump of the same areal extent were discovered in a different part of the sky . we estimate that the local density in the region of the sky that the clump was found is 3.5 times higher than average . therefore , following the prescription of appendix a4.4 , we multiplied the longitude width of the clump by a factor of 3.5 . it was not possible to account for every known peculiarity of the data . for instance , when we searched for clumps we divided the sky up into fixed @xmath42 boxes , so there are regions around the edge of each @xmath42 box that were not as thoroughly searched . however , it is possible that if nothing was found with these limits we would have adjusted the box sizes and positions , so this oversight ( which would underestimate the significance of a clump ) is probably justified . given the percentiles of the 4-d search space listed above , we expect 0.32 stars in a typical box with these dimensions . using the algorithm in appendix a , the number of clumps we should expect to find in this data that have seven or more stars in a 4-d box of these dimensions or smaller is 0.96 . since we found one clump , this is very close to random . if we eliminate one star from the sample that is an outlier in @xmath42 , so that the area of sky is now @xmath84 , @xmath85 , then we calculate that the e - value for finding six or more stars in a 4-d box with these ( smaller ) dimensions is 0.18 , so the probability of finding this moving group is not more than one in six . using only these dimensions , the clump of stars is not a significant detection . in order to claim a significant detection of the moving group represented by these seven stars , it is necessary to compare their metallicities with the metallicities of the background population . only 30% of the stars in the sample have metallicities as low as those in the clump of seven stars . if we extend the statistical calculation to five dimensions , including metallicity , then the probability of finding seven stars in a region of this size and shape ( or smaller ) , anywhere in the 5-d parameter space is not more than 1 in 244 ( e - value of 0.004 ) , and the probability of finding six stars in a smaller angular region in the sky ( in 5-d ) is not more than 1 in 490 ( e - value=0.002 ) . these are highly significant detections . since metallicity is critical to the detection of this substructure , we examine the sdss dr7 metallicity measurements of these stars , and bhb stars in general , in the next two sections . in section 8 , we use the information we have learned in the analysis of this moving group to re - select similar data from sdss dr7 , and make an even stronger argument for the statistical significance of our result . all seven of the candidate bhb stars in this moving group had unmeasured metallicities in sdss dr6 , but metallicities were assigned to these same spectra in the much improved sspp pipeline of sdss dr7 @xcite . our metallicities come from the dr7 , which became public in october 2008 . the sdss dr7 tabulates many measures of the metallicities of stars , of which the most commonly used is feha , the adopted " metallicity , which is derived from a comparison of all methods used to measure metallicity . in this paper we use instead the metallicity of @xcite , hereafter wbg , which is specifically designed to measure the metallicities of bhb stars . the wbg metallicities of the seven bhb stars in the moving group are given in table 1 and are shown in figure 4 . the mean of the distribution is [ fe / h]@xmath86 , and the width is @xmath87 . we now show that the metallicities of the moving group stars are not consistent with being drawn at random from the stellar halo . we selected all stars in dr7 with the same photometric constraints as the stars in the moving group , and with @xmath88 so that the bhb stars are likely to come from the same component ( the spheroid ) , and not be confused with disc populations of bhbs . to remove bhb stars in the sagittarius dwarf tidal stream from the sample , we also eliminated stars with @xmath89 . there were 298 stars with @xmath90 , @xmath18 , @xmath19 , and that passed the photometric low surface gravity cut . of these , 260 had measured metallicities that are shown in figure 4 . the mean of the distribution is [ fe / h]@xmath91 , and the width is @xmath92 . note that the error quoted here is a statistical error . because very few of the segue calibration stars are as blue as a stars , the systematic errors at this end are about @xmath93 . the comparison stars are about 45 kpc from the sun , and above @xmath94 , so they come from the distant halo . the blue straggler contaminants are 10 to 28 kpc away , which is also fairly far from the disc at high galactic latitudes . a comparison of the stars in the moving group with the similarly selected stars in the spheroid with a t - test gives a probability of 1 in @xmath95 that the two groups of stars were selected from the same stellar population . the mann - whitney test ( also known as the wilcoxon rank sum test ) gives a p - value of 0.00023 , or a 2 in 10,000 probability that the two samples are drawn from the same population . this leads us to conclude that the moving group is real . as a further test , we selected the other 8 stars in the lower panel of figure 1 that have the same @xmath38 and apparent magnitude as the blue horizontal branch stars in the moving group , and find that they have a mean metallicity of @xmath96 with a sigma of 0.86 . the t - test ( p=0.948 ) shows that this distribution is not distinguishable from the background population . ( performing the mann - whitney test for this sample is problematic since several of the stars are not part of our selected background distribution . ) in this section we explore the accuracies of the dr7 metallicity measurements of bhb stars as a function of s / n and in comparison with published metallicities for globular clusters . we also explore the metallicity of bhb stars in the halo as a function of apparent magnitude . since all of our moving group stars have low s / n , we need to test the accuracy of the metallicity measurements . we do this by selecting bhb stars in the leading tidal tail of the sgr dwarf galaxy that are about the same distance from the sun as the moving group . we selected all of the spectra from sdss dr7 that had elodierverr@xmath97 ( if the radial velocity error when matching to the stellar templates is less than zero , that indicates that the object is not a star ) , @xmath98 , @xmath99 , @xmath100 , @xmath101 , @xmath102 km s@xmath3 , logg9@xmath103 , and @xmath104 . the 94 stars that are likely bhbs in the sgr dwarf tidal stream were divided into three files based on s / n . there were 14 stars with @xmath105s / n@xmath106 , 49 stars with @xmath107s / n@xmath67 ( excluding three with unmeasured metallicities ) , and 28 with @xmath108s / n@xmath109 . the metallicity distributions of these three s / n groups is shown in figure 5 . the highest s / n group ( @xmath108s / n@xmath109 ) has a mean [ fe / h]@xmath110 , which is similar to the group with marginal s / n ( @xmath107s / n@xmath67 ) which has a mean [ fe / h@xmath111 . a mann - whittney test comparing these two samples gives a result of @xmath112 , @xmath113 , and a p - value of 0.165 . these distributions are not significantly different from each other . this gives us some confidence that the metallicities of the moving group stars , which have s / n measurements in this marginal range , are usefully measured . in contrast , the lowest s / n group ( @xmath114s / n@xmath106 ) has a mean of [ fe / h]@xmath115 . a mann - whittney test comparing this sample with the highest s / n sample produces @xmath116 , @xmath117 , and a p - value of 0.0349 . this is a statistically significant difference , and metallicities of bhb stars measured by this technique are probably not accurate . to understand the likely systematics in our metallicity determinations , we selected bhb stars from six globular clusters that had spectroscopic measurements of bhb stars in sdss dr7 . we selected stars that have the colors of bhb stars , are within half a degree of the centre of the globular cluster , and have radial velocities within 20 km s@xmath3 of the published value for the cluster . m2 , m3 , and m13 have metallicities very close to @xmath118 ; m53 has metallicity of @xmath119 ; and the two clusters m92 and ngc 5053 have metallicities very close to @xmath120 @xcite . because there were very few stars with spectroscopy in each globular cluster , we combined the data from clusters with similar metallicities before plotting the histograms in figure 6 . all of these globular clusters have horizontal branches near @xmath29 ( the magnitude range for which we have the most complete sample of bhb stars ) , and thus the spectra have high @xmath121 . = @xmath122 ) , m3 ( [ fe / h ] = @xmath123 ) , and m13 ( [ fe / h ] = @xmath124 ) . second panel : stars in m53 ( [ fe / h ] = @xmath119 ) . third panel : stars in m92 ( [ fe / h ] = @xmath125 ) and ngc 5053 ( [ fe / h ] = @xmath126 ) . lowermost panel : the seven hercules stream bhbs . note that the measured metallicities are compressed toward [ fe / h]=@xmath17 ( higher metallicity stars are measured systematically low , and lower metallicity stars are measured systematically high ) . also note that the hercules stream stars appear to be lower metallicity than m92 and ngc 5053 . ] figure 6 shows that the sdss dr7 wbg metallicities for bhb stars tend to push the metallicities toward [ fe / h]=@xmath17 . the measured metallicities are correlated with the published metallicities of the globular clusters ; however , stars that have higher metallicities are measured with metallicities that are too low , and stars with lower metallicities are measured systematically too high . we show for comparison the measured metallicities of the seven stream stars . the measured metallicity is lower than that of m92 and ngc 5053 . because the validity of our stream detection depends on our ability to separate the stream from the stellar halo in metallicity , we need to show that the metallicities in our background population are accurate . we selected all of the stars with spectra in dr7 that have photometry consistent with being a bhb star , as explained in 2 , @xmath88 , and @xmath127 . we further restricted the sample by insisting that the wbg estimate of @xmath66 , as measured in dr7 , was less than 3.75 . in figure 7 , we show the dr7 wbg metallicity as a function of apparent magnitude . we have spectra for stars with @xmath128 , which span distances of 6 to 50 kpc from the sun , all at high galactic latitude . the metallicity distributions for all of the stars brighter than @xmath129 are similar , with a mean near [ fe / h]=@xmath17 and a sigma of 0.45 . in the faintest set of stars , which have apparent magnitudes similar to that of the newly detected stream , the distribution appears considerably broader , but with a similar mean . we attribute the increased width to the lower s / n in many of these spectra . the mean is somewhat lower in the last panel in figure 7 than in figure 4 , probably due to a cleaner sample of bhb stars , with less contamination from bs , from the additional restriction in @xmath66 . . they are inconsistent with being drawn from the same distribution as the other stars in this panel . all of the stars have @xmath130 colors consistent with bhb stars , @xmath88 , @xmath127 , and @xmath131 . the [ fe / h ] and @xmath66 measurements are from the wbg @xcite techniques , as implemented in the sspp of dr7 . a gaussian with mean @xmath17 and sigma 0.45 , normalized to the number of stars in each panel , is shown for reference . the mean metallicity does not change as a function of distance from the sun . the distribution is wider in the lowest panel because the signal - to - noise is lower here . if one selects only the highest s / n spectra in this bin , the width is similar to the four other panels . ] we show for comparison the six stream stars that have @xmath132 in the last panel of figure 7 . the metallicity distribution of these stars is inconsistent with the distribution of the other stars in this figure at the 98% confidence level , as determined by the mann - whitney test . the value that we find for the metallicity of the outer halo , as measured from sdss bhb stars is on the low side of the distribution of previous measurements , but higher than the @xcite metallicity measurement of [ fe / h]=@xmath14 . ( we note , however that carollo et al . 2007 distribution of bhb stars in their paper , and the bhb star distributions peak closer to [ fe / h]=@xmath6 . ) the moving group was originally discovered by searching through data from sdss dr6 . in this section we show the moving group as selected from the now public sdss dr7 databases , and using our knowledge of metallicity and surface gravity determination at low s / n , as learned in the production of this paper . we selected from sdss dr7 all of the spectra with elodierverr @xmath97 ( which selects stellar spectra ) , magnitude @xmath133 , the colors of an a star [ @xmath134 , and a flag that indicates the spectrum is not from a segue plate . we did not select stars from segue plates since most of them have not been searched for moving groups of a stars , and because their non - uniform sky coverage , different section criterion , and different exposure lengths make it difficult to use them in a statistical calculation . we show the distribution of sdss dr7 a stars in equatorial coordinates in figure 8 . we then selected the low surface gravity ( logg9 @xmath135 ) stars with @xmath136 and @xmath137 , to match the region of the sky presented in figure 1 , and metallicity [ fe / h]@xmath138 , since we know we are looking for a low metallicity moving group . figure 9 shows plots like those used in the moving group discovery , but using data from this low metallicity dr7 dataset . from the upper panel , we identify a clump of stars with @xmath139 km s@xmath3 and @xmath140 . the lower panel shows the positions in galactic coordinates of the same objects from the top panel , with filled circles showing the six objects that are clumped in line - of - sight velocity and apparent magnitude . four of these objects are very tightly aligned in angular position on the sky . to get a sense for how common a moving group with these characteristics is , we selected from our original dr7 dataset all of the stars with surface gravity , metallicity , line - of - sight velocity , and apparent magnitude that are similar to the stars in the clump . in the entire northern sky portion of the sdss , there are twelve stars with these characteristics . four of the stars are in a tight line on the sky , and a fifth is only a few degrees away from the clump along the same line . because this fifth star is a candidate member of the moving group , and not included in the original list of seven , we have added it to the bottom of the list of candidate members in table 1 . the four highly probable members of the moving group , which were selected from both the dr6 and dr7 datasets , are marked in table 1 with a footnote . the four highly likely moving group members are a subset of the original seven , and have velocity and metallicity characteristics that are nearly identical to our previous results . the average @xmath141 is @xmath142 km s@xmath3 , with a sigma of @xmath143 km s@xmath3 . the average metallicity is [ fe / h]@xmath144 with a sigma of 0.39 . following a similar procedure to that used in section 5 , we now calculate the probability that this group of four stars is a random coincidence . there are 6118 stars with @xmath79 in the a star sample depicted in figure 8 . of these stars , 3335 have @xmath145 ( 54.5% ) , 1171 have @xmath146 [ fe / h]@xmath138 ( 28.9% ) , 224 have @xmath140 ( 3.7% ) , and 487 have @xmath147 km s@xmath3 ( 8.1% ) . the four stars in the clump have sky positions within @xmath148 ( 0.8% of longitude range ) and @xmath149 ( 4.2% of area - corrected latitude range ) . because the density of the stars in figure 8 is not uniform , we need to use a multiplier on the fraction of stars in the longitude range , since that is fractionally the smallest dimension . in the 3.0 square degree area that contains the four stars , there are 8 stars in the original sample of 6118 . so , in the region of the clump there are 2.7 stars per square degree , while there are @xmath150 stars per square degree in the whole sky region , a factor of 3.9 . using the procedure outlined in appendix a , we calculate that we would expect 0.0038 stars in a 6-d region that instead has four stars . the expected number of clumps of four stars is 0.0238 . the p - value is therefore less than 0.024 ( @xmath151 ) . since @xmath152 , this is a significant detection . note that this calculation is extremely conservative . first , the calculation assumed that we searched all 6118 stars for clumps , when in fact we only searched the ones that had low surface gravity , as determined by the photometric indices . second , the four stars are not randomly distributed within the @xmath42 box , they are tightly constrained to a line , as one would expect for a tidally disrupted globular cluster ; if our statistical analysis allowed a non - axis aligned region , the computed statistical significance would have been stronger . and third , when we calculated the density of stars in the region containing the moving group we included the moving group stars ( half of the sample of eight ) in the density calculation . this artificially inflates our galactic latitude multiplier . to estimate the effect of just this third factor , we counted the number of a stars in a slightly larger region of the sky with @xmath153 and @xmath149 and found 16 stars , of which 4 are members of the moving group . this gives us a local density of @xmath154 stars per square degree , which is a factor of 2.5 times the average star density . using this multiplier , we expect to find 0.0025 objects where we actually find four , and the expected fraction of the time a clump of four stars in this size region of 6-d space should be found in this size data set is not more than 0.0063 . we estimate the properties of a progenitor of this moving group , making the assumption that it was a globular cluster , and that we have detected the majority of the bhb stars from what was the core of the star cluster . we expect the progenitor was a globular cluster because the stars are very well aligned in the sky , have a velocity dispersion consistent with instrumental errors , and there are very few stars identified . if the progenitor was a dwarf galaxy , so that it also has younger or more metal rich components , or if we have found only a knot of increased stellar density along a longer tidal debris stream , then the inferred size of the progenitor is larger . we detected 4 - 8 bhb stars in the moving group . in the color - magnitude range in the vicinity of the moving group , we have spectra of 50 of the 73 candidate bhbs ( 68% ) . therefore , we expect that there are @xmath155 bhbs in the moving group . found five bhb stars in the globular cluster pal 13 ( [ fe / h]=@xmath156 , @xmath157 , harris 1996 ) . @xcite find at least five bhb stars in the globular cluster pal 5 ( [ fe / h]=@xmath158 , @xmath159 , harris 1996 ) . the moving group is consistent with a progenitor that is like one of the smaller globular clusters of the milky way , with a total integrated luminosity of about @xmath160 . to understand why this moving group could not be identified from photometry alone , we estimated the number of red giant and subgiant branch stars in the moving group by comparison with ngc 2419 . note that at 50 kpc , turnoff stars are at @xmath161 , which is close to the limiting magnitude of the sdss , so deeper photometry is required to clearly detect significant density enhancements expected from turnoff and main sequence stars fainter than these limits . we selected sdss dr6 stars within about nine arcminutes of the centre of ngc 2419 ( this does not include stars near the very centre of the gc , since individual stars are not resolved there ) . bhb stars were selected with @xmath162 and @xmath163 . giant stars were selected within a parallelogram with vertices @xmath164=[0.5,20 ] , [ 0.9 , 18 ] , [ 0.7 , 20 ] , [ 1.0 , 18]$ ] . subgiant stars were selected within a triangular area with vertices @xmath165= [ 0.5 , 21 ] , [ 0.1 , 23 ] , [ 0.6 , 23]$ ] . we found 101 bhbs , 98 red giants , and 302 subgiants over background in the region of sky that was searched . therefore , we expect about the same number of red giant stars as bhbs in our moving group , and about three times as many subgiants . we shifted the magnitudes of the color - magnitude boxes by 1.41 magnitudes ( the difference in distance modulus between our moving group and ngc 2419 ) and counted the number of stars in @xmath166 and @xmath167 . the moving group is expected to have 10 of 73 candidate bhb stars ( 1 sigma fluctuation ) , 10 of 2742 candidate giant stars ( 0.2 sigma ) , and 30 of 13,182 candidate subgiant stars ( 0.3 sigma ) . the background star counts are much too high for us to find this moving group in photometry alone . we have detected a moving group of at least four bhb stars in the corner ( toe " ) of the hercules constellation spilling into corona borealis . these stars are coincident in angular position [ @xmath168 ) ] , apparent magnitude ( @xmath1 ) , line - of - sight velocity ( @xmath4 km s@xmath3 , @xmath169 km s@xmath3 , @xmath170 km s@xmath3 ) , and metallicity ( [ fe / h]=@xmath171 ) . we expect that the progenitor of this moving group was a low metallicity globular cluster , with a luminosity like that of one of the smaller globular clusters in the milky way halo . we show that useful surface gravities and metallicities are measured for bhb stars with s / n@xmath172 in sdss dr7 . the mean metallicity of bhb stars in the outer halo is similar to m53 , which has a published metallicity of [ fe / h]=@xmath6 . the metallicity does not appear to change with distance from the sun ( @xmath173 kpc ) . our measurement of the spheroid metallicity is slightly higher than claimed by @xcite and somewhat lower than earlier studies of outer halo stars . the hercules moving group is one of many tidally disrupted stellar associations expected to comprise the spheroid of the milky way and could not have been identified from photometry alone ; more complete spectroscopic surveys are required to identify the component spheroid moving groups , and determine the merger history of our galaxy . we present a statistical technique that allows us to estimate the significance of clumps discovered in multidimensional data . this project was funded by the national science foundation under grant number ast 06 - 07618 . t.c.b and y.s.l . acknowledge partial support from grans phy 02 - 16783 and phy 08 - 22648 , physics frontier centers / jina : join institute for nuclear astrophysics , awarded by the national science foundation . p.r.f . acknowledges support through the marie curie research training network elsa ( european leadership in space astroometry ) under contract mrtn - ct-2006 - 033481 . many thanks to ron wilhelm , who answered our questions about potential rr lyrae stars and stellar metallicities . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the sdss web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , university of cambridge , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . 99 abazajian k. et al . , 2003 , aj , 126 , 2081 adelman - mccarthy j. k. et al . , 2008 , apjs , 175 , 297 allende prieto , c. et al . 2006 , apj , 636 , 804 an , d. , et al . 2009 , apjl , 707 , l64 belokurov , v. , et al . 2009 , mnras , 397 , 1748 belokurov v. et al . , 2007 , apjl , 657 , l89 belokurov v. et al . , 2007 , apj , 654 , 897 belokurov v. et al . , 2006 , apjl , 647 , l111 belokurov v. et al . , 2006 , apjl , 642 , l137 bok b. j. , 1934 , harvard college observatory circular , 384 , 1 borissova j. , markov h. , & spassova n. , 1997 , a&as , 121 , 499 carney b. w. , aguilar l. , latham d. w. , & laird j. b. , 1990 , aj , 99 , 201 carollo , d. et al . 2010 , apj , in press , arxiv:0909.3019 carollo d. et al . , 2007 , nature , 450 , 1020 clewley l. , & kinman t. d. , 2006 , mnras , 371 , l11 dehnen w. & binney j.j . , 1998 , mnras , 298 , 387 eggen o. j. , 1958 , mnras , 118 , 65 famaey b. , siebert a. , & jorissen a. , 2008 , a&a , 483 , 453 freeman k. c. , 1987 , ara&a , 25 , 603 fukugita m. , ichikawa t. , gunn j. e. , doi m. , shimasaku k. , & schneider d. p. , 1996 , aj , 111 , 1748 gilmore g. , wyse r. f. g. & kuijken k. , 1989 , ara&a 27 , 555 grillmair c. j. , 2006 , apjl , 645 , l37 grillmair c. j. , & dionatos o. , 2006 , apjl , 643 , l17 grillmair c. j. , & johnson r. , 2006 , apjl , 639 , l17 gunn j. e. et al . , 1998 , aj , 116 , 3040 gunn j. e. et al . , 2006 , aj , 131 , 2332 harris w. e. , 1996 , aj , 112 , 1487 helmi a. , white s. d. m. , & springel v. , 2003 , mnras , 339 , 834 hogg d.w . , finkbeiner d.p . , schlegel d.j . , & gunn j.e . , 2001 , aj , 122 , 2129 irwin m. j. et al . , 2007 , apjl , 656 , l13 ivezi . et al . , 2004 , astronomische nachrichten , 325 , 583 juri m. et al . , 2008 , apj , 673 , 864 koposov s. e. , yoo j. , rix h .- w . , weinberg d. h. , macci a. v. , & escud j. m. , 2009 , apj , 696 , 2179 klement , r. , et al . 2009 , apj , 698 , 865 kurucz , r. l. 1993 , kurucz cd - rom 13 , atlas9 stellar atmosphere programs and 2 km / s grid ( cambridge , ma : sao ) lee , y. s. et al . 2008 , aj , 136 , 2050 lee , y. s. et al . 2008 , aj , 136 , 2022 majewski s. r. , skrutskie m. f. , weinberg m. d. , & ostheimer j. c. , 2003 , apj , 599 , 1082 majewski s. r. , ostheimer j. c. , rocha - pinto h. j. , patterson r. j. , guhathakurta p. , & reitzel d. , 2004 , apj , 615 , 738 majewski s. r. , munn j. a. & hawley s. l. , 1996 , apjl , 459 , l73 morrison h. l. et al . , 2003 , aj , 125 , 2502 newberg , h. j. , yanny , b. , & willett , b. a. 2009 , apjl , 700 , l61 newberg , h. j. , yanny , b. , cole , n. , beers , t. c. , re fiorentin , p. , schneider , d. p. , & wilhelm , r. 2007 , apj , 668 , 221 newberg h. j. , & yanny b. , 2005 , astrometry in the age of the next generation of large telescopes , asp conf . ser . , 338 , 210 newberg h. j. et al . , 2002 , apj , 569 , 245 norris j. , 1986 , apjs , 61 , 667 odenkirchen m. et al . , 2003 , aj , 126 , 2385 padmanabhan n. et al . , 2008 , apj , 674 , 1217 pier j. r. , munn j. a. , hindsley r. b. , hennessy g. s. , kent s. m. , lupton r. h. , & ivezi . , 2003 , aj , 125 , 1559 pier , j. r. 1983 , apjs , 53 , 791 robin a. c. , reyl c. , & crz m. , 2000 , a&a , 359 , 103 saha a. , 1985 , apj , 289 , 310 schlaufman k. c. et al . , 2009 , apj , in preparation schlegel d. j. , finkbeiner d. p. , & davis m. , 1998 , apj , 500 , 525 searle l. , & zinn r. , 1978 , apj , 225 , 357 sirko e. et al . , 2004 , aj , 127 , 899 smith , m. c. , et al . 2009 , mnras , 399 , 1223 smith j. a. et al . , 2002 , aj , 123 , 2121 stoughton c. et al . , 2002 , aj , 123 , 485 suntzeff n. b. , kinman t. d. , & kraft r. p. , 1991 , apj , 367 , 528 tucker d. l. et al . , 2006 , astronomische nachrichten , 327 , 821 vivas a. k. et al . , 2001 , apjl , 554 , l33 walsh , s. m. , jerjen , h. , & willman , b. 2007 , apjl , 662 , l83 wilhelm r. , beers t. c. , & gray r. o. , 1999 , aj , 117 , 2308 ( wbg ) . willman b. et al . , 2005 , apjl , 626 , l85 willman b. et al . , 2005 , aj , 129 , 2692 xue , x. x. , et al . 2008 , apj , 684 , 1143 yanny b. et al . , 2009 , aj , 137 , 4377 yanny b. et al . , 2003 , apj , 588 , 824 yanny b. , newberg h. j. , et al . , 2000 , apj , 540 , 825 york d. g. et al . , 2000 , aj , 120 , 1579 zinn r. , 1985 , apj , 293 , 424 zucker d. b. et al . , 2006 , apjl , 650 , l41 zucker d. b. et al . , 2006 , apjl , 643 , l103 for this paper , we needed to estimate the probability of finding seven bhb stars in a restricted portion of @xmath174 parameter space . in this appendix , we present a solution to the more general problem of determining the probability of finding @xmath175 of @xmath176 data points in a small portion of a @xmath83-dimensional space . suppose one analyses of a set of @xmath176 data points in a large @xmath83-dimensional hypercube @xmath177 \times \ldots \times [ a_d , b_d]$ ] , and finds a small , axis - aligned , @xmath83-dimensional hypercube @xmath178 \times \ldots \times [ a_d , b_d]$ ] that has @xmath179 data points in it . the natural question is whether this `` clump '' of @xmath175 data points is statistically significant , or whether even random data would yield such a clump . here we precisely estimate an @xmath180-value for the expected number of clumps , thus providing an upper bound to the probability that random data would somewhere have @xmath175 or more data points within an axis - aligned , @xmath83-dimensional hypercube with dimensions @xmath181 . we start with the simple case where a full data set of @xmath176 points lies within a @xmath83-dimensional unit hypercube @xmath182^d$ ] . we further assume that a random ( or null ) model would distribute these @xmath176 points uniformly within the hypercube . additionally , we assume that the hypercube is not periodic / toroidal . we relax these simplifying assumptions in subsequent sections . we say that an axis - aligned , @xmath83-dimensional hypercube contained within @xmath182^d$ ] is a _ _ . we say that a set of @xmath179 of the @xmath176 points is a _ _ if there exists a box that contains the @xmath175 points without containing any of the remaining @xmath183 points . we define the _ _ for a @xmath175-boxed set as the intersection of all boxes that contain the @xmath175 points . a box of dimensions @xmath184 or smaller with @xmath175 or more points exists only if there is a @xmath175-boxed set that has a minimal box with dimensions not greater than @xmath184 . thus , we aim to compute the expected number of the latter . we use as an integrand the probability density that a box will be a minimal box for a set of @xmath175 points , and we integrate over all applicable boxes . consider a box @xmath185 \times \ldots \times [ f_d , g_d]$ ] with dimensions @xmath186 and hypervolume @xmath187 it will be a ( non - minimal ) box indicating that a set of @xmath175 points is a @xmath175-boxed set if exactly @xmath175 points fall within it . thus the probability that it indicates a @xmath175-boxed set is @xmath188 however , it is our goal to calculate the probability density for a _ minimal _ box for @xmath175 points . this box is minimal only if , for every dimension index @xmath189 , we have that @xmath190 is the minimum of the @xmath175 points @xmath189th coordinate values and @xmath191 is the maximum of these coordinate values . thus , for each dimension , we must have one of the @xmath175 points achieve the minimum and one of the remaining @xmath192 points achieve the maximum . the remaining @xmath193 coordinate values can be anywhere in the range @xmath194 . ( note that the possibility of having multiple points exactly achieve the minimum or maximum is an event that has a probability density of zero and thus is safely ignored . ) thus , the @xmath189th dimension contributes a factor of @xmath195 to the probability density . the probability density for the event that the box is a minimal box for a @xmath175-boxed set is @xmath196 = \left(\!\!\!\!\begin{array}{c}n\\k\end{array}\!\!\!\!\right ) k^d ( k-1)^d v^{k-2 } ( 1-v)^{n - k } \label{eqn : density } { \mbox{~.}}\ ] ] we note that a box with dimensions @xmath197 can have its minimum corner @xmath198 anywhere in @xmath199 \times \ldots \times [ 0 , 1-l_d]$ ] . when we restrict attention to minimal boxes with dimensions not greater than @xmath184 the expected number of @xmath175-boxed sets is @xmath200 . \label{eqn : evalue } { \mbox{~.}}\end{aligned}\ ] ] equation [ eqn : evalue ] provides an upper bound to the probability that random data will exhibit a clump of @xmath175 or more points in a box of dimensions @xmath184 or smaller , because such a box exists only if there is a @xmath175-boxed set that has a minimal box with dimensions not greater than @xmath184 . under many circumstances , when @xmath201 the probability of multiple @xmath175-boxed sets is small and the bound is quite tight . we observe that @xmath202 and , to speed the integration using maple 7 , we seek a way to truncate the sum at significantly fewer terms . we observe that @xmath203 is roughly a bell curve as a function of @xmath204 . the bell curve has mean and standard deviation given by @xmath205 where we have assumed that @xmath206 , _ i.e. _ , that a clump is under consideration only when the number of points it contains is more than the expected number . thus , we can include all terms of @xmath207 out to @xmath208 standard deviations with @xmath209 and the precise approximation @xmath210 we use this approximation in equation [ eqn : evalue ] and we evaluate the integral with maple 7 . in many circumstances , including the case presented in this paper , the set of @xmath176 data points does not fall uniformly into the unit hypercube under the random / null model . for instance , one does not expect the radial velocities of spheroid stars to be uniformly distributed in any velocity range , and they are certainly not limited to the range @xmath182 $ ] . we say that the density is _ _ if it can be written as the product of functions , each of which depends on a single dimension @xmath211 . for instance , the uniform spherical density , @xmath212 is separable . when a probability density for a coordinate vector @xmath213 \times \ldots \times [ a_d , b_d]$ ] is separable , we can perform a coordinate transformation to a coordinate vector @xmath214 that is uniformly distributed in the unit hypercube and which preserves axis - aligned hypercubes : @xmath215 the clumping question then becomes one of percentiles , _ i.e. _ , what is the probability that random data will show @xmath175 or more points in a axis - aligned hypercube with dimensions that do not exceed @xmath216 percentiles @xmath217 percentiles . in this paper , the radial velocities of the stars and the apparent @xmath39 magnitudes were treated as separable dimensions . in many circumstances , it is desirable to allow @xmath175-boxed sets to wrap around the boundaries in one or more dimensions . in this paper , it was desirable to allow a box to span the prime meridian in a longitude coordinate . in such situations the restriction that the minimum corner @xmath198 fall in @xmath199 \times \ldots \times [ 0 , 1-l_d]$ ] is relaxed . when the @xmath189th coordinate is periodic , @xmath190 can be any value in @xmath182 $ ] . equation [ eqn : evalue ] is thus modified by removal of the @xmath218 factor for each periodic dimension @xmath189 . in many circumstances , it is desirable to have the random / null model be poisson ; instead of requiring exactly @xmath176 points in the unit hypercube , the number of points @xmath219 is selected using a poisson distribution with mean @xmath176 : @xmath220 = e^{-n } \frac{n^n}{n ! } { \mbox{~.}}\ ] ] in our example , the poisson distribution applies if there are an average of 4149 bhb stars with spectra in the region @xmath88 , whereas the uniformly random model would put exactly 4149 bhb stars with spectra in that same region . in the poisson case , expression [ eqn : binomial ] becomes @xmath221 equation [ eqn : density ] becomes @xmath196 = \frac{n^{k}}{k ! } k^d ( k-1)^d v^{k-2 } e^{-n v } { \mbox{~,}}\label{eqn : poissondensity}\ ] ] and equation [ eqn : evalue ] is unmodified , except that it incorporates equation [ eqn : poissondensity ] instead of equation [ eqn : density ] . for integration , instead of truncating the series for @xmath207 , we truncate the series for @xmath222 we observe that @xmath223 is roughly a bell curve as a function of @xmath204 . the bell curve has mean and standard deviation given by @xmath224 thus , we can include all terms of @xmath225 out to @xmath226 standard deviations with @xmath209 and the precise approximation @xmath227 we use this approximation in equation [ eqn : poissondensity ] and we evaluate the integral of equation [ eqn : evalue ] with maple 7 . we determined that for this paper the distinction between random and poisson distributions was unimportant . when the dimensions are not separable , the assumption that the joint density function is the product of the individual dimension s density functions does not hold . however , if there is a set of individual density functions for which the assumption is approximately true , we define the _ _ to be the ratio of joint density function to the value implied by multiplying the approximating density functions together . the calculations for the separable case make a good approximation for the non - separable case if the percentile widths @xmath184 are adjusted by factors whose product is the overdensity . when the overdensity is greater than 1.0 , a conservative approach is to multiply the smallest of the widths of the non - separable dimensions by the overdensity .

the local quench of a fermi gas , giving rise to the fermi edge singularity and the anderson orthogonality catastrophe , is a rare example of an analytically tractable out of equilibrium problem in condensed matter . it describes the universal physics which occurs when a localized scattering potential is suddenly introduced in a fermi sea leading to a brutal disturbance of the quantum state . it has recently been proposed that the effect could be efficiently simulated in a controlled manner using the tunability of ultra - cold atoms . in this work , we analyze the quench problem in a gas of trapped ultra - cold fermions from a thermodynamic perspective using the full statistics of the so called _ work _ distribution . the statistics of work are shown to provide an accurate insight into the fundamental physics of the process . _ keywords _ : work distribution , non - equilibrium thermodynamics , quantum quench , orthogonality catastrophe .

1309.2669

in the past decade ultra - cold quantum gases have emerged as ideal candidates for clean and controllable simulation of condensed matter physics @xcite . ultra - cold quantum gases are now created in a variety of configurations in laboratories worldwide . in particular , both bosonic and fermionic atoms can be trapped and manipulated on optical lattice potentials @xcite . the lack of thermal phonons coupled with the tunability of the interactions by means of feschbach resonances @xcite has allowed for the detailed study of a multitude of phase diagrams , the most celebrated example is perhaps the bose - hubbard model @xcite . + equilibrium properties aside , over the past number of years there has been a surge in interest in the out of equilibrium behavior of closed quantum systems following a quench of a hamiltonian parameter . fundamentally , this is due to a series of spectacular experiments in ultra - cold atoms whereby the high degree of isolation and long coherence times permits the study of dynamics over long timescales @xcite . these experiments have raised a number of important theoretical issues such as the relationship between thermalisation and integrability and the universality of defect generation following evolution across a critical point @xcite . + given the controllability of ultra - cold atomic systems and the current interest in quench dynamics , it is a natural question to ask if there are any out of equilibrium condensed matter physics problems which could be simulated . unfortunately , due to their intrinsic complexity , there are very few examples of exactly solvable problems in the out - of - equilibrium domain . a worthy exception is the phenomenon of orthogonality catastrophe @xcite and the fermi - edge singularity . this problem was first pointed out by p. w. anderson over 40 years ago when he showed that the overlap of two many - body wave - functions , which describe deformed and undeformed fermi seas , vanishes in the thermodynamic limit @xcite . the corresponding ` quench ' problem , was investigated a few years later with the prediction of a universal absorption - edge singularity in the x - ray spectrum of metals , the ` fermi - edge ' singularity @xcite . + the universal physics of the anderson orthogonality catastrophe and the fermi - edge singularity was recently explored by goold _ _ in the context of ultra - cold quantum gases @xcite . in this work it was suggested that the physics maybe simulated in a controlled fashion by the appropriate embedding of a single probe qubit . the approach was further formalized in @xcite where the authors solved the dynamical problem in the inhomogeneous system by means of a linked cluster expansion . this qubit probe approach was further suggested as a mechanism to probe the physics of the orthogonality catastrophe in @xcite . + interestingly enough , a connection was made by heyl and kehrein @xcite between the absorption and emission spectrum in the original x - ray experiments and the so called quantum work distribution and corresponding fluctuation relations in classical and quantum statistical mechanics @xcite . treating a quench problem in manybody physics as a thermodynamic transformation and analyzing the statistics of work done has recently shown to be a useful approach to understand the intrinsic out of equilibrium dynamics in manybody systems @xcite . the approach is based on the study of the moments of a quantity known as the quantum work distribution @xcite which have been found to encode both thermodynamic and universal features of the model in question . + in this work this relationship will be explored in detail in the context of a locally quenched trapped fermi gas . in particular , in section [ sec100 ] the general problem of a system hamiltonian depending on a work - parameter will be introduced , adopting the description based on the grand canonical ensemble @xcite . the formalism provided by the work distribution and its characteristic function will be discussed , and the concept of irreversible work will be explored . in section [ sec200 ] the focus will be moved to a fermi gas in equilibrium with a harmonic trap , being suddenly perturbed by a spatially structure - less perturbation . in section [ lce ] the vacuum persistence amplitude and the linked cluster expansion will be used to reduce the calculation of the characteristic function of work to the sum of connected feynman diagrams . the relation of these diagrams to the characteristic function of work will be covered and an analytic approximation holding at low temperature will be presented . in section [ sec500 ] the first three cumulants of the work distribution will be computed and their link with thermodynamics will be discussed . finally , in section [ sec600 ] the irreversible work will be calculated using a perturbative and a numerical approaches , while section [ concluse ] will provide some further comments and conclusions . consider a system evolving according to the hamiltonian @xmath0 , which depends on some externally tunable parameter @xmath1 . the system is brought into weak contact with a heat reservoir , at inverse thermal energy @xmath2 , and allowed to equilibrate for each chosen value of @xmath1 , before the system - reservoir coupling is turned off . to each @xmath1 there corresponds a well defined gibbs state @xmath3}}{z({\eta } ) } , \quad { \mathrm{with } } \quad z({\eta})=\tr\left\{{\mathrm{e}}^{-\beta [ \hat{h}({\eta})-\mu \hat{n}]}\right\ } \label{rhog / zp}\ ] ] being the grand - canonical partition function and @xmath4 the particle number operator . suppose that @xmath1 is some explicitly time - dependent degree of freedom , which interacts directly with the system and not with the reservoir , i.e. , a _ work parameter _ @xmath5 . then , for an initial value @xmath6 , at the time @xmath7 , the state of the system is @xmath8 . the initial hamiltonian and particle number operators have , respectively , the spectral decompositions @xmath9 where @xmath10 is the @xmath11 simultaneous eigenstate of @xmath12 and @xmath4 , with eigenvalues @xmath13 and @xmath14 . next , some ` work ' is performed on the system taking the work parameter from @xmath6 to a final value @xmath15 , at a later time @xmath16 . the final hamiltonian , connected by the protocol @xmath17 , has the spectral decompositions @xmath18 where @xmath19 is the @xmath20 simultaneous eigenstate of @xmath21 and @xmath4 , with eigenvalues @xmath22 and @xmath23 . the definition of work in this scenario requires two projective measurements : the first projects onto the eigenbasis of the initial hamiltonian @xmath12 , with the system in thermal equilibrium . the system then evolves under the unitary dynamics @xmath24 , generated by the protocol @xmath17 , before the second measurement projects onto the eigenbasis of the final hamiltonian @xmath21 . the probability of obtaining @xmath13 for the first measurement outcome followed by @xmath22 for the second is then @xmath25 } \label{pnpnm}\ ] ] the work distribution is defined as @xcite @xmath26 , \label{eq : qworkdist}\ ] ] and the average work @xmath27 done on the system is given by the first moment of @xmath28 . it is useful to introduce the characteristic function of work @xcite as @xmath29with @xmath30 $ ] denoting the thermal equilibrium average over the initial state . in terms of @xmath31 the average work is expressed as @xmath32 . microscopically , the second law of thermodynamics is revised to the form @xmath33/\beta,\ ] ] with @xmath34/\beta$ ] being the thermodynamic grand potential , so as to encompass the explicit statistical nature of work . the deficit between average work and the variation in the grand potential can be accounted for by the introduction of the irreversible work contribution @xmath35 , being such that @xmath36 . when combined with the first law of thermodynamics , this relation can be rewritten as @xmath37 where @xmath38 is the change in entropy of the system and @xmath39 ( @xmath40 ) is the irreversible ( reversible ) entropy change . we note that for a closed quantum system , the heat transfer into the system is zero and the sole contribution to the entropy change is the irreversible entropy . to these concepts and their link to the irreversibility of a sudden transformation , we shall come back in sec . [ sec600 ] , after we have evaluated the characteristic function of the work distribution and its first moments for the specific problem at hand . consider a gas of non - interacting cold fermions , confined by a one - dimensional trapping harmonic potential . the trap has a characteristic length @xmath41 and a resonance frequency @xmath42 . then , each particle of the gas , in its unperturbed state , is described by the harmonic oscillator hamiltonian @xmath43 the unperturbed system hamiltonian @xmath44 is spanned by the fermion field in real space @xmath45 , with @xmath46 denoting the spin degrees of freedom , i.e. , @xmath46 ranging over @xmath47 values . let us now consider doing _ work _ on the fermi gas by the turning on of an external potential . let us further assume that the switch is done in a sudden way . in this paper we will take this external perturbation to be a localized potential at the centre of the trap which we can model by a dirac @xmath48 function with strength @xmath49 such that @xmath50 , where @xmath51 is the heaviside step function . then , the total hamiltonian for the gas after the switching on of the potential is @xmath52 , where @xmath53 consider the work parameter @xmath54 and the protocol @xmath55 , from @xmath56 ( with @xmath57 ) to @xmath58 ( with @xmath59 ) . then , the initial state of the system @xmath60 is given by eq . ( [ rhog / zp ] ) with the initial hamiltonian @xmath61 , introduced in eq . ( [ hhat0 ] ) , and the particle number operator @xmath62 . the partition function in the initial state will be denoted @xmath63 . the final hamiltonian @xmath64 includes the perturbation operator introduced in eq . ( [ vhat ] ) . in this particular case a sudden quench occurs , and the characteristic function of work ( [ eq : loschmidt ] ) reduces to @xcite @xmath65 in the dynamical response theory of many particle system , a key quantity to calculate is the so - called _ vacuum persistence amplitude _ @xcite @xmath66 i.e. , the probability amplitude that the gas will retrieve its equilibrium state at time @xmath67 , after the switching on of the perturbation . it is quite obvious that a simple relationship exists between the characteristic function ( [ eq : sudden ] ) and the vacuum persistence amplitude ( [ eq : vacamp ] ) : @xmath68 exploiting this relationship we have the full access to the statistics of work done via calculation of the vacuum persistence amplitude . initially the fermions lie in their equilibrium configuration , set by @xmath69 , until a sudden perturbation @xmath70 is felt by the gas . the initial equilibrium depends on the harmonic oscillator hamiltonian given in eq . ( [ h0 ] ) whose unperturbed eigenfunctions are @xmath71 with @xmath72 being the hermite polynomials of order @xmath73 . by the parity of these wave functions , the external potential induces excitations which connect only unperturbed one - fermion states labeled by even numbers @xmath74 , with @xmath75 . then , the matrix elements of the external potential in the unperturbed oscillator basis read @xmath76 where we have introduced the euler gamma function ratio @xmath77 . we express the fermion field in terms of the annihilation operator @xmath78 for the ( unperturbed ) @xmath73-th single particle state of energy @xmath79 and spin @xmath46 . then , using @xmath80 with @xmath81 being the spinor wave function , the unperturbed hamiltonian results in @xmath82 the particle number operator reads @xmath83 while the effect of the perturbation on the gas is represented by @xmath84 a typical approach in many - body physics to find a manageable expression for the vacuum persistence amplitude ( [ eq : vacamp ] ) is to turn to the interaction picture , where the impurity potential reads @xmath85 in this representation , we may use the identity ( [ chinu ] ) to express the characteristic function of work as @xmath86 then , we may perform a linked cluster expansion and reduce @xmath87 to an exponential sum of connected feynmann diagrams : @xmath88 these loops contain separable products of lines @xmath89 , \label{prop}\ ] ] connected to vertices ( @xmath90 ) , with level occupation numbers given by the fermi - dirac distributions @xmath91 as is standard in discrete - level systems , and intrinsic semiconductors , we let the chemical potential @xmath92 lie in the middle between the fermi energy @xmath93 and the lowest unoccupied one - fermion state at the absolute zero ( @xmath94 ) . without loss of generality , we may assume the fermi level number @xmath95 to be even , label it by @xmath96 ( @xmath97 being a positive integer ) so that @xmath98 . then , we can parametrize @xmath99 as @xmath100 , i.e. , @xmath101^{-1}$ ] , and compute @xmath102 for finite @xmath103 by constraining the average particle number to be : @xmath104from the plots of fig . [ figone ] , we observe that @xmath92 takes its maximum value @xmath105 at @xmath106 . then , it decreases with decreasing @xmath107 , and it becomes largely negative for very low @xmath108 where the classical limit applies . -12pt we now apply wick s theorem and focus on the lowest order loops in ( [ nubetf ] ) , i.e. , @xmath109 and @xmath110 . in terms of the auxiliary functions @xmath111 we express these contributions as @xmath112 it is useful for the following to parameterize @xmath49 in terms of the energy scale @xmath113 , which results from the geometric mean between the spacing of even energy levels and the fermi energy . to this end , we introduce @xmath114 as a sensible parameter , and to keep the interaction potential small , we investigate the range @xmath115 . it turns out that , in the free gas limit ( @xmath116 ) , this interaction strength coefficient reduces to the so called `` critical parameter '' of the mahan nozires de dominicis ( mnd ) theory of the edge singularity @xcite , which , in the x - ray absorption problem , also gives a measure of the asymmetry of the absorption spectrum . analogously , we will find below , for our trapped gas case , that the parameter @xmath117 determines the skewness of the work distribution . the full derivation of eqs . ( [ lambda1 ] ) and ( [ lambda2 ] ) , and the numerical calculations involved , can be found in a recent paper by the authors @xcite , and the reader interested in specific details is directed there . in the present context , we quote the main results and apply them to the determination of the work distribution @xmath118 and its characteristic function ( [ nubetf ] ) . the one - vertex loop is just the adiabatic response of the gas , being of the form @xmath119 , where @xmath120 represents the first - order correction to the equilibrium energy , given by @xmath121 eq . ( [ eq : e1 ] ) brings a phase factor to @xmath87 , which corresponds to shifting the work distribution @xmath118 by @xmath122 . the two - vertex loop can be split into three parts with well defined trends and physical meaning , namely , @xmath123 the first one is @xmath124 , where @xmath125 provides the second - order correction to @xmath126 and brings a further shift to @xmath118 . the second one @xmath127 includes the coefficient @xmath128 which produces a gaussian damping in @xmath87 and a gaussian broadening in @xmath118 . the third one @xmath129 accounts for the shake - up of the gas due to the sudden switching of the impurity . it is a periodic function of time with frequency @xmath130 ( as a direct consequence of the harmonic form of trapping potential ) , which provides the non trivial part of @xmath118 . -12pt the basic quantities of the problem , given in eqs . ( [ eq : e1 ] ) , ( [ delta2b ] ) , ( [ eq : gbeta ] ) , and ( [ lambda2e ] ) , contain summations running over all one - fermion eigenstates of the trap , weighted by the fermi factors @xmath131 , and enter the characteristic function of work ( [ chinu ] ) : @xmath132 by eq . ( [ eq : qworkdist ] ) , the work distribution is given by the convolution product @xmath133 numerical computations of @xmath118 are shown in fig . [ figttwo ] , where we recognize an asymmetric , broadened profile , signature of the singular behavior of the fermi gas . the monotonic structure turns into a satellite structure of sub - peaks , separated by @xmath134 and related to even - level transitions in the gas , as @xmath108 gets above @xmath135 . these secondary peaks are a direct manifestation of the single particle transitions occurring within the gas , with the fermions jumping between even harmonic oscillator states ( separated in energy by even multiples of @xmath136 ) , in response to the local perturbation . all of these transition constitute the so called `` shake - up '' process , which is , thus , explicitly witnessed by the work distribution . at high temperatures , this feature disappears , being hidden by the increased gaussian broadening of the main peak , which ( in our two - vertex approximation ) is found at @xmath137 , and which describes the transition between the equilibrium states induced by the switching of the scattering center . consider now a fermi gas with a large number of particles at sufficiently low temperatures , being such that we may approximate the chemical potential by eq . ( [ muinf ] ) . in this regime , we can expand @xmath131 in power series of @xmath138 and select the lowest order of this series . as detailed in the appendix , we can work in the @xmath139-limit and find some manageable expressions for the auxiliary functions @xmath140 introduced in eq . ( [ lambdas ] ) . by eq . ( [ lambda1 ] ) , the initial value @xmath141 determines the one - vertex loop and hence the first order energy shifts ( [ eq : e1 ] ) . on the other hand , the product @xmath142 enters a double time - ordered integral which gives ( [ lambda2 ] ) , together with the second - order energy shifts ( [ delta2b ] ) , the gaussian standard deviation ( [ eq : gbeta ] ) , and the shake up sub - diagram ( [ lambda2e ] ) . this integral can be carried out analytically , in the @xmath139-limit , by adding a small shift to the time domain on the imaginary axis . we , therefore , obtain the following analytical approximation for the characteristic function ( [ nust ] ) @xmath143 where @xmath144 here , @xmath145 is a regularization parameter , i.e. , an imaginary time shift , that we may interpret as the typical time interval needed by the system to respond to the abrupt switching on of the impurity potential at zero temperature . we see that the function ( [ nustinf ] ) is undefined in the @xmath146-limit , which is a consequence of the large @xmath97 expansion used in the thermal series for @xmath140 . the exact expressions provided by eqs . ( [ lambda1 ] ) and ( [ lambda2 ] ) are well set and can be used to compute ( [ nust ] ) and then ( [ pwork ] ) numerically , without any divergence problem , as we did in fig . [ figttwo ] . we remark that the imaginary time regularization leading to eq . ( [ nustinf ] ) is required to provide an analytical support to the theory . furthermore , it has been observed that @xcite @xmath147 correctly tends to the nozires and de dominicis core hole propagator @xcite when the harmonic trap frequency is lowered to zero , keeping the number of particles in the gas finite . in this limit , the regularization parameter @xmath148 , becomes exactly the one needed in the mnd theory @xcite . the mathematical details of the derivation of eqs . ( [ e12inf ] ) , ( [ gbetapp ] ) , and ( [ lambxx ] ) are discussed in [ appx1 ] . -12pt the leading behavior of @xmath149 in eq.([nustinf ] ) vs temperature is provided by the gaussian damping factor , whereas both the energy shifts and the periodic part of the characteristic function are well approximated by their absolute zero expressions in a wide range of temperatures , corresponding to @xmath150 for @xmath151 . as shown in fig . [ figthree]*a * , the first - order shift @xmath152 , given by eq . ( [ e12inf ] ) , is the correct large @xmath108-limit for @xmath153 , numerically calculated from eq . ( [ eq : e1 ] ) . -12pt to compute the low temperature form of @xmath154 , we truncate the series in eq . ( [ gbetapp ] ) and include the first @xmath155 terms . the corresponding approximations , for @xmath156 and @xmath157 , are plotted in fig . [ figthree]*b * together with the numerical form of @xmath158 calculated from eq . ( [ eq : gbeta ] ) . we see that already the @xmath156 curve accurately reproduces the numerical data for @xmath159 . the approximation with @xmath157 components works particularly well in the extended range @xmath160 , where the numerical @xmath161 curves are independent on @xmath97 . now , consider the regularised shift @xmath162 , as introduced in eq . ( [ e12inf ] ) , and the regularised shake - up diagram @xmath163 , reported in eq . ( [ lambxx ] ) . we need to adjust @xmath164 for each @xmath97 in such a way that these two quantities are the correct absolute zero limits for the corresponding numerical quantities , i.e. , @xmath165 it turns out that @xmath164 decreases with increasing @xmath97 , which makes sense because the more particles the system has the more allowed transitions are offered to respond to the sudden perturbation ( fig . [ ffour]*a * ) . however , the set of @xmath164 values that accurately fit the condition ( [ e2l2lima ] ) are slightly different from those that realize the condition ( [ e2l2limb ] ) . this is because the asymptotic forms of @xmath162 and @xmath166 contain terms proportional to @xmath167 , @xmath168 , @xmath169 , which we have neglected working in the large @xmath97 limit ( [ appx1 ] ) . as a reasonable compromise , we take the average between these two optimized sets ( fig . [ ffour]*a * ) , which is an agreement with both conditions ( fig . [ ffour]*a * , * c * , * d * ) within an error less than @xmath170 . interestingly , @xmath171 and @xmath153 have similar trends and absolute values for @xmath172 , while @xmath171 is more sensitive to temperature than @xmath153 for @xmath173 . on the other hand , the modulus of the sub - diagram @xmath174 presents zeroes at @xmath175 and maxima at @xmath176 , with @xmath177 . the intensities of such maxima ( fig . [ ffour]*c * ) increase with increasing the fermi number ( @xmath96 ) , the critical exponent ( @xmath178 ) , and the thermal energy ( @xmath179 ) . the phase of @xmath180 is discontinuous at the extremes of @xmath181 and less dependent on these parameters . we now have all the ingredients to start our discussion of the statistical properties of the work distribution of a non - interacting fermi gas held in a harmonic trap , following a sudden local quench of a point like scattering potential @xcite . in the present section we will provide a physical interpretation for the features of the work distribution . in particular , we will work out the cumulant expansion of the work distribution and look for the link between different cumulants and thermodynamical quantities . the characteristic function obtained in eq . ( [ eq : sudden ] ) has been introduced in eq . ( [ eq : loschmidt ] ) as the mean value of the random variable @xmath182 in the work distribution @xmath118 or , equivalently , as the fourier transform of the work distribution . the work distribution , then , is nothing but the absorption spectrum of the system due to the suddenly switched - on impurity potential @xcite : it has been set in eq . ( [ pwork ] ) and shown in fig . [ figttwo ] . we notice that because of the assumed normalization of @xmath118 , we have @xmath183 . we call moment of order @xmath73 , or equivalently @xmath184th moment of the distribution , the mean value @xmath185 . once the characteristic function is known , we can use the differentiation theorem of fourier transforms to evaluate each of the above defined moments as@xmath186this relation holds provided that @xmath87 is continuous and differentiable @xmath73-times , with all of the derivatives vanishing at @xmath187 ( that we will see not to be always the case ) . it will be more convenient for us to work with the cumulant expansion of @xmath188 instead of computing the moments of @xmath87 . the cumulants are defined analogously to the moments in eq . [ eq : moments ] with @xmath188 replacing @xmath87 . this will have two advantages . first it will be easier to characterize the distribution since , as we will see , @xmath189 makes it possible to calculate important quantities such as the mean value , variance and skewness straightforwardly . second , in our case @xmath190 so that we will be able to link the properties of the distribution ( as given by its mean value , variance and skewness ) to the dynamical functions @xmath191 . using the form given by eq . ( [ eq : sudden ] ) for the characteristic function , we express the cumulant generating function as @xmath192the first three cumulants are , then , given by @xmath193 where we defined the variance @xmath194 and the skewness @xmath195 . in our case , and by considering only two sets of diagrams , see eq . ( [ lnchi ] ) , the above quantities are simply calculated as @xmath196 we see that the two connected diagrams found previously allow us to calculate the first three cumulants in a straightforward manner . in particular they are determined by the analytical properties of the periodic sub - diagram ( [ lambda2e ] ) at @xmath197 . now , while @xmath198 is a continuous and differentiable function , its higher order time derivatives , i.e. , @xmath199 for @xmath200 , are ill - defined in the @xmath197-limit . as a corollary , we have that the higher - order time derivatives of the characteristic function are not defined at @xmath7 , then eq . ( [ eq : moments ] ) lacks formal justification for @xmath201 . this is a direct consequence of the point - like modeling of the impurity potential . a possible work around of the problem will be proposed in the following paragraphs by introducing suitable cut - off frequencies on the perturbation matrix elements ( [ vrrp ] ) . in particular , we shall use the fact that the regularized characteristic function ( [ nustinf ] ) admits cumulants of any order to renormalize the second and third ones given in eqs . ( [ k2 ] ) and ( [ k3 ] ) . indeed , the periodic subdiagram quoted in eq . ( [ nustinf ] ) has been proved to accurately reproduce the numerical expression ( [ lambda2e ] ) in the absolute zero - limit ( fig . [ ffour ] ) . then , it makes sense to match the cumulants , as given by eq . ( [ nustinf ] ) , with the corresponding quantities obtained from the numerical form of @xmath202 , eq . ( [ nust ] ) . such a condition does not alter the physics of the work distribution , while it will allow us to carry out the analysis of the physical behavior of the skewness , which will turn out to have an expression independent of this regularization procedure . first let us consider the mean value ( [ k1 ] ) . using the expressions for @xmath171 and @xmath203 given in eqs . ( [ delta2b ] ) and ( [ lambda2e ] ) , respectively , we may write@xmath204then , considering the identity@xmath205we find@xmath206interestingly eq . ( [ k1bet ] ) states that the mean work is given by the first - order energy shift of the quenched fermi gas . such a relation continues to hold at absolute zero with the regularized approximations given by eq . ( [ nustinf ] ) , ( [ e12inf ] ) , and ( gbetapp ) . indeed , the second order time derivative of the regularized periodic sub - diagram reads@xmath207so that , using eq . ( [ k1 ] ) again , we get : @xmath208 . although the calculations leading eq . ( [ k1bet ] ) have been derived from an approximated expression for the vacuum persistence amplitude , we may prove eq . ( [ k1bet ] ) to be formally exact . to see this let us consider the expression ( [ ip ] ) for the characteristic function together with the definition of the moments in eq . ( [ eq : moments]):@xmath209this result tells us that for a sudden quench the average work done is the mean value of the perturbation after the switch - on at instant @xmath7 . however , this is also the first - order energy reported in eq . ( [ eq : e1]):@xmath210 the behavior of @xmath153 , and hence of the average work @xmath211 vs the inverse thermal energy @xmath103 , as been thoroughly discussed in sec . [ sec4.2 ] and shown in fig . figttwo**a * * using eq . ( [ k2 ] ) , we may now evaluate the variance of the work distribution:@xmath212here we replace @xmath161 with its explicit form given in eq . ( eq : gbeta ) , and compute the second order time derivative@xmath213which leads the compact expression@xmath214to provide an interpretation to this relation , we replace the definition of @xmath178 and the expression ( [ vrrp ] ) for the impurity potential matrix elements:@xmath215by analogy with the fermi s golden rule , we can look at the quantity@xmath216with @xmath217 denoting the density of even fermion states in the continuous limit . we recognize @xmath218 to be the rate of transition from the occupied one - particle state @xmath219 to the empty state @xmath220 . following this interpretation , we rewrite the variance as@xmath221it is thus clear that @xmath194 gives information on the broadening due to fermions scattering in empty states and thus on the spectrum of the system . what is also evident from eq . ( [ sigsq ] ) is that the sum over unoccupied particle states , i.e. , the @xmath222-series , does not converge , because the perturbation @xmath223 is a weakly decreasing function of @xmath224 and @xmath222 . indeed , @xmath194 is proportional to the initial product of the auxiliary functions @xmath140 , introduced in eq . ( [ lambdas]):@xmath225now the asymptotic behaviors @xmath226 and @xmath227 for @xmath228 lead to @xmath229(see [ appx1 ] ) . on the other hand , we can substitute regularized characteristic function ( [ nustinf ] ) in eq . ( [ k2 ] ) to get the asymptotic trend@xmath230this result , combined with the fact that @xmath231 vanishes at the absolute zero , gives @xmath232we see that the divergent behavior of @xmath194 in eq . ( [ sigsql ] ) is absorbed by the regularization parameter @xmath164 . indeed , considering gases with very large particle numbers , i.e. , working in the @xmath233-limit , we find @xmath234 . to recover a consistent definitions of @xmath235 , as given by eq . ( [ k2 ] ) and hence eq . ( [ sigsql ] ) , we proceed similarly to the nozires and de dominicis @xcite work around of the fermi - edge singularity ; we introduce an exponential frequency cut - off on the impurity potential and change the @xmath236factors to @xmath237 . this is equivalent to adding a lifetime width @xmath238 to the unperturbed propagator ( [ prop ] ) . by doing so , eq . ( [ sigsq ] ) becomes well defined and factorisable as@xmath239and the auxiliary functions are evaluated on the complex time - domain . in appx1 we show the cuf - off frequencies @xmath240 is related to the regularization time @xmath164 by the condition that @xmath241 , as calculated from eq . ( [ sigsqr ] ) with @xmath242 , matches with the regularized expression ( [ k2inf ] ) . the skewness @xmath243 is related to both the second an the third cumulant of the distribution . to begin , we address our attention to the third cumulant ( k3 ) , i.e. , @xmath244keeping in mind that @xmath235 is a non negative quantity . we then compute the third order time derivative@xmath245and write@xmath246 next , we use the definitions for the impurity potential matrix elements and the transition rates , as in eq . ( [ sig0 ] ) , and express@xmath247 thus , for the skewness we get @xmath248 which encompasses a direct thermodynamical meaning . the first term is nothing but the energy taken from the system by emptying its states whereas the second is the energy given to the system by filling its empty states following the thermodynamic transformation , which in our case is a sudden quench . so this quantity tells us whether the transformation effect is to increase or decrease the internal energy of the gas and is clearly related to the asymmetry of the work distribution -12pt to provide a consistent definition of @xmath249 , we need to tackle the divergent behavior of the @xmath222-series in eq . ( kk3comp ) , which is more evident by expressing the third cumulant as@xmath250to avoid this divergence , we adopt the previous procedure once again and redefine eq . ( [ kk3comp ] ) by adding an imaginary time - shift to the auxiliary functions @xmath140 , i.e.,@xmath251 the new cuf - off frequency @xmath252 is fixed by the condition that @xmath253 , as calculated from eq . ( [ kk3cut - off ] ) with @xmath254 , tends to the regularized expression@xmath255computed with the absolute zero distribution ( [ nustinf ] ) . in [ appx1 ] we provide details on how the renormalization of the second and third cumulants of the work distribution , given by eqs . ( [ sigsqr ] ) and ( kk3cut - off ) , are carried out . in fig , * we show the cut - off times @xmath238 and @xmath256 that let us match the numerical behavior of @xmath235 and @xmath257 with the expressions given by eqs . ( [ k2inf ] ) and ( [ k3infreg ] ) , respectively . using such values we can obtain the behavior of the variance and the skewness vs @xmath258 for different values of @xmath97 ( fig , figcumul**b * * and [ figcumul]*d * ) . the dependence of the second and third moment on @xmath178 is trivial , as both @xmath259 and @xmath249 are directly proportional to the critical index . then , the skewness goes like @xmath260 . interestingly , we observe that by eqs . ( [ k2inf ] ) and ( [ k3infreg ] ) gives the following form to the skewness parameter@xmath261which does not suffer from the divergent behavior of @xmath241 and @xmath262 . furthermore , in the @xmath233-limit , the third cumulant behaves as @xmath263 and the skewness turns out to depend only on the critical exponent @xmath264 ( fig . [ figcumul]*d * ) . being completely independent of the regularization procedure , this is a physically meaningful result , showing how the critical parameter @xmath117 does indeed determine the asymmetry of the work distribution , ultimately due to the fermi edge behavior . in the previous section we have seen that the work distribution contains information about both the unperturbed system and the system following the thermodynamic transformation . in this section we shall first recover a jarzynski - like equality , which will help us in identifying the hypothetical final equilibrium state of the system to compare with . we shall then find an approximate and analytic expression , and discuss a numerical method for the computation of the irreversible work , which is a figure of merit of the irreversibility of the transformation . we are interested in highlighting the microscopic origin of irreversibility for the sudden quench studied above . first let us recover the jarzynski - like equality . we start from the evaluation of @xmath265 by eq . ( [ eq : qworkdist ] ) , i.e.,@xmath266 in going from the first to the second row we have used particle number conservation , @xmath267 , and the completeness relation @xmath268 , @xcite . we recognize that the @xmath73-sum in eq . ( [ eew1 ] ) is nothing but the grand canonical partition function for the gas in equilibrium with the impurity potential , with same temperature and chemical potential as the initial one . hence , we have:@xmath269where @xmath270 \label{omega0}\]]and @xmath271 \label{omegapp}\]]denote the grand potentials for the unperturbed and perturbed equilibrium states , respectively . the fact that the jarzynski relation makes explicit connection with a ( hypothetical ) final equilibrium state is meaningful as it gives us a reference for the ( hypothetical ) reversible version of the transformation encompassed by the switching on of the external perturbation . the reversible version of this protocol , thus , would involve a change in the number of particles , which is necessary in order to maintain the initial chemical potential even if the single particle energies are modified during the protocol @xcite . by means of the jensen inequality @xmath272 we can derive the statement of the second law of thermodynamics : @xmath273as expected , the average work done on the system is greater than the change in the grand potential of the initial state and the hypothetical final equilibrium state . this relation suggests us to define a new variable that we shall name @xmath274 , with @xmath275 . the new variable will have the very same distribution as the original one but for the mean value which will be shifted by an amount @xmath276 : @xmath277 in the following , we propose an analytic and a numerical approaches to compute this excess work ( [ wirr ] ) and discuss it in order to highlight the nature of irreversibility in our system . we have already set all the elements to compute @xmath278 , i.e. , the unperturbed level energies @xmath279 , distribution functions @xmath131 , and chemical potential @xmath99 , and we have already discussed the behavior of the average work @xmath280 ( see fig . [ figttwo]*a * ) . what is left to calculate is the perturbed grand potential @xmath281 , for which we will provide both an approximate analytic and a numerical computation . the energies entering the hypothetical final state belong to the spectrum of the hamiltonian@xmath282which describes a particle of the gas in equilibrium with both the harmonic trap and the impurity potential . as discussed in sec . [ sec200 ] , the odd single particle energy levels are left unperturbed @xmath283 . as for the even energies @xmath284 , which can not be calculated analytically , we resort to perturbation theory assuming the height @xmath285 of the @xmath48-potential to be small with respect to the unperturbed energies . then , we express : @xmath286where @xmath287 is computed from the non - degenerate rayleigh - schrdinger perturbation theory@xmath288based on the assumption@xmath289substituting ( [ e2rpert ] ) in the perturbed grand potential ( omegapp ) , and carrying out a power series expansion for small @xmath290 , to the second order we get:@xmath291 ^{2}\]]now , considering the assumption ( [ assumpt ] ) , we find@xmath292 ^{2}.\end{aligned}\]]by eq . ( [ rspertth ] ) , the first term at the right hand side of this relation is simply the average work @xmath211 calculated above ( see sec . [ work ] ) . the second term is the second order contribution @xmath293 to the change in the total energy of the system . the third one may be expressed in terms of the gaussian broadening ( [ eq : gbeta ] ) as @xmath294 . we are now able to calculate the excess work ( [ wirr ] ) as @xmath295notice that by eq . ( [ rspertth ] ) the second - order corrections @xmath296 are negative , which makes @xmath297 a positive quantity , and let the jensen inequality ( [ jensineq ] ) be always verified . ( [ wirrpert ] ) can be brought into a more interesting and general form , by noticing that@xmath298 ^{2}f_{r}^{+}f_{r}^{- } & = & \sum\limits_{r , r^{\prime } , \xi , \xi ^{\prime } } \varepsilon _ { 2r}^{(1)}\varepsilon _ { 2r^{\prime } } ^{(1)}\left ( f_{r}^{+}f_{r^{\prime } } ^{+}+f_{r}^{+}f_{r}^{-}\delta _ { rr^{\prime } } \delta _ { \xi \xi ^{\prime } } \right ) \\ & = & \sum\limits_{r , r^{\prime } , \xi , \xi ^{\prime } } \varepsilon _ { 2r}^{(1)}\varepsilon _ { 2r^{\prime } } ^{(1)}\left [ \langle \hat{c}_{2r\,\xi } ^{\dag } \hat{c}_{2r\,\xi } \hat{c}_{2r^{\prime } \,\xi ^{\prime } } ^{\dag } \hat{c}_{2r^{\prime } \,\xi ^{\prime } } \rangle -f_{r}^{+}f_{r^{\prime } } ^{+}\right ] \\ & = & \mathrm{var}\left ( \sum\limits_{r,\xi } \varepsilon _ { 2r}^{(1)}\hat{c}_{2r\,\xi } ^{\dag } \hat{c}_{2r\,\xi } \right)\end{aligned}\]]we thus conclude that the irreversible work thus takes the suggestive form:@xmath299this relation is independent on the model used to characterize the impurity potential . interestingly , we may look at the occupation numbers @xmath300 as independent random variables , distributed according to the probability distribution @xmath301 , the function @xmath302=\sum\limits_{r,\xi } \varepsilon _ { 2r}^{(1)}n_{2r\,\xi } \label{w1def}\]]of the configurations @xmath303 is a random variable too . as a rsult , the more peaked this random function is the smaller its contribution to the irreversible work . this means that for an adiabatic change there is no spread of the work distribution , since in that limit the average work done would be a dirac delta function of @xmath304 . we now turn the attention to the numerical calculation of the perturbed grand potential ( [ omegapp ] ) , which requires the knowledge of the eigenvalues of the single - particle hamiltonian ( [ ht ] ) . the odd harmonic oscillator eigenfunctions @xmath305 and eigenenergies @xmath306 are left unaffected by the @xmath48-potential , due to the fact that @xmath307 ( see sec . [ sec200 ] ) . on the other hand the perturbed even eigenfunctions of ( [ ht ] ) , with the physically correct asymptotic behavior , are the parabolic cylinder functions @xmath308with associated level energies @xmath309 . the latter are the just perturbed energies denoted @xmath310 above . as shown for example in ref . @xcite , the stationary shrdinger equation for the @xmath48-potential implies @xmath311where @xmath312 -12pt the last two relations yield an implicit condition between the strength of the @xmath48-function and the quantum numbers @xmath313 : @xmath314since the @xmath315-function has poles for negative integer values , eq . ( implicit ) leads to @xmath316 for @xmath317 , and @xmath318 for @xmath319 . then , the energy eigenvalues @xmath320 converge to the unperturbed energies @xmath284 when the potential barrier is set to zero , while they tend to @xmath321 for an infinite barrier . in the latter case , the perturbed energies become identical to the values of the odd eigenfunctions , leading to a double degeneracy of all eigenvalues . for arbitrary values of @xmath49 , which means for @xmath322 at any @xmath323 , ( [ alphapp ] ) , we need to solve eq . ( [ implicit ] ) numerically and get a sequence of quantum numbers @xmath324 , with @xmath325 . here , it is important then to notice that for each fixed value of @xmath178 and @xmath97 there exists a one to one correspondence between each unperturbed quantum number @xmath224 and a @xmath313 . once the @xmath313 are known , we can compute the perturbed energies @xmath320 and the normalization constants @xmath326 , which let us determine the perturbed wavefunctions @xmath327 . in fig . [ wk]*a * and [ wk]*b * , we show how the single particle ground state wave functions and the energy levels change with increasing both @xmath178 and @xmath97 . -12pt within this framework , the perturbed grand - canonical potential is given by@xmath328 \\ & & -\frac{{(2s+1)}}{\beta } \sum\limits_{r=0}^{\infty } \ln \left [ 1+{\mathrm{e}}^{-\beta \left ( { \varepsilon_{2{\tilde{r}}}}-\mu \right ) } \right].\end{aligned}\]]the irreversible work can be thus computed from ( [ wirr ] ) using@xmath329 . \label{numirr}\]]in fig . [ wowo]*a * and [ wowo]*c * we see that the second law ( [ jensineq ] ) is obeyed for any chosen valued of @xmath178 and @xmath97 . the amount of irreversible work , shown in fig . wowo**b * * and [ wowo]*d * , increases with decreasing temperature reaching a nearly constant saturation value for low enough temperatures , @xmath330 depending on the number of particles in the gas . a similar trend is followed by other crucial quantities discussed here , such as the chemical potential , the energy shifts , and the cumulants of the work distribution function . on the other hand , as suggested by the approximation ( wirrpert ) , the irreversible work is highly sensitive to the total particle number and the critical exponent following the proportionality relation @xmath331 , for @xmath332 . in this work , we explored the physics and thermodynamics of an inhomogeneous fermi gas perturbed by the sudden switch on of a local scattering potential . exploiting the direct relationship between the characteristic function of work and the vacuum persistence amplitude , and by means of the linked cluster expansion technique , we obtained the full statistics of work done by performing such a local quench on the gas , showing that the first , second and third moments of the work distribution encapsulate the salient thermodynamic features of the model . * indeed , features of the textbook fermi - edge singularity problem were found to be present in the higher moments . furthermore , we obtained analytic and perturbative expressions for the excess or irreversible work done on the system as well as a jarzynski like equality . * from a general perspective , our work demonstrates the potential of combining ideas coming from recent developed techniques in non - equilibrium statistical mechanics with traditional approaches from many body physics for the analysis of out of equilibrium problems in quantum systems . it is also worth pointing out that the approach developed here is by no means restricted to a specific system and can be extended to other types of quench problems , global and local , in non - interacting and interacting many - particle systems . in addition to the theoretical framework developed , one can imagine that in the future the work distribution of the fermi gas maybe extracted in an experimental setting by means of coupling to an auxiliary ancilla system which would then function as a probe , giving access to the relevant thermodynamic quantities by monitoring , e.g. its decoherence dynamics @xcite . this was first suggested in the context of fermi gases by goold et al in @xcite . in fact very recent proposals to verify the quantum fluctuation relations by means of interferometry of an ancillary probe qubit @xcite have been realised in a recent experiment in a nmr setting @xcite . the challenge for the future is to really push these experiments to the many - body domain . in the following we will present the details of the analytical approximations for the quantities introduced in sec . [ sec4.1 ] , i.e. , the energy shifts ( [ eq : e1 ] ) and ( [ delta2b ] ) , the gaussian standard deviation ( [ eq : gbeta ] ) , and the shake - up sub - diagram ( [ lambda2e ] ) , which determine the characteristic function of work ( [ nustinf ] ) . by the same methods , we will show how to find the asymptotic limits for the main properties of the work distribution , i.e. , its mean value , variance and skewness ( see sec . [ sec500 ] ) . to begin , we consider the power series expansions of the statistical fermi factors @xmath333 as well as their products @xmath334 with @xmath335 denoting the characteristic times @xmath336 induced by thermal fluctuations . by the relations ( [ frexp ] ) and ( [ f+f - exp ] ) we can convert eqs . ( [ eq : e1 ] ) , ( [ delta2b ] ) , ( [ eq : gbeta ] ) , and ( [ lambda2e ] ) into power series of @xmath337 . in particular , the auxiliary functions @xmath338 , entering the connected graphs ( [ lambda1 ] ) and ( [ lambda2 ] ) , are expanded as @xmath339 using eq . ( [ frexp ] ) , the coefficients of the series ( [ lambdasexp ] ) read : @xmath340 from the expansion ( [ coel2 ] ) we obtain the thermal series for the first - order energy shift , i.e. , the average work : @xmath341 in addition , we rewrite the two - vertex loop ( [ lambda2 ] ) as @xmath342 which by eq . ( [ l2decomp ] ) contains all other basic quantities of the problem , see eqs . ( [ delta2b ] ) , ( [ eq : gbeta ] ) , and ( [ lambda2e ] ) . to compute the gaussian term , however , it is more straightforward to work on expression given in eq . ( [ eq : gbeta ] ) and use the expansion ( [ f+f - exp ] ) . by doing so , we get @xmath343 in which @xmath344 as a first approximation , we focus on the temperature range where the chemical potential is approximated by eq . ( [ muinf ] ) , i.e. , @xmath150 for @xmath345 . in this regime , we apply the sum rules @xmath346 with @xmath347 being the regularised hypergeometric function @xmath348 substituting into eqs . ( [ coel1 ] ) and ( [ coel2 ] ) , we find @xmath349 \nonumber \\ & & \qquad \qquad \qquad \qquad + { \mathrm{e}}^{-3\tau_{m } { \omega}/2 } { { \setbox1=\hbox{\ensuremath{^ { } } } \setbox2=\hbox{\ensuremath{_{2 } } } \setbox5=\hbox{\ensuremath{\tilde{f } } } \hspace{\ifnum\wd1>\wd2\wd1\else\wd2\fi } \ensuremath{\copy5^{\hspace{-\wd1}\hspace{-\wd5 } \hspace{\wd5 } } _ { \hspace{-\wd2}\hspace{-\wd5}2\hspace{\wd5}1 } } } } \left [ { r_{{\hbox{\uppercase{\tiny f}}}}}+1 , { \mathrm{e}}^{2 { \omega}\left ( \pm { i } t-\tau_{m}\right ) } \right ] \bigg\ } \nonumber\end{aligned}\ ] ] as a second approximation , we take systems with large numbers of particles , and employ the large-@xmath97 expansions : @xmath350 so that we straightforwardly get @xmath351 . \label{coel2as}\end{aligned}\ ] ] now , the first - order energy shift ( [ wexp ] ) is approximated by eqs . ( [ coel1as ] ) and ( [ coel2as ] ) , as : @xmath352 this expression is largely dominated by the @xmath103-independent value , reported in eq . ( [ e12inf ] ) and shown in fig . [ figthree]*a*. turning to the gaussian coefficient , a simple change of summation indices , with @xmath353 , leads to @xmath354 the transformed summations in this last line are dominated by low @xmath224 terms . in a many fermion environment , by the asymptotic relation ( [ specfas ] ) , we approximate @xmath355 and neglect terms going like @xmath356 , to obtain @xmath357 so that @xmath358 this relation , being identical to eq . ( [ gbetapp ] ) , allows us to express express @xmath359 ^{1/2}. \label{deltabapp}\ ] ] then , the standard deviation at low temperatures is independent of the number of particles in the gas ( fig . [ figthree]*b * ) . as for the second - order shift @xmath360 and the fermi - edge component @xmath361 , we first consider the @xmath103-independent coefficients ( [ coel1as ] ) of the series ( [ lambdasexp ] ) , write down the product @xmath362 and plug it into the time - ordered integrals ( [ l2exp ] ) . then , we need to deal with the short - time singularity of eq . ( [ lambdasexpp ] ) by adding an imaginary time regularization to the @xmath363-integral , i.e. , we need to shift the @xmath363 integration domain by @xmath364 . the resulting integral is dominated by the fermi - edge terms reported in eq . ( [ lambxx ] ) : @xmath365 -12pt since at the absolute zero the gaussian part is absent , we may readily interpret @xmath366 as shown in figs . [ ffour]*b*-*d * , these results are in excellent agreement with the numerical calculations at @xmath106 by suitable adjustments of @xmath145 . such a regularization parameter depends on @xmath97 and takes the physical interpretation of the average time needed by the system to respond to the abrupt impurity perturbation at zero temperature ( fig . [ ffour]*a * ) . it is also interesting to note that if we let the harmonic frequency in eq . ( [ lambda0tt ] ) go to zero , by fixing @xmath178 and keeping the number of particles in the gas ( @xmath367 ) finite , we retrieve the nozieres - de dominicis result @xcite @xmath368 which leads to the propagator @xmath369 , originally calculated for a suddenly switched on core - hole in a free electron gas . such a limiting procedure is illustrated in fig . [ fapp1 ] , where we clearly see that as @xmath370 , the periodicity of the propagator @xmath371 vanishes and its modulus reduces to a single peak with tails going like @xmath372 . inclusion of finite temperature corrections @xcite may be done by considering all possible products of @xmath373 and @xmath374 , as given by eq . ( [ lambdhyp ] ) , then using the expansion ( [ hypinf ] ) , and finally performing the @xmath375 and @xmath363-integrals excluding terms proportional to @xmath376 . however , the zero temperature result ( [ lambda0tt ] ) is largely dominant in characteristic function @xmath377 , and in the system response @xmath378 , within the ranges of temperatures @xmath379 and particle numbers @xmath380 considered here . we now use the approximations ( [ wexp ] ) , ( [ deltabapp ] ) , and ( [ lambda0tt ] ) in eq . ( [ nust ] ) , so that the characteristic function of the work distribution reads@xmath381where the last line is equivalent to eq . ( [ nustinf ] ) of the main text . in sec . [ sec4.2 ] , we have observed that eq . ( [ chiinf2 ] ) is indeed a low temperature limit for the numerical expression ( [ chiinf1 ] ) once the regularisation parameter @xmath164 is adjusted in order to obey the conditions ( [ e2l2lima ] ) and ( [ e2l2limb ] ) . the adjusted values of @xmath164 have been also reported in fig . [ ffour]*a * vs @xmath97 . in sec . [ sec500 ] we have provided a method to determine the first three cumulants of the work distributions , see eqs . ( [ kappa1])-([kappa3 ] ) . in particular , we have seen that these quantities depend on the auxiliary functions @xmath140 and their first - order time - derivatives at the time the impurity potential is activated . the first cumulant is well defined , being proportional to @xmath141 , and coincides with the first - order energy shift ( [ wexp ] ) discussed above , as reported in eq . ( [ k1bet ] ) . on the other hand the second and third cumulants involve @xmath382 and @xmath383 , see eqs . ( [ sigsql ] ) and ( [ kk3lamb ] ) . these two quantities contain weighted sums of the @xmath384 and @xmath385-factors by the hole occupation numbers @xmath386 . for large @xmath224 , the two series diverge like @xmath387 and @xmath388 , respectively . it is not surprising that a normalizable distribution function , with a well defined mean value , has divergent higher order moments , and this is due to both the sudden dynamics and the spatial modelling of the impurity potential . however , the regularization procedure introduced in sec . sec4.2 , and applied above to derive eq . ( [ chiinf2 ] ) , leads to a well defined characteristic function admitting finite cumulants of any order . to have a consistent theory , we also want the cumulants of the work distribution , as computed from eq . ( [ chiinf1 ] ) , to correctly tend to the same values obtained from eq . ( [ chiinf2 ] ) . with this in mind , we have proposed in sec . [ sec500 ] a renormalization procedure in which the auxiliary functions are extended on the complex - time domain @xmath389 and shifted by a small imaginary time , i.e. , @xmath390 in eq . ( [ sigsqr ] ) and @xmath391 in eq . ( kk3cut - off ) . then , using eqs . ( [ coel1 ] ) , ( [ coel1inf ] ) and ( coel2inf ) above , we can use the extended auxiliary functions @xmath392 to obtain the absolute zero expressions @xmath393and @xmath394 { { \setbox1=\hbox{\ensuremath{^ { } } } \setbox2=\hbox{\ensuremath{_{2 } } } \setbox5=\hbox{\ensuremath{\tilde{f } } } \hspace{\ifnum\wd1>\wd2\wd1\else\wd2\fi } \ensuremath{\copy5^{\hspace{-\wd1}\hspace{-\wd5 } \hspace{\wd5 } } _ { \hspace{-\wd2}\hspace{-\wd5}2\hspace{\wd5}1 } } } } ( { r_{{\hbox{\uppercase{\tiny f}}}}}+1,e^{-2{\omega}/{\omega_{0}^{\prime } } } ) \nonumber \\ & & + \frac{4\sqrt{\pi } \alpha { \varepsilon_{{\hbox{\uppercase{\tiny f}}}}}e^{2{\omega}({r_{{\hbox{\uppercase{\tiny f}}}}}+2)/{\omega_{0}^{\prime}}}}{\sqrt{1-e^{-2{\omega}/{\omega_{0}^{\prime}}}}}\frac{d}{dz } { { \setbox1=\hbox{\ensuremath{^ { } } } \setbox2=\hbox{\ensuremath{_{2 } } } \setbox5=\hbox{\ensuremath{\tilde{f } } } \hspace{\ifnum\wd1>\wd2\wd1\else\wd2\fi } \ensuremath{\copy5^{\hspace{-\wd1}\hspace{-\wd5 } \hspace{\wd5 } } _ { \hspace{-\wd2}\hspace{-\wd5}2\hspace{\wd5}1 } } } } ( { r_{{\hbox{\uppercase{\tiny f}}}}}+1,z)\bigg|_{z = e^{-2{\omega}/{\omega_{0}^{\prime}}}}. \label{k3infapp}\end{aligned}\]]to link the these expressions with the regularized ones that we have derived in sec . [ sec500 ] , we have fixed the cut - off frequencies @xmath240 and @xmath252 by constraining : @xmath395 the adjusted values of the corresponding cut - off times @xmath238 and @xmath256 have been reported in fig . 99 i bloch , j dalibard and w zwerger ( 2008 ) rev . phys . * 80 * 885 . d andrieux , p gaspard , t.monnai , and s tasaki ( 2009 ) new j. phys . * 11 * 043014 ; j yi , p talkner , and m campisi ( 2011 ) phys . e * 84 * 011138 ; j yi , y .- w . kim , p talkner ( 2012 ) phys . e * 85 * 051107 .

a vertex coloring of a graph is _ nonrepetitive _ if there is no path in the graph whose first half receives the same sequence of colors as the second half . while every tree can be nonrepetitively colored with a bounded number of colors ( colors is enough ) , fiorenzi , ochem , ossona de mendez , and zhu recently showed that this does not extend to the list version of the problem , that is , for every there is a tree that is not nonrepetitively-choosable . in this paper we prove the following positive result , which complements the result of fiorenzi _ et al . _ : there exists a function such that every tree of pathwidth is nonrepetitively-choosable . we also show that such a property is specific to trees by constructing a family of pathwidth- graphs that are not nonrepetitively-choosable for any fixed .

1601.01886

a _ repetition _ of length @xmath7 ( @xmath8 ) in a sequence of symbols is a subsequence of consecutive terms of the form @xmath9 . a sequence is _ nonrepetitive _ ( or _ square - free _ ) if it does not contain a repetition of any length . in 1906 thue proved that there exist arbitrarily long nonrepetitive sequences over an alphabet of size @xmath10 ( see @xcite ) . the method discovered by thue is constructive and uses substitutions over a given set of symbols . a different approach to creating long nonrepetitive sequences was recently introduced by grytczuk , kozik , and micek @xcite : generate a sequence by iteratively appending a random symbol at the end , and each time a repetition appears erase the repeated block . ( for instance , if the sequence generated so far is @xmath11 and we add @xmath12 , then we erase the last two symbols , bringing us back to @xmath13 . ) by a simple counting argument one can prove that with positive probability the length of the constructed sequence eventually exceeds any finite bound , provided the alphabet has size at least @xmath0 . this is one more than in thue s result but the proof is more flexible and can be adapted to other settings . for instance , it led to a very short proof that for every @xmath14 and every sequence of sets @xmath15 , each of size at least @xmath0 , there exists a nonrepetitive sequence @xmath16 where @xmath17 for all @xmath18 ( see @xcite ) , a theorem first proved by grytczuk , przybyo , and zhu @xcite via an intricate application of the lefthanded local lemma . whether the analogous statement for lists of size @xmath10 is true remains an exciting open problem . in this paper we make use of the above - mentioned approach to color trees nonrepetitively . given an ( undirected , simple ) graph @xmath19 , we denote by @xmath20 and @xmath21 its vertex set and edge set , respectively . a coloring @xmath22 of the vertices of @xmath19 is _ nonrepetitive _ if there is no repetition in the color sequence of any path in @xmath19 ; that is , @xmath23 is nonrepetitive if for every path @xmath24 with an even number of vertices the sequence of colors on the first half of @xmath24 is distinct from the sequence of colors on the second half of @xmath24 . ( we remark that all paths in this paper are simple , that is , contain no repeated vertex . ) the minimum number of colors used in a nonrepetitive coloring of @xmath19 is called the _ thue chromatic number _ of @xmath19 and is denoted by @xmath25 . now , given a graph @xmath19 , suppose that each vertex @xmath26 has a preassigned list of available colors @xmath27 . a coloring of @xmath19 with these lists is a coloring @xmath23 of @xmath19 such that @xmath28 for each vertex @xmath26 . the _ thue choice number _ of @xmath19 , denoted by @xmath29 , is the minimum @xmath2 such that , for every list assignment @xmath30 with @xmath31 for each @xmath26 , there is a nonrepetitive coloring of @xmath19 with these lists . similarly as for many graph coloring parameters , the thue chromatic ( choice ) number can be bounded from above by a function of the maximum degree : alon , grytczuk , hauszczak , and riordan @xcite proved that for every graph @xmath19 with maximum degree @xmath32 we have @xmath33 for some absolute constant @xmath12 . a number of subsequent works @xcite focused on reducing the value of the constant @xmath12 , the current best bound being @xmath34 ( see @xcite ) . et al . _ @xcite also showed that there are graphs with maximum degree @xmath32 with @xmath35 . ( whether this can be improved by a @xmath36 factor remains an open problem . ) it is not difficult to show that every tree has thue chromatic number at most @xmath0 ( see @xcite ) , which is best possible . this result was generalized to graphs of bounded treewidth by kndgen and pelsmajer @xcite . they proved that @xmath37 for every graph @xmath19 of treewidth @xmath4 . it is not known whether this upper bound can be improved to a polynomial in @xmath4 . however , if one considers graphs of pathwidth @xmath4 instead , a polynomial bound is known : it was shown by dujmovi _ _ @xcite that @xmath38 for every graph @xmath19 of pathwidth @xmath4 . ( we note that quadratic might not be the right order of magnitude here . ) probably the most intriguing open problem regarding the thue chromatic number is whether it is bounded for all planar graphs , a question originally asked by grytczuk @xcite . a @xmath39 upper bound is known @xcite , and from below ochem constructed a planar graph requiring @xmath40 colors ( see @xcite ) . the main focus of this paper is the list version of the parameter , the thue choice number . as mentioned at the beginning of the introduction , we have @xmath41 for every path @xmath24 , and it is open whether this bound can be improved to @xmath10 . fiorenzi , ochem , ossona de mendez , and zhu @xcite gave the first example of a class of graphs where the thue chromatic and thue choice numbers behave very differently : while trees have thue chromatic number at most @xmath0 , they showed that the thue choice number of trees is unbounded . clearly , trees with large thue choice number must have large maximum degree , and in fact one can deduce from the proof in @xcite that there are trees with maximum degree @xmath32 and thue choice number @xmath42 . kozik and micek @xcite subsequently showed that a better - than - quadratic upper bound in terms of the maximum degree exists for trees : for every @xmath43 there exists @xmath44 such that @xmath45 for every tree @xmath46 of maximum degree @xmath32 . ( bridging the significant gap between the upper and lower bounds remains an open problem . ) note that graphs of bounded treewidth have unbounded thue choice number since this is already the case for trees . on the other hand , dujmovi _ et al . _ @xcite observed that @xmath29 is bounded when @xmath19 is a graph of pathwidth @xmath47 . this prompted the authors of @xcite to ask whether @xmath29 is bounded more generally when @xmath19 has bounded pathwidth ( which is the case for the thue chromatic number ) . also , since connected graphs @xmath19 of pathwidth @xmath47 are caterpillars , and thus trees in particular , they also asked the same question but with @xmath19 moreover required to be a tree . a second motivation for the latter question was that the trees with arbitrarily large thue choice number constructed by fiorenzi _ et al . _ @xcite also have unbounded pathwidth . in this paper we answer both questions . first , we give a simple construction showing that the thue choice number is unbounded for graphs of bounded pathwidth ; in fact , this is true even for graphs of pathwidth @xmath6 ( which is best possible as noted above ) : [ thm : pw2 ] for every @xmath48 , there is a graph @xmath19 of pathwidth @xmath6 with @xmath49 . next , we address the case of trees and prove that their thue choice number is bounded from above by a function of their pathwidth : [ thm : tree ] there is a function @xmath50 such that @xmath51 for every tree @xmath46 of pathwidth @xmath4 . the proof of theorem [ thm : tree ] combines an induction on the pathwidth with the algorithmic method of grytczuk _ et al . _ @xcite to produce arbitrarily long nonrepetitive sequences described at the beginning of the introduction . this method , which finds its roots in the celebrated algorithmic proof of the local lemma by moser and tardos @xcite , was extended to produce nonrepetitive colorings of graphs ( in @xcite ) and trees ( in @xcite ) . part of our proof consists in adapting the ideas from @xcite to the situation under consideration . we note that the bounding function @xmath52 in theorem [ thm : tree ] stemming from our proof is quite large , it is doubly exponential in @xmath4 . the paper is organized as follows : in section [ sec : definitions ] we introduce definitions and terminology . then we prove theorem [ thm : pw2 ] in section [ sec : pw2 ] , and theorem [ thm : tree ] in section [ sec : trees ] . for an integer @xmath14 , we let @xmath53 : = { \left\{1,\ldots , n\right\}}$ ] . also , given two integers @xmath54 with @xmath55 we let @xmath56:={\left\{a , a+1,\dots , b\right\}}$ ] , which we call an _ _ interval . graphs in this paper are finite , simple , and undirected . the vertex set and edge set of a graph @xmath19 are denoted @xmath20 and @xmath21 , respectively . note that , since only simple graphs are considered , resulting loops and parallel edges are removed when contracting edges in a graph . a graph @xmath57 is a _ _ minor of a graph @xmath19 if @xmath57 can be obtained from a subgraph of @xmath19 by contracting edges . a _ _ tree decomposition of a graph @xmath19 is a pair @xmath58 where @xmath46 is a tree and @xmath59 is a collection @xmath60 of non - empty subtrees of @xmath46 such that @xmath61 for every edge @xmath62 . the _ _ width of the tree decomposition @xmath58 is the maximum , over every @xmath63 , of the number of subtrees in @xmath59 containing @xmath64 , minus @xmath47 . the _ _ treewidth of @xmath19 is the minimum width of a tree decomposition of @xmath19 . _ _ path decompositions and _ _ pathwidth are defined analogously with the tree @xmath46 required instead to be a path . treewidth and pathwidth are minor - closed parameters , in the sense that every minor of a graph @xmath19 has treewidth ( pathwidth ) at most that of @xmath19 . we refer the reader to diestel s textbook @xcite for an introduction to the theory of treewidth and graph minors . the _ _ length of a path is the number of its edges . the _ _ height of a rooted tree @xmath46 is the maximum length of a path from the root to a leaf of @xmath46 . ( thus @xmath46 has height @xmath65 if it consists of a unique vertex . ) the _ _ height of a vertex @xmath66 of @xmath46 is the length of the path from the root to @xmath66 in @xmath46 . let @xmath67 be the graph constructed from the path on @xmath68 vertices where every second vertex is blown up to @xmath69 vertices forming an independent set . formally , @xmath70\right\ } } \cup { \left\{v_{2i}^j\mid i\in[n],\ j\in\left[\binom{\ell n}{\ell}\right]\right\}},\ ] ] and two vertices are adjacent in @xmath67 if and only if their lower indices differ by exactly @xmath47 . also , let @xmath71 for each odd index @xmath72 $ ] and @xmath73\right\}}$ ] for each even index @xmath72 $ ] . it is not difficult to check that @xmath67 has pathwidth at most @xmath6 ( with equality for @xmath74 and @xmath48 ) . thus theorem [ thm : pw2 ] follows from the following theorem . let @xmath2 and @xmath75 be integers such that @xmath48 and @xmath76 . then @xmath77 . consider the following list assignment for the vertices of @xmath67 . for each odd index @xmath78 $ ] , vertex @xmath79 is assigned the list @xmath80 thus these @xmath75 lists have size @xmath2 , are pairwise disjoint , and their union is @xmath81 $ ] . next , enumerate the @xmath2-subsets of @xmath81 $ ] in an arbitrary way . then , for each even index @xmath82 $ ] and index @xmath83 $ ] , vertex @xmath84 is assigned the list which is the @xmath85-th set in that enumeration . we claim that , because @xmath75 was chosen to be strictly larger than @xmath86 , there can not be a nonrepetitive coloring of @xmath67 with these lists . arguing by contradiction , let us suppose that @xmath23 is such a coloring . with a slight abuse of notation , for @xmath72 $ ] we use the shorthand @xmath87 for the set @xmath88 . consider an interval @xmath89 $ ] of the form @xmath90 $ ] with @xmath91 . suppose that the following two conditions are satisfied : @xmath92 , i \text { odd } \\[1ex ] \phi(v_{i+2k+1 } ) \in \phi(v_{i } ) & \text{for each } i \in [ a , a+2k ] , i \text { even . } \end{array}\ ] ] then it is easy to check that there exists a path @xmath93 in @xmath67 with @xmath94 for @xmath85 odd and @xmath95 for @xmath85 even , for all @xmath96 , such that the color sequence @xmath97 is a repetition ( of size @xmath98 ) . since this can not happen , it follows that there exists an index @xmath99 $ ] for which one of the above two conditions is not satisfied . for every such index @xmath18 , we say that the pair @xmath100 is a _ _ witness ( for interval @xmath101 ) , where @xmath102 with @xmath103 odd and @xmath104 even . next , consider an even index @xmath105 $ ] . observe that @xmath106 - \phi(v_{q})\right| { \leqslant}\ell -1 $ ] , since @xmath107 contains at least one color from each @xmath2-subset of @xmath81 $ ] . combining this with the fact that vertices @xmath108 with odd index @xmath109 $ ] have pairwise disjoint lists , we deduce that there are at most @xmath110 odd indices @xmath109 $ ] such that the pair @xmath100 is a witness . summing up over every even index @xmath105 $ ] , it follows that there are at most @xmath111 distinct witnesses in total . now consider a witness @xmath100 and let @xmath112 . the pair @xmath100 is a witness for at most @xmath98 intervals @xmath89 $ ] of the form @xmath90 $ ] . since there are exactly @xmath113 intervals @xmath101 of the latter form and each interval of that form must have a witness , it follows that the number of witnesses @xmath100 with @xmath112 is at least @xmath114 . summing up over every possible value of @xmath4 ( that is , @xmath115 ) , we obtain that the total number of witnesses is at least @xmath116 it follows that @xmath117 which contradicts the assumption that @xmath118 . a _ _ path - partition of a tree @xmath46 is a pair @xmath119 where @xmath120 is a rooted tree and @xmath121 is a collection @xmath122 of vertex - disjoint paths of @xmath46 which collectively partition the vertex set of @xmath46 and such that @xmath123 if and only if there is an edge between a vertex from @xmath124 and a vertex from @xmath125 in @xmath46 . observe that a consequence of this definition is that @xmath120 is a minor of @xmath46 . the _ _ root - path of @xmath119 is the path @xmath126 where @xmath64 is the root of @xmath120 . now , consider a path @xmath126 with @xmath64 distinct from the root . the path @xmath126 has a _ _ center , defined as the endpoint in @xmath124 of the edge in @xmath46 linking @xmath124 to @xmath125 where @xmath127 is the parent of @xmath64 in @xmath120 . the _ _ height of the path - partition @xmath119 is the height of @xmath120 . when considering a path - partition @xmath119 of a tree @xmath46 , it will be useful to embed @xmath46 itself in the plane in a way that is ` faithful ' to the path - partition . this leads to the following definition : an embedding of @xmath46 in the plane is _ _ faithful to the path - partition @xmath119 if each path in @xmath121 is drawn horizontally , and contracting each such path into one of its vertices we obtain some plane embedding of @xmath120 , with its root drawn at the bottom and its edges going up . see figure [ fig : representation ] for an illustration . left : a tree @xmath46 faithfully embedded according to some path - partition @xmath119 ( the paths in @xmath121 are drawn in bold ) . right : the rooted tree @xmath120 . , scaledwidth=93.0% ] as the paths in @xmath121 are drawn horizontally , they have a natural orientation from left to right . every edge @xmath128 of @xmath46 is either _ _ horizontal or _ _ vertical , depending on whether @xmath128 belongs to some path in @xmath121 or not . our motivation for considering path - partitions is the following lemma . [ lemma : representation ] every tree of pathwidth @xmath4 has a path - partition of height at most @xmath129 . we prove the following stronger statement : for every tree @xmath46 of pathwidth @xmath4 and every vertex @xmath130 , there is a path - partition of @xmath46 of height at most @xmath129 with @xmath131 in the root - path . the proof is by induction on @xmath4 . for @xmath132 , the tree @xmath46 consists of the single vertex @xmath131 . clearly , it has a path - partition of height @xmath65 with @xmath131 in the root - path . now suppose @xmath133 for the inductive case . let @xmath134 be a path decomposition of @xmath46 of width @xmath4 , where @xmath135 denotes the path associated to vertex @xmath136 . enumerate the vertices of the path @xmath24 indexing the path decomposition as @xmath137 , in order . we may assume without loss of generality that there are ( non - necessarily distinct ) vertices @xmath138 such that @xmath139 and @xmath140 ( otherwise @xmath24 could be shortened ) . let @xmath141 denote the unique path in @xmath46 between @xmath142 and @xmath143 in @xmath46 . let @xmath144 be the vertex of @xmath145 that is closest to @xmath127 in @xmath46 . let @xmath146 denote the unique path in @xmath46 between @xmath147 and @xmath148 in @xmath46 . notice that @xmath149 for each @xmath150 and that each @xmath151 induces a subpath of @xmath24 . similarly , @xmath152 for each @xmath153 and each @xmath154 induces a subpath of @xmath24 . thus we deduce that @xmath155 . consider the forest @xmath156 and let @xmath157 denote its components . observe that each tree @xmath158 ( @xmath159 ) has pathwidth at most @xmath160 . indeed , @xmath161 is a path decomposition of @xmath158 , and for each @xmath162 we have @xmath163 . for each @xmath159 , let @xmath164 denote the unique vertex of @xmath158 having a neighbor in @xmath165 in @xmath46 . by induction , each tree @xmath158 ( @xmath159 ) has a path - partition @xmath166 of height at most @xmath167 such that @xmath164 in the root - path . let @xmath168 denote the root of @xmath169 . we construct a path - partition @xmath170 of @xmath46 as follows : @xmath120 consists of the disjoint union of @xmath171 plus two extra vertices @xmath172 and @xmath173 with @xmath172 the root of @xmath120 and @xmath173 a child of @xmath172 . the paths associated to @xmath172 and @xmath173 are @xmath145 and @xmath174 , respectively . for each @xmath175 , we make @xmath172 or @xmath173 adjacent to @xmath168 , depending whether @xmath164 has a neighbor in @xmath145 or @xmath174 in @xmath46 . it is easy to verify that @xmath170 is a path - partition of @xmath46 of height at most @xmath176 . the fact that trees of bounded pathwidth have path - partitions of bounded height is a natural observation . it is thus likely that this observation was made before though we are not aware of any relevant reference . we also note that we made no effort to optimize the bound in lemma [ lemma : representation ] and we do not know whether the factor @xmath6 is unavoidable . let @xmath46 be a tree and fix a path - partition @xmath119 of @xmath46 . suppose further that @xmath46 is embedded in the plane faithfully to @xmath119 . we use the following terminology when discussing paths in @xmath46 . first , every path @xmath177 has a corresponding _ _ level , which is defined as the height of the corresponding vertex in @xmath120 . by extension , every vertex of @xmath46 has a level , the level of the path in @xmath121 it belongs to . define the _ _ base of an arbitrary path @xmath24 in @xmath46 as the subpath induced by the vertices of @xmath24 of minimum level . since the base of @xmath24 is a subpath of a path in @xmath121 , its vertices are ordered from left to right by the plane embedding of @xmath46 . the path @xmath24 is said to be _ _ ascending if at least one of its two endpoints belongs to its base . ( we note that in particular all paths in @xmath121 are ascending , even though they are drawn horizontally in the embedding of @xmath46 . ) each ascending path @xmath24 in @xmath46 has a _ _ source , defined as the endpoint of @xmath24 that is in the base of @xmath24 ; in case both endpoints are in the base , the left - most one is selected as the source . we typically think of ascending paths @xmath24 as being directed from their source to their other endpoint so that the notion of @xmath18th vertex of @xmath24 is well defined , the first vertex being the source . an ascending path @xmath24 with at least two vertices either _ _ goes right or _ _ goes left or _ _ goes up , depending on whether the second vertex of @xmath24 is on the base and to the right of the source , or on the base and to the left of the source , or is one level higher . next we generalize the notion of repetition as follows . a _ _ near repetition is a sequence of the form @xmath178 , where @xmath8 is its _ _ length , and @xmath179 is said to be its _ _ gap . ( thus for @xmath180 this is the usual notion of repetition . ) now , let us return to our tree @xmath46 from the previous paragraph , and let @xmath23 denote an arbitrary coloring of its vertices . a slightly technical but key definition for our purposes is the following : an ascending path @xmath24 of @xmath46 is said to be _ _ @xmath23-bad if , enumerating its vertices as @xmath181 starting from its source , the sequence @xmath182 forms a near repetition @xmath178 of length @xmath7 and gap @xmath183 where at most @xmath7 vertices from @xmath184 lie in the base of @xmath24 . ( that is , either @xmath185 , or @xmath186 but at most @xmath7 vertices from the ` gap ' section are in the base of @xmath24 . ) an ascending path that is not @xmath23-bad is said to be _ _ @xmath23-good . we sometimes drop @xmath23 when using these two adjectives if the coloring @xmath23 they refer to is clear from the context equipped with these definitions we may now state the following technical lemma , which turns out to be the heart of the proof . [ lemma : good ] there is a function @xmath187 such that , for every @xmath1 , every @xmath188 , and every tree @xmath46 faithfully embedded according to a path - partition @xmath119 of @xmath46 of height @xmath189 with lists @xmath190 @xmath191 of colors of size @xmath192 , one can find sublists @xmath193 @xmath191 of size @xmath2 such that , for every coloring @xmath23 of @xmath46 with these sublists , every ascending path of @xmath46 is @xmath23-good . in order to motivate lemma [ lemma : good ] , we show that with only a little extra effort ( greedy coloring from the sublists ) it implies theorem [ thm : tree ] . let @xmath46 be a tree of pathwidth @xmath4 and let @xmath190 @xmath191 be a list assignment for the vertices of @xmath46 where each list has size @xmath194 , where @xmath3 is the function from lemma [ lemma : good ] . by lemma [ lemma : representation ] there is a path - partition @xmath170 of @xmath46 of height at most @xmath129 . by lemma [ lemma : good ] there are sublists @xmath195 with @xmath196 and @xmath197 for each vertex @xmath136 , such that in any coloring @xmath23 of @xmath46 with these sublists , all ascending paths in @xmath46 are @xmath23-good . we define a nonrepetitive coloring @xmath23 of @xmath46 with the lists @xmath198 ( @xmath136 ) in a greedy manner . we color the vertices of @xmath46 one by one in non - decreasing order of their levels . let @xmath136 be a vertex under consideration . let @xmath199 denote the shortest path in @xmath46 from @xmath200 to the root - path ( thus it enters the root - path in vertex @xmath201 ) . recall that every edge of the form @xmath202 with @xmath203 is either horizontal or vertical in @xmath46 let @xmath204 each vertex in @xmath205 is said to be a _ _ guard for vertex @xmath66 . note that @xmath206 is exactly the level of vertex @xmath66 in @xmath46 , and thus in particular @xmath207 . we color @xmath66 as follows : let @xmath208 be an arbitrarily chosen color from the non - empty set @xmath209 . ( here , @xmath210 denotes the set of colors used for vertices in @xmath205 ; note that these vertices are already colored since they lie on lower levels . ) we claim that @xmath23 is a nonrepetitive coloring of @xmath46 . arguing by contradiction , suppose that there is a repetitively colored path @xmath211 . consider the edge @xmath212 . first we show that @xmath128 belongs to the base of @xmath24 . suppose not , and consider the shortest subpath of @xmath24 that includes the edge @xmath128 and has one endpoint in the base of @xmath24 . reversing @xmath24 if necessary , we may assume without loss of generality that this subpath is of the form @xmath213 , with @xmath214 being the only vertex on the base . observe that @xmath215 is a vertical edge . this implies that @xmath214 is a guard for all the vertices in @xmath216 . in particular , the color @xmath217 can not have been used for vertex @xmath218 since @xmath219 , contradicting the fact that @xmath220 . therefore , the edge @xmath128 must lie in the base of @xmath24 . let @xmath2 and @xmath7 be the number of vertices in @xmath221 and @xmath222 , respectively , that are in the base of @xmath24 . reversing @xmath24 if necessary , we may assume without loss of generality that @xmath223 . consider the path @xmath224 . observe that @xmath225 is an ascending path as one of its endpoints , namely @xmath226 , is in the base of @xmath225 . now , @xmath227 is a near repetition of length @xmath7 with gap @xmath228 , and exactly @xmath2 vertices from the gap section are in the base of @xmath225 . since @xmath229 , we deduce that @xmath225 is @xmath23-bad , contradicting the fact that every ascending path is @xmath23-good . an _ _ arborescence is a rooted directed tree where the edges are directed away from the root . it will be convenient to consider arborescences that are embedded in the plane without edge crossings in such a way that the root is drawn at the bottom and all arcs go up ( thus the source of an arc is drawn below its sink ) , which we simply call _ _ plane arborescences . the _ _ height of a vertex in an arborescence is defined as its distance to the root , thus in particular the root has height @xmath65 . the _ _ rightmost path of a plane arborescence is the path obtained by starting from the root and always taking the rightmost arc going up , until reaching a leaf . we classify directed paths in a plane arborescence @xmath230 as being good or bad w.r.t . a given coloring @xmath23 of @xmath230 , similarly as for ascending paths : say that a directed path @xmath24 is _ _ @xmath23-bad if , enumerating its vertices as @xmath231 in order , the sequence @xmath182 can be written as a near repetition @xmath178 of length @xmath7 and gap @xmath183 where at most @xmath7 vertices from @xmath184 lie on the rightmost path of @xmath230 . ( that is , either @xmath185 , or @xmath186 but at most @xmath7 vertices from the ` gap ' section are on the rightmost path . ) if the directed path @xmath24 is not @xmath23-bad then it is _ _ @xmath23-good . [ lemma : more - technical ] let @xmath1 , let @xmath230 be a plane arborescence , and let @xmath190 @xmath232 be lists of colors of size @xmath233 . then one can find sublists @xmath193 @xmath232 of size @xmath2 such that , for every coloring @xmath23 of @xmath230 with these sublists , every directed path starting on the rightmost path is @xmath23-good . the interest of lemma [ lemma : more - technical ] is that lemma [ lemma : good ] can be proved by iterated applications of lemma [ lemma : more - technical ] , as we now show . the function @xmath192 that will be used is defined inductively on @xmath189 as follows : @xmath234 , and @xmath235 for @xmath236 . let @xmath46 be a tree with a path - partition @xmath170 of height @xmath189 and let @xmath190 @xmath191 be lists of colors of size @xmath192 . suppose further that @xmath46 is faithfully embedded according to @xmath170 . we prove the lemma by induction on @xmath189 . for the base case of the induction , @xmath237 , we observe that @xmath46 is then a path and all ascending paths in @xmath46 are simply subpaths of @xmath46 . as @xmath238 , by lemma [ lemma : more - technical ] there are sublists @xmath196 for each vertex @xmath136 with @xmath239 such that , for every coloring @xmath23 of @xmath46 with these sublists , all ascending paths of @xmath46 are @xmath23-good , as required . ( as expected , when applying lemma [ lemma : more - technical ] we first turn the path @xmath46 into an arborescence by directing it from left to right . ) for the inductive case @xmath236 , let @xmath64 be the root of @xmath120 and let @xmath124 be the root - path of @xmath46 . let also @xmath157 be the components of the forest @xmath240 . ( note that there is at least one component . ) for each @xmath241 , the path - partition @xmath170 induces in a natural way a path - partition @xmath242 of @xmath243 of height at most @xmath244 , with @xmath245 rooted at the only vertex that is a neighbor of @xmath64 in @xmath120 . since @xmath246 , applying induction on @xmath243 we obtain for each vertex @xmath247 a sublist @xmath248 of size @xmath249 such that , for every coloring @xmath23 of @xmath250 with these sublists , every ascending path of @xmath250 is @xmath23-good . next , for each vertex @xmath251 let @xmath252 be an arbitrary subset of @xmath190 of size @xmath249 . thus , every vertex @xmath66 of @xmath46 now has a corresponding sublist @xmath253 of size @xmath249 . moreover , given any coloring @xmath23 of the tree @xmath46 with these sublists , the only ascending paths that could possibly be @xmath23-bad are those having their sources in @xmath124 . we shall refer to these ascending paths as the _ _ risky paths of @xmath46 . enumerate the vertices of the root - path @xmath124 as @xmath254 , from left to right . define two plane arborescences @xmath230 and @xmath255 from @xmath46 by rooting @xmath46 at @xmath256 and @xmath257 , respectively , and ensuring that @xmath124 is a prefix of the rightmost path in both instances . note that the rightmost path of @xmath230 could extend beyond @xmath124 ( in case @xmath257 is not a leaf of @xmath46 ) , and the same is true for the rightmost path of @xmath255 ( if @xmath256 is not a leaf ) . what is important for our purposes is to observe that each risky path of @xmath46 starts on the rightmost path in both @xmath230 and @xmath255 . observe also that each risky path of @xmath46 that goes right ( left ) is a directed path in @xmath230 ( respectively @xmath255 ) , and risky paths that go up are directed in both @xmath230 and @xmath255 . first , apply lemma [ lemma : more - technical ] on @xmath230 with list assignment @xmath252 ( @xmath258 ) , giving for each vertex @xmath136 a sublist @xmath259 of size @xmath233 . next , apply lemma [ lemma : more - technical ] on @xmath255 with list assignment @xmath260 ( @xmath258 ) , giving for each vertex @xmath136 a sublist @xmath261 of size @xmath2 . since every risky path of @xmath46 is mapped to a directed path starting on the rightmost path in @xmath230 or @xmath255 , by the properties of the sublists @xmath262 and @xmath198 ( @xmath136 ) guaranteed by lemma [ lemma : more - technical ] we know that , for every coloring @xmath23 of @xmath46 with the lists @xmath198 ( @xmath136 ) , all risky paths of @xmath46 are @xmath23-good . therefore , the lists @xmath198 ( @xmath136 ) have the desired properties . it remains to prove lemma [ lemma : more - technical ] . as alluded to in the introduction , we will do so by adapting the algorithmic method used in @xcite . let @xmath263 denote the size of the lists . for @xmath258 , let @xmath264 denote the set of vertices @xmath265 that can be reached via a directed path from @xmath66 in @xmath230 . ( note that @xmath266 . ) in the proof , we will often abbreviate ` subset of size @xmath4 ' and ` sublist of size @xmath4 ' into _ _ ` @xmath4-subset ' and _ _ ` @xmath4-sublist ' , respectively . we define a simple randomized algorithm , algorithm [ algo:1 ] , that tries to find an @xmath2-sublist @xmath198 of @xmath190 for each vertex @xmath258 such that , for every coloring @xmath23 of @xmath230 with these sublists , every directed path starting on the rightmost path of @xmath230 is @xmath23-good . the following informal description of the algorithm is complemented by the more formal description given in algorithm [ algo:1 ] . the algorithm explores the arborescence @xmath230 via a depth - first , left - to - right search starting from the root . the algorithm maintains at all time @xmath2-sublists @xmath196 for all vertices @xmath66 encountered _ before _ the current vertex @xmath131 in the depth - first search of @xmath230 . these sublists have the following property : for every coloring @xmath23 of these vertices with these sublists ( @xmath23 being thus a partial coloring of @xmath230 ) , every directed path starting on the rightmost path of @xmath230 that is fully colored is @xmath23-good . we say that such a partial sublist assignment is _ _ valid . next , the algorithm treats the current vertex @xmath131 and tries to maintain the above property . to do so , the algorithm first chooses an @xmath2-sublist @xmath267 uniformly at random . if this new sublist @xmath268 triggers the existence of a @xmath23-bad path in @xmath230 for some ( partial ) coloring @xmath23 with the current sublists that is , the current sublist assignment is no longer valid it erases some of these sublists as follows : say @xmath269 with @xmath270 is a @xmath23-bad path with color sequence @xmath271 of the form @xmath272 . the algorithm then erases the choice for the list @xmath198 for all vertices @xmath66 contributing to the second occurrence of the repeated sequence and their descendants , that is , for all @xmath273 . at the next iteration , @xmath274 becomes the new current vertex , that is , the next vertex to be treated . notice that this makes the algorithm backtrack a number of steps w.r.t . the depth - first left - to - right search of @xmath230 . if on the other hand , the new sublist @xmath268 does not trigger any such bad configuration , then the current sublist assignment remains valid . in this case , before proceeding to the next random choice the algorithm first tries to extend the current sublist assignment deterministically as much as possible . ( while it might not be clear at first glance why this deterministic extension step is needed , we remark that it is actually a key feature of the algorithm without which we could not do the analysis below . ) this is done as follows : the algorithm considers the children @xmath275 of @xmath131 one by one in left - to - right order , until a _ _ problematic child is identified : when considering @xmath276 , the algorithm checks whether there exist @xmath2-subsets @xmath196 for all @xmath277 such that , taken together , they extend the current sublist assignment in such a way that it remains valid . if these subsets exist , the current sublist assignment is extended in this way to the whole subtree rooted at @xmath276 , and the algorithm considers the next child of @xmath131 . ( if there are more than one valid choice for these sublists , the algorithm chooses one according to a deterministic rule . ) if no such extension of the current sublist assignment can be found for vertices in @xmath278 , then @xmath276 is identified as being a problematic child of @xmath131 , and @xmath276 becomes the next vertex to be treated . observe that this effectively makes the algorithm proceed with the depth - first left - to - right search of @xmath230 for some number of steps . let us make some observations concerning the algorithm : right at the beginning , after selecting a sublist for the root of @xmath230 , two situations can occur : ( 1 ) no child of the root is problematic . thus a valid sublist assignment for all vertices of @xmath230 has been found , and the lemma is proved . ( 2 ) some child of the root is problematic . in this case , it is important to observe that later on each vertex @xmath131 for which a random sublist @xmath268 is chosen was problematic when its parent was considered . it follows in turn that some child of @xmath131 will be problematic , since otherwise we could extend the sublist assignment to the whole subtree rooted at @xmath131 . to summarize , we may assume that we are in case ( 2 ) at the beginning , since otherwise we are done . this implies that the vertex @xmath131 that is currently being treated by the algorithm always has a problematic child . moreover , the algorithm _ will never stop _ , simply because while it can erase the choices of sublists for some vertices of @xmath230 it can not do so for the root , as is easily checked . our proof will then proceed in the following way : we run the algorithm until it made @xmath279 random choices of sublists and then stop it , where @xmath279 will be some large number which is a function of @xmath280 and @xmath2 . we then carefully set up a concise description ( called _ _ log ) of its execution that is precise enough to allow us to recover from it all random choices that were made by the algorithm . finally , we count the number of distinct logs that can occur after @xmath279 random choices , and show that , for sufficiently large @xmath279 , this number is strictly less than @xmath281 . from this we deduce that not all sequences of @xmath279 random choices of sublists can occur in case ( 2 ) . in other words , there is a choice for the sublist of the root of @xmath230 leaving us in case ( 1 ) , which then finishes the proof . this concludes our informal description of the algorithm , see algorithm [ algo:1 ] for the pseudo code . a few remarks about the latter are in order : first , we assume that the @xmath2-subsets of @xmath190 have been enumerated for each @xmath282 , so that the @xmath85-th @xmath2-subset of @xmath190 is well defined for @xmath283 $ ] . this ordering also induces an ordering on every subcollection of the collection of @xmath2-subsets of @xmath190 . we also use this enumeration in the proof . second , for simplicity we model the random choices made by the algorithm by a sequence @xmath284 of numbers given in input , each between @xmath47 and @xmath285 , where @xmath286 will be the number used for the @xmath18-th random choice . we call this sequence the _ _ random input . third , in line [ algo : choice - bad - path ] , the @xmath23-bad path is chosen according to some fixed rule . similarly , in line [ algo : choice - extension ] , the sublists @xmath198 are chosen according to some fixed rule . ( in each case , the actual rule is irrelevant , as long as it is deterministic . ) * input : * lists @xmath190 for all @xmath258 and random input @xmath284 + @xmath287 + @xmath288 + @xmath289 for each @xmath282 + * report failure * in the following , by the @xmath18-th iteration of the algorithm , we mean the @xmath18-th iteration of the ` while ` loop . we call operations in lines [ algo : def - of - g - and - k]-[algo : new - current ] a _ _ retraction of the near repetition @xmath178 . with a slight abuse of terminology , we will also say that the corresponding @xmath23-bad path has been _ _ retracted . from now on we argue by contradiction and suppose that the desired sublists for the vertices of @xmath230 do not exist . in other words , we assume that every choice for the sublist of the root at the beginning of the algorithm leaves us in case ( 2 ) described above . in particular , for all @xmath279 and all random inputs @xmath284 , algorithm [ algo:1 ] runs for @xmath279 steps and then reports failure . we are going to create a concise description of what algorithm [ algo:1 ] does during the @xmath279 steps of its execution . this description is completely determined by the lists and the random input . we see the lists @xmath190 ( @xmath258 ) as being fixed and thus treat the description as a function of the random input @xmath284 . the description , which we call an _ _ @xmath279-log , consists of a @xmath0-tuple @xmath290 defined as follows : @xmath291 and @xmath292 ( @xmath293 $ ] ) is the height of the vertex @xmath131 in algorithm [ algo:1 ] at the _ end _ of iteration @xmath18 , when reaching line [ algo : i - inc ] after @xmath131 was updated in the if else block . ( thus , for @xmath294 , @xmath292 is simply the height of vertex @xmath131 at the _ beginning _ of iteration @xmath295 . ) + @xmath296 \cup { \left\{{\textrm{\tt undefined}}\right\}}$ ] is a function encoding the final partial sublist assignment to vertices of @xmath230 at the end of iteration @xmath279 . more precisely , for each @xmath258 we have @xmath297 if sublist @xmath198 is defined at that moment and @xmath198 is the @xmath85-th @xmath2-subset of @xmath190 , and @xmath298 if @xmath198 is not defined . + @xmath299 and @xmath300 ( @xmath301 $ ] ) are defined as follows : if no bad path was retracted during the @xmath18-th iteration then @xmath302 . otherwise , @xmath300 is the number of vertices in the prefix @xmath303 of the retracted bad path @xmath304 that are on the rightmost path of @xmath230 . observe that @xmath305 in the latter case since at most @xmath7 vertices from the gap section of a bad path lie on the rightmost path of @xmath230 . + @xmath306 , where @xmath307 ( @xmath308 $ ] ) is defined as follows . if no bad path was retracted during the @xmath18-th iteration then @xmath309 . otherwise , letting @xmath304 be the retracted bad path , we set @xmath310 with @xmath311 ( @xmath312 $ ] ) defined as follows : let @xmath313 denote the collection of @xmath2-subsets of @xmath314 that have a _ non - empty intersection with @xmath315_. then @xmath311 is the index of the set @xmath316 in the collection @xmath313 . now , our aim is to bound from above the number of distinct @xmath279-logs @xmath290 by a relatively small function of @xmath317 , and @xmath2 . recall that the lists @xmath190 @xmath232 are fixed , thus @xmath2 and @xmath280 are fixed , and only @xmath279 and the random input @xmath318 vary . there are exactly @xmath319 distinct random inputs of length @xmath279 , and our goal in the following analysis is to deduce that there are @xmath320 distinct @xmath279-logs . ( the asymptotic notation is to be interpreted with respect to the variable @xmath279 of course . ) this is then a contradiction for @xmath279 large enough , as mentioned earlier . we start by estimating the number of @xmath279-tuples @xmath321 . each sequence @xmath322 can be injectively mapped to its sequence of differences @xmath323 . ( note that @xmath324 . ) all numbers in this new sequence belong to the set @xmath325 , as is easily seen . next we transform that sequence into yet another sequence by replacing each number @xmath4 by @xmath47 followed by @xmath326 consecutive occurrences of @xmath327 . for instance , the sequence of differences @xmath328 gets mapped to @xmath329 . it is easy to see that the second transformation is also injective . the resulting sequence @xmath330 is a sequence over the alphabet @xmath331 . the number of @xmath47 s in @xmath330 corresponds to the number of times the algorithm assigns a value to some variable @xmath332 in line [ algo : def - sj ] , and is thus equal to the number of iterations , that is , @xmath279 . the number of @xmath327 s in @xmath330 is the sum of all values of @xmath7 over all bad paths @xmath269 considered in lines [ algo : def - of - g - and - k]-[algo : erase - sj ] during the execution . one can see this as the number of times the algorithm ` erases ' a value of @xmath333 for some @xmath258 that was set earlier using the random input ( note that an execution of line [ algo : erase - sj ] erases @xmath7 such values ) . thus , this number is at most the total number of executions of line [ algo : def - sj ] , that is , the number of @xmath47 s in @xmath330 , which is @xmath279 . hence , @xmath330 has size between @xmath279 and @xmath334 , and there are at most @xmath335 such sequences @xmath330 . next we bound the number of different functions @xmath336 . note that this number depends only on @xmath337 , @xmath2 , and @xmath280 , so it can be treated as a constant w.r.t . we denote this number by @xmath12 ( its exact value being irrelevant for the analysis ) . now we turn our attention to the number of possible tuples @xmath338 in an @xmath279-log @xmath290 . recall that if @xmath339 then @xmath305 , where @xmath7 is the size of the repeated part in the near repetition retracted during the @xmath18-th iteration . hence , the sum of the @xmath300 s is at most twice the total number of @xmath327 s in @xmath330 . this implies that @xmath340 . the number of such sequences is easily seen to be at most @xmath341 , as one can encode each such sequence as a binary word @xmath342 of length @xmath343 ( the @xmath18-th section @xmath344 encodes the number @xmath300 for @xmath345 , and the padding @xmath346 at the end of the word ensures that the length is @xmath343 ) . it remains to estimate the number of possible tuples @xmath306 which can occur for fixed @xmath347 , @xmath336 , and @xmath348 . consider an index @xmath301 $ ] . if @xmath349 , then each number in the sequence @xmath307 is the index of some @xmath2-subset of some list @xmath190 ( @xmath258 ) among those that intersect some other list @xmath350 of size @xmath2 . clearly , the number of @xmath2-subsets of @xmath190 intersecting @xmath350 is at most @xmath351 . since @xmath337 is much bigger than @xmath2 ( recall that @xmath352 ) , each term in the previous sum is upper bounded by the first term @xmath353 , and we obtain that the sum is at most @xmath354 . ( of course , this is a rather crude upper bound but it is good enough for our purposes . ) hence all numbers in the sequence @xmath307 are between @xmath47 and @xmath355 . note also that the length of the sequence @xmath307 is exactly the length of the near repetition retracted during the @xmath18-th iteration . hence , given @xmath347 we know exactly which @xmath307 are defined and what are their lengths . the sum of these lengths is the total number of @xmath327 s in @xmath330 , which is at most @xmath279 . therefore , for a fixed @xmath347 there can be at most @xmath356 distinct sequences @xmath357 . putting all the previous observations together , we deduce that the number of distinct tuples @xmath290 is at most @xmath358 as desired . ( the @xmath359 follows from the fact that @xmath352 . ) this shows that , if @xmath279 is sufficiently large , then the number of possible @xmath279-logs is strictly smaller than @xmath319 , the number of random inputs of length @xmath279 . to obtain the desired contradiction , it remains to show that runs of the algorithm on different sources produce distinct @xmath279-logs , that is , that any @xmath279-log @xmath290 uniquely determines the random input used by the algorithm to produce it . this is exactly what we show next . consider an @xmath279-log @xmath290 and let @xmath318 be _ any _ random input that can lead to its production . we prove that @xmath318 are uniquely determined by induction on @xmath279 . this is clearly true if @xmath360 , since the function @xmath336 tells us explicitly which sublist was chosen for the root of @xmath230 . so assume @xmath361 for the inductive case . let @xmath362 , @xmath299 , and @xmath306 . first suppose that @xmath363 . then no near repetition was retracted during the @xmath279-th iteration . the vertex @xmath131 that was the current vertex at the beginning of the @xmath279-th iteration is determined by the function @xmath336 : it is the _ vertex @xmath364 in the depth - first left - to - right search order from the root such that @xmath365 that has a child @xmath366 with @xmath367 . ( note that the first such child @xmath366 is the problematic child @xmath276 of @xmath131 identified when exiting the inner while - loop . ) now , observe that @xmath368 is simply the index of @xmath369 among @xmath2-subsets of @xmath370 , and is thus completely determined by our @xmath279-log @xmath290 . having determined @xmath371 , we can use the inductive hypothesis to deduce that @xmath372 are also fully determined by the log as follows . let @xmath373 be the @xmath374-log resulting from the execution of the algorithm for @xmath375 iterations on random input @xmath372 . by induction , the latter sequence is uniquely determined by the @xmath374-log @xmath373 . hence , it is enough to show that @xmath373 is in turn uniquely determined by our initial @xmath279-log @xmath290 . clearly , @xmath376 as for @xmath377 , it is simply obtained from @xmath336 by letting the value of each vertex @xmath378 be @xmath379 . that is , @xmath380 { \mathcal{s}}(v ) & \textrm { otherwise } \end{array } \right.\ ] ] for each @xmath282 ( recall that @xmath381 ) . in the case when @xmath382 with @xmath383 , a near repetition was retracted during the @xmath279-th iteration with a repeated part of size @xmath7 . here we first show that @xmath372 are uniquely determined , and then we prove that the same holds for @xmath371 . let @xmath373 denote the @xmath374-log resulting from the execution of the algorithm for @xmath375 iterations on random input @xmath372 . we thus have : @xmath384 let us show that @xmath377 is completely determined by @xmath290 . let @xmath131 denote the current vertex at the beginning of the @xmath279-th iteration . ( remark : the vertex @xmath131 can be deduced from the log @xmath290 , as follows from the discussion below . ) during the @xmath279-th iteration the sublist @xmath332 of @xmath131 was assigned the @xmath371-th @xmath2-subset of @xmath385 . this triggered the existence of a bad path @xmath386 with @xmath270 , which was subsequently retracted , i.e. @xmath333 was then set to @xmath379 for all @xmath273 . now , we determine the first @xmath387 vertices @xmath388 of the bad path from the log . first , we show that @xmath274 is easily determined : since @xmath377 differs from @xmath336 only on the vertices from @xmath389 , vertex @xmath274 is the first vertex @xmath66 of @xmath230 in depth - first left - to - right order from the root with @xmath390 . ( we remark that @xmath274 would have been the next ` current vertex ' considered by the algorithm if it were run for an extra iteration . ) next , we determine the vertices @xmath391 from the log . let @xmath392 be the first vertex on the rightmost path of @xmath230 that is encountered when walking down towards the root of @xmath230 from @xmath274 ( note that we could have @xmath393 ) . by definition of the sequence @xmath348 , among the @xmath394 vertices @xmath395 the first @xmath396 vertices are on the rightmost path of @xmath230 , and none of the remaining @xmath397 vertices are . since @xmath398 , we may identify vertex @xmath256 by starting at vertex @xmath392 and walking down towards the root of @xmath230 either @xmath399 steps ( if @xmath393 ) or @xmath400 steps ( if @xmath401 ) . now , since we know vertices @xmath402 , @xmath274 and the integer @xmath7 , looking at the path from @xmath402 to @xmath274 gives us all intermediate vertices @xmath403 ( if any ) , as well as the size @xmath183 of the gap section . hence , @xmath404 are determined by the log @xmath290 , as claimed . building on this , we now complete the proof . let @xmath405 . we first consider a simple case , namely @xmath406 . then @xmath407 , and at the beginning of the @xmath279-th iteration , the @xmath371-th @xmath2-sublist @xmath408 of @xmath385 was assigned to variable @xmath332 , which triggered the existence of the bad path . by the definition of the sequence @xmath409 , the sublist @xmath408 is the @xmath410-th @xmath2-sublist of @xmath411 having a non - empty intersection with @xmath412 . we can thus deduce @xmath408 from the log @xmath290 , and obtain in turn @xmath371 from @xmath408 . notice that in this case the sublist assignments to vertices of @xmath230 at the end of the @xmath374-th and at the end of the @xmath279-th iterations are exactly the same , that is , @xmath413 . ( indeed , this is what makes the case @xmath406 simpler . ) hence , @xmath373 is completely determined by @xmath290 . by induction , the sequence @xmath372 is uniquely determined by @xmath373 , and therefore it is also uniquely determined by @xmath290 . since we have seen that @xmath371 is uniquely determined as well , this concludes the @xmath406 case . now , suppose that @xmath414 . we can obtain @xmath415 from @xmath290 , since @xmath415 is the @xmath410-th @xmath2-sublist of @xmath411 having a non - empty intersection with @xmath412 . knowing @xmath415 , we can identify vertex @xmath416 as follows . consider the last time the variable @xmath417 was modified during the execution of the algorithm before the @xmath279-th iteration , say this is during the @xmath418-th iteration . thus , during that iteration , @xmath417 was assigned the set @xmath415 , and this did not trigger the existence of a bad path . then , the children @xmath419 of @xmath274 were inspected one by one in order , until a problematic child @xmath420 was found . this problematic child @xmath420 is vertex @xmath416 . this process is completely deterministic , thus we can simulate it . ( indeed , we know the whole sublist assignment for vertices of @xmath230 at the beginning of the @xmath418-th iteration , and we know the sublist that was sampled for @xmath274 during that iteration , namely , @xmath415 ) . thus , for each vertex @xmath421 , we can figure out how the sublist @xmath198 was set during that iteration , and this is exactly the value of @xmath422 , since those sublists have not been modified afterwards prior to iteration @xmath279 . notice also that @xmath423 for all @xmath424 , so it only remains to determine @xmath422 for @xmath425 . we can iterate this argument and discover step by step vertices @xmath426 and the missing entries of @xmath377 . we spell out the general argument now , for the sake of completeness ( the reader who is already convinced that it can be done is invited to skip this paragraph ) . for each index @xmath427 , we proceed as follows . the set @xmath428 is the @xmath429-th @xmath2-sublist of @xmath430 having a non - empty intersection with @xmath431 , and is thus determined by @xmath290 . knowing @xmath428 , we now identify vertex @xmath432 . say @xmath433 was modified for the last time during the @xmath434-th iteration of the algorithm before the @xmath279-th iteration . during that iteration , variable @xmath433 was assigned the set @xmath428 , and this did not result in the existence of any bad path . next , the children @xmath419 of @xmath435 were inspected until a problematic child @xmath420 was found , which is vertex @xmath432 . we know the whole sublist assignment for vertices of @xmath230 at the beginning of the @xmath434-th iteration ; indeed , this is exactly the one at the _ end _ of the @xmath436-th iteration ( as is easily checked ) , which we already know . we also know the sublist that was sampled for vertex @xmath435 during the @xmath434-th iteration . hence , we can simulate the execution of lines [ algo : choice - extension - begin][algo : choice - extension - end ] of the algorithm for the @xmath434-th iteration . this implies that , for each vertex @xmath421 , we can determine how the sublist @xmath198 was set during that iteration , and that sublist is precisely the set @xmath422 . also , we have that @xmath423 for all @xmath424 . this way we completely determined @xmath422 for all vertices @xmath437 for each @xmath438 $ ] . for all other vertices @xmath66 of @xmath230 , we have @xmath439 . thus , @xmath377 is completely determined by @xmath290 , and hence so is the @xmath374-th log @xmath373 , as claimed . equipped with the knowledge of @xmath377 , we may now finish the proof in a manner similar to the @xmath406 case . let @xmath408 denote the @xmath371-th @xmath2-sublist @xmath408 of @xmath440 . the sublist @xmath408 is the @xmath441-th @xmath2-sublist of @xmath440 having a non - empty intersection with @xmath442 , which is thus determined by the log @xmath290 . hence , we can obtain @xmath368 from @xmath290 . moreover , as we have seen , @xmath373 is completely determined by the log @xmath290 . by induction , the sequence @xmath372 is uniquely determined by @xmath373 , and therefore it is uniquely determined by @xmath290 . this concludes the proof . we remark that no effort has been made to optimize the bound of @xmath443 in lemma [ lemma : more - technical ] . we thank the two anonymous referees for their insightful comments , which greatly helped us improve the exposition .

in this paper we develop the _ continuous averaging _ method of treschev to work on the simultaneous diophantine approximation and apply the result to give a new proof of the nekhoroshev theorem . we obtain a sharp normal form theorem and an explicit estimate of the stability constants appearing in the nekhoroshev theorem .

1210.3884

in the papers @xcite , treschev developed a new averaging method called continuous averaging . it is a powerful tool to derive sharp constants in the exponentially small splitting problems in hamiltonian systems with one and a half degrees of freedom . but the technicality becomes very heavy when we use the method to study hamiltonian systems of more degrees of freedom . for this reason , the method has not been applied to other problems yet . in this paper , we use the continuous averaging to give a new proof of the nekhoroshev theorem . we consider the following analytic nearly integrable hamiltonian system : @xmath0 the phase space is @xmath1 we complexify the variables and extend the domain of @xmath2 to a @xmath3 neighborhood and that of @xmath4 to a @xmath5 neighborhood of the original domains respectively . the extended phase space to the complex domain is @xmath6 where @xmath5 is the width of analyticity in @xmath4 and @xmath3 is that of the slow variables @xmath7 . as stated in @xcite , nekhoroshev theorem ensures that when the unperturbed hamiltonian @xmath8 is quasi - convex , by which we mean that the set @xmath9 is strictly convex , the following general estimate holds for sufficiently small @xmath10:@xmath11 for some constants @xmath12 independent of @xmath13 , where @xmath14 is the action variable component of any orbit associated to hamiltonian ( [ eq : main ] ) with initial condition in the set @xmath15 . there are many works studying the stability exponents @xmath16 and @xmath17 ( c.f.@xcite ) . their approaches are based on a careful study of the geometric and number theoretical aspects of resonances . instead , in this paper we try to sharpen the estimates in the analytic part of the proof using continuous averaging to obtain an improved normal form ( see theorem [ thm : main ] ) . then we apply the normal form to lochak s argument to get the nekhoroshev theorem ( see theorem [ thm : global ] ) where all the stability constants are estimated explicitly . in this paper , we only work on the case @xmath18 . but the normal form theorem can be easily applied to other prescribed @xmath16 and @xmath17 to get the corresponding @xmath19 . the method of lochak is called the simultaneous diophantine approximation , which turns out to be an important alternative to the classical approach via small divisor techniques , as explained in @xcite . its main idea is to do the averaging in a vicinity of a periodic orbit . so it is essentially an averaging procedure for systems with one fast angle . in general , we can kill the dependence on the fast angle up to exponential smallness . this makes the simultaneous diophantine approximation suitable to prove the nekhoroshev theorem . the work @xcite can be considered as a development of the continuous averaging to the small divisor case . in this paper , it is the first time that the continuous averaging has been developed to the simultaneous diophantine approximation . we point out the relation between continuous averaging and some important pdes . the idea of the continuous averaging is to study the averaging procedure using pde instead of iterations . the pde has the form @xmath20 , where @xmath21 is the hilbert transform of @xmath22 in some special cases ( see section 3 for more details ) . this type of equation has been studied ( c.f . @xcite ) as a simplified model for quasi - geostrophic equation ( c.f . @xcite ) , incompressible euler equation , etc . it would be interesting if we could apply some pde techniques to our problem . to state our theorems , we need the following definitions . 1 . we use @xmath23 to denote the @xmath24 norms for a vector in @xmath25 or @xmath26 . without causing confusion , we also use @xmath27 to denote the absolute value of a function whose range is in @xmath28 or @xmath29 . 2 . for a function @xmath30 , the @xmath31 is defined as:@xmath32 if we have the fourier expansion @xmath33 , and the other variables @xmath34 . 3 . @xmath35 . [ def : norm ] we also use the following definition to characterize the convexity of the unperturbed part @xmath36 . consider a hamiltonian @xmath36 defined on @xmath37 . then , we define the associated constants @xmath38 to characterize the convexity of @xmath36 . @xmath39 [ def : convex ] now we state a simplified version of our main theorem . the complete version is stated in the next section . [ thm : simple ] consider a hamiltonian system @xmath40 satisfying inequalities ( [ eq : ineq ] ) in definition [ def : convex ] and @xmath41 . for every orbit @xmath42 with initial condition @xmath43 and @xmath44 , we have the following estimates provided @xmath10 is small enough . @xmath45 for@xmath46 where @xmath47 satisfies * @xmath48 , * @xmath49 . + the norm @xmath50 is taken over @xmath51 . this theorem gives the estimate of the stability constant @xmath19 in . for a given system , we need to optimize @xmath52 under the constraints in the theorem . we see that the decomposition of @xmath5 can be qualitatively written as @xmath53 where the constants @xmath54 and @xmath55 . though not solved explicitly , we expect our estimate here improves the previous results @xcite since the continuous averaging method gives us an improved normal form theorem ( see theorem [ thm : main ] ) . a possible application of the result is to the 3-body problem in order to get long time stabilities . this direction is already pioneered in @xcite . but the mass ratio of jupiter to the sun obtained in @xcite is too small to be satisfactory . on the other hand , in @xcite , the authors construct diffusing orbits for restricted planar 3-body problem . the diffusion time there is polynomial w.r.t . @xmath56 . the paper is organized as follows . first we give a complete statement of the main theorem and compare it with previous results in section [ section : complete ] . then we state a normal form theorem [ thm : main ] about averaging in a vicinity of periodic orbits in section [ section : normal ] . this is the main result that we obtain using continuous averaging , which improves the corresponding one in @xcite . then we give a brief introduction to the continuous averaging method in section [ section : ca ] . after that we give a proof of theorem [ thm : main ] in section [ section : main ] . this section is a higher dimensional generalization of the case studied in section [ section : ca ] . we try to draw analogy between the two sections . with the normal form theorem , we first show local stability result of nekhoroshev theorem in section [ section : local ] , and then global stability in section [ section : global ] . here local stability means the stability result in a neighborhood of a periodic orbit and global stability means stability for all initial conditions . finally , we have two appendices [ section : appendix ] and [ section : majorant ] . the first one contains some technical estimates for the continuous averaging . the second one is some basics of majorant estimates . we give a complete statement of the main theorem as follows . [ thm : global ] under the same assumption as theorem [ thm : simple ] , we have @xmath45 @xmath57 where @xmath47 , provided the following restrictions are satisfied . * @xmath48 , * @xmath58 + @xmath59 * @xmath49 , + @xmath60 the norm @xmath50 is taken over @xmath51 . the constant @xmath61 plays the same role as the constant @xmath62 in @xcite . it is dual to @xmath10 since only the product @xmath63 enters the original hamiltonian . we need the smallness of @xmath61 to make the first bullet point in theorem [ thm : global ] satisfied . the same restriction is expressed in @xcite by introducing a constant @xmath64 . the second bullet point can be satisfied easily by taking @xmath10 small enough . to improve the stability time , we want @xmath52 to be as large as possible , but the third bullet point gives a restriction of @xmath52 so that we need to optimize among @xmath65 . this restriction appears due to the finiteness of the width of analyticity of the action variables @xmath66 and degenerate variables @xmath67 . it shows up in a different form in @xcite as item @xmath68 of theorem 2.1 , where the choice of @xmath69 there can be as small as @xmath70 . we will give more discussions in remark [ rk : rho ] and [ rk : global ] . we will see from the following theorem [ thm : main ] that our normal form theorem obtained from continuous averaging improves that obtained from the iteration method . therefore we see we also get improved @xmath19 here even though @xmath52 is not expressed explicitly . our main work in this paper is to obtain a normal form theorem using continuous averaging . following lochak , we do the averaging in a neighborhood of a periodic orbit . we define @xmath71 , and @xmath72 . this is the frequency vector of a periodic orbit of the unperturbed hamiltonian @xmath8 . the period @xmath73 of this vector is , @xmath74 . integer vectors @xmath75 with @xmath76 give us @xmath77 after a proper translation in the space of action variables , wlog , we assume @xmath78 we can split the hamiltonian ( [ eq : main ] ) into four parts @xmath79 where each of the terms is given in the next definition . we use the taylor expansion of @xmath8 to split it as @xmath80 where @xmath81 contains the higher order terms . for @xmath82 part , we use the fourier expansion @xmath83 to write @xmath84 where @xmath85[def : resonant ] the hamiltonian equations can be written as @xmath86 suppose @xmath87 for some @xmath88 . then there exists @xmath89 , such that for any @xmath90 , there exist @xmath91 such that @xmath92 and a symplectic change of variables , @xmath93 , for @xmath94 , @xmath95 , which sends the hamiltonian @xmath96 to the following normal form : @xmath97 with the nonresonant part @xmath98see definition [ def : resonant]@xmath99@xmath100 the resonant part @xmath101 and the change of variables @xmath102 the @xmath103 , @xmath104 , @xmath105 and @xmath65 satisfy the following restrictions , * @xmath106 * @xmath107 , * @xmath108 . moreover , @xmath52 can be arbitrarily close to @xmath5 if @xmath10 is sufficiently small ( see remark [ rk : rho ] ) . [ thm : main ] the exponential smallness obtained here improves that of @xcite . continuous averaging enables us to get rid of some extraneous numerical factors that worsen the estimates . moreover , our method has an advantage , that is we do not need to do a preliminary transform which is necessary in @xcite . the proof of this result is contained in section [ section : main ] . in this section , we give an introduction to the continuous averaging method . please see the chapter 5 of @xcite for more details . we try to explain the key points of the method that will be used in our later proof . we write the hamiltonian ( [ eq : main ] ) as @xmath109 . suppose we have a symplectic change of variables @xmath110 depending on parameter @xmath111 , where @xmath112 denotes the new variables . then we have @xmath113 @xmath114 if we choose the flow of @xmath115 to be the hamiltonian flow generated by a hamiltonian isotopy @xmath116 , i.e. + @xmath117,\label{eq : jdf}\ ] ] with initial value @xmath118 , then the change of variables is symplectic and we get @xmath119 where the subscript @xmath111 means partial derivative . the last equality follows from the fact that the poisson bracket is invariant under symplectic transformations . in the following , we only work with the variables @xmath120 . + to simplify our discussion , we consider a special case of ( [ eq : main ] ) with @xmath121 . a further simplification is to consider only time - periodic nonautonomous systems . this is equivalent to requiring that @xmath122 in equation ( [ eq : main ] ) and @xmath123 independent of @xmath66 . from equation ( [ eq : cont ] ) , we have : @xmath124 where @xmath125 stands for the @xmath67 part of the poisson bracket . + our goal is to show that _ if we choose a suitable hamiltonian isotopy @xmath21 and extend @xmath126 as large as possible , the dependence of @xmath22 on @xmath4 can be killed to be exponentially small , i.e. @xmath127 for some constant @xmath128 . _ suppose @xmath129 has fourier expansion @xmath130 where @xmath131 means the zeroth fourier coefficient of @xmath82 . + we choose the hamiltonian isotopy @xmath21 as the hilbert transform " : @xmath132 now equation ( [ eq : nonauto ] ) has the form in terms of fourier coefficients : @xmath133 we show this @xmath21 is the good choice that makes the dependence on @xmath134 decrease exponentially . following @xcite , we explain here the heuristic ideas that justify this choice of @xmath21 . if we set @xmath135 in ( [ eq : fourier ] ) , we get @xmath136 whose solution tends to zero as @xmath137 . if we neglect the third term in the rhs of ( [ eq : fourier ] ) , we have @xmath138 it has an exact solution of the form @xmath139 where @xmath64 means the hamiltonian flow generated by the hamiltonian @xmath140 . notice the imaginary unit @xmath141 here . it tells us that the flow is considered with purely imaginary time . as @xmath111 increases , the complex width of analyticity is lost gradually . so formula ( [ eq : exact ] ) has sense only if we take @xmath142 , where @xmath5 is the width of analyticity in @xmath4 . this is an obstacle for the extendability of @xmath111 . we see from the heuristic argument that this choice of @xmath21 gives us the exponential decay as well as a good guess for the stopping time . the lie method is used in the works @xcite . before working out the detailed proof of the above heuristic argument , we explain the hilbert transform " first . in fact this choice of @xmath21 is strongly related to the classical averaging theory . let us recall what we usually do in the lie method . define the linear operator of taking lie derivative along the hamiltonian flow generated by the hamiltonian function @xmath143 : @xmath144 . the time-@xmath145 map of ( [ eq : df ] ) and ( [ eq : cont ] ) is : @xmath146 in each step of iteration , we need to solve the cohomological equation @xmath147 in fact , we are only able to solve @xmath148 by comparing the fourier coefficients , we obtain the following @xmath149 now we can explain why we choose @xmath21 as the hilbert transform of @xmath22 in ( [ eq : hilbert ] ) . we select @xmath21 to inherit the most important information in @xmath143 , namely , the imaginary unit @xmath141 and @xmath150 . readers can check that we still get the heuristic argument above if we choose the @xmath143 whose fourier coefficients are ( [ eq : cohom2 ] ) to do the averaging in ( [ eq : cont ] ) . now we take into account the third term in the rhs of equation ( [ eq : fourier ] ) . we first remove the @xmath151 term in equation ( [ eq : fourier ] ) by setting @xmath152 to obtain @xmath153 if we define an operator @xmath154 , where @xmath155 is the flow generated by the hamiltonian @xmath156 , the exact solution of the truncated equation @xmath157 would be @xmath158 . + then using the variation of parameter method in ode , we can write the exact solution to equation ( [ eq : v ] ) in the following form . @xmath159 we will analyze this equation to study its solution . to do so , we need a good control of the non - homogeneous term , i.e. the second term in the rhs . to control the nonhomogeneous term , we use the majorant estimate . the majorant relation @xmath160 is defined as follows . for any two functions @xmath161 , @xmath162 , @xmath163 , analytic at the point @xmath164 , @xmath165 @xmath166 we say that @xmath64 is a majorant for @xmath167 @xmath168 if for any multi - index @xmath169 , we have @xmath170.[def : maj1 ] the proof is first to guess a majorant assumption , then show the function in the assumption satisfies an equation that majorates the integral equation ( [ eq : int ] ) . this checks the assumption and closes up the argument . + now we make a majorant assumption @xmath171 where @xmath172 is the maximal extension time determined by the homogeneous part of equation ( [ eq : fourier ] ) in the heuristic argument . the @xmath173 characterizes the way how the fourier coefficients decay in the case of analytic perturbation and @xmath174 . we choose @xmath175 to make it easier to calculate the derivatives since @xmath176 . + then the integrand of equation ( [ eq : int ] ) can be majorated by @xmath177 , where @xmath178 this @xmath179 depends on the smoothness and magnitude of @xmath82 and the number of combinations @xmath180 . the number of combinations of integers in one dimensional case is easy to estimate , but in higher dimensional case it becomes very difficult , which is the main difficulty that we need to overcome in this paper . if we can solve the equation @xmath181 , then equation ( [ eq : int ] ) can be viewed as@xmath182 this checks the majorant assumption ( [ eq : assumption ] ) . in order to solve the equation @xmath183 , we apply the operator @xmath184 to the equation . we get the following burgers equation by setting @xmath185 . + @xmath186 here @xmath187 majorates @xmath188 in the sense @xmath189 . the initial condition @xmath190 is due to lemma [ lm : digression ] in appendix [ section : majorant ] , where @xmath3 is the width of analyticity in the slow variables @xmath191 . the burgers equation can be solved explicitly using the characteristics method in pde . the solution is @xmath192 in order to ensure @xmath193 , we obtain the maximal flow time given by the slow variables is @xmath194 . in fact @xmath195 , so combined with @xmath196 we get the maximal flow time is @xmath197 . we also notice @xmath198 is always bounded provided @xmath199 is sufficiently small , so is @xmath200 . recall that we defined @xmath201 therefore each fourier coefficient @xmath202 after the continuous averaging would be less than @xmath203 for some constant @xmath128 . adding up all these fourier terms , we recover the hamiltonian after the averaging , which is of order @xmath127 . this is the result proved in @xcite . we will work out all the details in section [ subsection : closeup ] . now we prove theorem [ thm : main ] using the continuous averaging method . let us go back to the setup in section [ section : normal ] . since we are looking at a motion that is very close to periodic orbit in the region of the phase space , the continuous averaging explained in the previous section could be applied . the periodic orbit corresponds to the fast angle @xmath4 in equation ( [ eq : egham ] ) . the nonresonant part @xmath204 corresponds to the @xmath4 dependence term @xmath205 in equation ( [ eq : egham ] ) . the @xmath206 will produce the exponential decay in the same way as the term @xmath66 in equation ( [ eq : egham ] ) did in equation ( [ eq : exact ] ) . and the term @xmath81 will generate the imaginary flow in the same way as the term @xmath140 in equation ( [ eq : egham ] ) did in equation ( [ eq : exact ] ) . finally , the term @xmath207 leads to additional difficulties . we devote the remaining part of this section to the proof of theorem [ thm : main ] . the proof is organized as follows . * set up the continuous averaging in terms of hamiltonian and get some heuristic understanding of the averaging process in section [ subsection : ham ] . * apply it the hamiltonian vector field in section [ subsection : vec ] . * following procedures in section [ section : ca ] , we define the operator @xmath208 to write the differential equations as integral equations , then we write down the majorant equation and prove the majorant relations . * derive necessary estimates in the theorem from the majorant estimates in section [ subsection : closeup ] . . the diamond encloses integer vectors @xmath209 . @xmath210 is the frequency vector @xmath211 . the red line is a hyperplane perpendicular to @xmath210 . the region outside the diamond is @xmath212 . the red line splits the region inside the diamond into @xmath213.,scaledwidth=60.0% ] [ figure : def ] in this section , we write down the continuous averaging and get a heuristic understanding . we start with a definition . as we have seen in section [ section : ca ] , in the process of continuous averaging , we have different aspects like exponential decay , imaginary flow and nonhomogeneous terms . we define a partition of the width of analyticity @xmath5,@xmath214 and put @xmath215 for @xmath216 we also define the following sets to form a partition of the grid @xmath26 . @xmath217 finally , we define two functions of @xmath111 associated to the above sets . @xmath218[def : main def ] * we split the analyticity width @xmath5 of the fast angle @xmath4 into @xmath47 . this splitting is quite flexible . we will optimize it to make @xmath52 as large as possible in section [ section : local ] and [ section : global ] . here @xmath52 would be used to control the imaginary flow , @xmath219 is used to do averaging , and @xmath220 is the remaining width of analyticity in angular variables after averaging . these distinctions will be made clear in the course of the proof . * we choose the cut - off @xmath221 to make sure that if @xmath222 , then the corresponding fourier coefficient is smaller than @xmath223 , which we think to be sufficiently small . a fourier coefficient with @xmath224 will become smaller as @xmath111 increases . once it is smaller than @xmath223 , the vector @xmath75 enters @xmath212 . so @xmath225 keeps shrinking as @xmath111 increases . we stop running the continuous averaging once @xmath226 . + [ remark : main rk ] we also define @xmath227 as the stopping time . the stopping time @xmath228 satisfies @xmath229[lm : dt * ] from the definition of @xmath225 , we know that at time @xmath228 , we should have @xmath230 we know @xmath231 from equation ( [ eq : tbar ] ) . this implies @xmath232 . now let us build our continuous averaging . this part is analogous to section [ subsection : derivation ] . from equation ( [ eq : cont ] ) , we have @xmath233 if we define @xmath234 in the continuous averaging equation @xmath235 , where @xmath236 is defined in definition @xmath237 , then depending on the properties of the fourier mode @xmath75 , we have the following three groups of pdes . @xmath238 [ eq : 3 cont ] we have @xmath239 in the latter two cases.[lm : abc ] from the definition of @xmath21 we know the fourier harmonics of @xmath21 come only from @xmath225 . as a result for any @xmath240 , we must write @xmath241 for @xmath242 and some @xmath243 . the equation ( [ eq : 3 cont]b ) is straightforward . if @xmath244 , then @xmath245 . we know @xmath246 , then @xmath247 . so in equation ( [ eq : 3 cont]a ) , the @xmath248 terms do not appear . if @xmath249 , no fourier harmonics from @xmath250 in equation ( [ eq : main cont ] ) enter equation ( [ eq : 3 cont]c ) . following directly from definition [ def : convex ] , we have the lemma . if we define @xmath104 as the confinement radius of @xmath66 , i.e. @xmath251 , then we have the following estimates @xmath252[lm : g ] we first notice @xmath253 . for @xmath254 , we use the formula @xmath255 and definition [ def : convex ] to get the estimate in the lemma . for @xmath256 , we use @xmath257 the following lemma helps us to understand the heuristic ideas of the process of continuous averaging and definition [ def : main def ] . if we omit the @xmath258 terms in the rhs of equations @xmath259 , then equation ( [ eq : 3 cont ] ) can be solved explicitly and the solution satisfies @xmath260 @xmath261 moreover , at the stopping time @xmath228 we have @xmath262 where the domain of variables is @xmath34.[lm : homo ] it follows from definition [ def : norm ] that @xmath263 for @xmath264 if we truncate equations ( [ eq : 3 cont ] ) , then the first and the third become @xmath265 for @xmath266 . so we have the corresponding estimate of @xmath267 stated in the lemma . however , equation ( [ eq : b ] ) becomes @xmath268 this equation admits an explicit solution @xmath269 so we have the estimate @xmath270 using lemma [ lm : g ] , we get @xmath271 in the splitting @xmath47 , we use @xmath52 to bound the term @xmath272 . namely , we need @xmath273 it is enough to require that @xmath274 this also gives an upper bound for @xmath111 . we equate this upper bound with the one given in lemma [ lm : dt * ] to obtain the value of @xmath221 in definition [ def : main def ] . now we have @xmath275 the definition of @xmath225 implies that once this @xmath202 term is already @xmath276 , the @xmath75 will enter @xmath212 and not belong to @xmath225 any more . in order for the majorant estimates to be applicable to understand equations ( [ eq : 3 cont ] ) , we need to write the continuous averaging equations in terms of hamiltonian vector field . we introduce the following vector fields @xmath277 corresponding to different parts @xmath278 of the hamiltonian @xmath96 and @xmath167 corresponding to @xmath21 . @xmath279 we also use @xmath280 to denote the @xmath75-th fourier coefficients of @xmath281 and @xmath282 . moreover , corresponding to @xmath21 in equation @xmath98[eq : f]@xmath99 , we define @xmath283 with this definition , we can rewrite the continuous averaging equation ( [ eq : main cont ] ) as follows by replacing the poisson bracket by lie bracket and the upper case letters @xmath284 by the lower case letters @xmath285 respectively . @xmath286\ ] ] [ lm : abcv ] if we set @xmath287 @xmath98recall @xmath288 was defined in definition @xmath289 , then equations @xmath259 can be rewritten in the following form in terms of hamiltonian vector field . [ eq : abcv ] @xmath290e^{-\left(\langle\om^*,l_{\pm}\rangle s_{l_{\pm}}+\langle\om^*,l\rangle s_l\right)}.\\ & \textrm{for}\ k\in d_-(\dt)\cup d_+(\dt),\\ \nonumber & v^k_\dt = ie^{-i\langle k,\tt\rangle}\big(\sm_k\left[v^ke^{i\langle k,\tt\rangle},v_0\right]\\ \nonumber & + \sum_{l_{\pm}+l = k}\eps \sm_{l_{\pm}}\left[v^{l_{\pm}}e^{i\langle l_{\pm},\tt\rangle},v^le^{i\langle l,\tt\rangle}\right]e^{-(\langle\om^*,l_{\pm}\rangle s_{l_{\pm}}+\langle\om^*,l\rangle s_{l}-\langle\om^*,k\rangle s_{k})}\big).\\ & \textrm{for}\ k\in d_>(\dt),\\ \nonumber & v^k_\dt = ie^{-i\langle k,\tt\rangle}\sum_{l_{\pm}+l = k}\eps \sm_{l_{\pm}}\left[v^{l_{\pm}}e^{i\langle l_{\pm},\tt\rangle},v^le^{i\langle l,\tt\rangle}\right]e^{-(\langle\om^*,l_{\pm}\rangle s_{l_{\pm}}+\langle\om^*,l\rangle s_{l}-\langle\om^*,k\rangle s_k)}.\\ \nonumber\end{aligned}\ ] ] in equations ( [ eq : 3 cont ] ) , we replace the poisson bracket by lie bracket and the upper case letters @xmath284 by the lower case letters @xmath285 respectively . then we remove the @xmath291 in the second case as we did in section [ subsection : exact solution ] . we set @xmath287 . then direct computation proves the lemma . what we do next is to write the differential equations for @xmath292 s as integral equations . as we did in section [ subsection : exact solution ] , we first need to define an operator @xmath64 which solves the homogeneous part of equations ( [ eq : abcv ] ) , esp . ( [ eq : abcv]b ) . let @xmath155 be the hamiltonian flow of the hamiltonian vector field @xmath293 generated by the hamiltonian @xmath81 . we put @xmath294 for an arbitrary analytic function @xmath295 defined on @xmath296 and then define : @xmath297[def : g ] it is shown in section 5 and 7 of @xcite that @xmath208 has the following two properties . * @xmath298 solves @xmath299 $ ] . + * @xmath300\right)\\ = e^{-i\langle k_1+k_2,\tt\rangle}[e^{i\langle k_1,\tt\rangle}\mathbf{g}^t_{k_1}\hat{f}_1 , e^{i\langle k_2,\tt\rangle}\mathbf{g}^t_{k_2}\hat{f}_2].\ ] ] with this operator @xmath208 , we can write differential equations ( [ eq : abcv ] ) as integral equations . if we denote the @xmath258 terms in equations @xmath301 by @xmath302 respectively , then we have the following three integral equations equivalent to equations @xmath303 . @xmath304 [ eq : intv ] the equations ( [ eq : intv]a ) and ( [ eq : intv]c ) are straightforward . the equation ( [ eq : intv]b ) is an application of the first property of the operator @xmath208 above and the variation of parameter method in ode . we need the following majorant commutator to perform estimates . for any two functions @xmath305 , and any two vectors @xmath306 , we define the majorant commutator : @xmath307^{l , k}=(|l|+|k|)\hat{f}\hat{g}+(n+2m)\dfrac{\partial}{\partial y}(\hat{f}\hat{g}),\ ] ] where @xmath308.[def : commutator ] for this commutator , we have the following lemmas . suppose that @xmath309 , @xmath310 , and @xmath311 , @xmath312 . here the majorant relation @xmath313 means that @xmath314 majorates each component of the vector @xmath315 . then for any @xmath316 , @xmath317\ll[[\hat{f}_1,\hat{f}_2]]^{k_1,k_2}.\]][lm : majcomm1 ] suppose that @xmath318 . then @xmath319\right)\ll [ [ \hat{f}_1,\hat{f}_2]]^{k_1,k_2}.\]][lm : majcomm2 ] we first have majorant control on the initial value . for @xmath320 , and @xmath321 , @xmath322 , we have @xmath323[lm : initial ] we first consider @xmath324 . we know @xmath325 for @xmath326 from the definition of @xmath61 in definition [ def : norm ] . then we use lemma [ lm : digression ] ( 4 ) in appendix [ section : majorant ] to obtain the majorant control of @xmath327 . now we consider the effect of @xmath208 . the operator @xmath208 is defined by the hamiltonian flow generated by the hamiltonian @xmath328 in definition [ def : g ] . the variables @xmath329 are constants of motion of this hamiltonian flow . so @xmath208 only shrinks the width of analyticity in @xmath4 but has no influence on that of @xmath329 . from the definition of @xmath208 , we see @xmath330 we also have @xmath331 according to inequality ( [ eq : rho_1 ] ) . this tells us @xmath332 now use lemma [ lm : digression ] ( 4 ) in appendix [ section : majorant ] again to obtain the lemma . + the following construction is given in @xcite . consider a continuous function @xmath333 . we define the functions @xmath334 and @xmath335 as follows as solutions of pdes . @xmath336[def : w ] the solutions @xmath334 and @xmath335 are given explicitly by @xmath337 the solutions are defined up to time @xmath228 and for @xmath338 satisfying the restrictions . @xmath339[lm : restriction ] the fact that @xmath334 and @xmath335 are exact solutions can be checked directly . to obtain the restriction for @xmath228 , we need to ensure @xmath340 so that the square root makes sense . we want that when @xmath341 , we still have @xmath342 . we know @xmath343 [ rk : pde ] let us try to understand the pdes @xmath344 heuristically . consider @xmath345 the way to solve it is the characteristic method . the characteristics is given by @xmath346 . then we are able to write the pde in the form : @xmath347 . then @xmath348 . so we see that , @xmath334 determines how fast we approach the intersection of characteristics , while @xmath200 determines how @xmath198 grows . the main result of this section is summarized in the following proposition , which implies the solutions of equations ( [ eq : main maj ] ) majorate that of equation ( [ eq : intv ] ) . for any @xmath350 such that @xmath351 , we have the following majorant control of the solution @xmath352 of equation @xmath353 + @xmath354 and@xmath355 under the restriction [ eq : restriction ] coming from lemma @xmath356 . ( the expression of @xmath333 and @xmath357 will be given explicitly in lemma [ lm : estimate ] . ) moreover @xmath358 is given by @xmath359 [ lm : prop ] we first cite proposition @xmath360 in @xcite . consider the functions @xmath361 defined in definition [ def : w ] , then the following statements are true : 1 . @xmath362 2 . @xmath363 3 . @xmath364 4 . @xmath365 5 . @xmath366 [ lm : tr trick ] let us first divide equations ( [ eq : intv]a ) , ( [ eq : intv]b ) , ( [ eq : intv]c ) by the numerators of the initial condition in lemma [ lm : initial ] , i.e. @xmath367 , @xmath368 and @xmath367 respectively . then we use the expression @xmath369 to refer to any one of @xmath370 , @xmath371 or @xmath372 , ( see the integrands of equations ( [ eq : intv ] ) for the definitions of @xmath373 ) . + to carry out the proof , we substitute the majorant relation in proposition [ lm : prop ] into equations ( [ eq : intv ] ) to check that equations ( [ eq : intv ] ) are majorated by equations ( [ eq : main maj ] ) . this is the plan proposed in @xcite . we use the majorant commutator to majorate each of the lie bracket of @xmath374 according to lemma [ lm : majcomm1 ] , [ lm : majcomm2 ] , + @xmath375^{l_{\pm},l}. \nonumber\ ] ] here we also use that @xmath376 for @xmath377 and @xmath378 . for simplicity , we denote the exponential weight by @xmath379 applying the definition of the majorant commutator ( definition [ def : commutator ] ) , we get + @xmath380 here we use lemma [ lm : tr trick](3 ) . this gives @xmath381 . we introduce the notations + @xmath382 to obtain@xmath383 + the second term in the rhs is the most complicated one . we only consider this term . the other two terms are done similarly . @xmath384 here @xmath385 , because @xmath386 . we used lemma [ lm : tr trick](5 ) to decrease the exponent of @xmath387 . we also imposed a mild restriction:@xmath388 if @xmath222 , we get the majorant equation for the @xmath335 part in equation ( [ eq : main maj ] ) . + if @xmath209 , using lemma [ lm : tr trick](1 ) and @xmath389 , we replace the last @xmath390 " in ( [ eq : kn ] ) by @xmath391 this is the majorant equation for @xmath334 in equation ( [ eq : main maj ] ) . for @xmath392 and @xmath393 , we get the same majorant estimate with @xmath394 replaced by @xmath392 and @xmath393 . now the problem is to find @xmath333 to give bound for @xmath395 . we need to do some careful analysis for this and the result is summarized in the following lemma . + we have the following upper bound for @xmath396 , where @xmath397 are defined in @xmath398 , @xmath399 if we define @xmath400 , then @xmath401 is equation @xmath402 in proposition @xmath403 . [ lm : estimate ] the proof of this lemma is given in appendix [ section : appendix ] . this lemma gives the restriction @xmath402 in proposition [ lm : prop ] . what we have shown is that each integrand of equations ( [ eq : intv ] ) has majorant estimate @xmath404 where @xmath335 satisfies equations ( [ eq : main maj ] ) . combined with the majorant control on initial condition in lemma [ lm : initial ] , this implies the lhs of equations ( [ eq : intv ] ) is majorated by @xmath335 . now the proof of the proposition is complete . + the continuous averaging gives us the following information about the hamiltonian vector fields . at time @xmath341 , we have @xmath405 @xmath406[lm : after ] recall in lemma [ lm : abcv ] , we set @xmath287 . using the definition of @xmath288 in definition [ def : main def ] , we get @xmath407 for @xmath408 . then proposition [ lm : prop ] applies to such @xmath75 s . for @xmath409 , we must have @xmath410 according to the definition @xmath225 . then apply proposition [ lm : prop ] to this case . + now we use the information that we have obtained to prove theorem [ thm : main ] . let us define the change of variables @xmath411 obtained by the continuous averaging at the stopping time @xmath412 . then , the hamiltonian ( [ eq : hamiltonian ] ) in these new variables is of the form @xmath413 where @xmath414 is the resonant term and @xmath415 is the nonresonant term as defined in definition [ def : resonant ] . the following lemma gives estimates for the functions @xmath414 and @xmath415 . suppose @xmath416 , @xmath417 , @xmath418 and @xmath419 . denote after the averaging , @xmath420 and @xmath421 , then @xmath422 where @xmath423 after the averaging.[lm : phi ] notice the @xmath66 component of @xmath280 is @xmath424 . so lemma [ lm : after ] implies @xmath425 @xmath426 where @xmath427 s are fourier coefficients of @xmath414 and @xmath415 . first we get upper bound for @xmath334 . from lemma [ lm : restriction ] , for @xmath428 we have @xmath429 provided @xmath418 . ( we will see in section [ section : global ] that the confinement radius @xmath430 as @xmath431 , so this condition is easy to satisfy . ) the remaining width of analyticity for @xmath432 becomes @xmath433 . we can also replace the condition @xmath434 by @xmath435 the factor @xmath436 is absorbed in the @xmath437 norm . for @xmath409 , we are left with @xmath438 . for @xmath439 , we have @xmath440 according to ( [ eq : main maj ] ) . we only need to ensure @xmath441 so that @xmath442 . this is @xmath443 because @xmath444 and @xmath445 . ( we will also see in section [ section : global ] that @xmath446 as @xmath431 , so this condition is also easy to satisfy . ) + @xmath447@xmath448 recalling the definition of @xmath449 in definition [ def : main def ] , we find @xmath450 similarly , we have the estimates for @xmath451 . under the same hypothesis as lemma [ lm : phi ] , after the averaging the total deviation of the variables is @xmath452[lm : di ] here the norm @xmath50 is taken in the real domain . for simplicity , we consider only the @xmath66 component . the other components are similar . from equation ( [ eq : jdf ] ) , we have @xmath453 where the rhs is a real function . then @xmath454 we have @xmath455 since @xmath456 defined in lemma [ lm : abc ] . we also have @xmath457 hence we have the estimate ( recall @xmath458 . ) @xmath459 @xmath460 lemma [ lm : phi ] and [ lm : di ] complete the proof of the theorem . notice the conditions of the theorem coincide with that of lemma [ lm : phi ] and [ lm : di ] , where the last condition in the theorem is exactly @xmath461 . lemma [ lm : phi ] gives the estimate of @xmath462 and lemma [ lm : di ] gives the estimates for the deviation of the variables . in this section , we derive stability result using the normal form theorem [ thm : main ] . recall in section [ section : normal ] , we have set @xmath463 as the frequency vector of the periodic orbit that we are considering . we consider initial condition @xmath464 such that @xmath465 . ( local stability)[thm : local ] if the conditions of theorem [ thm : main ] and the following conditions are satisfied , * @xmath466 , * @xmath467 , then we have stability result : if the initial conditions @xmath465 , then one has @xmath468 , for all time @xmath469 . + the integrable part has the taylor expansion around the point @xmath470 , @xmath471 we obtain the following inequality using definition [ def : convex ] @xmath472 we use the energy conservation and lemma [ lm : phi ] for the first term of the rhs to get @xmath473 for the second term in the rhs of inequality ( [ eq : energy ] ) , we have @xmath474 for the first term in the rhs , we use the hamiltonian equation , lemma [ lm : phi ] and the fact that @xmath475 . @xmath476 for the second term , @xmath477@xmath478 where we use @xmath465 and @xmath479 following from lemma [ lm : di ] . we have a factor @xmath480 since we go from @xmath50 to @xmath481 . if we set @xmath482 , then we get an inequality of @xmath16 from ( [ eq : energy]):@xmath483 we choose @xmath484 to obtain @xmath485 we set @xmath486 then we have @xmath487 @xmath488 @xmath489 the proof is now complete . we introduce a restriction [ eq : r ] instead of introducing a constant @xmath64 as did in @xcite . the two restrictions of this theorem implies @xmath490 can be sufficiently small if @xmath10 is . then @xmath52 can be very close to @xmath5 . the restrictions of theorem [ thm : main ] are also satisfied for @xmath10 small enough . then we get improved stability time compared with @xcite . [ rk : rho ] + in this section , we consider stability result for arbitrary initial data and give a proof of theorem [ thm : global ] . we first prove the following lemma . [ lm : dirichlet ] let us fix @xmath491 and assume @xmath492 , then for any @xmath493 , there exists an integer @xmath494 , @xmath495 , and a point @xmath496 such that @xmath497 and @xmath498 is rational vector of period @xmath499 . first recall the dirichlet theorem for simultaneous approximation : + _ for any @xmath500 , there exists an integer @xmath494 , @xmath501 , such that @xmath502 . _ an improvement of the estimate can be obtained by rescaling @xmath503 to @xmath504 . then apply the dirichlet theorem to approximate the remaining @xmath505 components of @xmath503 with one of whose @xmath506 entries removed . we get the following : _ there exists a rational vector @xmath507 of period @xmath508 , @xmath509 , @xmath510 and @xmath511 @xmath98see ( the only ) proposition in @xcite@xmath99 . _ the frequency vector is @xmath512 . consider two points @xmath513 and @xmath66 such that @xmath498 is as stated in the lemma and approximates @xmath514 in the same way as @xmath507 approximates @xmath503 . @xmath515 hence from definition [ def : convex]@xmath516@xmath517 this implies @xmath518 in order to make sure the point @xmath513 can be found given @xmath66 , we need to show the frequency map @xmath514 can be inverted , which can be done within the ball @xmath519 centered at @xmath498 with radius @xmath520 using implicit function theorem ( see @xcite ) . so we assume @xmath521 . we define @xmath522 and @xmath523 as in lemma [ lm : dirichlet ] . if we set @xmath524 , then we have @xmath525 the stability time in theorem [ thm : local ] is @xmath526 now let us analyze the restrictions that we have . the restrictions are from theorem [ thm : main ] , theorem [ thm : local ] and lemma [ lm : dirichlet ] . the quantities @xmath527 satisfy the following : @xmath528 @xmath529 @xmath530 we substitute the bounds for @xmath527 into the restrictions that we have to obtain the restriction of theorem [ thm : global ] . the first restriction in theorem [ thm : global ] is from the first one of theorem [ thm : local ] . the second in theorem [ thm : global ] is a collection of the first two ones of theorem [ thm : main ] , the second of theorem [ thm : local ] and that of lemma [ lm : dirichlet ] . the last in theorem [ thm : global ] is from the third one of theorem [ thm : main ] . we break it into two inequalities by setting @xmath531 . the first restriction in theorem [ thm : global ] can be satisfied by making @xmath61 smaller while @xmath10 larger . this will lead to shorter stability time . the @xmath61 here plays the same role of the factor @xmath64 in @xcite . the second restriction can always be satisfied by making @xmath10 small . the third restriction can be satisfied by making @xmath61 or @xmath532 small . however , since @xmath533 grows very fast , for large @xmath480 , this restriction is easy to satisfy.[rk : global ] the proof is done in the following claim 1,2,3 , which estimates @xmath534 in lemma [ lm : estimate ] respectively . before the proof of the lemma , let us first analyze the geometry of numbers involved . let us look at the figure 2 . + [ figure : no ] * _ the diamond _ : the diamond in the figure encloses all the vectors @xmath75 with @xmath209 ( in 3-dim it is an octahedron . in general it is a ball of radius @xmath221 under the @xmath535 norm ) . the total number of integer vectors inclosed in the diamond is @xmath536 . indeed , in @xmath480-dim , the diamond consists of @xmath537 simplices . each of the simplices has volume @xmath538 . * _ the hyperplane _ : the small arrow indicates the rational frequency @xmath211 . @xmath539 is a hyperplane that is perpendicular to @xmath211 . @xmath540 and the @xmath541-dim volume of @xmath542 is less than @xmath543 , which is the @xmath541-dim volume of an @xmath541-dim diamond . any vector lies above @xmath539 has positive inner product with @xmath211 , while any vector below has negative inner products . moreover , if two vectors lie on the same hyperplane which is parallel to @xmath539 , they will have the same inner product with @xmath211 . let us denote @xmath544 . @xmath542 contains at most @xmath543 integer points . * _ the parallelogram _ : consider the vectors @xmath545 in the figure 2 . suppose we have the relation @xmath546 . then the three vectors together with the origin form a parallelogram . suppose @xmath547 , and @xmath548 , then @xmath549 . @xmath550 and @xmath551 can move on their corresponding hyperplane , but a parallelogram is always preserved . * _ the shape of the diamond under the averaging flow _ : in the definition of @xmath552 , we have the restriction @xmath553 for @xmath554 . when @xmath555 , this is our diamond . when @xmath111 increases , the diamond will collapse , i.e. the integer vectors becomes fewer on hp@xmath556 . @xmath557 the rate of decreasing depends on the inner product @xmath558 . the farther a hyperplane @xmath559 is away from @xmath539 ( the larger the @xmath560 ) , the faster it collapses ( with volume decreasing rate @xmath561 ) . @xmath539 does not change at all . when @xmath341 , the diamond would collapse to its intersection with @xmath539 . by then we would have successfully killed all the nonresonant terms up to the desired exponential smallness @xmath562 . we denote the collapsed diamond at time @xmath111 by @xmath563 . now we obtain estimates of @xmath564 for fixed @xmath75 . _ the sum @xmath565 defined in equation @xmath398 can be estimated as follows : _ @xmath566 in the proof , the vector @xmath75 is fixed . the @xmath392 is defined in equation ( [ eq : sum ] ) , @xmath567 which can be estimated as@xmath568 where we have dropped @xmath569 , since @xmath570 from the relation @xmath571 , and the inequality @xmath572 , we have the following two cases depending on the sign of @xmath573 . + if @xmath574 , then @xmath575 @xmath576 if @xmath577 , then @xmath578 @xmath579 moreover , when @xmath555 , the number of integer vectors contained in @xmath580 is no greater than @xmath543 . it is zero when @xmath581 . since @xmath75 is fixed , we can vary either @xmath550 or @xmath551 . the other one will be determined uniquely . according to the analysis above , we sum over @xmath550 if @xmath574 while over @xmath551 otherwise . we consider the @xmath550 case for instance . as @xmath111 increases , on each @xmath559 the number of integer vectors contained in @xmath582 is no greater than @xmath583 according to inequality ( [ eq : collapse ] ) . it is zero when @xmath581 or @xmath584 . now we have the estimate @xmath585 this estimate is poor when @xmath111 is close to zero . but in fact when @xmath555 , the upper bound is @xmath586 and the upper bound is monotonically decreasing w.r.t . so we can use the bound stated in claim @xmath145 . see figure 3 . .,scaledwidth=60.0% ] [ figure : cut ] _ the sum @xmath587 defined in equation @xmath398 can be estimated as follows : _ @xmath588 it is defined in equation ( [ eq : sum ] ) that @xmath589 we first show @xmath590 according to the definition of @xmath591 , the fourier term corresponding to @xmath592 is of the size @xmath593 but @xmath594 due to definition [ def : main def ] . the equality is achieved only when @xmath595 . so we know @xmath596 we drop the @xmath597 and @xmath598 to obtain @xmath599 . this is essentially the same as the case @xmath392 discussed above . so we have the bound stated in claim @xmath600 . + _ the sum @xmath602 defined in equation ( [ eq : sum ] ) can be estimated as follows : _ @xmath603 the term @xmath604 turns out to be the most troublesome term . again equation ( [ eq : sum ] ) tells us @xmath605 because of the relation @xmath606 and @xmath607 , we get @xmath608 and @xmath609 the reason is , we know that @xmath242 until time @xmath111 , but we do not know where @xmath75 is . this implies @xmath610 . the = " is achieved only if @xmath611 . + so we get @xmath612 now consider figure 2 . since @xmath613 , we get @xmath614 and @xmath75 must lie on the same hyperplane . so @xmath615 is bounded by the number of the possible @xmath616 s , which is @xmath617 this gives the claim @xmath601 . + we simply add up the upper bounds for @xmath618 to get an upper bound for @xmath619 . this proves lemma [ lm : estimate ] . in this appendix , we collect some basics about the majorant relation . the materials can be found in the chapter 5 of @xcite . the majorant relation @xmath160 is defined in definition [ def : maj1 ] . the relation @xmath390 " satisfies the following properties : 1 . if @xmath620 and @xmath621 , then @xmath622 and @xmath623 . 2 . if @xmath624 , then @xmath625 for any @xmath626 . if @xmath627 for any value of the parameter @xmath628 $ ] , then @xmath629 4 . let @xmath630 in the domain @xmath631 . then @xmath632 , where @xmath633 and @xmath634 [ lm : digression ] moreover , the majorant relation is also preserved by solving differential equations or integral equations . consider an ode system @xmath635 with some known functional @xmath636 and initial data @xmath637 . we call a system @xmath638 a majorant system associated with equation @xmath639 if + @xmath640 @xmath641 for any @xmath642 , and + @xmath643 @xmath644 for any @xmath645 , @xmath646 , and @xmath647 such that @xmath648 . + [ def : maj ] we have the theorem if @xmath649 is a solution of the majorant system @xmath650 associated with @xmath639 , then the system @xmath639 has a solution and @xmath651 for any @xmath652 $ ] , @xmath653 . the same is true if we rewrite systems @xmath654 in the integral form : @xmath655 [ thm : z ] with this theorem , we treat @xmath111 as a parameter instead of a variable . so we do not need to do the taylor expansion w.r.t . @xmath111 . the author would like to thank prof . vadim kaloshin to provide the idea of applying continuous averaging to the nekhoroshev theorem . he also would like to thank prof . treschev , neishtadt , dr . m. guardia and a. bounemoura for carefully reading the manuscript and giving many valuable suggestions . the last version of the paper is completed when the author is visiting ias and he would like to thank ias for its hospitality . dgo # 1 a. bounemoura , j .- marco , improved exponential stability for near - integrable quasi - convex hamiltonians , nonlinearity,*24 * , ( 2011 ) , 97 - 112 . a. crdoba , d. crdoba , m. fontelos , formation of singularities for a transport equation with nonlocal velocity . annals of mathematics , * 162 * , ( 2005 ) , 1377 - 1389 . j. fjoz , m. guardia , v. kaloshin , p. raldan , diffusion along mean motion resonance in the restricted planar three- body problem , http://arxiv.org/abs/1109.2892 . a. kiselev , f. nazarov , a. volberg , global well - posedness for the critical 2d dissipative quasi - geostrophic equation . inventiones math . * 167 * , ( 2007 ) , 445453 . p. lochak , canonical perturbation theory via simultaneous approximation , russian mathematical surveys , no . * 6 * , ( 1992 ) , 57 - 133 . p. lochak , simultaneous diophantine approximation in classical perturbation theory : why and what for ? progress in nonlinear science . vol.*1 * ras , inst . phys . , nizhnii ( novgorod , ( 2002 ) , 116 - 138 . p. lochak , ai . neishtadt , estimates of stability time for nearly integrable systems with a quasiconvex hamiltonian , chaos , no . * 4*,(1992 ) , 495 - 499 . p. lochak , a. i. neishtadt , l. niederman , stability of nearly integrable convex hamiltonian systems over exponentially long times . kuksin , s. ( ed . ) et al . , seminar on dynamical systems . basel : birkhuser . nonlinear diff . * 12 * , ( 1994 ) , 15 - 34 . l. niederman , stability over exponentially long times in the planetary problem , nonlinearity,*9 * , ( 1996 ) , 1703 - 1751 . n. nekhorochev , an exponential estimate of the time of stability of nearly integrable hamiltonian systems , russ . surv . * 32 * , 1 - 65 . j. pschel , nekhoroshev estimates for quasi - convex hamiltonian systems , mathematische zeitschrift , * 213 * , ( 1993 ) , 187 - 216 . a. pronin , d. treschev , continuous averaging in multi - frequency slow - fast systems . regular and chaotic dynamics , * 5:2 * , ( 2000 ) , 157 - 170 . d. treschev , oleg zubelevich . introduction to the perturbation theory of hamiltonian systems , springer , ( 2009 ) . d. v. treschev , the continuous averaging method in the problem of separation of fast and slow motions . regular and chaotic dynamics , * 2:3/4 * , ( 1997 ) , 9 - 20 . d. treschev , separatrix splitting for a pendulum with rapidly oscillating suspension point . russian j. math . * 5:1 * , ( 1997 ) , 63 - 98 .

in this paper we are concerned with the problem of data forwarding from a wireless body area network ( wban ) to a gateway when body shadowing affects the ability of wban nodes to communicate with the gateway . to solve this problem we present a new wban architecture that uses two communication technologies . one network is formed between on - body nodes , and is realized with capacitive body - coupled communication ( bcc ) , while an ieee 802.15.4 radio frequency ( rf ) network is used for forwarding data to the gateway . wban nodes that have blocked rf links due to body shadowing forward their data through the bcc link to a node that acts as a relay and has an active rf connection . for this architecture we design a network layer protocol that manages the two communication technologies and is responsible for relay selection and data forwarding . next , we develop analytical performance models of the medium access control ( mac ) protocols of the two independent communication links in order to be used for driving the decisions of the previous algorithms . finally , the analytical models are used for further optimizing energy and delay efficiency . we test our system under different configurations first by performing simulations and next by using real rf traces . wireless body area network ( wban ) , body sensor network ( bsn ) , ieee 802.15.4 , body shadowing , medium access control ( mac ) , capacitive body - coupled communication , cooperative communications , relay , performance analysis , delay , energy , optimization .

1506.08307

several applications where wireless body area networks ( wban ) are deployed around the human body , reliable and low delay communication is of paramount importance because of the critical nature of the collected data ( e.g. heart rate , blood pressure ) . energy consumption is also key for the prolonged operation of the devices attached to the human body . for their communication needs these wbans usually employ radio frequency ( rf ) technologies that operate in the industrial , scientific and medical ( ism ) radio band . one of the dominant solutions is the ieee 802.15.4 standard that is engineered specifically for low power devices @xcite . however , it is also possible to use the widely popular wireless lan ( wlan ) standard ieee 802.11 @xcite because of the readily available access point ( ap ) infrastructure . regardless of the specific wireless communication technology , when multiple sensors are deployed in the human body , a wban in a star topology is usually created so that all the on - body nodes communicate with a gateway for forwarding the collected data . [ fig : dual - rf - bcc - principle ] ( left ) depicts the organization of a wban in a star topology that is employed by practical systems @xcite . although rf is the only practical mechanism to forward data in this scenario , still several significant problems remain . rf signals suffer considerably from body shadowing in a highly variable way with respect to human body @xcite . this makes communication between on - body nodes , and also off - body , very unreliable . this inherent unreliability of rf communication around the human body is a critical problem for several real - life applications . we discuss two emerging application scenarios here to motivate our system design . one example is vital sign monitoring where multiple nodes need to be deployed in several different places of the human body @xcite . traffic is usually flowing in the uplink direction , i.e. from the wban nodes to the gateway . consider for example a human that has a node on the torso and one on the wrist for monitoring the heart rate . if this human is a patient lying in bed , then the rf link of the sensor on the torso might be completely blocked . however , it is possible that the node on the wrist can communicate perfectly with the gateway . or it can happen the other way around as fig . [ fig : dual - rf - bcc - principle](a ) indicates . since humans usually move frequently , any sensor can be potentially blocked and the data may not be communicated on - time to the gateway ( or it can be lost completely ) . besides health monitoring , the other important application is real - time media entertainment . with the rising popularity of wearable devices , it is even possible that high data rate video streams need to be forwarded from the human body ( uplink ) . one specific application example with downlink traffic , is audio streaming from a wifi access point ( ap ) to on - body earphones without wires @xcite . in these two examples rf connectivity between the ap and the earphones / glasses might be unreliable leading to problems in the audio / video transmission . therefore , the optimal node on the body must be found and act as a relay in order to transmit the data to and from the wifi ap . this node can be for example a smartphone . to address the problems identified in the aforementioned scenarios , in this paper we propose a novel wban architecture , a network layer ( nwk ) protocol that exploits the proposed architecture , and an optimization framework supported by an analytical performance model of the system . in our architecture that is depicted in fig . [ fig : dual - rf - bcc - principle](b ) , all the wban nodes on the human body are equipped with both rf and body - coupled communication ( bcc ) transceivers . when the rf link of a node is unreliable the node uses the bcc link for forwarding its data through the human body to a node with a better rf link . the rationale of this design choice is based on the extremely power - efficient bcc technology that can be used instead of rf communication . prototype bcc transceivers use simple baseband communication and have a very small form factor ( 1mm@xmath0 dye in @xcite ) . thus , the proposed wban architecture with two transceivers on board a wban node is an economically and technically viable option for real - life applications . * contributions . * the contributions of this paper are : * first , we propose a new cooperative wban architecture that employs two communication technologies namely bcc and rf . this architecture is orchestrated by a novel nwk relay selection protocol that identifies the optimal relay for forwarding the data off the body . the proposed scheme is unlike any existing relay selection protocols since two different communication technologies are used . * second , for each medium access control ( mac ) protocol of the rf and bcc subnetworks we develop accurate expressions of the delay and energy consumption as a function of specific protocol parameters . our novel contributions here include first the performance modeling of a nwk that concurrently uses two wireless mac / phy technologies . second , our model is a cross - technology model , i.e. it quantifies the impact of a specific parameter setting for one mac on the second and on the complete system . * third , we propose the use of the analytical performance models for further optimization of the duty cycling of the bcc transceivers and the retransmission strategy of the rf transceivers so that energy consumption and delay are minimized . * benefits . * the _ first _ advantage is that devices with a blocked rf link can use the services of the remaining nodes for forwarding their data to a gateway and improve thus the reliability of the data delivery . this idea is demonstrated with the help of fig . [ fig : dual - rf - bcc - principle](b ) while we have also outlined this generic architecture in @xcite . this is accomplished with c1 . _ second _ , rf transmissions are reduced to the absolute minimum since wban nodes do not communicate with each other through the rf link . the benefit is that the rf link is carrying a reduced load and the interference to surrounding rf devices is also minimized @xcite . this is accomplished with c1 . third _ advantage is that a node with an rf link that is not completely blocked , but operates inefficiently ( higher number of required retransmissions that increase delay and energy consumption ) , can communicate with other nodes through bcc and select another node for data forwarding . this is accomplished with c2 and c3 . our performance evaluation focuses on highlighting these benefits . performance is evaluated both through simulations and also through the use of real rf traces . results with ieee 802.15.4 rf traces interestingly support even more the need for our system in real scenarios due to the severity of human body shadowing . * paper organization . * the rest of this paper is organized as follows . related work is presented in section [ section : related - works ] . in section [ section : system - architecture ] we provide an overview of the system architecture and in section [ section : relay - selection ] we delve into the details of the relay selection and packet forwarding algorithm . section [ section : modeling - approach ] presents the motivation and an overview of our performance modeling approach . in section [ section : rf - radio - model ] we present the performance model of the rf network while in section [ section : bcc - model ] we present the analytical model of the bcc network . the mac protocol optimization that further exploits the previous analysis is described in section [ section : optimization ] . the performance of our system is evaluated with simulations and real measurements in section [ section : experiments ] . finally , in section [ section : conclusions ] we conclude this paper . in this paper we deal with the problem of optimizing data delivery from a wban to a gateway . the problem is challenging because communication between on - body and off - body nodes is very unreliable @xcite . the most promising way to attack the problem is through cooperation between wban nodes @xcite . with cooperation the nodes that aid in forwarding the data of other nodes are the _ relays_. there is a plethora of works that focus on selecting the optimal relay in the general case of wireless sensor networks ( wsn ) by considering different optimization objectives , while fewer works have focused in wbans . _ the majority of research works investigated cooperation in wsns / wbans with the objective to minimize power consumption . _ in @xcite the authors consider relay selection in wsns for minimizing the power consumption of individual nodes . they evaluate the impact of the sensor wake up schedules on the consumed power , and then they propose incentive mechanisms for participating in the cooperative network . this is an optimization approach that perceives the nodes as independent competing entities that may help each other for relaying data to a gateway . a number of authors investigated the impact of cooperation on the power consumption and lifetime of the complete wban introducing thus a more holistic approach @xcite . in these two works the cooperating nodes communicate with each other through the rf links opportunistically when they can not reach the gateway . however , a particular node may still not be able to reach another on - body node . to combat these intermittent problems , another parameter that was also investigated together with relay selection was power control . power control effectively means topology control in the wban @xcite , i.e. we can allocate more power to nodes that need it so that they can reach the rest of the network or the gateway . even though this approach ensures optimal power allocation for given channel conditions , power may be wasted in nodes with poor rf links . in the category of research efforts that target power minimization we can not identify solutions that eliminate the problem of a completely blocked rf link , but rather state - of - art solutions that allocate optimally the system resources to combat the rf inefficiencies . this is still unacceptable in scenarios where data from a specific node must be transmitted in real - time to the gateway . another significant number of research efforts focused on the problem of reliable communication and was motivated from experimental results . in @xcite the authors first focus on obtaining an experimental characterization of the channel . from the obtained results , the use of statically assigned relay nodes in specific body locations was proposed . nevertheless , specific on - body nodes may still fail or be occluded due to body shadowing . another aspect that affects the wban performance is the mobility of the user . mobility models for wbans driven from experiments were studied in @xcite . in this work the authors study the problem of user mobility and posture changes and how they affect wban performance . the solution is a multi - hop protocol that is aided by a single relay . in contrast to all the previous works that look into each specific node as a potential relay , there is an option to use multiple wban nodes for improving the reliability and/or energy / power . this class of protocols , that also targets generalized wireless cooperative networks , proposes the creation of _ clusters _ across groups of nodes in order to improve the diversity gain in the rf link by simultaneously transmitting the same information over different wireless paths @xcite . nevertheless , there is the major problem of node synchronization in this case . in cooperative schemes that involve many nodes we can also add recent works that apply advanced techniques like network coding @xcite . despite the architectural and protocol differences , all the previously described schemes share one common characteristic which is the use of a single rf communication technology . to alleviate the problems of a single technology two different ones can be used . the number of works in this area is quite limited . a communication protocol that uses two technologies was presented in @xcite . in that work the authors proposed the use of two different rf bands namely the 433mhz and 2.4ghz . the 433mhz band is used for data aggregation whereas the 2.4ghz band is used for data forwarding to the gateway . since the range of the 433mhz band is limited to approximately 2 meters around the node it is possible to improve the reliability and energy consumption by reducing the number of nodes competing for the same channel . still , both links use rf for on - body communications and two independent rf bands are required . _ thus , all the techniques we discussed in the last three paragraphs can not address completely the rf reliability problem because of the fundamental problems of the wireless channel . _ finally , we should mention that none of the previous works employed an in - depth analysis of the impact of the mac protocol parameters on energy / delay / reliability of the complete cooperative network . we could only trace a limited number of works for non - cooperative systems that use analytical protocol performance models for further optimization . these approaches use the mac model of ieee 802.15.4 for power minimization at an individual node @xcite . when compared to all the related works , our approach is significantly different . it focuses on using simultaneously the most reliable and most efficient rf link ( in terms of delay / energy ) for data forwarding since these two aspects are intertwined . our solution consists of a new wban architecture and a novel cooperative protocol design that is supported by an analytical performance model . with our system even when at least one node has an active rf link the data from every wban node will reach the gateway . to the best of our knowledge no other wban system can guarantee that . we consider a wban that consists of @xmath1 nodes where each node is equipped with an rf transceiver and a bcc transceiver . these two hardware components use different mac / phy protocols and are both controlled from a specialized nwk protocol . the nodes are organized into a single - hop bcc network , and a subset @xmath2 of them form a single - hop rf network as shown in fig . [ fig : dual - rf - bcc - principle ] . all the nodes that are part of the rf network can connect to a gateway for forwarding the collected data . each wireless link has different delay / energy characteristics due to channel variations . while only @xmath2 nodes are part of the rf network , all the @xmath1 nodes participate in the bcc network . non - relay nodes transmit their packets through the bcc link to one of the relays , while the selected relay transmits wirelessly to the gateway the locally generated and forwarded data packets . nodes that do not have a relay responsibility are allowed to put into sleep mode the rf transceiver for reducing the energy consumption . however , since it is possible that in the future they assume the role of an rf relay , they wake up periodically , they transmit an rf packet , and after they receive a response they update the received signal strength indication ( rssi ) value of their rf link . the ieee 802.15.4 is used for rf communication . a non - beacon - enabled and un - slotted carrier sense multiple access with collision avoidance ( csma / ca ) algorithm is assumed for channel access . all the nodes sense the channel status during the clear channel assessment ( cca ) slots . the basic idea of the un - slotted csma / ca algorithm is that backoff and packet transmissions are not aligned to specific slot boundaries . the un - slotted csma / ca operates as follows . each time a device generates a packet for transmission it waits for a random number of slots , called the backoff counter , ranging from 0 to @xmath3 , where @xmath4 denotes the backoff exponent . in ieee 802.15.4 , the backoff counter is decremented to zero regardless of the cca result and by default @xmath4 is initialized to 3 . when the backoff counter reaches zero , the device performs cca only once in order to check whether the channel is busy or not . if the channel is idle during the cca period that has duration @xmath5 ( fig . [ fig : rf - mac - protocol ] ) , the device transmits its data packet . when the channel is busy during the period @xmath5 , the backoff exponent @xmath4 is increased by 1 and the random backoff procedure is repeated . the backslash - shaped blocks in fig . [ fig : rf - mac - protocol ] depict the increase in the backoff time after a failed cca . more details regarding its operation are provided later while the complete algorithm is available in @xcite . bcc without requiring skin contact can be realized with two electrode transmitter / receiver devices capacitively coupled to the human body @xcite . the transmitter generates a variable electric field while the receiver senses the variable potential of the body with respect to the environment . a signal attenuation of less than 70db has been measured between devices placed at various positions of the human body @xcite . the human body channel is especially affected by interference below 1mhz while for higher frequencies the interference level is below -75dbm . in this paper a bpsk modulation scheme is assumed where the digital pulses are directly transmitted to the human body through the capacitive plates @xcite . digital pulses of 1.2v for transmission are used , and an receiving band of 1 - 30mhz was chosen for improving the ber in the presence of interference . the ber was measured to be lower than @xmath6 for the aforementioned conditions . the energy efficiency of the transceiver was measured at 0.32nj / bit . since the design of the phy is not the focus of this paper we do not delve into further details . the interested reader can find more details about a typical prototype transceiver in @xcite . regarding the mac layer it is a protocol that employs the well - known concept of low power listening ( lpl ) @xcite . lpl exploits a specialized _ wake up receiver _ hardware that is always in active mode and consumes very little energy while the main receiver is deactivated . [ fig : bcc - mac - protocol2 ] presents the channel access scheme for a single packet . a preamble packet is always transmitted before the transmission of an actual data packet which means that nodes contend with preambles . the contention for the transmission of preambles follows the same backoff procedure with the rf mac . nodes that are asleep can wake up with the preamble and prepare for the actual data reception . the preambles contain the destination address so that they wake up the appropriate node . when a preamble is successfully received , the node that wakes up sends an acknowledgment packet that is named _ ready - to - receive _ ( rtr ) since it also serves as an indication that the receiver can accept the actual data packet . after this transmission the receiver waits for the data packet to arrive and transmits the final ack . .main functions used in the protocol stack of the complete system . notation @xmath7 means that this function is executed when a message of type @xmath8 is transmitted / received in layer / technology @xmath9 . [ cols="<,<",options="header " , ] the first and main task of our distributed wban system is to calculate the optimal subset of @xmath2 rf relays from all the available nodes , while the second task is to forward the data through the optimal relay if needed . to accomplish the first task , our system is designed as follows . each wban node @xmath10 calculates the average delay and energy of the direct rf transmission and then it informs the remaining nodes in the network . the random variables of the delay and energy are denoted as @xmath11 and @xmath12 respectively . this functionality will be analyzed in subsections [ subsection:1 ] and [ subsection:2 ] . how this information is used for selecting a node as a relay will be examined in subsection [ subsection:3 ] . the process of making the actual forwarding decision is described in [ subsection:4 ] . all this functionality is concisely captured in the pseudo - code of fig . [ fig : coop - protocol ] . in the pseudo - code of fig . [ fig : coop - protocol ] we introduce two specific _ control _ messages at the nwk , namely @xmath13 for broadcasting the parameter estimates , and second @xmath14 that is used for relay selection . a summary of the most important functions that are used by our protocol stack are provided in table [ tab1 ] . finally , our protocol defines three data structures at each node in the form of linked lists , and are denoted as @xmath15 while their use will be explained in the next subsections . the parameter estimation and information exchange functionality is captured in the function param_est_node ( ) in the pseudo - code of fig . [ fig : coop - protocol ] and is executed periodically every @xmath16 seconds by every node . when an rf node is not a relay , the rf transceiver is deactivated by entering an idle state @xcite and so it has to enter an active state to execute this function . the channel estimation itself is a local task at each rf node that is executed by default in ieee 802.15.4 for every received packet and so no extra overhead is introduced by our scheme ( i.e. when param_est_node ( ) is invoked it reads the latest rssi measurement from the related register of the micro - controller ) . next , this measurement is used for calculating the average delay and energy that are denoted as @xmath17 $ ] and @xmath18 $ ] respectively ( the expectation of these random variables ) . this is accomplished by using the rssi measurement in the analytical model that we develop in sections [ section : rf - radio - model ] and [ section : bcc - model ] . next , node @xmath10 broadcasts the rf link performance estimates expressed though the average delay @xmath17 $ ] and average energy cost @xmath19 ^ 2/ e_{rem}(i)$],^2/ e_{rem}(i)$ ] . this is a simple metric but it captures the fact that the cost is increased as the remaining energy is reduced after continuous utilization , while if the energy of a single transmission is low then this has an impact on lowering the cost . our framework can support different and potentially better cost functions . ] with the special message @xmath13 . this message is transmitted at the bcc link and is broadcast which means that all the bcc transceivers of all the nodes will receive it . this concludes the description of the core functionality of the parameter estimation and information exchange functionality of our system . next we discuss some important details . * robustness of status_bcc transmissions . * the bcc phy offers a proven reliable link since it transmits with bpsk modulation and in addition it uses channel coding for eliminating bit errors @xcite . furthermore , @xmath13 messages are sent with a constant period @xmath16 even when the channel does not change . in case of a packet loss , this periodic propagation of information allows all nodes to reach the same consistent state . * communication overhead . * the forwarding of this information through the bcc link consumes a fraction of the bcc data rate . here we provide a simple analysis to demonstrate that this communication overhead is negligible for the bcc link . every @xmath16 the @xmath1 nodes transmit a @xmath13 message of length @xmath20 bits . and in the ( unrealistic ) worst case every @xmath16 all the current @xmath2 relays will also send a message @xmath14 in order to change their role . then the bcc data rate overhead is @xmath21 bps . for a typical update frequency of @xmath16=1 sec , and a payload of @xmath20=20 bytes for these short control messages , @xmath1=@xmath22 , and @xmath2=@xmath23 , we have an overhead 2.4kbps which is negligible for the bcc phy of 10mbps . the advantage of our system is that the bcc link offers highly reliable , low - energy , and high - data rate communication @xcite . therefore , the control messages that are exchanged across the different nodes and pass through the bcc link are more efficient in terms of both energy / bit and delay / bit when compared to rf . * frequency of channel estimate propagation . * when the rf channel changes rapidly , more frequent updates will be required since they will affect the relay selection . instead , when the channel is fairly static , the relay nodes will change in larger time scales . this is an important aspect that is a general problem that presents itself in wireless communication systems . in this paper we evaluate this aspect in our performance evaluation section . while collecting the information and broadcasting it is an important task , it is also critical to ensure that the incoming information at a node is properly processed . this functionality is implemented in the rx_status_bcc ( ) function of the pseudo - code in fig . [ fig : coop - protocol ] . this function is invoked when a new @xmath13 message is received . in this case this function first re - sorts the two linked lists named @xmath24 and @xmath25 that contain all the @xmath1 network nodes according to the delay and energy estimates of their rf links . next we explain why this simple message processing approach is enough . * exploiting the mac for serializing information updates . * the @xmath13 messages are transmitted and processed sequentially by all nodes and this ensures consistent and reliable flow of information in the network . [ fig : example - status - bcc - message ] has an example with five nodes that will clarify how these messages are processed . assume that the nodes start from the same consistently order lists @xmath24 and @xmath25 shown in the left of fig . [ fig : example - status - bcc - message ] . assume now that node 2 measures a local change in the rf performance . from the last known state shown in the left of fig . [ fig : example - status - bcc - message ] , it knows its new position in the list @xmath24 that it now updates locally . then , it broadcasts a @xmath13 . the transmission of this message is naturally serialized by the bcc mac in the sense that if there are more updates from other nodes they will be transmitted sequentially . thus , this list will be updated only once for every @xmath13 message transmission and all nodes make this update simultaneously . at this stage of our system description we have explained how all the necessary information is being updated . now the actual relay selection decision is carried out in the second part of the function rx_status_bcc ( ) . in lines 4 - 8 of the pseudo - code node @xmath10 checks how many nodes perform better than itself in terms of delay and energy costs . a relay node drops the role of being a relay if it estimates from these measurements that either its delay or energy performance of the rf link is worse from more than @xmath2 ( nodes that are allowed to be relays at all times ) . if this is the case it means that at least @xmath2 nodes have better performance characteristics depending on the delay or energy optimization objective . the node releases the token by transmitting a message @xmath14 to the currently best non - relay node in list @xmath24 ( lines 11 - 12 ) . similarly , in the case that a node estimates that it should become a relay , but currently it is not a relay , it waits for the broadcasted token that will be released in the way we explained before ( lines 14 - 16 ) . the rx_token_bcc ( ) function is responsible for processing the @xmath14 messages . when this message is received by the appropriate node , it starts this new relay role . * robustness of token_bcc transmissions . * [ subsection:4 ] @xmath14 messages are transmitted reliably , i.e. all nodes ack the packet . since a token is acknowledged by the new relay , the `` old '' relay that releases it , and the remaining network nodes , know that the transaction was executed . thus , all nodes know exactly which node is a relay and which is not . note also that the transmission of tokens is also serialized in the bcc network which means that a new token is released when the last transaction is successfully completed . at this stage in describing our system we have explained how it creates a list of the best relays . now the task of node @xmath10 is to check if packet forwarding through relay node @xmath26 should take place and this task is executed at the network layer by the function tx_data_nwk ( ) . when the application layer ( appl ) wants to send a packet it does so with this function that is executed on demand . this is where the algorithm determines if the bcc or the rf links will be used . in the pseudo - code in fig . [ fig : coop - protocol ] , the condition for forwarding data through a relay @xmath26 is that the expectation of the energy cost of the local rf transmission denoted as @xmath18 $ ] , must be higher than the energy cost of the combined use of the bcc link and the rf link of the relay @xmath26 : @xmath27 \geq \mathbb{e}[e_{bcc}]+ \mathbb{e}[e_{rf}(n)]^2/e_{rem}(n)\ ] ] the forwarding condition can also be formulated in terms of the delay and be expressed as @xmath28 \geq \mathbb{e}[d_{bcc}]+\mathbb{e}[d_{rf}(n)],\ ] ] or the forwarding decision can be based on a combined metric ( e.g forward only when both the delay and energy of the relayed transmission is better ) . in the previous expressions , @xmath29 $ ] and @xmath30 $ ] denote the expectation of the random variables of the energy and delay for one packet transmission in the bcc network . expressions and are evaluated by the main loop of the packet forwarding algorithm depicted in function tx_data_nwk(k ) of fig . [ fig : coop - protocol ] . in this function , the loop checks first the best relay node that is contained in the list @xmath31 , and if the condition is satisfied , then the message is forwarded to it . otherwise , if the condition is not satisfied by the best relay then we employ direct rf transmission since all the remaining relays will have worse energy transmission cost . * motivation . * in the previous section we explained that the rf relays are selected based on their delay and/or energy efficiency by using information that is distributed in the wban . if a node simply measures the average delay in the rf link and reports it to the wban , this does not account for the time consumed in retransmissions and channel access failures . the same is true for the bcc link . thus , a node needs to precisely account for the previous parameters in order to decide if it will use forwarding through bcc or not . therefore , before an actual forwarding decision is made by a node it must know that this is indeed the optimal course of action . _ to accomplish that we develop not a cross - layer , but a same - layer and cross - technology performance model of the two mac / phys that are controlled by the nwk . _ * traffic assumptions . * throughout this section we consider applications where a node @xmath26 generates locally and asynchronously data packets with an poission rate of @xmath32 packets per second . thus , we adopt an m / g/1 queueing model for modeling the random variables of the @xmath33 and @xmath34 for the rf and bcc networks respectively . several works that we reference in our subsequent analysis use a similar approach for capturing the service behavior of csma / ca algorithm . we also assume that the total load @xmath35 can be forwarded even by a single rf node . the rate of packets that are locally generated and forwarded _ directly _ from the relays is denoted as @xmath36 . we also denote with @xmath37 the total network load , it is not distributed equally to all the relay nodes since their link quality may be different instantaneously resulting into differences in and . ] that nodes receive through the bcc link and must _ forward _ through their rf links , i.e. it is @xmath38 . by decomposing the total load into the two components we have that @xmath39 * rf channel assumptions . * similar to other works , that focus primarily on accurate modeling of ieee 802.15.4 , we adopt the following assumptions for modeling the radio network @xcite . the wireless channel is a frequency - flat rayleigh fading wireless link that remains invariant per phy frame , but may vary between frames . this means that the received signal - to - noise ratio ( snr ) @xmath40 per frame is a random variable . snr estimation is performed from the rssi at the receiver . finally , we assume additive white gaussian noise ( awgn ) . modeling the delay and energy cost of wireless rf transmissions is important for making the optimal decisions . however , existing models can not be directly used @xcite . the reason is that we consider the aggregate traffic load that is flowing from the wban to the gateway and at the same time this traffic goes though the different packet erasure links of the involved rf relay nodes . thus , the basic idea of our model is that the packet loss probability is calculated by considering first the different packet erasure probability of each rf link , and second by using the aggregate traffic load for deriving the probability of having a cca failure . now according to the ieee 802.15.4 phy coding scheme , the code that is used is a pseudo - orthogonal scheme where 4 bits are encoded together into a 32 chip signal . the modulation scheme is offset quadrature phase shift keying with half sine shaping ( oqpsk - hss ) at a rate of 2mchip / s . the result is that the 1/8th rate coding scheme achieves a throughput of 250kb / s . now if @xmath41 is the gaussian @xmath42 function , then for oqpsk in a rayleigh channel , the instantaneous bit error rate ( ber ) is : @xmath43 the packet error probability can be calculated as @xmath44 where @xmath20 is the length of the packet in bits . since acks are used the probability that a phy frame transmission fails towards the gateway from relay node @xmath26 is : @xmath45[1-peb_{ack}],\ ] ] where @xmath46 and @xmath47 are the packet error probability for data packets and acknowledgments respectively . let now @xmath48 denote the probability that the rf channel is busy when cca is performed . the probability that the channel is free after @xmath49 failed cca attempts is simply @xmath50 . the value of @xmath48 depends only on @xmath2 , i.e. how many nodes contend for the rf channel . this probability is independent at each attempt while such an approximation is very accurate for saturated traffic as shown in @xcite , and has been widely adopted in the literature also for unsaturated traffic @xcite . we provide the methodology for the derivation of @xmath48 below . let @xmath51 denote the number of served / transmitted packets in a busy period of the m / g/1 rf queueing system . then it is @xmath52=\frac{1}{1-\rho}.\ ] ] note that for the m / g/1 queuing system we consider the total load that is transmitted in the rf network with an average service rate @xmath53 . this is because there is no need to differentiate the traffic from each node since the aggregate is what matters for the model as along as the arrival rate at each node is poisson . now for each packet it will be that the average service rate is : @xmath54 + t_{d } + t_{att } + t_{ack})\ ] ] in the above expression @xmath55 is the random variable of head of line ( hol ) delay while @xmath56 and @xmath57 are the time durations of the data packets and the acks . @xmath58 is the required turnaround time to switch from receiving ( rx ) to transmitting ( tx ) mode as specified in the standard @xcite . the hol delay is the duration from the time instant that the packet arrives at the head of the rf transmission queue to the time instant just before its transmission or final discard . to obtain the expression for @xmath48 we have to divide the time period that the channel is occupied by the remaining @xmath59 devices versus the total duration of the m / g/1 system busy period ( see fig . [ fig : rf - mac - protocol ] ) . the term @xmath60\mathbb{e}[d_{hol}]$ ] is the total duration of the busy period in the rf network . so we have that : @xmath61(t_{cca}+ t_{d } + t_{att } + t_{ack})}{\frac{1}{\lambda_f+\lambda_d}+\mathbb{e}[s_{rf}]\mathbb{e}[d_{hol}]}\ ] ] in the above equation we can replace @xmath62 $ ] that can be derived as a function of @xmath63 $ ] from and . in the last equation , @xmath64 is the average packet loss probability in the rf network that occurs both because of erasures and cca failures . since these two events are independent , we have that @xmath65 where @xmath66 and @xmath67 are the maximum number of transmissions and cca attempts respectively . note in the last equation that we account for the different @xmath68 that an rf relay might have . _ this is an important difference with related works on mac performance modeling @xcite that consider a single packet erasure probability but instead we consider the different impact of all the used relays . _ now we will also derive a second equation that along with can be solved numerically for the derivation of @xmath48 and @xmath63 $ ] . for calculating the delay @xmath55 we have to model the behavior of the backoff algorithm . we denote the value of the contention window as @xmath69 . for the successful cca after @xmath70 failed ones , the hol delay that accounts only for ccas will be : @xmath71 = \sum_{\upsilon=0}^{m_c-1}\pi_{cca}^{v}(1-\pi_{cca } ) \big \ { \sum_{i=0}^{\upsilon}\frac{w_i-1}{2 } t_{s}\nonumber\\ + ( \upsilon+1 ) t_{cca } \big \ } + \pi^{m_c}_{cca } \big \ { \sum_{i=0}^{m_c-1}\frac{w_i-1}{2 } t_{s}+(m_c+1 ) t_{cca } \big \}\end{aligned}\ ] ] the previous equation covers the case of @xmath72 failed cca attempts and a last successful one . in the case that there were @xmath72 times a failed cca , the total delay is attributed to the backoff algorithm that was executed at the sender , plus there is the time spent for the @xmath72 ccas and is denoted as @xmath5 ( plus one successful cca ) . the parameter @xmath73 is the duration of the slot time . the second summation term covers the case of reaching the maximum limit of @xmath67 failed ccas . if there is an erasure and the number of allowed transmissions attempts is @xmath66 , then the average hol delay becomes : @xmath74&=&\frac{1}{n_r}\sum_{n=1}^{n_r}\sum_{k=0}^{m_r-1}\pi_{e}^{k}(n)(1-\pi_{e}(n ) ) \ { \mathbb{e}[d^{cca}_{hol , rf } ] \nonumber\\ & + & k ( t_{att}+t_{d}+t_{ack } ) \}\end{aligned}\ ] ] the term @xmath75 is similarly with before the probability that a packet was erased @xmath76 times before a final successful transmission . if the data packet is lost then the lack of an ack incurs a delay @xmath77 . the transmitter observes the lack of an ack packet and assumes the packet was lost and proceeds with backoff and another round of possible multiple ccas ( depicted in fig . [ fig : rf - mac - protocol ] ) . now the average per packet delay of the rf link is : @xmath78=\mathbb{e}[d_{hol , rf}]+ t_{d } + t_{att } + t_{ack}\ ] ] in this subsection we characterize the average energy consumption of the rf transmissions . if the power consumption of the rf subsystem in active , cca , transmit , and receive modes are @xmath79 , and @xmath80 respectively , then the energy required for the rf transmission is calculated as follows : @xmath81&=&(\mathbb{e}[d_{hol , rf}]-a_{cca } t_{cca } ) p_{act}+p_{cca } a_{cca } t_{cca}\nonumber\\ & + & t_{d } p_{tx}+t_{att } p_{act}+t_{ack } p_{rx}\end{aligned}\ ] ] the first term is the energy consumed during contention while the node was simply in active mode , but we subtract the total time duration that ccas were executed during the total time that the packet was in the hol position . @xmath82 is the average number of ccas performed during this period and it is equal to @xmath83 although the transmission through the bcc network is very reliable , an increased number of contending nodes will eventually translate to an increase in the delay and energy consumption of the bcc subsystem . therefore , in this section we analyze the mac layer performance of the bcc network in order to calculate the impact of the different parameters on the packet delay and energy . we distinguish the delay to transmit a packet successfully on the bcc network as three random components . from fig . [ fig : bcc - mac - protocol2 ] we can identify those three components easily . first , @xmath84 $ ] is the expectation of the random hol delay spent by the transmitter node from the moment it decides to send a data packet , until the time instant it transmits a preamble packet . this delay does not include the processing time and the transmission time of the preamble . then @xmath85 is the random delay spent by the transmitter node until the receiver node is in the listening state and an rtr packet arrives at the transmitter node . finally , @xmath86 is the random delay spent by the transmitter node from the instant of the rtr reception until the transmission of the data packet . this component also includes the processing time , the transmission time of the data packet , and the transmission of the ack . * hol delay before preamble transmission . * with the used mac protocol , the preamble transmission mechanism is based on a binary exponential backoff algorithm similar to the ieee 802.15.4 rf mac . as we explained earlier , once the bcc channel is obtained through this preamble contention scheme , data packets are transmitted contention - free . if the channel is busy , a random backoff is executed before a further preamble transmission attempt . let @xmath87 be the maximum number of cca attempts for a preamble . also @xmath5 is the required channel sensing time from the bcc hardware and @xmath73 is the slot time . let also @xmath88 be the probability to sense the bcc channel busy , and @xmath89 is the number of transmitted packets in a busy period of the bcc queueing system . the parameter @xmath88 in this case is calculated in the same way with the rf network : @xmath90(t_{cca}+ t_{d } + t_{att } + t_{ack})}{\frac{1}{\sum_{n = n_r+1}^{n}\lambda_n}+\mathbb{e}[s_{bcc}]\mathbb{e}[d_{hol , bcc}]}\ ] ] recall that in this case there are no packet erasures since reliable transmission is assumed once the channel is occupied and so @xmath64 is calculated from a simpler version of . _ equation demonstrates another novel aspect of our model since it incorporates this cross - technology interaction : it quantifies the impact of different number of rf relays @xmath2 on the probability that there is a channel access failure in the bcc link . this expression also quantifies the impact of a different forwarding load @xmath37 on the bcc network . _ similar with the rf case we can write for the transmission of the preambles : @xmath91 & = & \sum_{\upsilon=0}^{m_{mp}-1 } \sigma_{cca}^{u}(1-\sigma_{cca } ) \big \ { \sum_{i=0}^{\upsilon}\frac{w_i-1}{2 } t_{s}\nonumber \\ & + & ( \upsilon+1 ) t_{cca}\big \ } + \sigma_{cca}^{m_{mp}}\big \ { \sum_{i=0}^{m_{mp}-1}\frac{w_i-1}{2 } t_{s}\nonumber\\ & + & ( m_{mp}+1 ) t_{cca } \big \}\end{aligned}\ ] ] since the bcc mac is also modeled as an m / g/1 queue , we can have a similar expression with for @xmath88 and @xmath84 $ ] . this second equation is solved jointly with , expressions for the hol delay and channel busy probability @xmath88 for the bcc network are obtained . * preamble and rtr transmission delay @xmath85 . * now we calculate @xmath85 , i.e. the random delay the transmitter node waits from the start of transmitting a preamble until the rtr transmitted by the receiving node is received by the transmitter node . [ fig : bcc - mac - protocol2 ] depicts this important detail for the calculation of @xmath85 . an important characteristic of the used bcc mac protocol is that the selection of the sleeping time @xmath92 is always such that the target sleeping node can be awake only from the successful transmission of a single preamble packet . the listening time @xmath93 on the other hand depends on the hardware . since the preamble has a fixed duration , the receiver will always wait until the end of the preamble @xcite . therefore it will be : @xmath94 although @xmath85 was derived with simple reasoning , there is a need to calculate another parameter that will be needed for the energy estimation in a later subsection . to this aim we define as @xmath95 the random time to wait until the wake up receiver and the actual receiver wakes up completely . in our system this time duration can be easily modeled since it falls with uniform distribution in the range @xmath96 $ ] . this is because such a time duration is computed from the beginning of the preamble transmission of the transmitting node that may uniformly fall somewhere in the interval @xmath96 $ ] . therefore , the average value for @xmath95 will be @xmath97$]=@xmath98 . we also denote with @xmath99 the random time to wait from the beginning of the preamble transmission until the start of the listening period ( see fig . [ fig : bcc - mac - protocol2 ] ) . now if the node is awake it is easy to see that @xmath99=@xmath100 , otherwise it will be @xmath97 $ ] on average . so for the average value of @xmath99 we can write that : @xmath101=\mathbb{e}[t_w ] \pr\{asleep\}=\frac{r_s}{2}\frac{r_s}{r_s+r_l}\ ] ] also from fig . [ fig : bcc - mac - protocol2 ] , @xmath102 is the delay from the wake up moment of the receiver due to the preamble , until the end of the preamble . this time duration will be simply @xmath103=t_{pream}-\mathbb{e}[t_a].\ ] ] * data transmission delay @xmath86 and total delay . * after the node is awakened and the rtr is sent , the transmission of the data packet with the associated ack consume a constant amount of time that depends only on the length of the data packet @xmath20 . since the probability of packet errors at the bcc phy is assumed to be negligible the bcc transmission is always successful and @xmath86 is constant . this constant delay component is @xmath104 where @xmath58 is the time required for the bcc transceiver to switch from rx to tx mode , and @xmath56 is the transmission time of the data packet of length @xmath20 . thus , we have that the total average delay from the moment a packet becomes the hol packet until it is acked is : @xmath105=\mathbb{e}[d_{hol , bcc}]+t_{2}+t_{3}\ ] ] the total energy consumption over a listening - sleeping time duration @xmath106 , is calculated by adding the energy consumed by a transmitting node to send a data packet ( @xmath107 $ ] ) , and the energy consumed by the receiving node to receive the data packet ( @xmath108 ) . the calculations use the value of power consumption for the bcc transceiver and the time duration that this power drainage occurred , while we again remove the subscript that indicates the bcc system for avoiding clogging the presentation . from fig . [ fig : bcc - mac - protocol2 ] we derive the time instants that the sender uses its transceiver in rx , and tx modes . @xcite . ] so it will be just for the contention round for a single packet transmission : @xmath109= \mathbb{e}[d_{hol , bcc}(act ) ] p_{act } + \mathbb{e}[d_{hol , bcc}(cca ) ] p_{cca}\ ] ] similarly we calculate @xmath110,\mathbb{e}[e_{3,bcc}]$ ] from and . the average energy consumption @xmath107 $ ] is the sum of these three expectations . at the receiver we calculate similarly the time instants that the hardware is in tx / rx mode , active , and in sleep mode . during the reception of a preamble we calculated that the average duration that the receiver is awake is @xmath111 $ ] . so we have that for a single packet transmission : @xmath112 & = & r_s p_{sleep}+\mathbb{e}[t_b ] p_{act}+t_{rtr } p_{tx}\nonumber\\ & + & t_{d } p_{rx}+t_{ack } p_{tx}\end{aligned}\ ] ] the average bcc energy @xmath29 $ ] for a single packet transmission is the sum of and @xmath107 $ ] that is derived as we explained a few lines above . from our derivations we observe from , , and that the total average energy consumption depends on several parameters . if we notice more carefully we see that for constant @xmath1 and @xmath2 the energy and delay expressions for the rf and bcc links are independent . this means that the energy for the rf link can be further optimized individually and locally by each node depending on the local rf channel while the parameters of the bcc link are optimized in the same way by all nodes . many optimization objectives exist for such a system . as a representative case , we examine energy minimization subject to an average packet delay constraint . for the rf part two parameters can be optimized for a given @xmath2 and packet length @xmath20 ( this usually depends on the application ) , and these are the maximum number of transmissions @xmath66 and cca attempts @xmath67 . we use the rf average energy expression in for optimally selecting the two aforementioned parameters at relay node @xmath26 : @xmath113(m_c , m_r , l , n_r,\pi_e(n))\\ \text{subject to } \pi_{loss } ( n ) \leq 15\% \nonumber\end{aligned}\ ] ] the constraint in the above is the packet loss rate ( plr ) for a specific node @xmath26 and is given from before the averaging over all the relays . an important note is that in theory the above optimization problem must be solved on - line every time a node calculates a different @xmath114 with the help of , , and the local rssi . for the energy of the bcc link more parameters can be optimized while the optimization problem has _ to be solved once for a given nework configuration_. the reason is that for a given number of nodes @xmath1 , relays @xmath2 , and locally generated load by each node @xmath32 , the bcc load @xmath37 that must be forwarded is constant regardless of who is the rf node that actually forwards it . this also means that @xmath88 is constant for a given network configuration . with this logic we can easily see that the following optimization problem indeed needs to be solved once : @xmath115(r_s , r_l , m_{mp},l,\lambda_f,\sigma_{cca},n , n_r)\\ \text{subject to } \mathbb{e}[d_{rf}]+\mathbb{e}[d_{bcc } ] \leq \tau \nonumber\end{aligned}\ ] ] the constraint denotes the average estimate of the delay that must be less that the maximum allowed delay @xmath116 . the reason that we need the delay constraint here unlike , is that the energy in the case of the bcc link can be minimized by increasing @xmath117 which will incur higher delay . regarding the solution and the implementation , for the parameters @xmath118 , @xmath119 , we produce off - line several analytical expressions for a given @xmath2 , @xmath1 , and @xmath20 . the result is that for the delay and energy of the rf network a linear expression is produced that only has as optimization parameters @xmath67 , @xmath66 , and accepts as input @xmath120 from the real measurements of each node @xmath26 . a linear equation with two unknowns can be solved in real - time even by underpowered micro - controllers . also for the bcc network the energy and delay formulas are similarly calculated off - line for specific sets of input parameters , while @xmath87,@xmath92,@xmath93 , are the optimization parameters . after the selection of the optimal values is made , they are enforced in next cycle of the node . are the same for the complete network . only one node is needed in practice to make the previous optimization and it informs the remaining ones about the result . ] * evaluated systems . * in the first part of our evaluation of the proposed system we used simulation due to the lack of hardware devices equipped with rf and bcc transceivers that are in a size that can be deployed in a human . for simulating our system we implemented the nwk protocol and the rf and bcc mac protocols . for the bcc phy system we used parameters available from the literature and are presented shortly , while for the rf link we considered a rayleigh channel and the ieee 802.15.4 phy that we described earlier in section [ section : system - architecture ] . we compare our system with a baseline system that employs only rf communication . in this system when a node is completely blocked from the gateway , it connects opportunistically with an on - body node that can reach the gateway . for all systems we measured energy consumption , average per packet delay , and plr . * parameters and topology . * the raw data rate of the bcc transceiver is approximately 8.5mbps with a ber of @xmath6 and a voltage setting of 1.2v , the power consumption is 2.1mw in rx mode , and 0.6mw in tx mode @xcite . for a typical ieee 802.15.4 rf transceiver the cc2420 from texas instruments that operates with the same voltage , the current consumption is at 19.7ma for the tx mode and 17.4ma for the rx mode @xcite . for this simulation every node starts with the same energy budget and there is a fixed number of 1000 packets that must be transmitted from each node to the gateway . we also set @xmath20=100 bytes , @xmath121=23us , @xmath122=@xmath123=0.1ms , @xmath124=0.1ms , @xmath125=0.2ms , @xmath124=0.384ms , @xmath126=0.192ms , @xmath127=0.25ms , @xmath128=1.12ms , and @xmath129=0.352ms . regarding the topology of the network , we consider _ scenario 1 _ where a human sits in a static position with randomly placed nodes in the body but with more than 50% of the nodes within line - of - sight ( los ) of the gateway . for _ scenario 2 _ we considered a more challenging case where a human has half of the nodes in the torso ( within los ) and half in the back while lying in bed ( no los ) . [ fig : energy - vs - load](a ) depicts the ratio of the consumed energy of the proposed system ( @xmath130 ) , versus the energy of the baseline system ( @xmath131 ) , for a different system load @xmath132 and _ scenario 1_. the delay constraint is very relaxed and equal to @xmath116=1s while @xmath66=@xmath67=3 . the maximum load corresponds to 250kbps which is the maximum data rate at the phy of the ieee 802.15.4 rf link . it is interesting to see from these results that with the proposed system the energy reduction is significant regardless of how many relays ( @xmath2 ) are used . the dominant factor that determines the reduction is the load which as it is increased , the performance gain becomes higher even with a single relay . this can be explained because for a higher load with the baseline method nodes with poor rf links are still used for transmission and successive failures ( allowed by the high @xmath116 ) result in a higher number of retransmissions . the higher number of packets in the rf medium increases also the contention that has minor impact in the overall plr as fig . [ fig : energy - vs - load](b ) shows . however , when the delay constraint is more tight , and equal to 5ms , the energy differences as we can see in fig . [ fig : energy - vs - load](c ) are reduced , but plr increases dramatically as fig . [ fig : energy - vs - load](d ) indicates . for a high load , plr becomes nearly 10% for the baseline case while it is increased slightly in the proposed system . this is because the average number of retransmissions and cca attempts are reduced significantly for the baseline case , which eventually leads to lower energy wastage but still high plr . _ this is a result that demonstrates that the proposed scheme optimizes the performance of each node individually and also jointly the performance of the complete network . _ [ fig : energy - vs - snr](a ) depicts the impact of the channel conditions on the energy reduction for @xmath132=0.5 and _ scenario 1_. this is also an interesting result since it shows that the energy is reduced drastically as the snr of the rf channel improves . however , at a certain point around 20db , we see clearly that the energy reduction is minimized . this is because the channel is improved and direct transmissions succeed without requiring retransmissions . therefore , relaying starts becoming less useful in this case since every node can optimize its own transmissions . on the other hand for poor channel quality , relaying can not help considerably because all the rf links are affected . in fig . [ fig : energy - vs - snr](b ) we present results for @xmath2=2 and @xmath1=4 , while different fixed values of maximum rf transmissions @xmath66 , cca attempts @xmath67 , and loads were simulated . here , @xmath116 was set to 200ms . as the load is increased we observe that there is a need for more cca attempts than the standard maximum 3 while the value of @xmath66 is not that important for _ scenario 1_. with the proposed scheme and _ scenario 1 _ , one or at most two transmissions are enough . however , for the cases with @xmath66=1,@xmath67=3 , and @xmath66=2,@xmath67=3 of _ scenario 2 _ , we notice that the ratio is reduced even more . the reason is that for higher @xmath66 the baseline method is very energy inefficient since half of the nodes suffer from high plr and several retransmissions are needed . on the other hand the proposed system does not use the maximum @xmath66=2 since packets are forwarded through other rf links . for this experiment , we reversed the optimization objective and the constraints of problem formulations explained in section [ section : optimization ] . more specifically the objective in this case was delay minimization subject to an energy constraint . the results correspond to different load @xmath132 and number of relays @xmath2 . the total energy constraint was set to 100 joule that can cover in theory the transmission of 1000 100-byte packets with the selected rf transceiver . for the optimization problem a proportional energy budget was assigned to each packet depending on the remaining energy of the node . the improvement in the average delay and plr of the transmitted packets can be seen in fig . [ fig : latency - vs - load ] . when every node operates with the baseline system the average number of retransmissions reaches a value between three and four . we observed that this difference is more significant if there is no energy constraint . * rf traces and setup . * for this part of our evaluation we used the real rssi traces from two mobile experiments that used the ieee 802.15.4 and involved two users @xcite . according to the scenario in @xcite , a single node is placed in each user and both users walk towards each other starting from a distance of 50 m . they continue walking after they cross each other for an additional distance of 10 m . in the first experiment both rf nodes are placed in the torso and in the second experiment the node is located in the pocket of a user s trousers . these measurements reflect exactly what we want to test and what @xcite also showed : for the larger duration of the experiment connectivity between the two torso nodes is very good and better than the experiment where nodes are placed in the pocket . however , when the two users cross each other , connectivity between nodes in the torso is lost completely , while for the experiment with nodes placed in the pocket connectivity is still good . note also that these traces correspond to low mobility scenarios that is generally the focus of wban monitoring applications . * energy model of the complete system . * in this case we considered additional sources of power consumption . we include first the power consumption for powering - up / down the rf transceiver as specified in the cc2420 datasheet @xcite . we also approximated the computational overhead of the relay selection algorithm by compiling and profiling in terms of executed instructions our code in the cc2420 simulator . based on the number of required execution cycles , we used the additional delay and power consumption of the micro - controller in that mode . so in this way we include both communication and computational energy and delay costs . * results . * in our results presented in fig . [ fig : real - traces ] we see the average packet delay for different @xmath66 . it is interesting to note that even when both nodes are assumed to be used simultaneously , helping each other as the baseline system we defined earlier ( this system here is denoted as `` rf relay '' ) , this is not always the optimal case even when compared with just `` torso '' and `` pocket '' . the reason is that there is significant body shadowing between the node in torso and the node in the pocket and so there is significant packet loss that has to be compensated with higher number of retransmissions . energy and delay results follow a similar trend in fig . [ fig : real - traces](b ) . regarding the plr , when @xmath66 is increased this is decreased as fig . [ fig : real - traces](c ) indicates . the proposed system reaches perfect plr because the best link is selected and there is no need for many retransmissions . finally , we also evaluated the impact of a slower frequency for propagating rf channel estimates . we measured all the previous parameters and we only present delay in fig . [ fig : real - traces](d ) . this higher update frequency mainly affects the `` rf relay '' system because it makes suboptimal rf relay decisions . our system is affected with a difference that is barely noticeable . of course for an even slower update frequency of more than 10 seconds , we observed that similarly the performance of all systems was becoming proportionally worse . an interesting detail is that the used traces consider backlogged traffic at each node which is inconsistent with the poisson assumption used in our performance model . thus , the performance improvement clearly shows that our approach is not limited by the traffic model and it can work under more general settings . in traditional rf - based wbans in the face of body shadowing there is a fundamental tradeoff : either consume more energy for retransmissions or channel coding in order to increase reliability and suffer also higher delay , or reduce delay and energy at the cost of reduced reliability . however , we demonstrated that the above does not have to be the case if nodes cooperate through a delay / energy - efficient secondary link ( in this case bcc ) in order to select the most efficient rf link for forwarding wban data to a gateway . the performance gains are materialized not only with the novel wban architecture , but also with a nwk protocol that uses an algorithm for rf relay selection and packet forwarding driven by a performance model . furthermore , the cross - technology performance models are exploited for local optimization of the mac parameters of the protocols that operate in the bcc and rf subnetworks . the performance results indicate that the proposed system is more efficient than a state - of - the - art scheme in terms of energy and delay under different realistic channel conditions , application loads , and performance constraints . the potential concern regarding the proposed system is the need for two different technologies . however , existing ic technology already allows the integration of multiple transceivers in a single chip . our future work will be focused first on the optimal configuration of other system parameters like the number of used relays and the content of the transmitted packets . next , we intend to investigate the potential benefits of the proposed system when it is used with an ieee 802.11 rf communication link for multimedia transmission scenarios . the authors would like to acknowledge the anonymous reviewers for their thorough comments that helped improve this manuscript considerably . d. cavalcanti , r. schmitt , and a. soomro , `` performance analysis of 802.15.4 and 802.11e for body sensor network applications , '' in _ 4th international workshop on wearable and implantable body sensor networks ( bsn 2007 ) _ , ser . ifmbe proceedings.1em plus 0.5em minus 0.4em springer berlin heidelberg , 2007 , pp . 914 . e. reusens , w. joseph , g. vermeeren , l. martens , b. latre , i. moerman , b. braem , and c. blondia , `` path loss models for wireless communication channel along arm and torso : measurements and simulations , '' in _ ieee ap - s _ , 2007 . h. baldus , a. argyriou , k. klabunde , t. falck , v. buil , l. geurts , t. schenk , and s. ouzounov , `` method for operating an acoustic system and wearable acoustic device , '' nov . 25 2010 , wo patent app . pct / ib2010/051,970 . [ online ] . available : http://www.google.com/patents/wo2010133991a1?cl=en s. corroy , a. argyriou , z. wadood , and h. baldus , `` a body - coupled communication and radio frequency dual technology cooperation protocol for body - area networks , '' in _ second international workshop on medical applications networking ( man ) , in conjunction with ieee icc _ , 2010 . b. braem , b. latre , i. moerman , c. blondia , e. reusens , w. joseph , l. martens , and p. demeester , `` the need for cooperation and relaying in short - range high path loss sensor networks , '' in _ sensorcomm _ , 2007 . c. fischione , s. c. ergen , p. park , k. h. johansson , and a. sangiovanni - vincentelli , `` medium access control analytical modeling and optimization in unslotted ieee 802.15.4 wireless sensor networks , '' in _ secon _ , 2009 . antonios argyriou received the diploma in electrical and computer engineering from democritus university of thrace , greece , in 2001 , and the m.s . and ph.d . degrees in electrical and computer engineering as a fulbright scholar from the georgia institute of technology , atlanta , usa , in 2003 and 2005 , respectively . currently , he is a tenure - track faculty member at the department of electrical and computer engineering , university of thessaly , greece . from 2007 until 2010 he was a senior research scientist at philips research , eindhoven , the netherlands . from 2004 until 2005 , he was a senior engineer at soft.networks , atlanta , ga . dr . argyriou currently serves in the editorial board of the _ journal of communications_. he has also served as guest editor for the _ ieee transactions on multimedia _ special issue on quality - driven cross - layer design , and he was also a lead guest editor for the _ journal of communications _ , special issue on network coding and applications . argyriou serves in the tpc of several international conferences and workshops in the area of communications , networking , and signal processing . his current research interests are in the areas of wireless communication systems and networks , and video communications . he is a member of ieee . alberto c. breva received the diploma in electrical and computer engineering from the university of seville , spain , in 2008 , and the m.s . degree in 2010 from the same university . he was an intern at philips research during 2009 - 2010 where he worked on the topic of body coupled communication . marc aoun received his bachelor degree in computer and communications engineering from the american university of beirut , lebanon , in 2004 , and his master s degree in communications engineering from rwth aachen , germany , in 2007 . since 2007 he is a research scientist at philips research , the netherlands . his current research interests are in the areas of wireless communication systems , wireless sensor networks , real - time systems and segmentation of large - scale networks .

dust extinction can be determined from the number of distant field galaxies seen through a spiral disk . to calibrate this number for the crowding and confusion introduced by the foreground image , and developed the `` synthetic field method '' ( sfm ) , which analyses synthetic fields constructed by adding various deep exposures of unobstructed background fields to the candidate foreground galaxy field . the advantage of the sfm is that it gives the average opacity for area of galaxy disk without assumptions about either the distribution of absorbers or of the disk starlight . however it is limited by low statistics of the surviving field galaxies , hence the need to combine a larger sample of fields . this paper presents the first results for a sample of 32 deep hst / wfpc2 archival fields of 29 spirals . the radial profiles of average dust extinction in spiral galaxies based on calibrated counts of distant field galaxies is presented here , both for individual galaxies as well as for composites from our sample . the effects of inclination , spiral arms and hubble type on the radial extinction profile are discussed . the dust opacity of the disk apparently arises from two distinct components ; an optically thicker ( ) but radially dependent component associated with the spiral arms and a relatively constant optically thinner disk ( ) . these results are completely in agreement with earlier work on occulted galaxies . the early type spiral disks in our sample show less extinction than the later types . lsb galaxies and possibly sd s appear effectively transparent . the average color of the field galaxies seen through foreground disks does not appear to change with radius or opacity . this grey behaviour is most likely due to the patchy nature of opaque clouds . the average extinction of a radial annulus and its average surface brightness seem to correlate for the brighter regions . this leads to the conclusion that the brighter parts of the spiral disk , such as spiral arms , are also the ones with the most extinction associated with them .

astro-ph0411663

the optical depth of spiral disks has been the topic of many and varied studies since the claim by @xcite that they are transparent . the subject became controversial when @xcite and @xcite argued that disks were virtually opaque . at the cardiff meeting @xcite many possible methods to attack the problem were proposed . the dust disks of spirals may obscure objects in the high redshift universe @xcite or conceal mass in their disks @xcite . an excellent review of the current state of knowledge on the opacity of spiral disks is given by @xcite . early approaches to this subject were presented in @xcite and more recent developments are : + ( 1 ) disks are more opaque in the blue @xcite , ( 2 ) they are practically transparent in the near infrared @xcite , making these bands the best mass - to - luminosity estimator @xcite , ( 3 ) disks are practically transparent in the outer parts but show significant absorption in the inner regions @xcite , ( 4 ) the extinction correlates with galaxy luminosity @xcite , and ( 5 ) spiral arms are more opaque than the disk @xcite the majority of these studies are based on either inclination effects on the light distribution of a large sample of disks or a dust and light model to fit the observed profiles . while there is some agreement on the view that spiral disks are substantially optically thick in their central regions and become optically thin in their outer parts , the exact radial extinction profile remains uncertain . most measurements to date use the disk light itself to measure the extinction , and consequently require an assumption on the relative distribution of dust and light in a spiral disk . the extinction in a disk can be derived from far infrared and sub - mm emission arising from the cold dust in disks . however , these methods make the assumption that the emission characterises the dust in the disk . but the far infrared and sub - mm emission is likely to be dominated by the warmest component of the dust , which tends to be the smaller grains on the outside of molecular clouds facing an energy source @xcite . in this case the far - infrared and sub - mm emission will underestimate the average opacity . @xcite mention this in their estimate of gas - to - dust ratio in spiral disks based on iras observations . to obtain a better characterisation of the absorption in a spiral disk without knowledge of the distribution of stars and dust in the disk , a known background source is needed . @xcite proposed to use an occulted galaxy for this purpose , assuming it has a symmetric light distribution @xcite . @xcite and the companion paper of this work @xcite use the number of distant galaxies in the field as their background source , calibrating this number with simulations following their `` synthetic field method '' ( sfm ) . in @xcite , we describe the details of this method . both the `` occulting galaxy method '' and the sfm have the benefit of not using the disk s own light to measure the extinction . the drawbacks of the keel and white method are the assumption of symmetry of both galaxies and the few suitable pairs available . the sfm is more universally applicable but limited by the poor statistics . it does however not need to assume any distribution of either the disk s light or the absorbers in it . the sfm does require high - resolution images from hst and remains limited to the arms and disk of spiral galaxies due to crowding . in this paper we report radial extinction profiles for spiral disks of different hubble types based on 32 hubble space telescope , wide field planetary camera 2 ( hst / wfpc2 ) fields in 29 galaxies of hubble types sab and later . in section 2 we describe the sample and its selection in detail . the sfm is briefly outlined in section 3 ; a more complete description of the method and the recent improvements we have made to it is given in the companion paper @xcite . in section 4 we discuss the radial profile of average opacity for individual galaxies and , in section 5 , the composite average radial extinction profile for our entire sample . the effects of inclination , spiral arm prominence , and hubble type on the profiles are discussed as well . section 6 discusses the average ( v - i ) color of the field galaxies we found and section 7 , the tentative relation between average surface brightness and opacity . we discuss some of the implications of our results , and end with the conclusions on disk opacity from the numbers and colors of field galaxies seen through them . our sample of hst / wfpc2 fields was selected from the mast archive at stsci , based on criteria for both the target galaxy and the hst data . the total exposure times and original proposal i d and reference are listed in table 1 , the basic data on the galaxies in table 2 . the total solid angle of this sample is 146 arcmin@xmath2 . the foreground galaxy should be a spiral , ideally face - on , spanning enough sky to cover a significant number of field galaxies . this solid angle constraint limits the maximum distance for application of the sfm to approximately 30 mpc . hst starts to resolve the disk population of spirals at close distances , making the field too crowded for field galaxy identification ( see @xcite ) . this imposes a minimum distance of a few mpc . the galaxies are type sab and later , plotted in figure 1 , according to the rc3 @xcite . no limit on inclination was imposed as long as spiral arms could be discerned . the majority of this sample is from the distance scale key project . the project s observing strategy @xcite was geared towards maximizing the number of cepheid variables detected . this resulted in the selection of fields in the optical disk of face - on later - type spirals with a prominent arm visible . these selection criteria are to be considered when interpreting the opacity measurements of these fields . the sample of spiral disks has multi - epoch imaging in 2 photometric band available , as the original science driver for many of these data sets was to sample the cepheid light curves . the reacquisition resulted in a slight shift in the pointing at each epoch . the unintentional dither allowed us to drizzle the combined images to a pixel scale of 0.05 . the data reduction is described in detail by @xcite . @xcite predicted that improved resolution would mainly benefit the statistics of distant galaxy counts in the local group . however , as the sample spans a range of distances and the statistics are likely to be poor , the maximum possible sampling was selected . the pc chip was not used for analysis because it often lies on the most crowded region , its noise characteristics are different from the wf chips , and there are fewer reference fields available for the sfm . this sample of hst pointings was selected as it is reasonably uniform , which allows similar processing using hubble deep field ( hdf ) background fields . the hst / wfpc2 pointing should neither be on the center of the galaxy where crowding is too much of a problem , nor be outside the disk of the target galaxy where the expected opacity is likely too low to be measured with the sfm . we selected pointings with the f814w ( i ) and f555w ( v ) filters as the identification of objects is based on several parameters characterising structure and on the v - i color . a minimum exposure time of about 2000 seconds in both filters was adopted . the choice of filters and exposure time was based on our earlier experiences with spiral and irregular galaxies @xcite . and to maximize the number of suitable fields . @xcite describe in detail the data - reduction and the automated sfm , so we will give only a short summary here . to calibrate the numbers of field galaxies found in the science fields , simulated or `` synthetic '' fields are made . these are the original science fields with an extincted hubble deep field added . the numbers of simulated field galaxies found suffer from the same confusion and crowding as the number from the science field and therefore depend solely on the dimming applied . a series of simulations give us the relation between field galaxy numbers and applied dimming . the following relation is fitted to the simulated numbers : @xmath3 @xmath4 is the dimming in magnitudes , @xmath5 the number of field galaxies in a simulation , @xmath6 and @xmath7 are the normalization and slope of the fit ( i.e. the number of field galaxies expected with no extinction and how this number diminishes with increasing extinction ) . the slope c characterises the effects of crowding and confusion on the relation between opacity and galaxy numbers . it is usually slightly over unity but the unique character of each field led us to characterise c using the simulations in each field or combination of fields . the intersection between this relation and the _ real _ number of field galaxies gives us the average extinction ( @xmath8 ) typical for the solid angle where these field galaxies were found . the field galaxies are identified in the science fields , using automated selection , based on structural parameters and ( v - i ) color , together with a visual verification . in the synthetic fields , the visual check of objects was substituted by anti - correlating the automatically selected objects with the automatic selection in the science field , removing both the real distant galaxies as well as contaminants and leaving only added objects . the uncertainty in the number of field galaxies from the science field is a combination of the poisson uncertainty and the uncertainty due to field galaxy clustering . the uncertainty in the simulated numbers is poisson only as they come from a known background , the hubble deep fields . the uncertainty in average opacity is derived from the uncertainties in field galaxy numbers based on the poisson uncertainty as expressed by @xcite , and the clustering uncertainty based on the 2-p correlation function found by @xcite and equation 1 . the crowding and confusion bias is calibrated with the simulations . for a detailed error discussion of the sfm see @xcite , 4.3 . a difference in dust extinction from our own galaxy between the reference fields ( hdf - n / s ) and the pointing at the foreground galaxy will introduce a bias in the extinction measurement of the specific disk . @xcite used galaxy counts to measure the extinction towards grb 970228 . and found excellent agreement with other measurements of galactic extinction @xcite . @xcite produced an all - sky map of galactic extinction based on cobe and iras maps and we use their values for galactic extinction ( table 2 ) . most galaxies in the sample do not show a significant difference in galactic extinction compared to the average of the galactic extinction towards the hdf - n / s ( @xmath8 = 0.039 mag ) . however , the numbers of galaxies from each science field were nevertheless corrected for the difference in galactic extinction using equation 1 . @xcite presented results based on the sfm for individual wide field chips , characterised as `` arm '' or `` disk '' regions based on predominance in the chip . @xcite segmented the wfpc2 mosaics based on morphological component ( arm , inter - arm , disk ) or projected radius from the center . however , the statistics from individual wfpc2 fields barely allow any meaningful opacity measurements for solid angles smaller than a single wf field . in table 3 we present opacity values for the projected radial annuli in each foreground galaxy in our sample . the radii are expressed in @xmath9 , half the @xmath10 from rc3 catalog @xcite . the error - bars are computed from the uncertainties in the numbers of real and simulated field galaxies from counting and clustering @xcite . for the galaxies for which we have two wfpc2 pointings ( ngc5194 , ngc3621 and ngc4414 ) , the radial extinction from the combined counts is also shown . estimates of extinction in a galaxy disk based on a single wfpc2 chip suffer from poor statistics @xcite , which make a radial dependence hard to establish ( see table 3 and @xcite ) . this led us to apply the sfm to a large sample of foreground galaxies in order to estimate the general extinction properties of galaxy disks from the combined numbers of field galaxies seen through these disks . we caution against averaging the values in table 3 to derive average profiles . figure 2 shows the radial opacity plot from all our fields combined . the numbers of field galaxies from both the science fields and the simulations from all fields were combined based on their projected radial distance from the respective galaxy s centers , expressed in @xmath9 . the top panel in figure 2 shows the total combined solid angle for each radial bin for which the average opacity was determined . the middle panel shows the number of field galaxies from the science field and the average number found in the simulations without any dimming ( a=0 ) . the bottom panel shows the opacity for each bin derived from the intersection of equation 1 fitted to the simulations and the real number of galaxies . the solid angle or the number of simulated galaxies without any dimming is a good indicator for the reliability of our opacity estimate , as reflected by the error bars in the bottom panel of figure 2 . the estimates are limited at small radii by the high surface brightness and crowding of the foreground galaxy center , effectively masking some of the solid angle available at those radii . at higher radii the uncertainty comes from a lack of solid angle covered ( see figure 2 , top panel ) . this is a selection effect of our sample , as most of the wfpc2 fields were pointed at the optical disk of the galaxy . in order to determine the effects on average radial opacity of disk inclination , prominence of spiral arms or hubble type , radial opacity plots of subsets of our sample were constructed . the solid angle used is then some fraction of those in the top panel in figure 2 , which consequently increases the uncertainty in the opacity measurement . to counter this , instead of the radial bins of 0.1 @xmath9 ( figure 2 ) , radial binning of 0.2 @xmath9 was applied . the inclination of the foreground disks affects the measured opacity , but the amount of this effect depends on the dust geometry . in the case of a uniform thick screen , the path length attenuating the field galaxies is increased . however in the case of a screen of dark clouds in the disk , the effect is on the apparent filling factor of clouds . because the correction depends on which dust geometry is assumed , we present the radial results ( figure 2 ) without any correction and explore corrections using several different models , illustrated in figure 3 . the homogeneous screen results in a multiplicative factor cos(i ) to be applied to the opacity value ( a ) or in a correction to the number of field galaxies found in the science field , as follows : @xmath11 the other models consider a screen with fully opaque clouds or patches . depending on the thickness of these clouds , the apparent filling factor depends differently on inclination . in these models , the clouds have an average oblateness @xmath12 ( @xmath13 ) with major axis a and minor axis b of the clouds . following the geometry of figure 3 , the relation between the number of field galaxies , seen through the foreground disk face - on ( @xmath14 ) , at an inclination i ( @xmath15 ) , and the average number of field galaxies in the field behind the foreground galaxy ( @xmath6 ) can then be expressed as : @xmath16 \ [ 1 \ - \ cos(i)\ ] \ n_0 + \left\ { \ \epsilon \ [ 1-cos(i ) ] \ + \ cos(i ) \right\ } \ n_i\ ] ] where the extreme cases for @xmath12 are spherical clouds ( @xmath17 ) and flat patches ( @xmath18 ) . the oblateness @xmath12 parameterises the ratio between the scale height of the dust and its extent in the plane of the disk . in images of edge - on disks , the visible dust lanes are confined to a thinner disk than the stars . it is therefore likely that @xmath12 is not 0 . for purposes of illustration we use a value of 0.5 in figure 8 . figure 4 shows the radial profile from figure 2 corrected for both extreme cases and uniform screens in all galaxies . the contributions from each galaxy to the composite radial profile were corrected according to equation 2 and equation 3 before addition . figure 5 shows the total uncorrected radial profile for four subdivisions of our sample based on inclination . from figure 5 it seems clear that other effects are much more important than the inclination of the foreground disk . for this reason we ignoring the effects of inclination on our measurements . as the effects of inclination are debatable on the basis of preferred dust geometry , we present further results without inclination corrections except where noted . another effect on the average radial opacity profile in figure 2 is due to the presence of spiral arms . as our sample is predominantly from the cepheid distance project , the wfpc2 images all feature spiral arms . if these are more opaque than the disk proper @xcite , then the radial profile presented in figure 2 could be biased towards higher opacities . separate radial plots for the arm , the inter - arm part of the disk and outside any spiral arm part of the disk are shown in figure 7 based on the counts from the typical regions in the entire sample , with the exception of the lsb galaxies ugc 2302 and ugc 6614 . they were left out , as not much spiral structure can be discerned ; they appear completely transparent ( see figure 8) . in order to differentiate between the effects of spiral arms and disks in the opacity plots , the images were segmented into crowded , arm , disk ( inter - arm ) and disk ( outside arm ) . these regions were flagged in the mosaiced wfpc2 fields using the gipsy ( groningen image processing system ) function _ blot _ , in the same way as ngc1365 in @xcite . the choice of typical regions was made in order to compare the arm and inter - arm results of @xcite and @xcite . a typical mask is presented in figure 6 and masks for all galaxies in the sample are presented in @xcite . it should be noted that this segmentation into typical regions is subjective , and based on those sections of the foreground galaxies covered by the wfpc2 observations we use . radial extinction profiles of the typical regions are presented in figure 7 . the arm regions show much more opacity and a much more pronounced radial dependence of that opacity . there is a radial dependence as well for the inter - arm parts of the disk , although not as steep . the outside parts of the disk of the spiral galaxy , however , show little or no relation between opacity and radius . the opaque components of a spiral disk appear to be the spiral arms while the disk itself is more transparent but much more extended . figure 7 also shows the total solid angle of the radial annuli over which the opacities are determined . from these we can conclude that the regions deemed ` arm ' are not dominating the whole of the fields . the radial effect of the arms only becomes visible in the total radial opacity plot ( figure 2 and top left panel figure 7 ) , at the lower radii where the arms and the inter - arm region dominate . in figure 8 we present the profiles corrected for inclination assuming an @xmath12 of 0.5 . the general trends remain but now for somewhat lower opacity values . @xcite and @xcite presented their extinction values from occulting galaxy pairs as a function of the radius , scaled with @xmath9 . @xcite compared ground - based photometry , and @xcite used spectroscopic measurements of the occulted galaxy light . these extinction points are plotted in figure 9 for arm and inter - arm regions . the extinction curves from figure 7 and the inclination corrected one from figure 8 are plotted as well . both the arm and the inter - arm extinction values as a function of radius agree well with the values obtained from the occulting galaxy technique . it is remarkable how well the results compare , considering they were obtained from completely different samples of spiral galaxies and using different techniques . the values from the occulting galaxy technique are slightly lower than ours . there are several possible reasons for this . the spectroscopic results from @xcite ( triangles in figure 9 ) possibly favour the more transparent regions in a disk ( domingue 2004 , private communication ) . but more importantly , the sample of occulting foreground galaxies consists of a different makeup of spiral galaxy subtypes than that of this paper . it should be noted that there are no galaxies common to both our sample and that of the occulting galaxy technique . single galaxy in both our sample and the occulting galaxy method s . @xcite noted that their later types ( sbc ) seem more opaque , as do we ( figure 11 ) . figure 10 compares the results for the most prevalent hubble subtypes in the occulting galaxy method ( sb and sbc ) with our results for those . the arm values seem to match up but there is a difference in the inter - arm results . it is unclear to us whether this points to a structural effect in either technique . a likely explanation is that the definition of `` inter - arm '' applies to slightly different regions in the spiral disks for @xcite , and his paper . this paper s definition of typical regions is based on the high - resolution mosaic , whereas the @xcite is on their ground - based imaging . there is therefore the possibility that we include sections in the interarm regions which the occulting techniques would not resolve as `` inter - arm '' , raising our values of opacity for those regions with respect to the occulting galaxy technique . the sfm provides an independent verification of the occulting galaxy technique using a fundamentally different approach . in addition , a component was added to the distinction between arm and inter - arm parts of the disk , the `` outside '' , meaning not directly enclosed by spiral arms . the fact that this component is not fully transparent raises the possibility of a dust disk extending beyond the spiral arms ( figure 8) . with the effect of spiral arms on the opacity profile , the hubble type is likely to be of influence on the profile . figure 10 shows the extinction profiles averaged per hubble type . hubble types sab through sd are presented , as well as the average opacity profile of the two lsb galaxies in our sample . hubble types sab through scd in our sample show disk extinction up to the @xmath9 . the sab result is tentative , owing to the poor statistics from only 2 wfpc2 fields . the sd and lsb galaxies however appear effectively transparent . lsb galaxies were in fact assumed to be transparent by @xcite when discussing the morphology of field galaxies seen though them . however both profiles , sd and lsb , are based on only two wfpc2 fields , accounting for the higher uncertainties . comparing early type galaxies with later types , it appears that the later type galaxies ( sbc - sc ) show more extinction and at larger radii . the sb galaxies show a bump which appears to be associated with higher extinction from the more tightly wound spiral arms . figures 12 and 13 show the radial profiles for arm and disk -both inter - arm and outside regions- for our sample divided into early and late spirals . a finer separation in hubble type would have resulted in even higher uncertainties , due to lack of solid angle and hence statistics . the solid angle of each radial annulus is also plotted . purely arm regions do not dominate the opacity profiles . however , the interarm regions are similar in behaviour ( figures 7 and 8) . the inner parts of the profile are therefore more arm - like , while the remainder is disk dominated . the radial dependence of arm and inter - arm regions is more pronounced in the case of earlier galaxies than for later types . the opacity in the spiral arms is substantially higher then in the disk , for both early and late types . the total profiles show bumps at 0.9 and 1.1 @xmath9 for the early and late types , respectively . in the case of the early types this seems due to the spiral arm contribution at that radius . the bump for the late types is not as significant , but might be related to the general position of spiral arms as well . the average color of the field galaxies found in the science fields can , in principle , tell us something about the actual dust geometry responsible for the drop in numbers . if there is a correlation between the average reddening of the field galaxies and the average opacity of the foreground spiral , then the dust extinction responsible for the drop in number of field galaxies is , at least in part , in the form of a diffuse screen , reddening the visible galaxies . if , however , the average color of the field galaxies does not change with opacity , then the drop in their number is likely due to fully opaque clouds with transparent sections between them to allow for the detection of the unreddened surviving field galaxies . + our method influences the average color of the detected field galaxies in several ways . the v band ( f555w ) for the synthetic field background was constructed from the original hdf images in the f606w and f450w filters . however , @xcite estimated that this introduces a negligible error . crowding introduces blended objects in the synthetic fields . the synthetic counts are corrected for this effect but it will influence the average color . the simulated dust extinction we used was grey , so no preferred reddening was introduced in the synthetic fields . however , our automated selection procedure for field galaxy candidates selects against very blue objects , introducing a preference for red galaxies . a similar selection effect may take place in the visual check of the science fields , as blue objects are treated as suspected foreground objects . overall , these selection biases will cause the average color of the field galaxies in the science and synthetic fields to be redder than the average for an unobstructed field of distant galaxies . the effects of blended objects on the average colors of science and synthetic fields are not identical , as not all blends have been removed from the synthetic counts . nonetheless , we can compare the trends of both these average colors with radius . the average ( v - i ) color of the field galaxies found in the science fields , the simulation without extinction ( a = 0 ) , and the opacity derived from the galaxy numbers are plotted as a function of radius in figure 14 . the average ( v - i ) color of the field galaxies in the science fields does not change with radius and hence average opacity . the average color of the synthetic field objects appears to become bluer with radius . however , beyond 1.4 @xmath9 the number of objects is very small and the averages should be treated with caution . comparing the average color of distant galaxies from science and synthetic fields for the inner part of the disk , the average of the science field objects is redder than the synthetic field average . this difference in average color is likely the result of blends with blue foreground objects being inadvertently included in the the selection of synthetic field objects . the number of objects from the synthetic fields was corrected for this effect ( see also @xcite but the average color was not . the science field objects do not suffer from this as they were checked visually for blended objects . the field galaxies seen in the science fields are likely visible in parts of the disk which are nearly transparent or , alternatively , the dust screen in the disk behaves according to the `` grey '' extinction law . this is remarkable , considering the bias towards redder objects throughout our method . figure 15 shows the average color - extinction measurements for the field galaxies in the science fields . the galactic extinction law is shown for comparison . each point has been determined in a radial annulus , for all the fields combined ( figure 2 ) and for the typical regions ( figure 7 , but with the finer radial sampling of figure 2 . ) . without distinction between regions , no trend with opacity can be discerned for the average color . the average color in the arm regions does not seem to rise much with opacity . the arm extinction is decidedly grey , something also found by @xcite . for the disk regions , no distinct trend can be seen , and they do not seem to follow the galactic law very well . @xcite found the disk region of ngc4536 to be more galactic in its reddening law . there is an average reddening with respect to the average ( v - i ) color of the hdf objects identified as galaxies by our algorithm without a foreground disk . this reddening is likely to be the result of the effects of the visual check and contamination of the color measurement by stray disk light . these color - extinction relations for both arms and disk seem to be greyer than the galactic extinction slope , favouring the possibility that at least part of the extinction is in opaque clouds . uncertainties are such , however , that even for our increased statistics no good relation between opacity and color can be found . the counts of distant galaxies were added per radial annulus , expressed in @xmath9 . additionally , the flux and solid angle from each field can be added for each radial annulus . the averaged surface brightness ( @xmath19 ) and the average opacity ( a ) from each radial bin in figure 2 , are plotted in figure 16 . these values for the radial bins per typical region in figure 7 are plotted as well , but with finer sampling . there is a hint of a relationship between luminosity and extinction . this is consistent with the relation found by @xcite and @xcite between overall disk opacity and the galaxy s total luminosity . the solid angles over which the surface brightness and opacity were averaged were selected by radius and not luminosity . areas with different surface brightnesses are therefore combined at each radius , smoothing out any relation between opacity and surface brightness . hence , most of the points in figure 15 are around the same opacity and the range of surface brightness values is small . nevertheless , the values of high extinction and brightness ( @xmath20 ) do show some relationship between surface brightness and opacity , with arm regions displaying more extinction at the same surface brightness levels than the inter - arm disk . the disk values appear to show no correlation , but these values are per definition not for high surface brightness levels . and yet , together with the higher values of opacity found for the spiral arms , the points are not inconsistent with a relation between the opacity and luminosity . opacity measurements in partitions of the wfpc2 images based on average surface brightness instead of radius should reveal any relation more clearly . this comparison between extinction and emission will be presented , in more detail , in a later paper . from the number of field galaxies found through the disks of spiral galaxies , a quantative picture of extinction as a function of radius can be found . the sfm is too limited by statistics to obtain a good result for individual fields . however a meaningful measurement can be derived from a combination of several galaxies . the effect of spiral arms on the radial extinction is quite distinct and dependent on hubble type . from these radial plots it becomes clear that the dust in the disk is not one smooth layer sandwiched between the disk s stars but two separate components : radially dependent spiral arms and a more transparent but also extended disk . in addition , there is a highly opaque central bulge component . whether the average opacity is a result of dust clouds or a smooth screen of grey dust , is difficult to determine from the numbers , colors and luminosities of the field galaxies found through the disks of our sample . however , the relative independence on inclination of the average opacity measured from numbers of spiral galaxies and the grey relation between opacity and average field galaxy color , both point towards a patchy distribution of the absorbing dust in the disk . assuming that the optical depth of the disk can be expressed as @xmath21 , where f is the area filling factor , the average opacity of @xmath22 would require a filling factor of 40% of the disk , a figure that rises to 85 % in the spiral arms . while the visible dust lanes can account for at least a part of these clouds , other dust clouds are likely embedded in the disks and arms . this patchy coverage explains the occasional distant galaxy seen through a spiral disk @xcite , which has sometimes been used as anecdotal evidence for transparent disks . the disk is in fact relatively transparent where background galaxies are seen . however , by calibrating the number of the distant galaxies found , a very different picture emerges . while this result is consistent with earlier findings , these values are likely upper limits , not lower ones . any inclination correction will lower the face - on value for extinction . however , the patchy nature of dust extinction is likely responsible for a high variety of extinction values in the disk . integrated measurements such as light profiles , however , will be affected according to the presented opacity values . a possible relation between extinction and brightness would also cast doubt on the fixed mass - to - light ratios generally assumed when modelling the kinematics of a spiral disk . the effects of dust extinction on the number of field galaxies found in our fields lead us to the following conclusions : + ( 1 ) the sfm gives an unbiased but uncertain measure of opacity for single wfpc2 fields ( table 3 ) . + ( 2 ) on average , the disk of a spiral galaxy has an opacity in @xmath23 of @xmath241 magnitude ( figure 2 ) . + ( 3 ) the extinction measured from the number of field galaxies seems to be independent of the inclination of the foreground disk over the range in @xmath25 covered by our sample ( @xmath26 ) ( figure 5 ) . this is consistent with fully opaque flattened clouds covering a fraction of the area as the cause of the observed average opacity . + ( 4 ) the absorption profile for typical regions in the disk is strongly influenced by the spiral arms . spiral arm regions are the most opaque and display a radial dependence . the disk regions enclosed by a spiral arm are also more opaque than other disk regions and display a similar radial dependence ( figure 7 ) . + ( 5 ) the radial extinction curves derived from numbers of background galaxies and those reported by the occulting galaxy technique agree reasonably well ( figure 9 and 10 ) , although a systematic effect in either ( or both ) technique seems to be present in the interarm results ( figure 10 ) . + ( 6 ) sc galaxies show much more opacity in their central regions than other types ( figure 11 ) . + ( 7 ) all hubble types earlier than scd in our sample show substantial disk extinction in @xmath23 up to @xmath9 ( figure 11 ) . + ( 8) the numbers from sd and lsb galaxies are consistent with a transparent disk , but these measurements are limited by statistics . ( figure 11 ) . + ( 9 ) both early and late type spirals exhibit the extinction profiles of the two distinct components : a radially dependent one ( arm and inter - arm ) , and a more extended disk ( outside arms ) ( figures 12 and 13 ) . + ( 10 ) the average color of the distant galaxies identified in the science fields does not change with either radius or the opacity derived from their numbers ( figure 14 ) . + ( 11 ) the grey nature of the absorption derived from field galaxy counts and colors holds for all typical regions ( figure 15 ) . the absence of inclination effects and the grey nature of the absorption is consistent with strongly absorbing dense clouds masking off the distant galaxies . their covering factor in the disk would be around 40% , regardless of cloud sizes ( which our technique can not provide ) . + ( 12 ) the average surface brightness in radial annuli and the corresponding average opacity derived from distant galaxy counts appear correlated . they are consistent with a rise of opacity with surface brightness . ( figure 16 ) . although the range of surface brightness is small and the uncertainty in opacity rises with higher surface brightness , this is in good agreement with earlier results for bright galaxies . this would constitute a relation between dust mass and light . + the `` synthetic field method '' has proven itself to be a useful , model - independent , technique to measure the total opacity of spiral disks . it can be applied to any spiral disk at intermediate distance for which there is high resolution imaging available . we will present further results for opacity as a function of surface brightness in a future paper . the authors would like to thank the anonymous referee for his or her comments and harry ferguson for insightful discussions on the `` synthetic field method '' . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jpl , caltech , under contract with nasa . this work is primarily based on observations made with the nasa / esa hubble space science telescope ( stsci ) and obtained from the data archive at the space telescope institute . support for this work was provided by nasa through grant number hst - ar-08360 from stsci . stsci is operated by the association of universities for research in astronomy , inc . , under the nasa contract nas 5 - 26555 . we are also grateful for the financial support of the stsci director s discretionary fund ( grant numbers 82206 and 82304 ) , and of the kapteyn institute of groningen university , l. , bresolin , f. , kennicutt , r. c. , saha , a. , stetson , p. b. , freedman , w. l. , mould , j. r. , madore , b. f. , sakai , s. , ford , h. c. , gibson , b. k. , graham , j. a. , han , m. , hoessel , j. g. , huchra , j. , hughes , s. m. , illingworth , g. d. , phelps , r. , prosser , c. f. , and silbermann , n. a. : 1998 , , 655 , l. , freedman , w. l. , hill , r. j. , saha , a. , madore , b. f. , kennicutt , r. c. , stetson , p. b. , ford , h. c. , graham , j. a. , hoessel , j. g. , han , m. , huchra , j. , hughes , s. m. , illingworth , g. d. , kelson , d. , mould , j. r. , phelps , r. , silbermann , n. a. , sakai , s. , turner , a. , harding , p. , and bresolin , f. : 1996 , , 568 , w. l. , hughes , s. m. , madore , b. f. , mould , j. r. , lee , m. g. , stetson , p. , kennicutt , r. c. , turner , a. , ferrarese , l. , ford , h. , graham , j. a. , hill , r. , hoessel , j. g. , huchra , j. , and illingworth , g. d. : 1994 , , 628 , w. l. , madore , b. f. , gibson , b. k. , ferrarese , l. , kelson , d. d. , sakai , s. , mould , j. r. , kennicutt , r. c. , ford , h. c. , graham , j. a. , huchra , j. p. , hughes , s. m. g. , illingworth , g. d. , macri , l. m. , and stetson , p. b. : 2001 , , 47 , b. k. , hughes , s. m. g. , stetson , p. b. , freedman , w. l. , kennicutt , r. c. , mould , j. r. , bresolin , f. , ferrarese , l. , ford , h. c. , graham , j. a. , han , m. , harding , p. , hoessel , j. g. , huchra , j. p. , illingworth , g. d. , kelson , d. d. , macri , l. m. , madore , b. f. , phelps , r. l. , prosser , c. f. , saha , a. , sakai , s. , sebo , k. m. , silbermann , n. a. , and turner , a. m. : 1999 , , 48 , j. a. , ferrarese , l. , freedman , w. l. , kennicutt , r. c. , mould , j. r. , saha , a. , stetson , p. b. , madore , b. f. , bresolin , f. , ford , h. c. , gibson , b. k. , han , m. , hoessel , j. g. , huchra , j. , hughes , s. m. , illingworth , g. d. , kelson , d. d. , macri , l. , phelps , r. , sakai , s. , silbermann , n. a. , and turner , a. : 1999 , , 626 , j. a. , phelps , r. l. , freedman , w. l. , saha , a. , ferrarese , l. , stetson , p. b. , madore , b. f. , silbermann , n. a. , sakai , s. , kennicutt , r. c. , harding , p. , bresolin , f. , turner , a. , mould , j. r. , rawson , d. m. , ford , h. c. , hoessel , j. g. , han , m. , huchra , j. p. , macri , l. m. , hughes , s. m. , illingworth , g. d. , and kelson , d. d. : 1997 , , 535 , b. w. , allen , r. j. , and van der kruit , p. c. : 2002 , in _ asp conf . ser . 273 : the dynamics , structure & history of galaxies : a workshop in honour of professor ken freeman ed . da costa & h. jerjen . san francisco , astronomical society of the pacific _ , pp 337+ , s. m. g. , han , m. , hoessel , j. , freedman , w. l. , kennicutt , r. c. , mould , j. r. , saha , a. , stetson , p. b. , madore , b. f. , silbermann , n. a. , harding , p. , ferrarese , l. , ford , h. , gibson , b. k. , graham , j. a. , hill , r. , huchra , j. , illingworth , g. d. , phelps , r. , and sakai , s. : 1998 , , 32 , d. d. , illingworth , g. d. , saha , a. , graham , j. a. , stetson , p. b. , freedman , w. l. , kennicutt , r. c. , mould , j. r. , ferrarese , l. , huchra , j. p. , madore , b. f. , prosser , c. f. , bresolin , f. , ford , h. c. , gibson , b. k. , hoessel , j. g. , hughes , s. m. g. , macri , l. m. , sakai , s. , and silbermann , n. a. : 1999 , , 614 , d. c. , filippenko , a. v. , gates , e. l. , li , w. , eastman , r. g. , barth , a. j. , bus , s. j. , chornock , r. , coil , a. l. , frink , s. , grady , c. a. , harris , a. w. , malkan , m. a. , matheson , t. , quirrenbach , a. , and treffers , r. r. : 2002a , , 35 , l. m. , huchra , j. p. , stetson , p. b. , silbermann , n. a. , freedman , w. l. , kennicutt , r. c. , mould , j. r. , madore , b. f. , bresolin , f. , ferrarese , l. , ford , h. c. , graham , j. a. , gibson , b. k. , han , m. , harding , p. , hill , r. j. , hoessel , j. g. , hughes , s. m. g. , kelson , d. d. , illingworth , g. d. , phelps , r. l. , prosser , c. f. , rawson , d. m. , saha , a. , sakai , s. , and turner , a. : 1999a , , 155 , l. m. , huchra , j. p. , stetson , p. b. , silbermann , n. a. , freedman , w. l. , kennicutt , r. c. , mould , j. r. , madore , b. f. , bresolin , f. , ferrarese , l. , ford , h. c. , graham , j. a. , gibson , b. k. , han , m. , harding , p. , hill , r. j. , hoessel , j. g. , hughes , s. m. g. , kelson , d. d. , illingworth , g. d. , phelps , r. l. , prosser , c. f. , rawson , d. m. , saha , a. , sakai , s. , and turner , a. : 1999b , , 155 , j. r. , hughes , s. m. g. , stetson , p. b. , gibson , b. k. , huchra , j. p. , freedman , w. l. , kennicutt , r. c. , bresolin , f. , ferrarese , l. , ford , h. c. , graham , j. a. , han , m. , hoessel , j. g. , illingworth , g. d. , kelson , d. d. , macri , l. m. , madore , b. f. , phelps , r. l. , prosser , c. f. , rawson , d. , saha , a. , sakai , s. , sebo , k. m. , silbermann , n. a. , and turner , a. m. : 2000 , , 655 , d. m. , macri , l. m. , mould , j. r. , huchra , j. p. , freedman , w. l. , kennicutt , r. c. , ferrarese , l. , ford , h. c. , graham , j. a. , harding , p. , han , m. , hill , r. j. , hoessel , j. g. , hughes , s. m. g. , illingworth , g. d. , madore , b. f. , phelps , r. l. , saha , a. , sakai , s. , silbermann , n. a. , and stetson , p. b. : 1997 , , 517 , s. , ferrarese , l. , kennicutt , r. c. , graham , j. a. , silbermann , n. a. , mould , j. r. , freedman , w. l. , bresolin , f. , ford , h. c. , gibson , b. k. , han , m. , harding , p. , hoessel , j. g. , huchra , j. p. , hughes , s. m. , illingworth , g. d. , kelson , d. , macri , l. , madore , b. f. , phelps , r. l. , saha , a. , sebo , k. m. , stetson , p. b. , and turner , a. : 1999 , , 540 , n. a. , harding , p. , ferrarese , l. , stetson , p. b. , madore , b. f. , kennicutt , r. c. , freedman , w. l. , mould , j. r. , bresolin , f. , ford , h. , gibson , b. k. , graham , j. a. , han , m. , hoessel , j. g. , hill , r. j. , huchra , j. , hughes , s. m. g. , illingworth , g. d. , kelson , d. , macri , l. , phelps , r. , rawson , d. , sakai , s. , and turner , a. : 1999 , , 1 , n. a. , harding , p. , madore , b. f. , kennicutt , r. c. , saha , a. , stetson , p. b. , freedman , w. l. , mould , j. r. , graham , j. a. , hill , r. j. , turner , a. , bresolin , f. , ferrarese , l. , ford , h. , hoessel , j. g. , han , m. , huchra , j. , hughes , s. m. g. , illingworth , g. d. , phelps , r. , and sakai , s. : 1996 , , 1 , a. , ferrarese , l. , saha , a. , bresolin , f. , kennicutt , r. c. , stetson , p. b. , mould , j. r. , freedman , w. l. , gibson , b. k. , graham , j. a. , ford , h. , han , m. , harding , p. , hoessel , j. g. , huchra , j. p. , hughes , s. m. g. , illingworth , g. d. , kelson , d. d. , macri , l. , madore , b. f. , phelps , r. , rawson , d. , sakai , s. , and silbermann , n. a. : 1998 , , 207 l l l l l l ngc 925 & 26400.0 & 9000.0 & & 5397 & @xcite + ngc 1365 & 66560.0 & 16060.0 & & 5972 & @xcite + ngc 1425 & 58800.0 & 29700.0 & & 5972/6431 & @xcite + ngc 1637 & 26400.0 & 13200.0 & & 9155 & @xcite + ngc 2541 & 28760.0 & 12760.0 & & 5972 & @xcite + ngc 2841 & 26400.0 & 11000.0 & & 8322 & @xcite + ngc 3031 ( m81 ) & 2000.0 & 2000.0 & & 9073 & @xcite + ngc 3198 & 27760.0 & 12560.0 & & 5972 & @xcite + ngc 3319 & 26400.0 & 10400.0 & & 6431 & @xcite + ngc 3351 ( m95 ) & 31900.0 & 9830.0 & & 5397 & @xcite + ngc 3621 - 1 & 5200.0 & 7800.0 & & 8584 & @xcite + ngc 3621 - 2 & 20759.0 & 7380.0 & & 5397 & @xcite + ngc 3627 ( m66 ) & 58800.0 & 25000.0 & & 6549 & @xcite + ngc 4321 ( m100 ) & 32750.0 & 17150.0 & & 5397 & @xcite + ngc 4414 - 1 & 1600.0 & 1600.0 & & 8400 & hubble heritage + ngc 4414 - 2 & 32430.0 & 10230.0 & & 5397 & @xcite + ngc 4496a & 68000.0 & 16000.0 & & 5427 & @xcite + ngc 4527 & 60000.0 & 25000.0 & & 7504 & @xcite + ngc 4535 & 48800.0 & 31200.0 & & 5397/6431 & @xcite + ngc 4536 & 68000.0 & 20000.0 & & 5427 & @xcite + ngc 4548 ( m98 ) & 48500.0 & 30900.0 & & 6431 & @xcite + ngc 4559 & 2000.0 & 2000.0 & & 9073 & @xcite + ngc 4571 & 10400.0 & 26400.0 & & 6833 & @xcite + ngc 4603 & 58800.0 & 14800.0 & & 6439 & @xcite + ngc 4639 & 58800.0 & 13000.0 & & 5981 & @xcite + ngc 4725 & 32430.0 & 10230.0 & & 5397 & @xcite + ngc 6946 & 2000.0 & 2000.0 & & 9073 & @xcite + ngc 7331 & 40660.0 & 9860.0 & & 5397 & @xcite + ugc 2302 & 15000.0 & 15600.0 & & 8255 & @xcite + ugc 6614 & 10100.0 & 10100.0 & & 8213 & @xcite + ngc 5194 - 1 ( m51 ) & 2000.0 & 2000.0 & & 9073 & @xcite + ngc 5194 - 2 ( m51 ) & 2000.0 & 2000.0 & & 9073 & @xcite + l l l l l l l l l l ngc 925 & sabd & 36.820469 & 33.578880 & -70 & 62.48 & 11.22 & 9.16 & 0.147 & + ngc 1365 & sbb & 53.401909 & -36.140659 & 49 & 34.41 & 10.47 & 17.95 & 0.039 & + ngc 1425 & sbb & 55.548061 & -29.893511 & -51 & 57.32 & 5.75 & 21.88 & 0.025 & + ngc 1637 & sab(rs)c & 70.367622 & -2.858040 & 27 & 30.68 & 3.98 & 8.2 & 0.078 & ( 1 ) + ngc 2541 & sacd & 123.666969 & 49.061440 & -7 & 56.63 & 6.31 & 11.22 & 0.097 & + ngc 2841 & sab & 140.511063 & 50.976479 & -30 & 60.33 & 8.13 & 14.1 & 0.030 & ( 2 ) + ngc 3031 ( m81 ) & sa(s)ab & 148.88826 & 69.06526 & -38.5 & 58.87 & & 3.63 & 0.155 & + ngc 3198 & sbc & 154.979126 & 45.549690 & 37 & 61.77 & 8.51 & 13.80 & 0.024 & + ngc 3319 & sb(rs)cd & 159.789719 & 41.686871 & 39 & 69.51 & 6.17 & 13.30 & 0.028 & + ngc 3351 ( m95 ) & sbb & 160.990555 & 11.703610 & -2 & 14.53 & 7.41 & 10.00 & 0.054 & + ngc 3621 & sac & 169.567917 & -32.812599 & -16 & 58.13 & 12.3 & 6.64 & 0.156 & + ngc 3627 ( m66 ) & sab(s)b & 170.062607 & 12.991290 & 1 & 57.38 & 9.12 & 10.05 & 0.063 & + ngc 4321 ( m100 ) & sabbc & 185.728745 & 15.822380 & -57 & 39.65 & 7.41 & 15.21 & 0.051 & + ngc 4414 & sac & 186.612869 & 31.223545 & -20 & 48.70 & 3.63 & 17.70 & 0.038 & + ngc 4496a & sbm & 187.913361 & 3.939467 & 65 & 58.67 & 3.98 & 14.86 & 0.048 & + ngc 4527 & sab(s)bc & 188.535400 & 2.653810 & 67 & 70.73 & 6.17 & 14.1 & 0.043 & ( 3 ) + ngc 4535 & sabc & 188.584625 & 8.197760 & 27 & 28.36 & 7.08 & 15.78 & 0.038 & + ngc 4536 & sab(rs)bc & 188.613037 & 2.187880 & -70 & 62.61 & 7.08 & 14.93 & 0.035 & + ngc 4548 ( m98 ) & sbb & 188.860123 & 14.496320 & -60 & 22.48 & 5.37 & 16.22 & 0.074 & + ngc 4559 & sab(rs)cd & 188.990372 & 27.959761 & -37 & 61.77 & 10.72 & 10.88 & 0.034 & + ngc 4571 & sa(r)d & 189.234879 & 14.217357 & 75 & 23.07 & 3.63 & 14.9 & 0.091 & ( 4 ) + ngc 4603 & sa(rs)bc & 190.229980 & -40.976402 & 30 & 50.21 & 3.39 & 33.3 & 0.325 & ( 5 ) + ngc 4639 & sabbc & 190.718140 & 13.257536 & -25 & 43.95 & 2.75 & 21.98 & 0.050 & + ngc 4725 & sabab & 192.610886 & 25.500759 & 47 & 45.25 & 10.72 & 12.36 & 0.023 & + ngc 5194 ( m51 ) & sa(s)bc & 202.46957 & 47.19526 & 45.0 & 25.58 & 11.22 & 8.4 & 0.067 & + ngc 6946 & sab(rs)cd & 308.718048 & 60.153679 & 70 & 39.65 & 11.48 & 11.48 & 0.663 & + ngc 7331 & sab & 339.267090 & 34.415920 & -12 & 57.38 & 10.47 & 14.72 & 0.176 & + ugc 2302 & sb(rs)m & 42.285831 & 2.127265 & 60.0 & 1.0 & 4.79 & 14.7 & 0.156 & ( 6 ) + ugc 6614 & sa(r)a & 174.811844 & 17.143578 & -65 & 40.54 & 1.66 & 84.68 & 0.055 & ( 7 ) + ngc 925 & @xmath27 & @xmath28 & @xmath29 & @xmath30 & & & & & & & + ngc 1365 & & @xmath31 & @xmath32 & & & & & & & & + ngc 1425 & & & @xmath33 & @xmath34 & @xmath35 & @xmath36 & @xmath37 & & & & + ngc 1637 & & @xmath38 & @xmath39 & @xmath40 & @xmath41 & & & & & & + ngc 2541 & @xmath42 & @xmath43 & @xmath44 & @xmath45 & & & & & & & + ngc 2841 & @xmath46 & @xmath47 & @xmath48 & @xmath49 & @xmath50 & & & & & & + ngc 3031 & @xmath51 & @xmath52 & & & & & & & & & + ngc 3198 & & @xmath53 & @xmath54 & @xmath55 & & & & & & & + ngc 3319 & @xmath56 & @xmath57 & @xmath58 & @xmath59 & & & & & & & + ngc 3351 & & @xmath60 & @xmath61 & & @xmath62 & @xmath63 & @xmath64 & @xmath65 & & @xmath66 & + ngc 3621 & @xmath67 & @xmath68 & @xmath69 & & & & & & & & + ngc 3621 - 1 & & @xmath70 & & & & & & & & & + ngc 3621 - 2 & @xmath71 & @xmath72 & @xmath73 & & & & & & & & + ngc 3627 & & @xmath74 & @xmath75 & & & & & & & & + ngc 4321 & & @xmath76 & @xmath77 & @xmath78 & & & & & & & + ngc 4414 & & @xmath79 & @xmath80 & @xmath81 & @xmath82 & @xmath83 & & & & & + ngc 4414 - 1 & & @xmath84 & @xmath85 & @xmath86 & @xmath87 & @xmath88 & & & & & + ngc 4414 - 2 & & @xmath89 & @xmath90 & @xmath91 & @xmath92 & @xmath93 & & & & & + ngc 4496a & & @xmath94 & @xmath95 & @xmath96 & & & & & & & + ngc 4527 & @xmath97 & @xmath98 & @xmath99 & @xmath100 & & & & & & & + ngc 4535 & @xmath101 & @xmath102 & @xmath103 & @xmath104 & @xmath105 & & & & & & + ngc 4536 & & @xmath106 & @xmath107 & @xmath69 & & & & & & & + ngc 4548 ( m98 ) & & @xmath108 & @xmath109 & @xmath110 & @xmath111 & @xmath112 & @xmath113 & @xmath114 & @xmath115 & @xmath116 & + ngc 4559 & @xmath117 & @xmath118 & @xmath119 & @xmath120 & & & & & & & + ngc 4571 & @xmath121 & @xmath122 & @xmath123 & @xmath124 & & & & & & & + ngc 4603 & @xmath125 & @xmath126 & @xmath127 & & & & & & & & + ngc 4639 & @xmath128 & @xmath129 & @xmath130 & @xmath131 & & & & & & & + ngc 4725 & & @xmath132 & @xmath133 & @xmath134 & & & & & & & + ngc 5194 & & @xmath135 & @xmath136 & @xmath137 & & & & & & & + ngc 5194 - 1 & & @xmath138 & @xmath139 & @xmath140 & & & & & & & + ngc 5194 - 2 & & @xmath141 & @xmath142 & @xmath143 & & & & & & & + ngc 6946 & & @xmath144 & @xmath145 & @xmath146 & @xmath147 & & & & & & + ngc 7331 & & @xmath148 & @xmath149 & @xmath150 & & & & & & & + ugc 2302 & @xmath151 & @xmath91 & @xmath152 & @xmath153 & & & & & & & + ugc 6614 & & & @xmath154 & @xmath155 & @xmath156 & @xmath41 & @xmath157 & @xmath158 & @xmath159 & @xmath160 & @xmath161 +

the bridge index and superbridge index of a knot are important invariants in knot theory . we define the bridge map of a knot conformation , which is closely related to these two invariants , and interpret it in terms of the tangent indicatrix of the knot conformation . using the concepts of dual and derivative curves of spherical curves as introduced by arnold , we show that the graph of the bridge map is the union of the binormal indicatrix , its antipodal curve , and some number of great circles . similarly , we define the inflection map of a knot conformation , interpret it in terms of the binormal indicatrix , and express its graph in terms of the tangent indicatrix . this duality relationship is also studied for another dual pair of curves , the normal and darboux indicatrices of a knot conformation . the analogous concepts are defined and results are derived for stick knots .

1205.5248

the bridge index and superbridge index of a knot are important invariants in knot theory . they are defined as the minimum and maximum number of local maxima respectively in the projection of a conformation of a knot onto an axis , minimized over all conformations of the knot . the bridge index was introduced by schubert @xcite in 1954 to study companionship in satellite knots . the superbridge index was introduced by kuiper @xcite in 1987 , and has proven to be a useful invariant in obtaining lower bounds on the stick number of knots , which is defined as the least number of straight line segments needed to be placed end - to - end to form the knot in space . for example , jin @xcite used the superbridge index to prove that the stick number of the @xmath0-torus knot @xmath1 is equal to @xmath2 for @xmath3 . for a fixed embedding @xmath4 of a knot into @xmath5 and any choice of axis , we can count the number of stationary points in the projection of the knot onto that axis . this allows us to define a map on the unit sphere , the bridge map of the knot conformation @xmath4 , by assigning to each vector @xmath6 one half of the number of stationary points in the projection of the knot onto the axis defined by @xmath7 . the bridge index is then just one half of the minimum value of the bridge map , minimized over all conformations of the knot . the superbridge index is one half the maximum value of the bridge map , minimized over all conformations of the knot . in this paper , we study the graph of the bridge map , the ( minimal ) set of points in @xmath8 that separates the sphere into open regions on which the bridge map is constant . by interpreting the bridge map in terms of intersections of the tangent indicatrix of @xmath4 with great circles and using the concepts of dual and derivative curves as introduced by arnold @xcite in the study of the geometry of spherical curves , we show that the bridge graph is the union of the binormal indicatrix of the knot conformation with its antipodal curve and a number of great circles . we also define another map on @xmath8 , the inflection map of the knot conformation @xmath4 , which counts the number of inflection points and cusps in planar projections of @xmath4 , and show that its graph is the union of the tangent indicatrix of the knot conformation with its antipodal curve and a number of great circles . this duality relationship is studied for another dual pair of curves , the normal and darboux indicatrices of a knot conformation . analogous concepts are defined and results are derived for stick knots . this paper is organized as follows . in sec . [ sec : preliminaries_smooth ] , we review the necessary background from the differential geometry of closed space curves , define the bridge and inflection maps and their respective graphs for smooth knot conformations , and relate these maps to intersections of the spherical indicatrices with great circles . in sec . [ sec : smooth_knots ] , we introduce the concepts of dual and derivative curves as introduced by arnold @xcite and prove the smooth knot versions of our main results . in sec . [ sec : preliminaries_polygonal ] , we review mcrae s @xcite definition of dual and derivative curves of spherical polygons and banchoff s @xcite construction of the spherical indicatrices of space polygons . in sec . [ sec : polygonal_knots ] , we derive analogues of the results in sec . [ sec : smooth_knots ] for stick knots . finally , in sec . [ sec : discussion ] , we discuss possible methods of solving existing open problems using the ideas developed in this paper . throughout this paper , we shall use the notation @xmath4 to denote a particular embedding of a knot in @xmath5 , and @xmath9 $ ] to denote a knot type , that is , the equivalence class of embeddings that can be obtained from a particular one under ambient isotopy . however , as in common parlance , we shall often refer to a knot type as a knot , and an embedding of a knot as a knot conformation . we shall also use the terms `` smooth knot '' and `` stick knot '' when referring to smooth and polygonal conformations of a knot respectively . we follow fenchel s @xcite definition of the frenet trihedral and curvature . let @xmath10 , @xmath11 , denote the arc length and @xmath12 the position vector of a varying point on a space curve @xmath4 . we assume that the coordinates of @xmath12 are smooth . each point @xmath12 has an associated osculating plane , that is , a plane containing the tangent vector @xmath13 and the vector @xmath14 . we assume that a suitably oriented unit vector normal to the plane , @xmath15 , the binormal vector of @xmath4 , has smooth coordinates . we also assume that the vectors @xmath16 and @xmath17 do not vanish simultaneously , and in addition , that they vanish at only a finite number of points . this implies that no arc of @xmath4 is contained in a plane . this allows us to avoid the usual assumption that @xmath14 , and hence the curvature , never vanishes . finally , we make the additional assumption that all of the vectors in the collection of vectors @xmath18 at the set of points where @xmath16 vanishes are pairwise distinct , and similarly for @xmath19 at the set of points where @xmath17 vanishes . the geometric meaning of this assumption will be explained in sec . [ sec : smooth_knots ] . we proceed to define the unit normal vector by @xmath20 and the curvature @xmath21 and torsion @xmath22 by @xmath23 which is possible since the derivatives are orthogonal to both @xmath24 and @xmath25 . the definition of the normal vector @xmath26 then yields @xmath27 thus we obtain the usual frenet - serret formulas for the movement of a point along a smooth curve : @xmath28 note , however , that in this circumstance , the curvature may vanish and even be negative . however , only a change in the sign of @xmath21 , and not the sign itself , has a geometric significance , since we may replace @xmath25 and hence @xmath26 by its opposite vector ( for all @xmath10 ) . points where @xmath21 and @xmath22 change sign will be called @xmath21- and @xmath22-inflection points respectively . the vectors @xmath29 form a trihedral known as the _ frenet trihedral _ or _ frenet frame_. if this trihedral is parallel translated to the origin , the endpoints of the translated vectors describe three curves @xmath30 , @xmath31 , and @xmath32 on the unit sphere @xmath33 , which are called the _ tangent _ , _ normal _ , and _ binormal indicatrices _ of the curve respectively ( or tantrix , notrix , and binotrix respectively for short ) . the lengths of these indicatrices are given by the formulae @xmath34 note that the above formulae differ from the regular formulae by the addition of the absolute value symbol around @xmath35 , which arises because we have allowed @xmath21 to take on negative values . the frenet trihedral defines a rigid motion around the origin called _ frenet motion _ , with angular velocity @xmath36 . the instantaneous axis of rotation of frenet motion , which we call the _ darboux axis _ , lies in the normal plane containing @xmath25 and @xmath24 because it is perpendicular to the velocity vectors @xmath17 and @xmath16 . the unit darboux vector @xmath37 is defined to be the unit vector on the darboux axis such that its sense , together with the sense of rotation of the frenet frame , form a right - handed screw . from the orthogonality of @xmath37 and @xmath38 , we obtain @xmath39 as @xmath10 varies , the endpoint of the darboux vector describes yet another curve on @xmath8 , the darboux indicatrix @xmath40 of the curve . next , we define the bridge map and bridge graph of a knot conformation and relate it to two knot invariants , the bridge index and superbridge index of a knot . we also define the inflection map of a knot conformation and its corresponding graph . given a smooth knot conformation @xmath4 in @xmath5 , the _ bridge map _ of the knot conformation @xmath4 is the map defined on the unit sphere @xmath8 by @xmath41 for every vector @xmath6 . [ def : bridgemap ] the bridge map is closely related to two knot invariants , the bridge index and superbridge index of a knot . the _ bridge index _ of a knot type @xmath9 $ ] is defined by @xmath42 = \min_{k \in [ k ] } \min_{\mathbf{v } \in s^2 } \#\{\mbox{local maxima in the projection of $ k$ onto the axis defined by $ \mathbf{v}$}\}.\ ] ] it is easy to see that the bridge index of a knot is related to the bridge map by @xmath42 = \frac{1}{2 } \min_{k \in [ k ] } \min_{\mathbf{v } \in s^2 } \mathfrak{b}_\mathbf{v}(k).\ ] ] the bridge index has been extensively studied . schubert @xcite proved that bridge index is additive minus one under composition : @xmath43 = b[k_1 ] + b[k_2 ] - 1 $ ] . knots with bridge index 2 are precisely the rational knots . kuiper @xcite introduced a related knot invariant , the superbridge index of a knot . the _ superbridge index _ of a knot type @xmath9 $ ] is defined by @xmath44 = \min_{k \in [ k ] } \max_{\mathbf{v } \in s^2 } \#\{\mbox{local maxima in the projection of $ k$ onto the axis defined by $ \mathbf{v}$}\}.\ ] ] similarly , the superbridge index of a knot type is related to the bridge map by @xmath42 = \frac{1}{2 } \min_{k \in [ k ] } \max_{\mathbf{v } \in s^2 } \mathfrak{b}_\mathbf{v}(k).\ ] ] the superbridge index is a useful invariant in obtaining lower bounds on the stick number of knots . jin @xcite used the superbridge index to show that the stick number of the @xmath0-torus knot @xmath1 is equal to @xmath2 for @xmath3 . given a smooth knot conformation @xmath4 in @xmath5 , the _ inflection map _ of the knot conformation @xmath4 is the map defined on the unit sphere @xmath8 by @xmath45 for every vector @xmath6 . [ def : inflectionmap ] observe that a point @xmath46 on the knot conformation @xmath4 projects to a stationary point along the axis defined by @xmath7 if and only if the tangent vector to @xmath4 at @xmath46 is orthogonal to @xmath7 . hence the stationary points in the projection of @xmath4 onto the axis defined by @xmath7 are in bijective correspondence with the points of intersection of the tantrix of @xmath4 with the great circle orthogonal to @xmath7 . similarly , since @xmath46 projects to an inflection point or a cusp in the plane orthogonal to @xmath7 if and only if the osculating plane at @xmath46 is projected onto a line , which occurs if and only if the binormal vector at @xmath46 is orthogonal to @xmath7 , it follows that the inflection points and cusps in the projection of @xmath4 onto the plane orthogonal to @xmath7 are in bijective correspondence with the points of intersection of the binotrix of @xmath4 with the great circle orthogonal to @xmath7 . we shall make extensive use of this observation in the proof of theorem [ thm : bridgeinflection ] . [ rem : intersectionswithgreatcircles ] the length of a curve @xmath47 on the unit sphere is equal to @xmath48 times the average over all great circles @xmath49 of the number of times that @xmath47 intersects @xmath49 . [ lem : intersectionswithgreatcircles ] we interpret the word `` average '' in the above lemma as follows . corresponding to each great circle @xmath49 , there is a unique pair of unit vectors that are perpendicular to the plane containing @xmath49 . the usual lebesgue measure on the unit sphere applied to them then induces a measure on the set of all great circles . ( the average is then interpreted as a lebesgue integral . ) lemma [ lem : intersectionswithgreatcircles ] , together with remark [ rem : intersectionswithgreatcircles ] and the fact that the bridge map is not constant for any conformation @xmath4 of a non - trivial knot , implies that @xmath50,\ ] ] where @xmath51 denotes the average of the bridge map @xmath52 over all @xmath6 under the usual lebesgue measure . since the bridge index of any non - trivial knot is at least @xmath53 , this , as noted by milnor in @xcite , yields the well - known theorem of fry @xcite and milnor @xcite that the total ( absolute ) curvature of a conformation of any non - trivial knot is greater than @xmath54 . the following definition will be used extensively in the statement of our main results . the _ graph _ of a map on the unit sphere @xmath8 is the set of points @xmath55 such that there does not exist an open neighborhood @xmath56 of @xmath46 on @xmath8 ( with the standard euclidean topology ) such that the value of the map is constant for all points @xmath57 at which the map is defined . [ def : graph ] this definition can be interpreted as follows . label each point @xmath6 with the value of the map at that point . this divides the sphere into several regions with the same label , and the interiors of those regions are separated by the graph of the map . the graphs of the bridge map and the inflection map , which we call the _ bridge graph _ and _ inflection graph _ , as well as two other graphs that we shall define later on , the tantrix - bridge graph and tantrix - inflection graph , form the subject of the remainder of this paper . in this section , we review the concepts of the dual curve and derivative curve of a co - oriented curve on a sphere as introduced by arnold @xcite and use them to determine the bridge graph and inflection graph of smooth knots . we also use the concepts of dual and derivative curves to study the relationships between another dual pair of curves , the notrix and darboux indicatrix . for our purposes , a co - orientation of a vector in a plane is a choice of one of the two unit vectors perpendicular to it in the plane , and a co - orientation of a spherical curve is a continuous choice of co - orientations of tangent vectors in their tangent planes . a wave front is a curve obtained from a smooth co - oriented curve by moving each point of the curve by a constant distance along the co - orienting normal . we refer the reader to @xcite for a more general definition of co - orientations and wave fronts in the context of contact geometry . the _ dual curve _ @xmath58 to a given co - oriented curve @xmath59 on the sphere is the curve obtained from the original curve by moving a distance of @xmath60 along the normals on the side determined by the co - orientation . the dual curve @xmath58 inherits its co - orientation from that of @xmath59 . [ def : dual ] this definition applies not only to smoothly immersed curves , but also to wave fronts having semi - cubical cusps . the cusps on the original curve correspond to points of inflection on the dual curve , while points of inflection on the original curve correspond to cusps on the dual curve . the _ derivative curve _ @xmath61 of a co - oriented curve @xmath59 on the oriented standard sphere @xmath8 is the curve obtained by moving each point a distance @xmath60 along the great circle tangent to the original curve at that point . the direction of motion along the tangent is chosen so that the orientation of the sphere , given by the direction of the tangent and the direction of the co - orienting normal , is positive . [ def : derivative ] for example , if the tantrix is co - oriented such that the co - orienting normal is obtained from the derivative of the tangent vector via a counterclockwise rotation of @xmath60 on the surface of the sphere when the curvature @xmath21 is positive and via a clockwise rotation of @xmath60 when @xmath21 is negative , then it is easy to see that the derivative curve of the tantrix is the notrix and the dual curve of the tantrix is the binotrix . note that the co - orientation of the binotrix is thus induced by the co - orientation on the tantrix . in what follows , we shall always assume that the tantrix is oriented as such . we refer the reader to arnold s @xcite paper for a more thorough discussion of the geometry of spherical curves . a cusp of the tantrix and a spherical inflection point of the binotrix corresponds to a @xmath21-inflection point of the knot conformation @xmath4 , while a spherical inflection point of the tantrix and a cusp of the binotrix corresponds to a @xmath22-inflection point of @xmath4 . ( see , e.g. , fenchel @xcite for more details . ) our assumption in sec . [ sec : preliminaries_smooth ] that the vectors in the collection of vectors @xmath18 at the set of points where @xmath16 vanishes are pairwise distinct and the vectors in the collection of vectors @xmath19 at the set of points where @xmath17 vanishes are pairwise distinct simply says that no two cusps of the tantrix coincide and a cusp of the tantrix never coincides with a cusp of the anti - tantrix , and similarly for the binotrix . the dual and derivative curves of a co - oriented spherical curve satisfy the following properties . the dual curve is formed from the centers of the great circles tangent to the original curve . the second dual of a curve is antipodal to the original curve : @xmath62 . [ lem : seconddual ] the derivative of a curve coincides with the derivative of any curve equidistant from it and is a smoothly immersed curve on @xmath8 even if the original curve has generic singularities . [ lem : smoothderivative ] since the notrix and the darboux indicatrix have the same derivative curves and the normal and darboux vectors are mutually orthogonal , it follows from lemma [ lem : smoothderivative ] that the notrix and darboux indicatrix are dual curves . in what follows , we shall always orient the notrix such that the co - orienting normal is obtained from the derivative of the normal vector via a counterclockwise rotation of @xmath60 on the surface of the sphere ; then the darboux indicatrix is the dual curve of the notrix . since the notrix is the derivative curve of the tantrix , lemma [ lem : smoothderivative ] tells us that it has no cusps ( this also follows from the fact that @xmath63 because @xmath21 and @xmath22 do not vanish simultaneously ) , and hence the darboux indicatrix has no spherical inflection points . inflection points of the notrix and cusps of the darboux indicatrix correspond to points where the geodesic curvature @xmath64 of the tantrix is stationary and the knot behaves locally like a helix . ( see fenchel @xcite or uribe - vargas @xcite for further details . ) in what follows , we shall also assume that no two cusps of the darboux indicatrix coincide and a cusp of the darboux indicatrix never coincides with a cusp of the anti - darboux indicatrix . since the notrix is the curve of normalized derivatives of the tantrix and the binotrix , we may view it as the `` tantrix of the tantrix '' or the `` tantrix of the binotrix '' up to a sign . similarly , since the darboux indicatrix is the dual curve to the notrix , we may view it as the `` binotrix of the tantrix '' or the `` binotrix of the binotrix . '' in what follows , we shall focus our interpretation of the notrix and darboux indicatrix as the tantrix and binotrix respectively of the tantrix ; similar results hold if we interpret them as the tantrix and binotrix respectively of the binotrix . this allows us to define a `` bridge map '' and `` inflection map '' for the tantrix analogous to definitions [ def : bridgemap ] and [ def : inflectionmap ] . given a smooth knot conformation @xmath4 in @xmath5 , the _ tantrix - bridge map _ of the knot conformation @xmath4 is the map defined on the unit sphere @xmath8 by @xmath65 for every vector @xmath6 . [ def : tanbridgemap ] given a smooth knot conformation @xmath4 in @xmath5 , the _ tantrix - inflection map _ of the knot conformation @xmath4 is the map defined on the unit sphere @xmath8 by @xmath66 for every vector @xmath6 . [ def : taninflectionmap ] in analogy to remark [ rem : intersectionswithgreatcircles ] , we can reinterpret these maps in terms of intersections of the corresponding great circles with the notrix and darboux indicatrix respectively . moreover , we can define the _ tantrix - bridge graph _ and _ tantrix - inflection graph _ as in definition [ def : graph ] . however , some care must be taken at cusps of the tantrix : when do we count a cusp of the tantrix as contributing to a stationary point in height to the tantrix - bridge map ? returning to our interpretation of the tantrix - bridge map as the number of intersections of the notrix with a great circle , we see we want the derivative of the tantrix to lie in the plane of that great circle , and hence a cusp of the tantrix should be counted as a stationary point in height only if the derivative at that point is perpendicular to the height axis . similar arguments show that the only cusps that should contribute to an inflection point or cusp in the tantrix - inflection map are those that lie on the great circle in the projection . in the following two theorems , we determine the bridge graph , the inflection graph , the tantrix - bridge graph and the tantrix - inflection graph for smooth knots . the bridge graph of a smooth knot is the union of the binotrix , the anti - binotrix , and the great circles tangent to the binotrix and anti - binotrix at points corresponding to @xmath21-inflection points of the knot . the inflection graph of a smooth knot is the union of the tantrix , the anti - tantrix , and the great circles tangent to the tantrix and anti - tantrix at points corresponding to @xmath22-inflection points of the knot . [ thm : bridgeinflection ] we prove the first statement . there are three types of intersections of the tantrix with a great circle , in the vicinity of which the number of intersections of the tantrix with a great circle may change : ( i ) the great circle is tangential to the tantrix neither at a spherical inflection point of the tantrix nor at a cusp ( fig [ fig : tantrix_tangential ] ) , ( ii ) the great circle is tangential to the tantrix at a spherical inflection point of the tantrix ( fig [ fig : tantrix_sphericalinflection ] ) , and ( iii ) the great circle intersects the tantrix at a semi - cubical cusp of the tantrix ( fig [ fig : tantrix_cusp ] ) . intersections of the first type contribute a portion of the bridge graph that is formed from the centers of the great circles tangent to the tantrix at these points , which by lemma [ lem : seconddual ] is the dual curve to the tantrix and its antipodal curve , the binotrix and anti - binotrix , excluding cusp points on both curves . next , a change in the number of intersections of the tantrix with a great circle in the vicinity of a spherical inflection point of the tantrix corresponds to moving the center of a great circle from outside a cusp of the binotrix to inside ( and similarly for the anti - binotrix ) , or vice versa , as shown in fig . [ fig : binotrix_cusp ] , as the limiting tangent at a cusp of the binotrix is obtained from the tangent to the tantrix at a spherical inflection point by moving a distance of @xmath60 along the normal on the side determined by the co - orientation . finally , the number of intersections of the tantrix with a great circle changes as we move the great circle over a cusp in almost every direction ( by our assumption that no two cusps of the tantrix coincide and a cusp of the tantrix never coincides with a cusp of the anti - tantrix ) . hence the portion of the bridge graph that such intersections contribute to are great circles whose centers are at those cusps , and which are therefore great circles tangent to the binotrix and anti - binotrix at points corresponding to @xmath21-inflection points of the knot . the proof of the second statement follows analogously upon noting that cusps in the binotrix correspond to @xmath22-inflection points of the knot . in fact , a little additional work enables us to identify the direction in which the value of the bridge map changes whenever we cross the bridge graph . in general , the value of the bridge map increases when we move across the bridge graph from a region that locally `` bulges out '' to an adjacent region that locally `` bulges in . '' the change in value of the bridge map as we move across a great circle defining the bridge graph can be determined from the shape of the binotrix near a @xmath21-inflection point . similar observations apply to the inflection graph . the details are relatively straightforward and will be omitted . since the notrix has no cusps while the darboux indicatrix has cusps precisely when the geodesic curvature @xmath64 of the tantrix is stationary , a similar argument yields the following theorem . the tantrix - bridge graph is the union of the darboux indicatrix and the anti - darboux indicatrix . the tantrix - inflection graph is the union of the notrix , the anti - notrix , and the great circles tangent to the notrix and the anti - notrix at the points corresponding to where the geodesic curvature @xmath64 of the tantrix is stationary . [ thm : tantrix - bridgeinflection ] in this section , we review mcrae s @xcite definition of the dual curve of a spherical polygon @xmath67 . we also review banchoff s @xcite construction of the tantrix , notrix , and binotrix of an oriented space polygon @xmath68 in @xmath5 and propose a definition for the darboux indicatrix of the oriented space polygon @xmath68 . this section provides the necessary background for sec . [ sec : polygonal_knots ] . the fundamental idea behind duality for spherical polygons is that the dual of a point on the unit sphere @xmath8 is the co - oriented great circle at a distance of @xmath60 away from that point , with the co - orienting normal pointing away from the point , while the dual of a co - oriented great circle is the point @xmath60 away in the direction of the co - orienting normal . thus , composing the dual with itself gives the antipodal map on @xmath8 . let @xmath67 be a spherical polygon determined by a cycle of @xmath69 vertices @xmath70 , and co - oriented geodesic segments @xmath71 , where @xmath72 is the shorter of the great circle arcs joining @xmath73 to @xmath74 ( we consider vertex indices modulo the number of vertices , so that , in particular , @xmath75 ) . the _ dual curve @xmath76 of the spherical polygon @xmath67 _ is defined to be the co - oriented polygon determined by the cycle of vertices @xmath77 and sides @xmath78 , where @xmath79 and @xmath80 is the geodesic segment joining @xmath81 to @xmath82 whose length is equal to the exterior angle at @xmath73 and with co - orientation induced from @xmath83 . observe that the relation @xmath84 continues to hold in this setting . for a co - oriented spherical polygon @xmath67 , we can define a second spherical polygon in terms of @xmath67 and its dual @xmath76 , as follows . let @xmath67 be a co - oriented spherical polygon . we define the _ direct sum _ of @xmath67 and its dual , denoted @xmath85 , to be the co - oriented spherical polygon constructed in the following manner : we regard @xmath86 as the north pole , so that the co - oriented segment @xmath87 lies on its equator . this segment is rotated counterclockwise around @xmath86 through an angle of @xmath60 . next , we regard the endpoint @xmath88 of @xmath87 as the south pole and rotate the segment @xmath89 counterclockwise through an angle of @xmath60 . we repeat this process , alternating between angles of the form @xmath90 and @xmath91 , until we arrive back at the vertex @xmath86 . mcrae proves the following proposition about the direct sum @xmath85 . consecutive sides of @xmath85 meet at alternating right angles . more will be said about the direct sum @xmath85 later in this section . next , we review banchoff s @xcite construction of the tantrix , notrix , and binotrix of an oriented space polygon @xmath68 in @xmath5 and define the darboux indicatrix of @xmath68 . we consider @xmath68 to be determined by a cycle of vertices @xmath92 , where vertex indices are taken modulo @xmath69 , that is , @xmath68 is an oriented closed curve @xmath93 defined on some closed interval @xmath94 $ ] , @xmath95 , with @xmath96 and @xmath68 linear in each subintervcal @xmath97 $ ] . we say that @xmath68 is in _ general position _ if no four consecutive vertices of @xmath68 lie in a plane . in what follows , we consider only space polygons in general position . moreover , we shall assume that no two ( undirected ) edges of @xmath68 and no two osculating planes of @xmath68 are parallel . milnor @xcite defined the curvature of the space polygon @xmath68 at the vertex @xmath98 to be the angle @xmath99 , @xmath100 , between the vectors @xmath101 and @xmath102 , and the total curvature of the polygon @xmath68 to be the angle sum @xmath103 . in particular , note that the curvature of a space polygon is positive at every vertex . we further define the torsion @xmath104 of the space polygon @xmath68 at the edge @xmath105 to be the directed angle @xmath106 , @xmath107 , between the projections of the directed edges @xmath108 and @xmath109 when we project down the directed edge @xmath105 , and the total absolute torsion of @xmath68 to be the sum @xmath110 . the tantrix , notrix , binotrix , and darboux indicatrix of @xmath68 are defined to be co - oriented polygons on the unit sphere with vertices as follows . the vertices @xmath111 and @xmath112 of the tantrix and binotrix respectively are defined by @xmath113 @xmath114 so that both of these indicatrices have the same number of vertices as @xmath68 . ( it will often expedite calculations to use an unnormalized version of the vertices of the binotrix , which we denote by @xmath115 . ) this definition can be interpreted as follows . each directed edge of @xmath68 gives rise to a vertex of the tantrix , and each vertex of @xmath68 , together with the preceding and following directed edges , gives rise to a vertex of the binotrix . at the vertex @xmath98 , the tangent line sweeps counterclockwise through an angle of @xmath99 in the oriented osculating plane , which corresponds to a great circle arc of length @xmath99 connecting two adjacent vertices of the tantrix , and along the edge @xmath105 , the binormal vector sweeps through an angle of @xmath116 , which corresponds to a great circle arc of length @xmath116 connecting two adjacent vertices of the binotrix . hence these definitions preserve the property that the total curvature of the space polygon @xmath68 is equal to the length of its tantrix and its total absolute torsion is equal to the length of the binotrix . note also that the relation @xmath117 continues to hold under these definitions ( recall that we defined the tantrix to be co - oriented such that the co - orienting normal is oriented towards the left when viewed from the exterior of the sphere ) . observe that the segments of the direct sum @xmath85 obtained by rotating segments @xmath118 of @xmath67 behave like derivative curves of the arcs of @xmath67 in the sense of definition [ def : derivative ] , while those obtained by rotating segments @xmath119 of @xmath76 behave like tantrix arcs corresponding to the vertices of a stick knot . hence it makes sense to think of @xmath85 as the derivative curve of the spherical polygon @xmath67 , and we write @xmath120 . we say a few words about how the torsion of a space polygon can be interpreted in terms of the tantrix and the binotrix . since the torsion @xmath104 is defined along the edge @xmath105 of the space polygon @xmath68 , we can also think of it as a property of the vertex @xmath111 of the tantrix , or of the edge @xmath121 of the binotrix . if we orient the unit sphere such that the directed tantrix arc @xmath122 lies on the equator and turns counterclockwise when viewed from the north pole , it is easy to see that @xmath104 is positive if the tantrix arc @xmath123 lies above the equator , and negative if @xmath123 lies below the equator . moreover , by regarding the torsion @xmath104 as a property of the edge @xmath121 of the binotrix , we can speak about a change of sign in torsion at a vertex of the binotrix . further relations between the sign of the torsion , the tantrix , and the binotrix will be elucidated later in this section . the notrix @xmath31 of @xmath68 should take into account that associated to each edge of @xmath68 there is one tangent direction but two binormal directions , and associated to each vertex of @xmath68 there is one binormal direction but two tangent directions . consequently , we define the vertices @xmath124 of @xmath31 by @xmath125 note that the notrix has twice as many vertices as @xmath68 . it is easy to see from the definition of the notrix that @xmath126 , thus yielding an analogue of the result @xmath127 for space polygons . as in sec . [ sec : smooth_knots ] , we co - orient the notrix such that the co - orienting normal is oriented towards the left when viewed from the exterior of the sphere . finally , the darboux indicatrix @xmath40 of @xmath68 should take into account that the tangent direction is constant along each edge of @xmath68 and the binormal direction is constant at each vertex of @xmath68 . to this end , we define the vertices @xmath128 of @xmath40 by @xmath129 like the notrix , the darboux indicatrix also has twice as many vertices as @xmath68 . it is easy to see that under this definition , the vertices of the darboux indicatrix are the axes of rotation of the frenet frame . the sign of the tangent vector at the vertices of the darboux indicatrix is chosen so that its direction , together with the direction of rotation of the frenet frame , form a right - handed screw , as we shall show in the course of proving theorem [ thm : inflection_polygonal ] . in this section , we discuss stick knot analogues of the maps defined by the various indicatrices and show that the graphs of these maps satisfy similar properties as in sec . [ sec : smooth_knots ] . we begin with a discussion of how intersections of the various indicatrices of a stick knot with a great circle can be interpreted in terms of properties of the knot conformation . throughout , we shall use the convention that an intersection of an indicatrix with a great circle over an interval is counted as a single intersection . first , the tantrix arc @xmath123 of a stick knot intersects the equatorial @xmath130-plane , with neither of its vertices lying on the equator , if and only if the vertex @xmath98 of the knot is a local extremum along the @xmath131-axis and neither of the edges @xmath105 and @xmath109 have stationary height along the @xmath131-axis , since a vertex of the tantrix above the equator corresponds to an edge of the knot directed upwards , while a vertex of the tantrix below the equator corresponds to an edge of the knot directed downwards . next , a vertex @xmath111 of the tantrix lies on the equator if and only if the edge @xmath105 of the knot has stationary height along the @xmath131-axis . it follows that the number of intersections of the tantrix of a stick knot with a great circle counts the number of stationary points in the projection onto the axis perpendicular to the plane of the great circle , where a point that is stationary over an interval counts as a single stationary point . this allows us to define a direct analogue of the bridge map of a smooth knot conformation , as follows . the _ bridge map _ of a stick conformation @xmath4 is a map defined on the unit sphere @xmath8 by @xmath41 for each @xmath6 , where we use the convention that a point that is stationary over an interval counts as a single stationary point . [ def : stick_bridge ] our argument above immediately yields the following lemma . [ lem : stick_tantrix ] the bridge map is related to intersections of the tantrix with great circles by the following formula : @xmath132 we require the following definition in studying intersections of the binotrix with great circles . given a projection of a stick conformation , we define an _ inflection stick _ to be ( i ) an edge of the stick conformation such that the two edges adjacent to it are projected to opposite sides of the line obtained by extending the projection of the edge infinitely ( fig . [ fig : inflection_stick1 ] ) , ( ii ) a pair of adjacent edges projected to collinear sticks ( fig [ fig : inflection_stick2 ] ) , or ( iii ) an edge projected down to a point ( fig [ fig : inflection_stick3 ] ) . [ def : inflectionstick ] inflection sticks of the first two types may be viewed as the stick knot equivalent of inflection points , while inflection sticks of the third type may be viewed as the stick knot equivalent of cusps . the _ inflection map _ of a stick conformation @xmath4 is the map defined on the unit sphere @xmath8 by @xmath133 for each @xmath6 . [ def : stick_inflection ] [ lem : stick_binotrix ] the inflection map is related to intersections of the binotrix with great circles by the following formula : @xmath134 orient the sphere such that the plane orthogonal to @xmath7 is the equatorial @xmath130-plane and the projection of the edge @xmath105 of the knot conformation @xmath4 is in the direction of the positive @xmath135-axis . we shall consider the effect of the position of the projections of the edges @xmath108 and @xmath109 on the binotrix of @xmath4 . let the tangent vectors @xmath136 , @xmath111 and @xmath137 be given by @xmath138 where @xmath139 . then @xmath140 the binotrix arc @xmath121 intersects the equatorial plane with neither of its vertices lying on the equator if and only if @xmath141 and @xmath142 and @xmath143 have the same sign , so that we have the configuration in fig . [ fig : inflection_stick1 ] . next , the vertex @xmath112 of the binotrix lies on the equator but both of its adjacent vertices do not if and only if @xmath144 , @xmath145 , @xmath146 , and @xmath147 ( the final inequality arises from the fact that otherwise , @xmath148 would also lie on the equator ) . hence the edges @xmath105 and @xmath109 are projected down to parallel sticks , as in fig . [ fig : inflection_stick2 ] . finally , @xmath149 and @xmath112 both lie on the equator if and only if @xmath150 ( we can not have @xmath151 because of our assumption that the knot conformation is in general position ) , and thus the edge @xmath105 is projected to a point , as in fig . [ fig : inflection_stick3 ] . this proves the lemma . the _ tantrix - bridge map _ of a stick conformation @xmath4 is a map defined on the unit sphere @xmath8 by @xmath152 for each @xmath6 , where we use the convention that a point that is stationary over an interval counts as a single stationary point . [ def : stick_tantrix - bridge ] the tantrix - bridge map is related to intersections of the notrix with great circles by the following formula : @xmath153 first , the tantrix arc @xmath123 has an extrema in height along the axis defined by @xmath7 at a point @xmath46 in its interior if and only if the notrix arc @xmath154 intersects the great circle orthogonal to @xmath7 at the point @xmath155 on @xmath154 obtained by moving @xmath46 by a distance of @xmath60 along the great circle that @xmath123 lies on . next , the tantrix arc @xmath122 is going up ( respectively down ) locally in the vicinity of @xmath111 if and only if the notrix vertex @xmath156 lies above ( respectively below ) the equator , and the tantrix arc @xmath123 is going down ( respectively up ) locally in the vicinity of @xmath111 if and only if the notrix vertex @xmath157 lies below ( respectively above ) the equator . it follows that the tantrix vertex @xmath111 is an extrema in height if and only if @xmath158 intersects the equator . since this accounts for all types of extrema in height and all types of intersections of the notrix with the equator , the lemma follows . finally , we wish to interpret intersections of the darboux indicatrix with a great circle in terms of projections of the tantrix to the plane defined by the great circle . for convenience , we shall consider only projections where no vertex of the darboux indicatrix lies on the great circle ; we call such projections _ regular projections _ and the corresponding intersections with the darboux indicatrix _ regular intersections_. the reader can easily generalize our formulae to include non - regular projections , if desired . to this end , we consider the case where the great circle lies in the equatorial plane and study the effect of the position of vertices of the binotrix and tantrix on the projection of the the tantrix . this is tabulated in table [ tab : stick_darboux ] . note that the direction of rotation never changes along the projection of an arc of the tantrix because the arcs of the tantrix are great circle arcs . .relation between the location of vertices of the binotrix and tantrix and the projection of the tantrix . [ cols="^,^,^,^ " , ] this allows us to formulate the following definition . let @xmath159 , @xmath160 and @xmath161 be the number of pairs of adjacent arcs of the tantrix whose projection onto the plane orthogonal to @xmath7 appear as in figs . [ fig : tantrix_projection_a ] , [ fig : tantrix_projection_b ] , and [ fig : tantrix_projection_c ] respectively . the _ tantrix - inflection map _ of a stick conformation @xmath4 is the map defined almost everywhere on the unit sphere @xmath8 by @xmath162 for each @xmath6 corresponding to a regular projection . [ def : stick_tantrix_inflection ] the reason for the coefficient ` 2 ' in the above definition is explained in the following lemma . [ lem : stick_darboux ] the tantrix - inflection map is related to intersections of the darboux indicatrix with great circles by the following formula : @xmath163 as illustrated in table [ tab : tantrix_projection ] , all regular intersections of the darboux indicatrix with a great circle correspond to one of these diagrams . each of the diagrams in figs . [ fig : tantrix_projection_a ] and [ fig : tantrix_projection_b ] are counted by only one type of intersection of the darboux indicatrix with a great circle ( e.g. , the upper diagram in fig . [ fig : tantrix_projection_a ] is counted only by an arc of the darboux indicatrix of the form @xmath164 that goes from above the equator to below the equator , while the lower diagram in fig . [ fig : tantrix_projection_b ] is counted only by an arc of the form @xmath165 that goes from below the equator to above the equator ) . on the other hand , both diagrams in fig . [ fig : tantrix_projection_c ] are counted by two types of intersections of the darboux indicatrix with a great circle ( e.g. the upper diagram is counted by an arc of the form @xmath165 that goes from below the equator to above the equator , as well as an arc of the form @xmath164 that goes from above the equator to below the equator ) . hence the number of intersections of the darboux indicatrix with the equator ( and hence , with any great circle ) is indeed given by the formula in definition [ def : stick_tantrix_inflection ] . having defined stick knot analogues of the maps defined by the various indicatrices , we now show that their graphs satisfy similar properties as in sec . [ sec : smooth_knots ] . as in sec . [ sec : smooth_knots ] , we define the graph of a map on @xmath8 to be the set of points @xmath55 such that there does not exist an open neighborhood @xmath56 of @xmath46 on @xmath8 ( with the standard euclidean topology ) such that the value of the map is constant for all points @xmath57 at which the map is defined . the bridge graph of a stick knot is the union of the binotrix and the anti - binotrix . [ thm : bridge_polygonal ] the inflection graph of a stick knot is the spherical polygon obtained by connecting vertices of the tantrix and the anti - tantrix according to the following rule : we connect @xmath111 to @xmath137 and @xmath166 to @xmath167 if the torsions @xmath104 and @xmath168 associated to the edges @xmath105 and @xmath109 have the same sign , and we connect @xmath111 to @xmath167 and @xmath166 to @xmath137 if @xmath104 and @xmath168 have opposite signs . [ thm : inflection_polygonal ] although the proofs of theorems [ thm : bridge_polygonal ] and [ thm : inflection_polygonal ] are more involved than the proof of theorem [ thm : bridgeinflection ] , they also provide more insight into the relationship between the tantrix , the binotrix and the sign of the torsion . first , consider an intersection of the tantrix with a great circle in the vicinity of a vertex @xmath111 of the tantrix , as shown in figs . [ fig : tantrix_vertex ] and [ fig : tantrix_vertex2 ] . recall that we assumed that no two ( undirected ) edges of the knot are parallel ; this implies that no two vertices of the tantrix coincide and no vertex of the tantrix coincides with a vertex of the anti - tantrix . by the definition of the binotrix , the vertices @xmath149 and @xmath112 lie on the great circle orthogonal to @xmath111 and always lie on the left side of the directed tantrix arcs @xmath122 and @xmath123 respectively as seen from from the exterior of the sphere . the number of intersections of the tantrix with a great circle changes exactly when the center of the great circle moves across the open arc @xmath121 and its antipodal arc , by our assumption on the vertices of the tantrix , thus contributing to the bridge graph all the open arcs of the binotrix . next , consider the situation when the great circle moves across an arc of the tantrix while parallel to it , as shown in figs . [ fig : tantrix_edge_up ] and [ fig : tantrix_edge_down ] . orient the sphere such that the tantrix arc @xmath123 lies on the equator , rotates counterclockwise as viewed from the north pole , and intersects the negative @xmath135-axis . since @xmath149 is perpendicular to @xmath111 , it lies on the great circle @xmath169 dual to @xmath111 , and since @xmath170 is perpendicular to @xmath137 , it lies on the great circle @xmath171 dual to @xmath137 , as illustrated in fig . [ fig : binotrix_top ] , which shows a projection onto the equatorial plane . depending on whether the tantrix arcs @xmath122 and @xmath172 lie above or below the equator , the vertices @xmath149 and @xmath170 of the binotrix lie on different sides of @xmath112 , as shown in fig . [ fig : binotrix_top ] ( this can easily be seen by computing cross products and considering their @xmath135-coordinates ) . in particular , we see that the binotrix arcs @xmath121 and @xmath173 lie on the same side of the @xmath174-plane if and only if @xmath122 and @xmath172 lie on the same side of the equator ( see figs . [ fig : binotrix_top_ex ] and [ fig : binotrix_top_ex2 ] for examples ) . hence the number of intersections of the tantrix with a great circle changes only when we move into a different region defined by the binotrix . this completes the proof . and @xmath170 depending on whether the tantrix arcs @xmath122 and @xmath172 lie above or below the equator , as seen from a projection onto the equatorial plane . note that the great circles @xmath169 and @xmath171 are projected onto line segments and that @xmath149 and @xmath170 can lie on either side of the equator.,scaledwidth=30.0% ] theorem [ thm : bridge_polygonal ] can be re - interpreted in the following manner . let @xmath67 be a directed spherical polygon with sides @xmath71 , such that no two of its vertices coincide and none of its vertices coincides with a vertex of its antipodal spherical polygon . for each geodesic segment @xmath118 , let @xmath175 be the point on @xmath8 obtained by moving a distance of @xmath60 normal to the geodesic segment and towards its left as viewed from the exterior of the sphere . let @xmath176 be the spherical polygon with vertices @xmath177 . then the graph of the map on @xmath8 that assigns to each vector @xmath6 the number of intersections of @xmath67 with the great circle orthogonal to @xmath7 is given by the union of @xmath176 and its antipodal spherical polygon . in particular , if @xmath67 is a directed co - oriented spherical polygon such that the co - orienting normal points towards the left when viewed from the exterior of the sphere , then the vertices of the dual spherical polygon are precisely the @xmath175 s and hence the graph of the map on @xmath8 will simply be the union of @xmath76 and its antipodal spherical polygon . [ rem : reinterpretation ] recall that we assumed that no two osculating planes of @xmath68 are parallel , so that no two vertices of the binotrix coincide and no vertex of the binotrix coincides with a vertex of the anti - binotrix . by remark [ rem : reinterpretation ] , it suffices to show that the vertex obtained by moving a distance of @xmath60 normal to the binotrix arcs and towards their left as viewed from the exterior of the sphere is @xmath111 when the torsion @xmath104 is positive , and @xmath166 when the torsion @xmath104 is negative . orient the sphere such that the tantrix arc @xmath122 lies on the equatorial @xmath130-plane and rotates counterclockwise as viewed from the north pole , and @xmath111 lies on the negative @xmath135-axis . since @xmath112 is orthogonal to @xmath111 , it lies on the unit circle in the @xmath178-plane . if @xmath104 is positive , then @xmath137 lies above the equator and thus the @xmath179-coordinate of @xmath112 is negative , as illustrated in fig . [ fig : sideofbinotrix_up ] . hence @xmath111 lies on the left side of @xmath121 . on the other hand , if @xmath104 is negative , then @xmath136 lies below the equator and thus the @xmath179-coordinate of @xmath112 is positive , as shown in fig . [ fig : sideofbinotrix_down ] . hence @xmath111 lies on the right side of @xmath121 , that is , @xmath166 lies on the left side of @xmath121 , as required . alternatively , this can also be seen from the vector triple product formula . we have @xmath180 since the sign of @xmath181 is precisely the sign of @xmath104 , we see that the vertex obtained by moving a distance of @xmath60 normal to @xmath121 and towards the left of @xmath121 is @xmath111 when @xmath104 is positive , and @xmath166 when @xmath104 is negative , as required . theorem [ thm : inflection_polygonal ] provides an alternative interpretation of the great circles tangent to the tantrix and anti - tantrix at @xmath22-inflection points of a smooth knot in theorem [ thm : bridgeinflection ] . as we remarked earlier , we can talk about a change in the sign of the torsion at a vertex @xmath112 of the binotrix of a stick knot if @xmath104 and @xmath168 have different signs . the arcs of the inflection graph corresponding to such a vertex @xmath112 join @xmath111 to @xmath182 and @xmath166 to @xmath137 . similarly , @xmath22-inflection points of a smooth knot are points where the sign of the torsion changes , and the great circles tangent to the tantrix and anti - tantrix can be viewed as the union of two antipodal great circle arcs joining the tantrix to the antitantrix . in the case of the bridge graph of a stick knot , all the arcs connect two adjacent vertices of the binotrix and two adjacent vertices of the anti - binotrix . this reflects the fact that the sign of curvature , as defined in sec . [ sec : preliminaries_smooth ] , unlike that of torsion , has no geometric meaning , and serves merely to allow us to have isolated points of zero curvature on a smooth knot . [ rem : greatcirclestangent ] the proof of theorem [ thm : inflection_polygonal ] easily yields the following proposition . the darboux indicatrix of a stick knot is the dual of its notrix . [ prop : notrix_darboux_dual ] we only need to verify that the points dual to the arcs of the notrix are the vertices of the darboux indicatrix . this follows immediately from the equations @xmath183 and @xmath184 and the fact that @xmath181 has the same sign as the torsion @xmath104 . remark [ rem : reinterpretation ] immediately yields the following corollary to proposition [ prop : notrix_darboux_dual ] . the tantrix - bridge graph of a stick knot is the union of the darboux indicatrix and the anti - darboux indicatrix . [ cor : tantrixbridge_polygonal ] finally , the following theorem is the stick equivalent of the second half of theorem [ thm : tantrix - bridgeinflection ] . the tantrix - inflection graph of a stick knot is the spherical polygon obtained by connecting vertices of the notrix and the anti - notrix according to the following rule : we always connect @xmath156 to @xmath185 and @xmath186 to @xmath157 , we connect @xmath157 to @xmath187 and @xmath185 to @xmath188 if @xmath104 and @xmath168 have the same sign , and we connect @xmath157 to @xmath188 and @xmath185 to @xmath187 if @xmath104 and @xmath168 have different signs . [ thm : tantrixinflection_polygonal ] that is , if two adjacent edges of the darboux indicatrix share a vertex of the tantrix , then we always join the notrix vertex corresponding to one edge to the anti - notrix vertex corresponding to the other edge , and if two adjacent edges of the darboux indicatrix share a vertex of the binotrix , then we join the notrix vertex corresponding to one edge to either the notrix vertex or the anti - notrix vertex corresponding to the other edge depending on whether or not the sign of the torsion changes at that vertex of the binotrix . since the darboux indicatrix is the dual of the notrix , each of the points obtained by moving a distance of @xmath60 normal to each arc of the darboux indicatrix and towards its left as viewed from the exterior of the sphere is either a vertex of the notrix or the anti - notrix . by remark [ rem : reinterpretation ] , it suffices to check that this sequence of vertices satisfies the conditions above . we verify this by explicitly computing this sequence of vertices : @xmath189 it is straightforward to see that when two adjacent edges @xmath190 , @xmath165 of the darboux indicatrix share a vertex @xmath111 of the tantrix , we join @xmath156 to @xmath185 and @xmath186 to @xmath157 , and when two adjacent edges @xmath165 , @xmath164 share a vertex @xmath112 of the binotrix , we join @xmath157 to @xmath188 if the sign of the torsion changes at that vertex of the binotrix , and to @xmath187 if the sign of the torsion does not change at that vertex of the binotrix . as in remark [ rem : greatcirclestangent ] , theorem [ thm : tantrixinflection_polygonal ] provides an alternative interpretation of the great circles tangent to the notrix and anti - notrix of a smooth knot at points where the geodesic curvature @xmath64 of the tantrix is stationary in theorem [ thm : tantrix - bridgeinflection ] . if we interpret arcs of a great circle as having zero geodesic curvature , a vertex of the tantrix with negative torsion as having positive geodesic curvature , and a vertex of the tantrix of positive torsion as having negative geodesic curvature , then we have the equivalent of a `` point of stationary geodesic curvature of the tantrix '' at every vertex of the tantrix and along edges of the tantrix ( which correspond naturally with vertices of the binotrix ) where the sign of the torsion is the same at both vertices . these are precisely the vertices of the darboux indicatrix where we move from a vertex of the notrix to a vertex of the anti - notrix . in this sense , the great circles tangent to the notrix and anti - notrix at points where the geodesic curvature @xmath64 of the tantrix is stationary can be viewed as the union of two antipodal great circle arcs joining the notrix to the anti - notrix . let @xmath4 be a stick knot with @xmath69 edges . since each arc of a spherical indicatrix of @xmath4 is a geodesic , it can intersect any great circle at most once . it follows that the bridge and inflection maps of @xmath4 satisfy @xmath191 and @xmath192 and the tantrix - bridge and tantrix - inflection maps of @xmath4 satisfy @xmath193 and @xmath194 for all @xmath6 . we believe that for any stick knot @xmath4 , it is possible to use the methods in a paper by wu @xcite to show that there exists a smooth knot @xmath195 with the same knot type such that the maps defined by the various indicatrices of @xmath195 are combinatorially the same as those of @xmath4 , that is , there exists an isotopy of @xmath8 that sends a map on @xmath195 to that on @xmath4 ( note that we do not require the same isotopy for all maps ) . this would enable us to obtain bounds on the stick number of knots by studying the maps defined by the various indicatrices for smooth knots . also , since each arc of the darboux indicatrix of a stick knot connects a vertex of the tantrix to a vertex of the binotrix and thus has length @xmath60 , the total length of the darboux indicatrix of @xmath4 is @xmath196 . it would be interesting to see if one could relate the length of the darboux indicatrix of a stick knot to the length of the darboux indicatrix of smooth knot conformations of the same knot type , as this would also enable us to obtain bounds on the stick number of knots by studying their darboux indicatrices . observe that the bridge index and superbridge index of a knot can always be realized in smooth conformations with non - vanishing curvature . in such a case , the bridge graph is simply the union of the binotrix and the anti - binotrix . it follows that @xmath197 = b[k ] + 1 $ ] if and only if there exists a knot conformation @xmath198 $ ] such that the binotrix does not intersect itself or the anti - binotrix and this conformation realizes the lowest possible value of the bridge map over all knots @xmath198 $ ] . this is because at a point where the binotrix intersects itself or the anti - binotrix , there will be two regions adjacent to that point where the value of the bridge map differs by @xmath199 , and hence the superbridge index would be at least two greater than the bridge index . in particular , jeon and jin @xcite conjectured that the only knots with superbridge index 3 are the trefoil knot and the figure eight knot . this conjecture would be proved if it could be shown that the trefoil knot and the figure eight knot are the only non - trivial knots with conformations such that the binotrix does not intersect itself or the anti - binotrix and the bridge map realizes a value of @xmath199 in such a conformation .

recursively compressed inverse preconditioning ( rcip ) is a numerical method for obtaining highly accurate solutions to integral equations on piecewise smooth surfaces . the method originated in 2008 as a technique for solving laplace s equation in two - dimensional domains with corners . in a series of subsequent papers the technique was then refined and extended in various directions as to cover a broad range of boundary value problems in physics and engineering . the purpose of the present paper is threefold : first , to review the rcip method in a simple setting . second , to show how easily the method can be implemented in matlab . third , to present new applications of rcip to integral equations of scattering theory on planar curves with corners .

1207.6737

this paper is about an efficient numerical solver for elliptic boundary value problems in piecewise smooth domains . such a solver is useful for applications in physics and engineering , where computational domains of interest often have boundaries that contain singular points such as corners and triple junctions . furthermore , elliptic boundary value problems in non - smooth domains are difficult to solve irrespective of what numerical method is used . the reason being that the solution , or the quantity representing the solution , often exhibits a non - smooth behavior close to boundary singularities . that behavior is hard to resolve by polynomials , which underly most approximation schemes . mesh refinement is needed . this is costly and may lead to artificial ill - conditioning and the loss of accuracy . the numerical solver we propose takes its starting point in an integral equation reformulation of the boundary value problem at hand . more precisely , we assume that the problem can be modeled as a fredholm second kind integral equation with integral operators that are compact away from singular boundary points and whose solution is a layer density representing the solution to the original problem . at the heart of our solver lie certain integral transforms . their inverses , constructed in a recursive manner and on locally refined temporary meshes , modify the kernels of the integral operators in such a way that the layer density becomes piecewise smooth and simple to resolve by polynomials . conceptually , this corresponds to applying a fast direct solver @xcite locally to regions with troublesome geometry . we then apply a global iterative method . this gives us many of the advantages of fast direct methods , for example the ability to deal with certain classes of operators whose spectra make them unsuitable for iterative methods . in addition , the approach is typically much faster than using only a fast direct solver . our method , or scheme , has previously been referred to as _ recursive compressed inverse preconditioning _ @xcite and there is a good reason for that name : the scheme relies on applying a relieving right inverse to the integral equation ; on compressing this inverse to a low - dimensional subspace ; and on carrying out the compression in a recursive manner . still , the name _ recursive(ly ) compressed inverse preconditioning _ is a bit awkward and we will here simply use the acronym rcip . a strong motivation for writing the present paper is that the original references @xcite are hard to read for non - experts and that efficient codes , therefore , could be hard to implement . certain derivations in @xcite use complicated intermediary constructions , application specific issues obscure the general picture , and the notation has evolved from paper to paper . in the present paper we focus on the method itself , on how it works and how it can be implemented , and refer to the original research papers for details . demo codes , in matlab , are a part of the exposition and can be downloaded from : http://www.maths.lth.se/na/staff/helsing/tutor/ section [ sec : back ] provides some historical background . the basics if the rcip method are then explained by solving a simple model problem in sections [ sec : model][sec : rec ] . sections [ sec : sb][sec : sing ] review various extensions and improvements . some of this material is new , for example the simplified treatment of composed operators in section [ sec : compose ] . the last part of the paper , sections [ sec : helmdiri][sec : helmfield ] , contain new applications to scattering problems . the line of research on fast solvers for elliptic boundary value problems in piecewise smooth domains , leading up to the rcip method , grew out of work in computational fracture mechanics . early efforts concerned finding efficient integral equation formulations . corner singularities were either resolved with brute force or by using special basis functions , see @xcite for examples . such strategies , in combination with fast multipole @xcite accelerated iterative solvers , work well for simple small - scale problems . real world physics is more complicated and , for example , the study @xcite on a high - order time - stepping scheme for crack propagation ( a series of biharmonic problems for an evolving piecewise smooth surface ) shows that radically better methods are needed . special basis functions are too complicated to construct and brute force is not economical merely storing the discretized solution becomes too costly in a large - scale simulation . a breakthrough came in 2007 , when a scheme was created that resolves virtually any problem for laplace s equation in piecewise smooth two - dimensional domains in a way that is fully automatic , fast , stable , memory efficient , and whose computational cost scales linearly with the number of corners in the computational domain . the resulting paper @xcite constitutes the origin of the rcip method . unfortunately , however , there are some flaws in @xcite . the expressions in ( * ? ? ? * section 9 ) are not generally valid and the paper fails to apply rcip in its entirety to the biharmonic problem of ( * ? ? ? * section 3 ) , which was the ultimate goal . the second paper on rcip @xcite deals with elastic grains . the part ( * ? ? ? * appendix b ) , on speedup , is particulary useful . the third paper on rcip @xcite contains improvement relative to the earlier papers , both in the notation and in the discretization of singular operators . the overall theme is mixed boundary conditions , which pose similar difficulties as do piecewise smooth boundaries . the paper @xcite , finally , solves the problem of ( * ? ? ? * section 3 ) in a broad setting and one can view the formalism of @xcite as the rcip method in its most general form , at least in two dimensions . further work on rcip has been devoted to more general boundary conditions @xcite , to problems in three dimension @xcite , and to solving elliptic boundary value problems in special situations , such as for aggregates of millions of grains , where special techniques are introduced to deal with problem specific issues @xcite . we end this retrospection by pointing out that several research groups in recent years have proposed numerical schemes for integral equations stemming from elliptic partial differential equations in piecewise smooth domains . see , for example , @xcite . there is also a widespread notion that a slight rounding of corners is a good idea for numerics . while rounding may work in particular situations , we do not believe it is a generally viable method . for one thing , how does one round a triple junction ? of ( [ eq : gamma ] ) with a corner at the origin . the solid curve corresponds to opening angle @xmath0 . the dashed curve has @xmath1.,height=279 ] let @xmath2 be the closed contour of figure [ fig : sno1 ] with the parameterization @xmath3\ , . \label{eq : gamma}\ ] ] let @xmath4 be the fundamental solution to laplace s equation which in the plane can be written @xmath5 we shall solve the integral equation @xmath6 for the unknown layer density @xmath7 . here @xmath8 is the exterior unit normal of the contour , @xmath9 is an element of arc length , @xmath10 is a parameter , and @xmath11 is a unit vector . the equation ( [ eq : inteq1a ] ) models an electrostatic inclusion problem @xcite . using complex notation , where vectors @xmath12 , @xmath13 , and @xmath8 in the real plane @xmath14 correspond to points @xmath15 , @xmath16 , and @xmath17 in the complex plane @xmath18 , one can express ( [ eq : inteq1a ] ) as @xmath19 where the ` bar ' symbol denotes complex conjugation . equation ( [ eq : inteq2 ] ) is a simplification over ( [ eq : inteq1a ] ) from a programming point of view . from an algorithmic point of view it is advantageous to write ( [ eq : inteq1a ] ) in the abbreviated form @xmath20 where @xmath21 is the identity . if @xmath2 is smooth , then ( [ eq : inteq3 ] ) is a fredholm second kind integral equation with a compact , non - self - adjoint , integral operator @xmath22 whose spectrum is discrete , bounded by one in modulus , and accumulates at zero . we also need a way to monitor the convergence of solutions @xmath23 to ( [ eq : inteq3 ] ) . for this purpose we introduce a quantity @xmath24 , which corresponds to polarizability @xcite @xmath25 of ( [ eq : gamma ] ) with opening angle @xmath26 . a subset of @xmath2 , called @xmath27 , covers four coarse panels as indicated by the dashed curve . right : a fine mesh created from the coarse mesh by subdividing the panels closest to the corner @xmath28 times.,height=113 ] we discretize ( [ eq : inteq3 ] ) using the original nystrm method based on composite 16-point gauss legendre quadrature on two different meshes : a _ coarse mesh _ with @xmath29 quadrature panels and a _ fine mesh _ which is constructed from the coarse mesh by @xmath30 times subdividing the panels closest to the corner in a direction toward the corner . the discretization is in parameter . the four panels on the coarse mesh that are closest to the corner should be equisized in parameter . these panels form a subset of @xmath2 called @xmath27 . see figure [ fig : sno2 ] . the linear systems resulting from discretization on the coarse mesh and on the fine mesh can be written formally as @xmath31 where @xmath32 and @xmath33 are square matrices and @xmath34 and @xmath35 are column vectors . the subscripts fin and coa indicate what type of mesh is used . discretization points on a mesh are said to constitute a grid . the coarse grid has @xmath36 points . the fine grid has @xmath37 points . the discretization of ( [ eq : inteq3 ] ) is carried out by first rewriting ( [ eq : inteq2 ] ) as @xmath38\ , . \label{eq : inteq2b}\ ] ] then nystrm discretization with @xmath39 points @xmath40 and weights @xmath41 on @xmath2 gives @xmath42 the program demo1.m sets up the system ( [ eq : inteq3b ] ) , solves it using the gmres iterative solver @xcite incorporating a low - threshold stagnation avoiding technique ( * ? ? ? * section 8) , and computes @xmath24 of ( [ eq : q ] ) . the user has to specify the opening angle @xmath43 , the parameter @xmath10 , the number @xmath29 of coarse panels on @xmath2 , the unit vector @xmath11 and the number of subdivisions @xmath30 . the opening angle should be in the interval @xmath44 . we choose @xmath45 , @xmath26 , @xmath46 , and @xmath47 . the quantity @xmath24 converges initially as @xmath30 is increased , but for @xmath48 the results start to get worse . see figure [ fig : conv1 ] . this is related to the fact , pointed out by bremer @xcite , that the original nystrm method captures the @xmath49 behavior of the solution @xmath50 , while our @xmath50 is unbounded . see , further , appendix d. let us split the matrices @xmath51 and @xmath52 of ( [ eq : inteq3a ] ) and ( [ eq : inteq3b ] ) into two parts each @xmath53 here the superscript @xmath54 indicates that only entries of a matrix @xmath55 whose indices @xmath56 and @xmath57 correspond to points @xmath40 and @xmath58 that both belong to the boundary segment @xmath27 are retained . the remaining entries are zero . now we introduce two diagonal matrices @xmath59 and @xmath60 which have the quadrature weights @xmath41 on the diagonal . furthermore , we need a prolongation matrix @xmath61 which interpolates functions known at points on the coarse grid to points on the fine grid . the construction of @xmath61 relies on panelwise 15-degree polynomial interpolation in parameter using vandermonde matrices . we also construct a weighted prolongation matrix @xmath62 via @xmath63 the matrices @xmath61 and @xmath62 share the same sparsity pattern . they are rectangular matrices , similar to the the identity matrix , but with one full @xmath64 block . let superscript @xmath65 denote the transpose . then @xmath66 holds exactly . see appendix a and ( * ? ? ? * section 4.3 ) . equipped with @xmath61 and @xmath62 we are ready to compress ( [ eq : inteq3b ] ) on the fine grid to an equation essentially on the coarse grid . this compression is done without the loss of accuracy the discretization error in the solution is unaffected and no information is lost . the compression relies on the variable substitution @xmath67 here @xmath68 is the discretization of a piecewise smooth _ transformed density_. the compression also uses the low - rank decomposition @xmath69 which should hold to about machine precision . the compressed version of ( [ eq : inteq3b ] ) reads @xmath70 where the compressed weighted inverse @xmath71 is given by @xmath72 see appendix b for details on the derivation . the compressed weighted inverse @xmath71 , for @xmath2 of ( [ eq : gamma ] ) , is a block diagonal matrix with one full @xmath73 block and the remaining entries coinciding with those of the identity matrix . after having solved ( [ eq : inteq4 ] ) for @xmath68 , the density @xmath74 can easily be reconstructed from @xmath68 in a post - processor , see section [ sec : recon ] . it is important to observe , however , that @xmath74 is not always needed . for example , the quantity @xmath24 of ( [ eq : q ] ) can be computed directly from @xmath68 . let @xmath75 be a column vector which contains the absolute values of the boundary derivatives @xmath76 multiplied with weights @xmath41 , positions @xmath40 , and the complex scalar @xmath77 . then @xmath78 the compressed weighted inverse @xmath71 is costly to compute from its definition ( [ eq : r ] ) . as we saw in section [ sec : disc ] , the inversion of large matrices @xmath79 on highly refined grids could also be unstable . fortunately , the computation of @xmath80 can be greatly sped up and stabilized via a recursion . this recursion is derived in a roundabout way and using a refined grid that differs from that of the present tutorial in ( * ? ? ? * section 7.2 ) . a better derivation can be found in ( * ? ? ? * section 5 ) , but there the setting is more general so that text could be hard to follow . here we focus on results . let @xmath81 be a prolongation matrix , performing panelwise 15-degree polynomial interpolation in parameter from a 64-point grid on a four - panel mesh to a 96-point grid on a six - panel mesh as shown in figure [ fig : pbc ] . let @xmath82 be a weighted prolongation matrix in the style of ( [ eq : pw ] ) . if t16 and w16 are the nodes and weights of 16-point gauss legendre quadrature on the canonical interval @xmath83 $ ] , then @xmath81 and @xmath82 can be constructed as .... t32=[t16 - 1;t16 + 1]/2 ; w32=[w16;w16]/2 ; a = ones(16 ) ; aa = ones(32,16 ) ; for k=2:16 a(:,k)=a(:,k-1).*t16 ; aa(:,k)=aa(:,k-1).*t32 ; end ip = aa / a ; ipw = ip.*(w32*(1./w16 ) ' ) ; % pbc = zeros(96,64 ) ; pbc ( 1:16 , 1:16)=eye(16 ) ; pbc(17:48,17:32)=ip ; pbc(49:96,33:64)=flipud(fliplr(pbc(1:48,1:32 ) ) ) ; % pwbc = zeros(96,64 ) ; pwbc ( 1:16 , 1:16)=eye(16 ) ; pwbc(17:48,17:32)=ipw ; pwbc(49:96,33:64)=flipud(fliplr(pwbc(1:48,1:32 ) ) ) ; .... see ( * ? ? ? * appendix a ) for an explanation of why high - degree polynomial interpolation involving vandermonde systems gives accurate results for smooth functions . , @xmath84 , and @xmath85 along with their corresponding type b meshes for @xmath28.,height=151 ] let @xmath86 , @xmath87 , be a sequence of subsets of @xmath27 with @xmath88 and @xmath89 . let there also be a six - panel mesh and a corresponding 96-point grid on each @xmath86 . the construction of the subsets and their meshes should be such that if @xmath90 , @xmath91 $ ] , is a local parameterization of @xmath86 , then the breakpoints ( locations of panel endpoints ) of its mesh are at @xmath92 and the breakpoints of the mesh on @xmath93 are at @xmath94 . we denote this type of nested six - panel meshes _ type b_. the index @xmath56 is the _ level_. an example of a sequence of subsets and meshes on @xmath27 is shown in figure [ fig : subsets ] for @xmath28 . compare ( * ? ? ? * figure 2 ) and ( * ? ? ? * figure 5.1 ) . let @xmath95 denote the discretization of @xmath22 on a type b mesh on @xmath86 . in the spirit of ( [ eq : split1],[eq : split2 ] ) we write @xmath96 where the superscript @xmath54 indicates that only entries with both indices corresponding to points on the four inner panels are retained . now , let @xmath97 denote the full @xmath73 diagonal block of @xmath71 . the recursion for @xmath97 is derived in appendix c and it reads @xmath98 where the operator @xmath99 expands its matrix argument by zero - padding ( adding a frame of zeros of width 16 around it ) . note that the initializer ( [ eq : rstart ] ) makes the recursion ( [ eq : recur ] ) take the first step @xmath100 the program demo2.m sets up the linear system ( [ eq : inteq4 ] ) , runs the recursion ( [ eq : recur],[eq : rstart ] ) , and solves the linear system using the same techniques as demo1.m , see section [ sec : disc ] . in fact , the results produced by the two programs are very similar , at least up to @xmath101 . this supports the claim of section [ sec : comp ] that discretization error in the solution is unaffected by compression . figure [ fig : conv2 ] demonstrates the power of rcip : fewer unknowns and faster execution , better conditioning ( the number of gmres iterations does not grow ) , and higher achievable accuracy . compare figure [ fig : conv1 ] . we emphasize that the number @xmath30 of recursion steps ( levels ) used in ( [ eq : recur ] ) corresponds to the number of subdivisions @xmath30 used to construct the fine mesh . the recursion ( [ eq : recur ] ) can be sped up using the schur banachiewicz inverse formula for partitioned matrices @xcite , which in this context can be written ( * ? ? ? * appendix b ) @xmath102 \left [ \begin{array}{cc } { \bf a}^{-1 } & { \bf u } \\ { \bf v } & { \bf d } \end{array } \right]^{-1 } \left [ \begin{array}{cc } { \bf p}^{\star } & { \bf 0 } \\ { \bf 0 } & { \bf i } \end{array } \right]=\ ] ] @xmath103\ , , \label{eq : cbana}\ ] ] where @xmath104 plays the role of @xmath105 , @xmath106 and @xmath107 are submatrices of @xmath81 and @xmath82 , and @xmath108 , @xmath109 , and @xmath110 refer to blocks of @xmath111 . the program demo3.m is based on demo2.m , but has ( [ eq : cbana ] ) incorporated . besides , the integral equation ( [ eq : inteq1a ] ) is replaced with @xmath112 which has the same solution @xmath7 but is more stable for @xmath10 close to one . for the discretization of ( [ eq : inteq1b ] ) to fit the form ( [ eq : inteq4 ] ) , the last term on the left hand side of ( [ eq : inteq1b ] ) is added to the matrix @xmath113 of ( [ eq : inteq4 ] ) . the execution of demo3.m is faster than that of demo2.m . figure [ fig : conv3 ] shows that a couple of extra digits are gained by using ( [ eq : inteq1b ] ) rather than ( [ eq : inteq1a ] ) and that full machine accuracy is achieved for @xmath114 . let us introduce a new discrete density @xmath115 via @xmath116 rewriting ( [ eq : inteq4 ] ) in terms of @xmath115 gives @xmath117 which resembles the original equation ( [ eq : inteq3a ] ) . we see that @xmath118 , which is discretized using gauss legendre quadrature , acts on @xmath115 . therefore one can interpret @xmath115 as pointwise values of the original density @xmath7 , multiplied with weight corrections suitable for integration against polynomials . we refer to @xmath119 as a _ weight - corrected density_. compare ( * ? ? ? * section 5.4 ) . assume now that there is a square matrix @xmath120 which maps @xmath68 to discrete values @xmath121 of the original density on the coarse grid @xmath122 the matrix @xmath120 allows us to rewrite ( [ eq : inteq4 ] ) as a system for the original density @xmath123 we can interpret the composition @xmath124 as a matrix of multiplicative weight corrections that compensate for the singular behavior of @xmath7 on @xmath27 when gauss legendre quadrature is used . let @xmath125 denote the rectangular matrix @xmath126 and let @xmath127 be a restriction operator which performs panelwise 15-degree polynomial interpolation in parameter from a grid on the fine mesh to a grid on a the coarse mesh . we see from ( [ eq : subst ] ) that @xmath125 is the mapping from @xmath128 to @xmath74 . therefore the columns of @xmath125 can be interpreted as discrete basis functions for @xmath7 . it holds by definition that @xmath129 the quantities and interpretations of this section come in handy in various situations , for example in 3d extensions of the rcip method @xcite . an efficient scheme for constructing @xmath120 will be presented in section [ sec : s ] . the action of @xmath125 on @xmath68 , which gives @xmath74 , can be obtained by , in a sense , running the recursion ( [ eq : recur ] ) backwards . the process is described in detail in ( * ? ? ? * section 7 ) . here we focus on results . the backward recursion on @xmath27 reads @xmath130{\bf p}_{\rm bc}\tilde{\boldsymbol{\rho}}_{{\rm coa},i}\ , , \quad i = n_{\rm sub},\ldots,1\ , . \label{eq : back}\ ] ] here @xmath131 is a column vector with @xmath132 elements . in particular , @xmath133 is the restriction of @xmath68 to @xmath27 , while @xmath131 are taken as elements @xmath134 of @xmath135 for @xmath136 . the elements @xmath137 and @xmath138 of @xmath139 are the reconstructed values of @xmath74 on the outermost panels of a type b mesh on @xmath86 . outside of @xmath27 , @xmath74 coincides with @xmath68 . when the recursion is completed , the reconstructed values of @xmath74 on the four innermost panels are obtained from @xmath140 should one wish to interrupt the recursion ( [ eq : back ] ) prematurely , at step @xmath141 say , then @xmath142 gives values of a weight - corrected density on the four innermost panels of a type b mesh on @xmath143 . that is , we have a part - way reconstructed weight - corrected density @xmath144 on a mesh that is @xmath145 times refined . this observation is useful in the context of evaluating layer potentials close to their sources . if the memory permits , one can store the matrices @xmath146 and @xmath147 in the forward recursion ( [ eq : recur ] ) and reuse them in the backward recursion ( [ eq : back ] ) . otherwise they may be computed afresh . the program demo4.m builds on the program demo3.m , using ( [ eq : inteq4 ] ) for ( [ eq : inteq1b ] ) . after the main linear system is solved for @xmath68 , a postprocessor reconstructs @xmath74 via ( [ eq : back ] ) . then a comparison is made with a solution @xmath74 obtained by solving the un - compressed system ( [ eq : inteq3b ] ) . figure [ fig : conv45 ] shows that for @xmath148 the results are virtually identical . this verifies the correctness of ( [ eq : back ] ) . for @xmath149 the result start to deviate . that illustrates the instabilities associated with solving ( [ eq : inteq3b ] ) on a highly refined mesh . compare figure [ fig : conv1 ] . the program demo5.m investigates the effects of premature interruption of ( [ eq : back ] ) . the number of recursion steps is set to @xmath150 and the recursion is interrupted at different levels . the density @xmath74 is reconstructed on outer panels up to the level of interruption . then a weight - corrected density is produced at the innermost four panels according to ( [ eq : premat ] ) . finally @xmath24 of ( [ eq : q ] ) is computed from this part - way reconstructed solution . the right image of figure [ fig : conv45 ] shows that the quality of @xmath24 is unaffected by the level of interruption . this section discusses the construction of @xmath120 and other auxiliary matrices . note that in many applications , these matrices are not needed . the entries of the matrices @xmath61 , @xmath62 , @xmath127 , @xmath80 , @xmath120 , and @xmath125 can only differ from those of the identity matrix when both indices correspond to discretization points on @xmath27 . for example , the entries of @xmath71 only differ from the identity matrix for the @xmath73 block denoted @xmath151 in ( [ eq : recur ] ) . in accordance with this notation we introduce @xmath152 , @xmath153 , @xmath154 , @xmath155 and @xmath156 for the restriction of @xmath61 , @xmath62 , @xmath127 , @xmath120 and @xmath125 to @xmath27 . in the codes of this section we often use this restricted type of matrices , leaving the identity part out . we observe that @xmath155 is a square @xmath73 matrix ; @xmath152 , @xmath157 and @xmath156 are rectangular @xmath158 matrices ; and @xmath154 is a rectangular @xmath159 matrix . furthermore , @xmath154 is very sparse for large @xmath30 . all columns of @xmath160 with column indices corresponding to points on panels that result from more than eight subdivisions are identically zero . the program demo6.m sets up @xmath152 , @xmath161 and @xmath154 , shows their sparsity patterns , and verifies the identities ( [ eq : pwtp ] ) and ( [ eq : qp ] ) . the implementations for @xmath152 and @xmath157 rely on repeated interpolation from coarser to finer intermediate grids . the implementation of @xmath160 relies on keeping track of the relation between points on the original coarse and fine grids . output from demo6.m is depicted in the left image of figure [ fig : conv67 ] . note that the matrices @xmath152 and @xmath157 are never needed in applications . we are now ready to construct @xmath120 . section [ sec : recon ] presented a scheme for evaluating the action of @xmath162 on discrete functions on the coarse grid on @xmath27 . the matrix @xmath163 , itself , can be constructed by applying this scheme to a @xmath73 identity matrix . the matrix @xmath154 was set up in demo6.m . composing these two matrices gives @xmath164 , see ( [ eq : qy ] ) . this is done in the program demo7.m , where the identity part is added as to get the entire matrix @xmath120 . in previous work on rcip we have found use for @xmath165 in complex situations where ( [ eq : inteq6 ] ) is preferable over ( [ eq : inteq4 ] ) , see ( * ? ? ? * section 9 ) . if one merely needs @xmath121 from @xmath68 in a post - processor , setting up @xmath120 and using ( [ eq : smap ] ) is not worthwhile . it is cheaper to let @xmath125 act on @xmath68 and then let @xmath127 act on the resulting vector . anyhow , demo7.m builds on demo4.m and gives as output @xmath166 computed via ( [ eq : smap ] ) , see the right image of figure [ fig : conv67 ] . for comparison , @xmath167 , computed from the un - compressed system ( [ eq : inteq3b ] ) , is also shown . it often happens that @xmath168 is wedge - like . a corner of a polygon , for example , has wedge - like @xmath168 at all levels . if @xmath27 is merely piecewise smooth , then the @xmath168 are wedge - like to double precision accuracy for @xmath169 . wedge - like sequences of @xmath168 open up for simplifications and speedup in the recursion ( [ eq : recur],[eq : rstart ] ) . particularly so if the kernel of the integral operator @xmath22 of ( [ eq : inteq3 ] ) is scale invariant on wedges . then the matrix @xmath170 becomes independent of @xmath56 . it can be denoted @xmath171 and needs only to be constructed once . furthermore , the recursion ( [ eq : recur],[eq : rstart ] ) assumes the form of a fixed - point iteration @xmath172 the iteration ( [ eq : fixed ] ) can be run until @xmath147 converges . let @xmath173 be the converged result . one need not worry about predicting the number of levels needed . choosing the number @xmath174 of levels needed , in order to meet a beforehand given tolerance in @xmath173 , is otherwise a big problem in connection with ( [ eq : recur],[eq : rstart ] ) and non - wedge - like @xmath27 . this number has no general upper bound . assume now that the kernel of @xmath22 is scale invariant on wedges . if all @xmath86 are wedge - like , then ( [ eq : fixed],[eq : fstart ] ) replaces ( [ eq : recur],[eq : rstart ] ) entirely . if @xmath27 is merely piecewise smooth , then ( [ eq : fixed],[eq : fstart ] ) can be run on a wedge with the same opening angle as @xmath27 , to produce an initializer to ( [ eq : recur ] ) . that initializer could often be more appropriate than ( [ eq : rstart ] ) , which is plagued with a very large discretization error whenever ( [ eq : inteq2c ] ) is used . this fixed - point recursion initializer is implemented in the program demo8b.m , which is a simple upgrading of demo3b.m , and produces figure [ fig : conv8 ] . a comparison of figure [ fig : conv8 ] with figure [ fig : conv3 ] shows that the number @xmath30 of levels needed for full convergence is halved compared to when using the initializer ( [ eq : rstart ] ) . there are , generally speaking , several advantages with using the fixed - point recursion initializer , rather than ( [ eq : rstart ] ) , in ( [ eq : recur ] ) on a non - wedge - like @xmath27 : first , the number of different matrices @xmath175 and @xmath170 needed in ( [ eq : recur ] ) and in ( [ eq : back ] ) is reduced as the recursions are shortened . this means savings in storage . second , the number @xmath30 of levels needed for full convergence in ( [ eq : recur ] ) seems to always be bounded . the hard work is done in ( [ eq : fixed ] ) . third , newton s method can be used to accelerate ( [ eq : fixed ] ) . that is the topic of section [ sec : newton ] . when solving integral equations stemming from particularly challenging elliptic boundary value problems with solutions @xmath7 that are barely absolutely integrable , the fixed - point iteration ( [ eq : fixed],[eq : fstart ] ) on wedge - like @xmath27 may need a very large number of steps to reach full convergence . see ( * ? ? ? * section 6.3 ) for an example where @xmath176 steps are needed . fortunately , ( [ eq : fixed ] ) can be cast as a non - linear matrix equation @xmath177 where @xmath173 , as in section [ sec : init ] , is the fixed - point solution and @xmath178 the non - linear equation ( [ eq : newton ] ) , in turn , can be solved for @xmath173 with a variant of newton s method . let @xmath179 be a matrix - valued perturbation of @xmath173 and expand @xmath180 to first order in @xmath179 . this gives a sylvester - type matrix equation @xmath181 for the newton update @xmath179 . one can use the matlab built - in function dlyap for ( [ eq : sylvester ] ) , but gmres seems to be more efficient and we use that method . compare ( * ? ? ? * section 6.2 ) . , [ eq : fstart ] ) is compared to newton s method for ( [ eq : newton]).,height=279 ] figure [ fig : conv9 ] shows a comparison between the fixed - point iteration ( [ eq : fixed],[eq : fstart ] ) and newton s method for computing the fixed - point solution @xmath173 to ( [ eq : newton ] ) on a wedge - like @xmath27 . the program demo9.m is used and it incorporates schur banachiewicz speedup in the style of section [ sec : sb ] . the wedge opening angle is @xmath26 , the integral operator @xmath22 is the same as in ( [ eq : inteq3 ] ) , and @xmath45 . the relative difference between the two converged solutions @xmath173 is @xmath182 . figure [ fig : conv9 ] clearly shows that ( [ eq : fixed],[eq : fstart ] ) converges linearly ( in 68 iterations ) , while newton s method has quadratic convergence . only four iterations are needed . the computational cost per iteration is , of course , higher for newton s method than for the fixed - point iteration , but it is the same at each step . recall that the underlying size of the matrix @xmath183 , that is inverted according to ( [ eq : r ] ) , grows linearly with the number of steps needed in the fixed - point iteration . this example therefore demonstrates that one can invert and compress a linear system of the type ( [ eq : r ] ) in sub - linear time . the integral equation ( [ eq : inteq1a ] ) comes from a boundary value problem for laplace s equation where the potential field @xmath184 at a point @xmath12 in the plane is related to the layer density @xmath7 via @xmath185 see ( * ? ? ? * section 2.1 ) . , left image , but with the program demo3c.m . the potential @xmath184 at @xmath186 is evaluated via ( [ eq : uofr ] ) and npan=14 is used.,height=249 ] figure [ fig : conv3c ] shows how the field solution @xmath184 of ( [ eq : uofr ] ) converges with mesh refinement at a point @xmath186 inside the contour @xmath2 of ( [ eq : gamma ] ) . we see that the accuracy in @xmath184 , typically , is one digit better than the accuracy of the dipole moment @xmath24 of ( [ eq : q ] ) . one can therefore say that measuring the field error at a point some distance away from the corner is more forgiving than measuring the dipole moment error . it is possible to construct examples where the difference in accuracy between field solutions and moments of layer densities are more pronounced and this raises the question of how accuracy best should be measured in the context of solving integral equations . we do not have a definite answer , but think that higher moments of layer densities give more fair estimates of overall accuracy than do field solutions at points some distance away from the boundary . see , further , appendix e. assume that we have a modification of ( [ eq : inteq3 ] ) which reads @xmath187 here @xmath22 and @xmath188 are as in ( [ eq : inteq3 ] ) , @xmath189 is a new , bounded , integral operator , and @xmath190 is an unknown layer density to be solved for . this section shows how to apply rcip to ( [ eq : inteq7 ] ) using a simplified version of the scheme in @xcite . let us , temporarily , expand ( [ eq : inteq7 ] ) into a system of equations by introducing a new layer density @xmath191 . then @xmath192 and after discretization on the fine mesh @xmath193 + \left [ \begin{array}{cc } { \bf 0}_{\rm fin } & { \bf m}_{\rm fin } \\ -{\bf k}_{\rm fin } & { \bf 0}_{\rm fin } \end{array } \right ] \right ) \left[\begin{array}{c } \boldsymbol{\rho}_{1{\rm fin } } \\ \boldsymbol{\rho}_{2{\rm fin } } \end{array}\right ] = \left[\begin{array}{c } { \bf g}_{\rm fin } \\ 0 \end{array } \right]\ , . \label{eq : disc1}\ ] ] standard rcip gives @xmath194 + \left [ \begin{array}{cc } { \bf 0}_{\rm coa } & { \bf m}_{\rm coa}^{\circ } \\ -{\bf k}_{\rm coa}^{\circ } & { \bf 0}_{\rm coa } \end{array } \right ] \left [ \begin{array}{cc } { \bf r}_1 & { \bf r}_3 \\ { \bf r}_2 & { \bf r}_4 \end{array } \right ] \right ) \left[\begin{array}{c } \tilde{\boldsymbol{\rho}}_{1{\rm coa } } \\ \tilde{\boldsymbol{\rho}}_{2{\rm coa } } \end{array}\right ] = \left[\begin{array}{c } { \bf g}_{\rm coa } \\ 0 \end{array } \right]\ , , \label{eq : disc2}\ ] ] where the compressed inverse @xmath71 is partitioned into four equisized blocks . now we replace @xmath195 and @xmath196 with a single unknown @xmath68 via @xmath197 the change of variables ( [ eq : single1],[eq : single2 ] ) is chosen so that the second block - row of ( [ eq : disc2 ] ) is automatically satisfied . the first block - row of ( [ eq : disc2 ] ) becomes @xmath198\tilde{\boldsymbol{\rho}}_{\rm coa}={\bf g}_{\rm coa}\ , . \label{eq : disc3}\end{gathered}\ ] ] when ( [ eq : disc3 ] ) has been solved for @xmath68 , the weight - corrected version of the original density @xmath190 can be recovered as @xmath199 . [ fig : conv10 ] figure [ fig : conv10 ] shows results for ( [ eq : inteq7 ] ) with @xmath189 being the double layer potential @xmath200 the left image shows the convergence of @xmath24 of ( [ eq : q ] ) with @xmath30 using the inner product preserving discretization scheme of appendix d for ( [ eq : inteq7 ] ) as implemented in demo10.m . the right image shows @xmath24 produced with rcip according to ( [ eq : disc3],[eq : hat1 ] ) as implemented in demo10b.m . the reference value for @xmath24 is computed with the program demo10c.m , which uses inner product preserving discretization together with compensated summation @xcite in order to enhance the achievable accuracy . one can see that , in addition to being faster , rcip gives and extra digit of accuracy . actually , it seems as if the scheme in demo10.m converges to a @xmath24 that is slightly wrong . in conclusion , in this example and in terms of stability , the rcip method is better than standard inner product preserving discretization and on par with inner product preserving discretization enhanced with compensated summation . in terms of computational economy and speed , rcip greatly outperforms the two other schemes . the nystrm scheme of section [ sec : disc ] discretizes ( [ eq : inteq3 ] ) using composite 16-point gauss legendre quadrature . this works well as long as the kernel of the integral operator @xmath22 and the layer density are smooth on smooth @xmath2 . when the kernel is not smooth on smooth @xmath2 , the quadrature fails and something better is needed . see @xcite for a comparison of the performance of various modified high - order accurate nystrm discretizations for weakly singular kernels and @xcite for a recent high - order general approach to the evaluation of layer potentials . we are not sure what modified discretization is optimal in every situation . when logarithmic- and cauchy - singular operators need to be discretized in sections [ sec : helmdiri][sec : helmfield ] , below , we use two modifications to composite gauss legendre quadrature called _ local panelwise evaluation _ and _ local regularization_. see ( * ? ? ? * section 2 ) for a description of these techniques . let @xmath201 be the domain enclosed by the curve @xmath2 and let @xmath202 be the exterior to the closure of @xmath201 . the exterior dirichlet problem for the helmholtz equation @xmath203 can be modeled using a combined integral representation ( * ? ? ? * chapter 3 ) @xmath204 where @xmath205 is the fundamental solution to the helmholtz equation in two dimensions @xmath206 here @xmath207 is a hankel function of the first kind . insertion of ( [ eq : helmrep1 ] ) into ( [ eq : helmbv1 ] ) gives the combined field integral equation @xmath208 where @xmath209 figure [ fig : conv11 ] shows the performance of rcip applied to ( [ eq : dirihelm ] ) for 1000 different values of @xmath210 $ ] . the program demo11.m is used . the boundary @xmath2 is as in ( [ eq : gamma ] ) with @xmath26 and the boundary conditions are chosen as @xmath211 with @xmath212 inside @xmath2 . the error in @xmath184 of ( [ eq : helmrep1 ] ) is evaluated at @xmath213 outside @xmath2 . since the magnitude of @xmath184 varies with @xmath214 , peaking at about unity , the absolute error is shown rather than the relative error . the number of panels on the coarse mesh is chosen as npan=0.6*omega+18 rounded to the nearest integer . the exterior neumann problem for the helmholtz equation @xmath215 can be modeled as an integral equation in several ways . we shall consider two options : an `` analogy with the standard approach for laplace s equation '' , which is not necessarily uniquely solvable for all @xmath214 , and a `` regularized combined field integral equation '' which is always uniquely solvable . see , further , @xcite . let @xmath216 be the adjoint to the double - layer integral operator @xmath217 of ( [ eq : koper ] ) @xmath218 insertion of the integral representation @xmath219 into ( [ eq : helmbv2 ] ) gives the integral equation @xmath220 figure [ fig : conv12 ] shows the performance of rcip applied to ( [ eq : neuhelm1 ] ) . the program demo12.m is used and the setup is the same as that for the dirichlet problem in section [ sec : helmdiri ] . a comparison between figure [ fig : conv12 ] and figure [ fig : conv11 ] shows that the number of gmres iterations needed for full convergence now grows much faster with @xmath214 . furthermore , the relative error in the solution to the neumann problem is larger , particularly so when @xmath214 happens to be close to values for which the operator @xmath221 in ( [ eq : neuhelm1 ] ) has a nontrivial nullspace . recall that ( [ eq : dirihelm ] ) is always uniquely solvable while ( [ eq : neuhelm1 ] ) is not . the literature on regularized combined field integral equations for the exterior neumann problem is rich and several formulations have been suggested . we shall use the representation @xcite @xmath222 which after insertion into ( [ eq : helmbv2 ] ) gives the integral equation @xmath223 where @xmath224 the hypersingular operator @xmath225 of ( [ eq : t ] ) can be expressed as a sum of a simple operator and an operator that requires differentiation with respect to arc length only @xcite @xmath226 this makes it possible to write ( [ eq : neuhelm2 ] ) in the form @xmath227 where @xmath228 , @xmath229 , and the action of the operators @xmath230 , @xmath231 , and @xmath232 is given by @xmath233 all integral operators in ( [ eq : neuhelm3 ] ) are such that their discretizations admit the low - rank decomposition ( [ eq : decomp ] ) . we use the temporary expansion technique of section [ sec : compose ] for ( [ eq : neuhelm3 ] ) , with two new layer densities that are later eliminated , to arrive at a single compressed equation analogous to ( [ eq : disc3 ] ) . that equation involves nine equisized blocks of the compressed inverse @xmath71 . solving the problem in the example of section [ sec : analogy ] again , we now take the number of panels on the coarse mesh as npan=0.6*omega+48 rounded to the nearest integer . figure [ fig : conv13 ] shows results from the program demo13b.m . the resonances , visible in figure [ fig : conv12 ] , are now gone . it is interesting to observe in figure [ fig : conv13 ] that , despite the presence of several singular operators and compositions in ( [ eq : neuhelm3 ] ) , the results produced with rcip are essentially fully accurate and the number of gmres iterations needed for convergence grows very slowly with @xmath214 . the program demo13c.m differs from demo13b.m in that it uses local regularization for the cauchy - singular operators of ( [ eq : c1 ] ) and ( [ eq : c2 ] ) rather than local panelwise evaluation . the results produced by the two programs are virtually identical , so we do not show yet another figure . to the exterior dirichlet helmholtz problem of section [ sec : helmdiri ] evaluated at 62392 points on a cartesian grid using demo11b.m with @xmath234 . the source at @xmath212 is shown as a blue star.,height=321 ] , but the exterior neumann helmholtz problem is solved using demo13d.m . the accuracy is even higher than in figure [ fig : dirifield].,height=321 ] strictly speaking , a boundary value problem is not properly solved until its solution can be accurately evaluated in the entire computational domain . the program demo11b.m is a continuation of demo11.m which , after solving ( [ eq : dirihelm ] ) for @xmath68 with rcip and forming @xmath119 via ( [ eq : wcd ] ) , computes the solution @xmath184 via ( [ eq : helmrep1 ] ) using three slightly different discretizations : * when @xmath12 is away from @xmath2 , 16-point gauss legendre quadrature is used in ( [ eq : helmrep1 ] ) on all quadrature panels . * when @xmath12 is close to @xmath2 , but not close to a panel neighboring a corner , 16-point gauss legendre quadrature is used in ( [ eq : helmrep1 ] ) on panels away from @xmath12 and local panelwise evaluation is used for panels close to @xmath12 . * when @xmath12 is close to a panel neighboring a corner , the density @xmath68 is first used to reconstruct @xmath144 according to section [ sec : recon ] . then 16-point gauss legendre quadrature is used in ( [ eq : helmrep1 ] ) on panels away from @xmath12 and local panelwise evaluation is used for panels close to @xmath12 . the first two discretizations only use the coarse grid on @xmath2 . the third discretization needs a grid on a partially refined mesh on @xmath2 . the program demo13d.m is a continuation of demo13b.m which , after solving ( [ eq : neuhelm2 ] ) with rcip as described in section [ sec : regul ] , computes the solution @xmath184 via ( [ eq : helmrep3 ] ) using the three discretizations of the previous paragraph . figure [ fig : dirifield ] and figure [ fig : neufield ] show that rcip in conjunction with the modified quadrature rules of ( * ? ? ? * section 2 ) is capable of producing very accurate solutions to exterior helmholtz problems in , essentially , the entire computational domain . the idea to make this tutorial came up during a discussion with alex barnett and adrianna gillman of 5/20/12 at the facm12 conference at njit . input and feedback from these colleagues and from rikard ojala has been of great value . the work was supported by the swedish research council under contract 621 - 2011 - 5516 . * * * * * * let @xmath236 and @xmath237 be two column vectors , corresponding to the discretization of two panelwise polynomials with panelwise degree 15 on the coarse mesh of @xmath2 . then @xmath238 because composite 16-point gauss legendre quadrature has panelwise polynomial degree 31 . the diagonal matrix @xmath59 has size @xmath239 . since there are @xmath240 linearly independent choices of @xmath241 and of @xmath237 it follows from ( [ eq : a1 ] ) that @xmath242 which , using ( [ eq : pw ] ) , can be rewritten @xmath243 the compression of ( [ eq : inteq3b ] ) , leading up to ( [ eq : inteq4 ] ) , was originally described in ( * ? ? ? * section 6.4 ) . here we give a summary . the starting point is ( [ eq : inteq3 ] ) which , using the operator split analogous to ( [ eq : split1],[eq : split2 ] ) @xmath244 and the variable substitution @xmath245 gives the right preconditioned equation @xmath246 now , let us take a close look at ( [ eq : inteq4a ] ) . we observe that @xmath247 is an operator whose action on any function gives a function that is smooth on the innermost two panels of the coarse grid on @xmath27 . this is so since @xmath248 is constructed so that its action on any function gives a function that is smooth on the innermost two panels of the coarse grid on @xmath27 . furthermore , the right hand side @xmath249 of ( [ eq : inteq4a ] ) is assumed to be panelwise smooth . using an argument of contradiction we see that @xmath250 has to be a panelwise smooth function on the innermost two panels of the coarse grid on @xmath27 . having concluded that @xmath250 is panelwise smooth close to the corner we can write @xmath251 we also have @xmath252 the discrete version of ( [ eq : subst2 ] ) on the fine grid @xmath253 and the relations ( [ eq : split2 ] ) and ( [ eq : decomp ] ) which we now repeat : @xmath254 substitution of ( [ eq : b1],[eq : b2],[eq : b3],[eq : b4],[eq : b5 ] ) into ( [ eq : inteq3b ] ) , which we now repeat : @xmath255 gives @xmath256 applying @xmath257 ( or @xmath127 ) to the left in ( [ eq : b7 ] ) and using the identities ( [ eq : pwtp ] ) ( or ( [ eq : qp ] ) ) gives the compressed equation ( [ eq : inteq4 ] ) . the recursion ( [ eq : recur ] ) for the rapid construction of the diagonal blocks of the compressed weighted inverse @xmath71 was originally derived in ( * ? ? ? * section 7 ) using different notation and different meshes than in the present tutorial . the recursion was derived a second time in ( * ? ? ? * section 7 ) using new meshes . better notation was introduced in ( * ? ? ? * section 6 ) . a third derivation , in a general setting , takes place in ( * ? ? ? * section 5 ) and it uses the same notation and meshes as in the present tutorial . a problem when explaining the derivation of ( [ eq : recur ] ) is that one needs to introduce intermediate meshes and matrices whose appearance may cause enervation at a first glance . particularly so since these meshes and matrices are not needed in the final expression ( [ eq : recur ] ) . we emphasize that the underlying matrix property that permits the recursion is the low rank of certain off - diagonal blocks in discretizations of @xmath248 of ( [ eq : ko ] ) on nested meshes . for @xmath258 and @xmath28 . the type a mesh has @xmath259 panels . the type b mesh has six panels . the type c mesh has four panels . the type a mesh is the restriction of the fine mesh to @xmath86 . for @xmath260 , the type c mesh is the restriction of the coarse mesh to @xmath27 . the type a mesh and the type b mesh coincide for @xmath261.,height=124 ] the recursion ( [ eq : recur ] ) only uses uses one type of mesh explicitly the type b mesh of figure [ fig : subsets ] . on each @xmath86 there is a type b mesh and a corresponding discretization of @xmath248 denoted @xmath262 . here we need two new types of meshes denoted type a and type c , along with corresponding discrete operators . for example , @xmath263 is the discretization of @xmath22 on a type a mesh on @xmath86 . the three types of meshes are depicted in figure [ fig : types ] . actually , a straight type c mesh was already introduced in figure [ fig : pbc ] . now we define @xmath147 as @xmath264 where @xmath265 and @xmath266 are prolongation operators ( in parameter ) from a grid on a type c mesh on @xmath86 to a grid on a type a mesh on @xmath86 . note that @xmath147 for @xmath267 , according to the definition ( [ eq : ridef ] ) , is identical to the full diagonal @xmath73 block of @xmath71 of ( [ eq : r ] ) . note also that @xmath268 comes cheaply . the rest of this appendix is about finding an expression for @xmath147 in terms of @xmath105 that is cheap to compute . let us split @xmath269 into two parts @xmath270 where @xmath271 and @xmath272 is such that @xmath273 holds to about machine precision , compare ( [ eq : decomp ] ) . the prolongation operators @xmath274 and @xmath275 act from a grid on a type b mesh to a grid on a type a mesh . it holds that @xmath276 summing up , we can rewrite ( [ eq : ridef ] ) as @xmath277 the subsequent steps in the derivation of ( [ eq : recur ] ) are to expand the term within parenthesis in ( [ eq : sumup ] ) in a neumann series , multiply the terms in this series with @xmath278 from the left and with @xmath274 from the right , and bring the series back in closed form . the result is @xmath279^{-1 } { \bf p}_{\rm bc}\ , , \label{eq : result}\ ] ] which , in fact , is ( [ eq : recur ] ) in disguise . to see this , recall from ( [ eq : ridef ] ) that @xmath280 then @xmath281 where the second equality uses @xmath282 , see appendix a. substitution of ( [ eq : final ] ) in ( [ eq : result ] ) gives the recursion in the familiar form @xmath283 in @xcite , bremer describes a scheme that stabilizes the solution to the discretized system ( [ eq : inteq3b ] ) on the fine mesh . the scheme can be interpreted as an inner product preserving discretization . in practice it corresponds to making a similarity transformation of the system matrix . while inner product preserving nystrm discretization elegantly solves problems related to stability ( the condition number of the system matrix is improved ) it does not reduce the number of discretization points ( unknowns ) needed to achieve a given precision in the solution . neither does it affect the spectrum of the system matrix ( similarity transformations preserve eigenvalues ) and hence it does not in any substantial way improve the convergence rate of the gmres iterative method ( * ? ? ? * lecture 35 ) . for completeness , we have implemented inner product preserving nystrm discretization in the program demo1d.m . the program is a continuation of demo1b.m where we also have replaced ( [ eq : inteq1a ] ) with the more stable integral equation ( [ eq : inteq1b ] ) . this should facilitate comparison with the program demo3b.m and the results shown in figure [ fig : conv3 ] . figure [ fig : conv1de ] shows results produced by demo1d.m . beyond @xmath284 one now achieves essentially full machine precision in @xmath24 of ( [ eq : q ] ) . despite this success , inner product preserving nystrm discretization can perhaps not quite compete with the rcip method in this example . the differences in performance relate to issues of memory and speed . the rcip method uses a much smaller linear system ( @xmath240 unknowns ) than does inner product preserving nystrm discretization ( @xmath285 unknowns ) . besides , the rcip method converges in only eight gmres iterations , irrespective of @xmath30 . see figure [ fig : conv3 ] . it is considered difficult to solve integral equations on piecewise smooth boundaries . one reason being that solutions ( layer densities ) often have complicated asymptotics close to corner vertices . also stable numerical schemes , such as that of appendix d , are burdened with the task of resolution . it takes a lot of mesh refinement to resolve non - polynomial - like layer densities with polynomial basis functions . sometimes layer densities diverge , sometimes they stay bounded , but they are seldom polynomial - like close to corner vertices . while we refrain from giving a characterization of what properties of fredholm second kind integral equations give rise to unbounded solutions we observe the following : * the solution @xmath7 to ( [ eq : inteq1a ] ) diverges in corners whenever @xmath286 and @xmath10 is such that @xmath7 exists . figure [ fig : conv67 ] illustrates this for @xmath45 . * the size of the zone where the asymptotic shape of a layer density is visible depends on the smoothness of the solution to the pde that underlies the integral equation . see ( * ? ? ? * figure 3 ) . * the divergence of layer densities close to corner vertices can be very strong . in classical materials science applications it is not unusual with layer densities that barely lie in @xmath287 ( * ? ? ? * section 2 ) . in metamaterial applications layer densities may barely lie in @xmath288 ( * ? ? ? * section 3 ) . furthermore , it is likely that @xmath289 spaces are not the most natural function spaces for the characterization of layer densities ( * ? ? ? * section 5 ) . variable separation is often used to predict the asymptotic behavior of layer densities close to corner vertices . see , for example , ( * ? ? ? * section 2 ) and ( * ? ? ? * section 2 ) . but perhaps asymptotic studies of layer densities are not that important ? assume that there are two different fredholm second kind integral equations for the same underlying pde : one where the layer density is bounded and one where it is unbounded . if the rcip method is to be used , it really does not matter which equation is chosen . the recursion ( [ eq : recur],[eq : rstart ] ) might need fewer steps for the equation with a bounded solution . the achievable accuracy for functionals of the layer density , on the other hand , is often slightly higher if computed from the unbounded solution . the reason being , loosely speaking , that more information can be contained in a rapidly varying function than in a slowly varying function .

quantum escape of a particle via a time - dependent confining potential in a semi - infinite one - dimensional space is discussed . we describe the time - evolution of escape states in terms of scattering states of the quantum open system , and calculate the probability for a particle to remain in the confined region at time in the case of a delta - function potential with a time - oscillating magnitude . the probability decays exponentially in time at early times , then decays as a power later , along with a time - oscillation in itself . we show that a larger time - oscillation amplitude of the confining potential leads to a faster exponential decay of the probability , while it can rather enhance the probability decaying as a power . these contrastive behaviors of the probability in different types of decay are discussed quantitatively by using the decay time and the power decay magnitude of the probability .

1211.5212

quantum escape , as a quantum mechanical leakage of materials from a confined area , is an important concept to study dynamical properties of quantum open systems . it appears in many physical phenomena , such as the @xmath2 decay of a nucleus @xcite , radiative decay of molecules @xcite , transition of states in chemical reactions @xcite , etc . studies on quantum escapes have been done in one - dimensional systems with a stationary potential barrier @xcite , billiard systems with leads @xcite , kicked rotator models @xcite , network systems @xcite , and many - particle systems @xcite , and so on . in escape phenomena , materials continue escaping from a confined area as an irreversible process , so various quantities decay in time . as one of such decaying quantities , the probability for particles to remain in the confined area , which we call the survival probability @xcite in this paper , has been widely investigated in escape phenomena . in quantum escapes , the survival probability decays exponentially in time @xcite , by tunneling via a potential barrier , etc . moreover , it also shows a power decay after a long time @xcite . as decay properties of the survival probability , time scales of its exponential decay @xcite , and changes of the power of decay under different conditions @xcite , etc . have been studied . the principal aim of this paper is to investigate effects of a time - dependent confining potential in quantum escape phenomena . quantum dynamics of systems with a time - periodic potential is analyzed by the floquet theory , etc . @xcite , and recent experimental and theoretical developments in microscopic or mesoscopic systems have pushed to study quantum dynamical behaviors caused by a time - dependent external manipulation @xcite . in this paper , we consider quantum escapes of a particle via a time - periodic confining potential in a semi - infinite one - dimensional space , as a model of a time - dependent manipulation of particle escapes . in the case of a time - dependent delta - function potential , we obtain an analytical form of the quantum propagator to describe a time - evolution of the wave function , leading to a calculation of the survival probability of the system at any time . we show that a time - oscillation of the confining potential has opposite effects in different decay behaviors of the survival probability . in the exponential decay of the survival probability , a larger time - oscillation amplitude of the confining potential leads to a faster decay of the survival probability . in contrast , the time - oscillation of the confining potential can rather enhance the survival probability decaying as a power . we calculate the decay rate of the survival probability and the magnitude of its power decay , by which these reduction and enhancement of the survival probability by a time - oscillation of the confining potential are discussed quantitatively . the system , which we consider in this paper , is a quantum particle system with a time - dependent potential in the semi - infinite one - dimensional space . the wave function @xmath3 of the system at position @xmath4 and time @xmath1 satisfies the schrdingier equation @xmath5 . here , @xmath6\partial^2/\partial x^{2 } + u(x , t)$ ] is the hamiltonian operator with the mass @xmath7 of particle , planck s constant @xmath8 , and the potential @xmath9 at position @xmath4 and time @xmath1 . we impose the hard - wall boundary condition at the end @xmath10 of the semi - infinite one - dimensional space @xmath11 , so that the wave function @xmath3 satisfies the condition @xmath12 at any time @xmath1 . we assume that the potential @xmath9 is a periodic function of time , so that @xmath13 with the time period @xmath14 , in which @xmath15 is a positive constant and @xmath16 is a function of position @xmath4 only ( @xmath17 ) . we impose the condition @xmath18^{*}$ ] for the function @xmath16 of @xmath4 in eq . ( [ tdepenpoten1 ] ) , so that the potential @xmath9 is a real function of @xmath4 and @xmath1 . here , @xmath19 with the asterisk @xmath20 denotes the complex conjugate of @xmath21 for any complex number @xmath21 . we consider the scattering state @xmath22 at position @xmath4 and time @xmath1 , which is induced by the incident plain wave from the infinite region @xmath23 with the energy @xmath24 and the wave number @xmath25 . as a solution of the schrdingier equation with the time - dependent potential ( [ tdepenpoten1 ] ) , the scattering state is represented as @xmath26 where the function @xmath27 of @xmath4 satisfies the time - independent differential equation @xcite as @xmath28 with @xmath29 defined by @xmath30 from eq . ( [ schruequat1 ] ) we derive the lippmann - schwinger equation @xmath31 where @xmath32 is given by @xmath33 as a green function satisfying the differential equation @xmath34 with the boundary condition @xmath35 . here , in the scattering state ( [ scattstate1 ] ) we imposed the condition that if there is no potential energy then the scattering state @xmath22 is represented as the state @xmath36 induced by the incident plain wave with the wave number @xmath25 , with the orthogonal relation @xmath37^{*}\psi_{k'}(x , t)= \delta(k - k')$ ] for @xmath38 and the boundary condo @xmath39=0 . this condition determines the form @xmath40 of the first term in the right - hand side of eq . ( [ lippmschwi1 ] ) . we also defined the square root of @xmath29 as @xmath41 for the case of @xmath42 in eq . ( [ greenfunct1 ] ) . more explanations of eq . ( [ lippmschwi1 ] ) and the green function ( [ greenfunct1 ] ) are given in appendix [ greenfunctionhardwall ] . from now on we especially consider the case of @xmath43 with constants @xmath44 and @xmath45 , satisfying the condition @xmath46^{*}$ ] , @xmath17 . under the condition ( [ deltapoten1 ] ) the potential is represented as @xmath47 , i.e. the delta - function potential with a time - dependent magnitude . by eq . ( [ deltapoten1 ] ) , eq . ( [ lippmschwi1 ] ) is transformed into @xmath48 without any spatial integration . we introduce the matrix @xmath49 with the @xmath50 matrix element @xmath51 using the matrix element ( [ functlambda1 ] ) , eq . ( [ lippmschwi2 ] ) at @xmath52 leads to the relation @xmath53\phi_{k}^{(\mu)}(l ) = \delta_{n0 } \sqrt{\frac{2}{\pi}}\sin(k l)$ ] , namely @xmath54^{(n0 ) } \label{valuephil1}\end{aligned}\ ] ] with the identical matrix @xmath55 . here and hereafter we put @xmath56 as the @xmath50 element of @xmath21 for any matrix @xmath21 in this paper . by inserting eq . ( [ valuephil1 ] ) into eq . ( [ lippmschwi2 ] ) and using eq . ( [ functlambda1 ] ) we obtain the quantity @xmath27 as @xmath57^{(n0 ) } . \nonumber \\ & & \label{lippmschwi3}\end{aligned}\ ] ] the scattering state @xmath22 induced by the incident plain wave with the wave number @xmath25 is given by eq . ( [ scattstate1 ] ) with eq . ( [ lippmschwi3 ] ) . it is essential to note for contents of this paper that the scattering state @xmath22 satisfies the orthogonal relation as @xmath58^{*}\phi_{k'}(x , t ) = \delta(k - k ' ) \label{orthorelat1}\end{aligned}\ ] ] for any time @xmath1 and almost all positive numbers @xmath25 and @xmath59 . the proof and explanations of eq . ( [ orthorelat1 ] ) are given in appendix [ orthogonalrelation ] . in this paper we restrict ourselves our considerations into the quantum state @xmath60 which is expanded by the scattering state @xmath22 as @xmath61 with the coefficient @xmath62 . by using eqs . ( [ orthorelat1 ] ) and ( [ wavefunct1 ] ) , the inner product @xmath63^{*}\psi(x,0 ) $ ] of @xmath64 and @xmath65 leads to the coefficient @xmath62 , namely @xmath66^{*}\psi(x,0 ) . \label{coeffa1}\end{aligned}\ ] ] by inserting eq , ( [ coeffa1 ] ) into eq . ( [ wavefunct1 ] ) , the wave function @xmath60 at position @xmath4 and time @xmath1 is rewritten as @xmath67 where @xmath68 is defined by @xmath69 as the propagator for the time - evolution of the wave function @xmath60 . now we consider the initial condition in which the particle exists only in the finite region @xmath70 at the initial time @xmath71 , namely @xmath72 with a function @xmath73 of @xmath4 . staring from this initial condition , we investigate the probability @xmath0 for a particle in the finite region @xmath70 at time @xmath1 as @xmath74 which we call the survival probability . furthermore , from now on we assume that the initial state is given by the energy eigenstate of the particle confined in the finite region @xmath75 $ ] , namely , @xmath76 for @xmath77 with an integer value @xmath78 . under this specification of the initial conditions ( [ initicondi1 ] ) and ( [ initicondi2 ] ) we can carry out analytically the integral over the position @xmath4 in eq . ( [ surviproba1 ] ) . we give an explicit form of the survival probability @xmath0 in appendix [ survivalprobability ] under this initial condition . as a concrete time - dependence in the potential @xmath9 , in the next subsection [ timedependentpotentialeffects ] we consider a time - oscillating potential , i.e. @xmath79\delta(x - l ) \label{tdepenpoten2}\end{aligned}\ ] ] with real constants @xmath80 and @xmath81 . this potential is derived from @xmath82 for the constants @xmath45 to specify the potential @xmath9 by eqs . ( [ tdepenpoten1 ] ) and ( [ deltapoten1 ] ) . as functions of time @xmath1 for the cases of @xmath83 ( the solid red line ) , @xmath84 ( the dashed green line ) and @xmath85 ( the dotted blue line ) . the main figure is the linear - log plots of @xmath0 , and the inset is their log - log plots as well as the asymptotic power decay @xmath86 corresponding to the case of @xmath85 shown in the straight shin black line . here , and in all figures hereafter , we use dimensionless units with @xmath87 , @xmath88 and @xmath89 . ] in this subsection we show graphs of the survival probabilities @xmath0 as functions of time @xmath1 under the initial conditions ( [ initicondi1 ] ) and ( [ initicondi2 ] ) with the time - dependent potential ( [ tdepenpoten2 ] ) for various values of the parameters @xmath80 and @xmath81 . for calculations of such survival probabilities we use the parameter values as @xmath87 , @xmath90 [ corresponding to the time - period @xmath91 of the potential ( [ tdepenpoten2 ] ) ] , @xmath92 , @xmath88 and @xmath89 . in actual numerical calculations we replaced the integral with respect to the wave number @xmath25 up to @xmath93 by the integral with respect to the wave number @xmath25 up to @xmath94 , checking that the absolute value of the actual integrand is extremely small at @xmath95 so that the contribution of the integral over @xmath96 would be negligible . for numerical calculations of the probability @xmath0 we also have to replace the infinity times infinity matrix @xmath97 with a finite @xmath98 times @xmath98 matrix @xmath99 with @xmath100 and @xmath101 . this implies that we take into account of effects up to the @xmath102-th closest energy scattering states from the state with the wave number @xmath25 for a small magnitude of the time - depending part of the potential ( [ tdepenpoten2 ] ) . numerical calculations of the probability @xmath0 with a large dimensional matrix @xmath99 is very time - consuming , and we used the @xmath103 times @xmath103 matrix @xmath99 , i.e. @xmath104 , for the numerical results in this subsection [ timedependentpotentialeffects ] . for some cases discussed in this subsection we calculated the probabilities @xmath0 not only for @xmath104 and but also for @xmath105 with the same values of the other parameters . differences between these calculated results for @xmath104 and @xmath105 were almost negligible comparing to values of the probabilities @xmath0 themselves , although these results were partly different from the ones for @xmath106 . these facts would justify to calculate the probability @xmath0 by @xmath104 for values of the parameters @xmath80 and @xmath81 , etc . , shown in this subsection [ timedependentpotentialeffects ] . figure [ fig2asurprob1d5 ] is the survival probabilities @xmath0 as functions of time @xmath1 for the cases of @xmath83 ( the solid red line ) , @xmath84 ( the dashed green line ) and @xmath85 ( the dotted blue line ) , as their linear - log plots ( the main figure ) and their log - log plots ( the inset ) @xcite . here , in its inset we added the shin black line showing the asymptotic power decay of the survival probability in time for the time - independent potential with @xmath107 , as given analytically by @xmath108 with @xmath109 $ ] @xcite . it is shown in this figure that the survival probability @xmath0 itself oscillates in time for the case of @xmath110 by a time - oscillation of the potential ( [ tdepenpoten2 ] ) . figure [ fig2asurprob1d5 ] also shows that there are two types of decay behaviors of the survival probability @xmath0 ; one is an exponential decay at early times , and the other is a power decay later , apart from its time - oscillatory behavior . moreover , a comparison of the time - independent potential case of @xmath111 with the cases of @xmath112 and @xmath113 in fig . [ fig2asurprob1d5 ] suggests that the time - oscillation of the potential reduces values of the survival probability @xmath0 decaying exponentially , while it can increase values of the probability @xmath0 decaying as a power . in the following subsections [ exponentialdecaypart ] and [ powerdecaypart ] we will discuss these behaviors caused by a time - oscillating potential in different types of decay of the survival probability separately in more details . as functions of time @xmath1 for the cases of @xmath114 ( the solid red line ) , @xmath115 ( the dashed green line ) , @xmath116 ( the dotted blue line ) , @xmath117 ( the dash - dotted purple line ) and @xmath118 ( the dash - double - dotted cyan line ) . the main figure is the linear - log plots of @xmath0 , and the inset is their linear - linear plots up to the first oscillating period @xmath119 of the potential . ] of the survival probability decaying exponentially as a function of the parameter @xmath81 of the potential @xmath120\delta(x - l)$ ] for the case of @xmath121 . the dashed line is a fitting quadratic curve @xmath122 for the numerical data of @xmath123 with fitting parameters @xmath124 and @xmath125 . ] figure [ fig2bsurprob5d0 ] is the survival probabilities @xmath0 as functions of time @xmath1 for the cases of @xmath114 ( the solid red line ) , @xmath115 ( the dashed green line ) , @xmath116 ( the dotted blue line ) , @xmath117 ( the dash - dotted purple line ) and @xmath118 ( the dash - double - dotted cyan line ) . here , the main figure is the linear - log plots of @xmath0 , and the inset is their linear - linear plots at beginning times up to the first oscillating period @xmath119 of the potential ( [ tdepenpoten2 ] ) . for this figure , we chose a larger value of the parameter @xmath80 than that in fig . [ fig2asurprob1d5 ] , so that the survival probabilities show their exponential decays more clearly . the almost straight lines in the main figure of fig . [ fig2bsurprob5d0 ] as linear - log plots of @xmath0 show that at the presented times the survival probabilities @xmath0 decay exponentially in time , apart from their small time oscillations with the period @xmath126 . these exponential decays continue for a longer time for a larger value of the parameter @xmath80 of the potential ( [ tdepenpoten2 ] ) , suggesting that it is caused by a quantum tunneling via the delta - function potential . the survival probability @xmath0 for @xmath110 oscillates in time with a larger amplitude for a larger time - oscillation amplitude @xmath127 of the potential ( [ tdepenpoten2 ] ) . the survival probabilities with the same absolute values of the parameter @xmath81 , such as @xmath128 or @xmath129 , decay exponentially in time with the same decay rate but with the opposite phase of their time - oscillations . the time - dependent potential barrier magnitude @xmath130 of the potential ( [ tdepenpoten2 ] ) is larger for a larger value of @xmath81 for @xmath131 ( and for @xmath132 ) , so that the survival probability @xmath0 decays slower for a larger value of @xmath81 , as shown in the inset of fig . [ fig2bsurprob5d0 ] . after it , for @xmath133 the potential barrier magnitude becomes to be smaller for a larger value of @xmath81 , so that the survival probability @xmath0 should decay faster for a larger value of @xmath81 at this time . as shown in the inset of fig . [ fig2bsurprob5d0 ] , the magnitude of such an enhancement of the decay speed of the survival probability @xmath0 in @xmath133 is larger than that of its suppression in @xmath131 and @xmath132 for a larger value of @xmath81 . this kind of unbalance of decay speeds of the survival probability @xmath0 for each oscillating period @xmath126 of the potential ( [ tdepenpoten2 ] ) causes a faster decay of the probability @xmath0 on average for a larger value of the time - oscillation amplitude @xmath127 of the potential ( [ tdepenpoten2 ] ) as shown in the main figure of fig . [ fig2bsurprob5d0 ] . in order to discuss time - oscillating effects of the potential ( [ tdepenpoten2 ] ) in exponential decays of the survival probabilities , we fit each exponential decay of the survival probability @xmath0 to an exponential function @xmath134 with the fitting parameters @xmath135 and @xmath123 , and introduce the decay time @xmath123 as its fitting value . figure [ fig2cdecayrate5d0 ] is a graph of such a decay time @xmath123 as a function of the parameter @xmath81 for the case of @xmath121 . this figure suggests that the decay time @xmath123 is a decreasing function of the time - oscillation amplitude @xmath127 of the potential , and is also an even function of @xmath81 . as shown in the dashed line of this figure , the @xmath81-dependence of the decay time @xmath123 is well fitted to a quadratic curve @xmath122 with constants @xmath136 and @xmath137 near @xmath107 , although a deviation from it is recognized for a large value of @xmath127 . as functions of time @xmath1 for the cases of @xmath138 ( the solid red line ) , @xmath139 ( the dashed green line ) , and @xmath140 ( the dotted blue line ) , @xmath141 ( the dash - dotted purple line ) and @xmath142 ( the dash - double - dotted cyan line ) on log - log plots . the straight shin black line shows the asymptotic power decay @xmath86 for the case of @xmath107 . ] of the survival probability decaying as a power as a function of the parameter @xmath81 of the potential @xmath120\delta(x - l)$ ] for the case of @xmath143 on a linear - log plot . the dashed line is a fitting curve given by @xmath144 $ ] for the numerical data of @xmath145 with fitting parameters @xmath146 , @xmath147 and @xmath148 . ] in fig . [ fig2dsurprob0d5 ] we show the graphs of the survival probabilities @xmath0 as functions of time @xmath1 for the cases of @xmath138 ( the solid red line ) , @xmath139 ( the dashed green line ) , @xmath140 ( the dotted blue line ) , @xmath141 ( the dash - dotted purple line ) and @xmath142 ( the dash - double - dotted cyan line ) . this figure is their log - log plots to show their power decays as straight lines , and also shows the asymptotic power decay @xmath86 obtained analytically for the case of @xmath107 . here , we chose a smaller value of the parameter @xmath80 than that in fig . [ fig2asurprob1d5 ] , so that power decay behaviors of the survival probabilities appear more clearly . as shown in fig . [ fig2dsurprob0d5 ] , after a finite time ( approximately @xmath149 ) the survival probabilities @xmath0 decay as a power @xmath150 , apart from their time - oscillations . the survival probability in such a power decay has a tendency to take a large value for a large value of the time - oscillation amplitude @xmath127 of the potential ( [ tdepenpoten2 ] ) , but such increasing rate of the survival probability for @xmath151 is different from that for @xmath152 . the time - oscillation of the potential also causes a time - oscillation of the survival probability @xmath0 , whose amplitude is larger for a larger value of @xmath127 . the phase of time - oscillations of the survival probabilities for @xmath151 is opposite to that for @xmath152 . it may be noted that the time - oscillations of the survival probabilities in their power decays involve clear temporal increases of the value of @xmath0 as the time @xmath1 passes , differently from the ones in their exponential decays . in order to investigate quantitatively enhancement of the survival probabilities by the time - oscillation of the potential ( [ tdepenpoten2 ] ) in their power decays , we fit them to the power decay function @xmath153 with the fitting parameter @xmath145 as the power decay magnitude , apart from their time - oscillations . figure [ fig2epoweramp0d5 ] is a graph of such a fitting value of @xmath145 as a function of the parameter @xmath81 of the potential ( [ tdepenpoten2 ] ) . here , we used the analytical value @xmath154 only for the case of @xmath107 . this figure clearly demonstrates a tendency in which the power decay magnitude @xmath145 of the survival probability decaying as a power takes a large value for a large time - oscillation amplitude @xmath127 of the potential ( [ tdepenpoten2 ] ) . however , it is important to note that the power decay magnitudes @xmath145 is not an even function of @xmath81 , and it takes the minimum value at a non - zero negative value of @xmath81 . to clarify this point , we fitted the numerical data of @xmath145 around its minimum point to the function @xmath144 $ ] with fitting parameters @xmath146 , @xmath147 and @xmath148 . here , we used a quadratic fitting function to the data of @xmath155 rather than of @xmath145 itself for a better fit for a wider values of @xmath81 , and the used values of the fitting parameters are @xmath156 , @xmath157 and @xmath158 . this function fits our numerical results of @xmath145 reasonably well around its minimum point , and suggests that the power decay magnitudes @xmath145 takes its minimum value at @xmath159 approximately , for the case of @xmath143 . in this paper we discussed quantum escapes of a particle via a confining potential with a time - dependent magnitude in a semi - infinite one - dimensional space . we calculated scattering states of the quantum open system , and based on their orthogonal relations we derived the quantum propagator for the time - evolution of wave function of the system with a time - dependent potential . our special interest in such a quantum open system was decay properties of the survival probability , i.e. the probability for a particle to remain in the finite region between the wall and the confining potential . by using the propagator represented by the scattering states , we calculated the survival probability for the case of a delta - function confining potential with a time - oscillating magnitude , and the initial state as the energy eigenstate of a particle confined in the finite region . the survival probability shows two different types of decay behaviors : an exponential decay at early times and a power decay later . the time - oscillation of the potential magnitude reduces a decay time of the survival probability decaying exponentially . in contrast , a large time - oscillation of the confining potential enhances a magnitude of the survival probability decaying as a power . the power decay magnitude of the survival probability is not symmetric under change of the sign of the time - oscillating part of the potential , different from the decay time of the survival probability decaying exponentially . in this paper , we calculated numerically the double integrals over wave numbers for the survival probability . it may be noted that such numerical integrations for a time - dependent potential are much harder than the one for the corresponding time - independent potential because of more complicate time - dependence of the propagator and matrix treatments . this kind of difficulties required us some additional numerical techniques ( such as a calculation of the matrix @xmath160 as solutions of linear algebraic equations , etc . ) , and actually restricted us parameter values under which we could calculate the survival probability within a reasonable calculation time and a numerical accuracy . it would be a future problem to improve further numerical calculation methods , as well as analytical ones , to calculate the survival probability for wide range of values of various parameters . one of the authors ( t.t . ) is grateful to t. okamura for many valuable comments and suggestions on scattering states and decay behaviors in quantum open systems . in this appendix , we explain the lippmann - schwinger equation ( [ lippmschwi1 ] ) with the outgoing green function ( [ greenfunct1 ] ) as the solution of the differential equation ( [ greendiffe1 ] ) . we note that the function @xmath161 of @xmath25 and @xmath162 , proportional to the energy eigenstate with a wave number @xmath25 at position @xmath4 for the case of no potential in the semi - infinite one - dimensional space @xmath163 , satisfies the relation @xmath164 for any real numbers @xmath4 and @xmath165 satisfying the condition @xmath166 . by eq . ( [ deltafunct1 ] ) the green function @xmath32 satisfying eq . ( [ greendiffe1 ] ) , corresponding to outgoing scattering states , is given by @xmath167 we further note the integral formula @xmath168 where we define the square root of @xmath29 as @xmath41 for the case of @xmath42 . by carrying out the integral in eq . ( [ greenfunct2 ] ) using the formula ( [ integformu1 ] ) we obtain eq . ( [ greenfunct1 ] ) . for any function @xmath169 satisfying the equation @xmath170 , the function @xmath171 satisfies eq . ( [ schruequat1 ] ) . by taking @xmath172 in this function @xmath27 as a special case , we obtain eq . ( [ lippmschwi1 ] ) . in this appendix , we give a derivation of eq . ( [ orthorelat1 ] ) on the orthogonalization of the scattering state @xmath22 of the system with the time - dependent potential given by eqs . ( [ tdepenpoten1 ] ) and ( [ deltapoten1 ] ) . as preparations of later calculations we note the mathematical identity @xmath173 for any complex number @xmath174 whose imaginary part is non - negative . here , @xmath175 and @xmath176 are defined by @xmath177 respectively . the function ( [ functdel1 ] ) is an even function of @xmath174 , while the function ( [ functchi1 ] ) is an odd function of @xmath174 . it is also noted that the function @xmath175 becomes the delta function @xmath178 for any real number @xmath179 . we also note @xmath180 for any complex number @xmath174 . we note the relation @xmath181 \nonumber \\ \label{orthobasic1}\end{aligned}\ ] ] for any real numbers @xmath2 and @xmath182 . moreover , by using eqs . ( [ matheident1 ] ) and ( [ matheident2 ] ) we can show @xmath183 -\sin(\alpha l ) \ ; w\!\left(\alpha,\sqrt{\upsilon_{\beta}^{(\nu)}}\right ) , \;\;\ ; \label{orthobasic2}\end{aligned}\ ] ] @xmath184^ { * } g_{\beta}^{(\nu)}(x , l ) \nonumber \\ & & { \hspace{2em}}= \pi e^{i\left[\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } -\sqrt{\upsilon_{\beta}^{(\nu)}}\right]l } \left[g_{\alpha}^{(\mu)}(l , l)\right]^{*}g_{\beta}^{(\nu)}(l , l ) d\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } -\sqrt{\upsilon_{\beta}^{(\nu ) } } \right ) \nonumber \\ & & { \hspace{2em}}{\hspace{2em}}- \left[g_{\alpha}^{(\mu)}(l , l)\right]^ { * } w\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } , \sqrt{\upsilon_{\beta}^{(\nu)}}\right ) - g_{\beta}^{(\nu)}(l , l ) w\!\left(\sqrt{\upsilon_{\beta}^{(\nu ) } } , \left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^{*}\right ) . \;\;\ ; \label{orthobasic3}\end{aligned}\ ] ] here , we used the function @xmath185 of @xmath21 and @xmath186 defined by @xmath187 so that we obtain @xmath188 using the relation @xmath189 . by eq . ( [ scattstate1 ] ) the inner product @xmath190^{*}\phi_{\beta}(x , t ) $ ] of the scattering states is represented as @xmath191^{*}\phi_{\beta}(x , t ) = \sum_{\mu=-\infty}^{+\infty } \sum_{\nu=-\infty}^{+\infty } e^{i[e_{\alpha}-e_{\beta}+(\mu-\nu)\hbar \omega]t/\hbar } \int_{0}^{+\infty } dx \ ; \left[\phi_{\alpha}^{(\mu)}(x)\right]^{*}\phi_{\beta}^{(\nu)}(x ) . \label{innercalcu1}\end{aligned}\ ] ] further , by eqs . ( [ lippmschwi3 ] ) , ( [ orthobasic1 ] ) , ( [ orthobasic2 ] ) and ( [ orthobasic3 ] ) the inner product @xmath192^{*}\phi_{\beta}^{(\nu)}(x)$ ] in the right - hand side of eq . ( [ innercalcu1 ] ) is given by @xmath193^{*}\phi_{\beta}^{(\nu)}(x ) \nonumber \\ & & { \hspace{2em}}= \delta_{\mu 0}\delta_{\nu 0 } \frac{2}{\pi } \int_{0}^{+\infty } dx \ ; \sin(\alpha x)\sin(\beta x ) \nonumber \\ & & { \hspace{2em}}{\hspace{2em}}+ \delta_{\mu 0 } \frac{\left[\lambda_{\beta}(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0)}\sin(\beta l ) } { g_{\beta}^{(\nu)}(l , l ) } \frac{2}{\pi } \int_{0}^{+\infty } dx \ ; \sin(\alpha x)g_{\beta}^{(\nu)}(x , l ) \nonumber \\ & & { \hspace{2em}}{\hspace{2em}}+ \delta_{\nu 0 } \left\{\frac{\left[\lambda_{\alpha}(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0)}\sin(\alpha l ) } { g_{\alpha}^{(\mu)}(l , l ) } \right\}^ { * } \frac{2}{\pi}\int_{0}^{+\infty } dx \ ; \left[g_{\alpha}^{(\mu)}(x , l)\right]^ { * } \sin(\beta x ) \nonumber \\ & & { \hspace{2em}}{\hspace{2em}}+ \left\{\frac{\left[\lambda_{\alpha}(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0)}\sin(\alpha l ) } { g_{\alpha}^{(\mu)}(l , l ) } \right\}^ { * } \frac{\left[\lambda_{\beta}(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0)}\sin(\beta l ) } { g_{\beta}^{(\nu)}(l , l ) } \frac{2}{\pi}\int_{0}^{+\infty } dx \ ; \left[g_{\alpha}^{(\mu)}(x , l)\right]^ { * } g_{\beta}^{(\nu)}(x , l ) \nonumber \\ & & { \hspace{2em}}= \delta_{\mu 0}\delta_{\nu 0 } \left [ d ( \alpha -\beta ) - d ( \alpha + \beta)\right ] + \sin(\alpha l)\sin(\beta l)\left({\mathcal{a}}_{(\alpha,\beta)}^{(\mu,\nu ) } + \frac{2}{\pi } { \mathcal{b}}_{(\alpha,\beta)}^{(\mu,\nu)}\right ) \label{innercalcu2}\end{aligned}\ ] ] where @xmath194 and @xmath195 are defined by @xmath196l } \left\{\left[\lambda_{\alpha}(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0)}\right\}^ { * } \left[\lambda_{\beta}(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0 ) } d\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } -\sqrt{\upsilon_{\beta}^{(\nu ) } } \right ) \nonumber \\ & & { \hspace{2em}}+ i \delta_{\mu 0 } \frac{e^{-i\sqrt{\upsilon_{\beta}^{(\nu)}}l}}{\sin(\alpha l ) } \left[\lambda_{\beta}(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0 ) } \left[d\!\left(\alpha-\sqrt{\upsilon_{\beta}^{(\nu)}}\right ) - d\!\left(\alpha+\sqrt{\upsilon_{\beta}^{(\nu)}}\right ) \right ] \nonumber \\ & & { \hspace{2em}}- i\delta_{\nu 0 } \frac{e^{i\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^{*}l}}{\sin(\beta l ) } \left\{\left[\lambda_{\alpha}(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0 ) } \right\}^ { * } \left[d\!\left(\beta-\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^{*}\right ) - d\!\left(\beta+\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^{*}\right ) \right ] , \nonumber \\ \label{functa1 } \\ { \mathcal{b}}_{(\alpha,\beta)}^{(\mu,\nu ) } & \equiv & -\left\{\left[\lambda_{\alpha}(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0 ) } \right\}^ { * } \frac{\left[\lambda_{\beta}(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0 ) } } { g_{\beta}^{(\nu)}(l , l ) } w\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } , \sqrt{\upsilon_{\beta}^{(\nu)}}\right ) \nonumber \\ & & { \hspace{2em}}- \left\{\frac{\left[\lambda_{\alpha}(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0 ) } } { g_{\alpha}^{(\mu)}(l , l ) } \right\}^ { * } \left[\lambda_{\beta}(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0 ) } w\!\left(\sqrt{\upsilon_{\beta}^{(\nu ) } } , \left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^{*}\right ) \nonumber \\ & & { \hspace{2em}}-\delta_{\mu 0 } \frac{\left[\lambda_{\beta}(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0 ) } } { g_{\beta}^{(\nu)}(l , l ) } w\!\left(\alpha,\sqrt{\upsilon_{\beta}^{(\nu)}}\right ) - \delta_{\nu 0 } \left\{\frac{\left[\lambda_{\alpha}(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0 ) } } { g_{\alpha}^{(\mu)}(l , l ) } \right\}^ { * } w\!\left(\beta,\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^{*}\right ) . \nonumber \\ \label{functb1}\end{aligned}\ ] ] here , the quantity @xmath194 includes the type of @xmath197 and the quantity @xmath195 includes the terms with the function @xmath198 . now , we note that for the function ( [ functdel1 ] ) we have @xmath199 for any function @xmath200 of complex number @xmath174 because of @xmath201 for any @xmath202 . noting eq . ( [ dfunctident1 ] ) we obtain @xmath203^{*}g_{\beta}^{(\nu)}(l , l ) d\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } -\sqrt{\upsilon_{\beta}^{(\nu ) } } \right ) \nonumber \\ & & { \hspace{2em}}= \left\ { \frac{\left[g_{\alpha}^{(\mu)}(l , l ) \right]^{*}}{2i\sqrt{\upsilon_{\beta}^{(\nu ) } } } - \frac{g_{\beta}^{(\nu)}(l , l)}{2i\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } } \right\ } d\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } -\sqrt{\upsilon_{\beta}^{(\nu ) } } \right ) \label{greendeltasup2a } \\ & & { \hspace{2em}}= -i\left\{\left[g_{\alpha}^{(\mu)}(l , l ) \right]^ { * } - g_{\beta}^{(\nu)}(l , l ) \right\ } d\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } , \sqrt{\upsilon_{\beta}^{(\nu ) } } \right ) . \label{greendeltasup2b } \end{aligned}\ ] ] here , we used the relation @xmath204^ { * } - g_{\beta}^{(\nu)}(l , l ) \right\ } d\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } + \sqrt{\upsilon_{\beta}^{(\nu ) } } \right ) = 0 \label{greendeltaident1}\end{aligned}\ ] ] to derive eq . ( [ greendeltasup2b ] ) from eq . ( [ greendeltasup2a ] ) by eqs . ( [ greenfunct1 ] ) and ( [ dfunctident1 ] ) , and also the function @xmath205 of @xmath21 and @xmath186 defined by @xmath206 and for eq . ( [ greendeltasup1 ] ) we adopt the convention that any equation containing the symbols @xmath207 on the left- and right - hand sides , denote _ two _ equations , one with only the upper symbol ( @xmath208 in @xmath207 ) and the other with only the lower symbol ( @xmath209 in @xmath207 ) . moreover , we note the relation @xmath210^{(\mu 0 ) } = -\delta_{\mu 0 } + \left[(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0 ) } . \label{identappen1}\end{aligned}\ ] ] by eqs . ( [ greendeltasup1 ] ) , ( [ greendeltasup2b ] ) and ( [ identappen1 ] ) the quantity ( [ functa1 ] ) is rewritten as @xmath211^{(\mu 0)}\right\}^ { * } \frac{\left[\lambda_{\beta}(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0 ) } } { g_{\beta}^{(\nu)}(l , l ) } \right.\nonumber \\ & & { \hspace{2em}}{\hspace{2em}}{\hspace{2em}}\left . - \left\{\frac{\left[\lambda_{\alpha}(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0 ) } } { g_{\alpha}^{(\mu)}(l , l)}\right\}^ { * } \left[(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0)}\right\ } d\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } , \sqrt{\upsilon_{\beta}^{(\nu ) } } \right ) \nonumber \\ & & { \hspace{2em}}+ 2i\delta_{\nu 0 } \left\{\frac{\left[\lambda_{\alpha}(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0 ) } } { g_{\alpha}^{(\mu)}(l , l)\sqrt{\upsilon_{\alpha}^{(\mu)}}}\right\}^ { * } d\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^{*}+\beta\right ) \label{functbsup1}\end{aligned}\ ] ] for any positive numbers @xmath2 and @xmath182 . here , we used the relation @xmath212 for any positive number @xmath2 . by eq . ( [ identappen1 ] ) we also rewrite the quantity ( [ functb1 ] ) as @xmath213^{(\mu 0 ) } \right\}^ { * } \frac{\left[\lambda_{\beta}(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0 ) } } { g_{\beta}^{(\nu)}(l , l ) } w\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } , \sqrt{\upsilon_{\beta}^{(\nu)}}\right ) \nonumber \\ & & { \hspace{2em}}- \left\{\frac{\left[\lambda_{\alpha}(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0 ) } } { g_{\alpha}^{(\mu)}(l , l ) } \right\}^ { * } \left[(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0 ) } w\!\left(\sqrt{\upsilon_{\beta}^{(\nu ) } } , \left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^{*}\right ) \label{functasup1}\end{aligned}\ ] ] for any positive numbers @xmath2 and @xmath182 . by the definition of @xmath214 the real part of @xmath214 is non - negative , so that we have @xmath215 , as well as @xmath216 and @xmath217 , for any positive numbers @xmath2 and @xmath182 . by these facts and eqs . ( [ innercalcu2 ] ) , ( [ functbsup1 ] ) and ( [ functasup1 ] ) we obtain @xmath218^{*}\phi_{\beta}^{(\nu)}(x ) \nonumber \\ & & { \hspace{2em}}= \delta_{\mu 0}\delta_{\nu 0 } \delta ( \alpha -\beta ) \nonumber \\ & & { \hspace{2em}}{\hspace{2em}}-\frac{2i}{\pi } \sin(\alpha l)\sin(\beta l ) \bigg\{\left\{\left[(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0)}\right\}^ { * } \frac{\left[\lambda_{\beta}(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0)}}{g_{\beta}^{(\nu)}(l , l ) } \nonumber \\ & & { \hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}\left.\times \left[\pi d\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } , \sqrt{\upsilon_{\beta}^{(\nu ) } } \right ) - i w\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } , \sqrt{\upsilon_{\beta}^{(\nu)}}\right ) \right ] \right.\nonumber \\ & & { \hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}\left . - \left\{\frac{\left[\lambda_{\alpha}(i-\lambda_{\alpha})^{-1 } \right]^{(\mu 0 ) } } { g_{\alpha}^{(\mu)}(l , l ) } \right\}^ { * } \left[(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0 ) } \right.\nonumber \\ & & { \hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}\times \left[\pi d\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } , \sqrt{\upsilon_{\beta}^{(\nu ) } } \right ) + iw\!\left(\sqrt{\upsilon_{\beta}^{(\nu ) } } , \left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^{*}\right ) \right]\bigg\ } \label{innercalcu3}\end{aligned}\ ] ] for any positive numbers @xmath2 and @xmath182 . for the quantity @xmath219 we have @xmath220 where we used eqs . ( [ upsi1 ] ) , ( [ functdel1 ] ) and ( [ functd1 ] ) , the relation @xmath221 , @xmath222 , @xmath223/(2|y|)$ ] and @xmath224 for any real number @xmath4 and @xmath165 @xmath225 . by eqs . ( [ upsi1 ] ) , ( [ functchi1 ] ) and ( [ functw1 ] ) we further note @xmath226 for any numbers @xmath2 and @xmath182 except in the zero - measure region satisfying the condition @xmath227 , i.e. for almost all numbers @xmath2 and @xmath182 . by introducing the quantity @xmath228 as @xmath229 , \label{functc1}\end{aligned}\ ] ] and by eqs . ( [ functlambda1 ] ) , ( [ innercalcu1 ] ) , ( [ innercalcu3 ] ) , ( [ functdprope1 ] ) , ( [ functwprope1 ] ) and @xmath230^ { * } = \lambda^{(-n)}$ ] , we obtain @xmath231^{*}\phi_{\beta}(x , t ) \nonumber \\ & & { \hspace{2em}}= \delta ( \alpha -\beta ) -\frac{4 m i}{\pi \hbar^{2 } } \sin(\alpha l)\sin(\beta l ) \sum_{n=-\infty}^{+\infty}\left\{\left[(i-\lambda_{\alpha})^{-1 } \right]^{(n 0)}\right\}^ { * } \nonumber \\ & & { \hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}\times \sum_{\mu=-\infty}^{+\infty } \sum_{\nu=-\infty}^{+\infty } \bigg\ { c_{\alpha\beta}^{(n-\mu)}(t ) \lambda^{(-(\mu-\nu))}\left[(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0 ) } \nonumber \\ & & { \hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}{\hspace{2em}}- \lambda^{(-(n-\mu ) ) } c_{\alpha\beta}^{(\mu-\nu)}(t ) \left[(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0)}\bigg\ } \label{innercalcu4 } \end{aligned}\ ] ] for almost all positive numbers @xmath2 and @xmath182 . now , we introduce the discretized fourier transformation @xmath232 as @xmath233 for any function @xmath234 of integer @xmath235 , noting the formula @xmath236 for any integer @xmath235 . by using this transformation and the formula @xmath237 for any @xmath238 $ ] , the discretized fourier transformation @xmath239 of the convolution @xmath240 for any functions @xmath234 and @xmath241 of @xmath235 is represented as @xmath242 for @xmath243 , as known as the convolution formula of the fourier transformation . by using this property we have @xmath244^{(\nu 0 ) } = \sum_{\mu=-\infty}^{+\infty } \sum_{\nu=-\infty}^{+\infty } \lambda^{(-(n-\mu ) ) } c_{\alpha\beta}^{(\mu-\nu)}(t ) \left[(i-\lambda_{\beta})^{-1 } \right]^{(\nu 0 ) } \label{fouritransident1}\end{aligned}\ ] ] because the discretized fourier transformation of the left - hand side of eq . ( [ fouritransident1 ] ) is equal to the one of the right - hand side of eq . ( [ fouritransident1 ] ) . by inserting ( [ fouritransident1 ] ) into eq . ( [ innercalcu4 ] ) we obtain eq . ( [ orthorelat1 ] ) for almost all positive numbers @xmath2 and @xmath182 . in this appendix we give an explicit form of the survival probability @xmath0 under the initial conditions ( [ initicondi1 ] ) and ( [ initicondi2 ] ) . @xmath247^ { * } g_{\beta}^{(\nu)}(x , l ) = \frac{l e^{-i\left[\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } -\sqrt{\upsilon_{\beta}^{(\nu)}}\right]l } } { 2\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^{*}\sqrt{\upsilon_{\beta}^{(\nu ) } } } { \mathcal{s}}\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } l , \sqrt{\upsilon_{\beta}^{(\nu)}}l\right ) \label{coeffaint3}\end{aligned}\ ] ] where @xmath248 is the function of @xmath21 and @xmath186 , defined by @xmath249 \label{functs1}\end{aligned}\ ] ] and satisfies the relation @xmath250 . by using eqs . ( [ scattstate1 ] ) , ( [ lippmschwi3 ] ) , ( [ coeffaint1 ] ) and ( [ coeffaint2 ] ) , for the initial state given by eqs . ( [ initicondi1 ] ) and ( [ initicondi2 ] ) the coefficient ( [ coeffa1 ] ) is represented as @xmath251^ { * } , \label{coeffa2}\end{aligned}\ ] ] where we define the function @xmath252 of @xmath25 , @xmath235 and @xmath1 by @xmath253^{(n0)}\sin(kl ) } { g_{k}^{(n)}(l , l)\sqrt{\upsilon_{k}^{(n ) } } } e^{i \left(\sqrt{\upsilon_{k}^{(n)}}l - n\omega t\right ) } . \label{functr1}\end{aligned}\ ] ] by eqs . ( [ scattstate1 ] ) , ( [ lippmschwi3 ] ) , ( [ wavefunct1 ] ) , ( [ coeffaint1 ] ) , ( [ coeffaint2 ] ) , ( [ coeffaint3 ] ) , ( [ coeffa2 ] ) and ( [ functr1 ] ) , the survival probability ( [ surviproba1 ] ) is represented as @xmath254 where @xmath255 is defined by @xmath256^ { * } + \sum_{\nu=-\infty}^{+\infty } { \mathcal{r}}(\beta,\nu;t ) { \mathcal{s}}\!\left(\alpha l,\sqrt{\upsilon_{\beta}^{(\nu)}}l\right ) \nonumber \\ & & { \hspace{2em}}+ \sum_{\mu=-\infty}^{+\infty } \sum_{\nu=-\infty}^{+\infty } \left[{\mathcal{r}}(\alpha,\mu;t ) \right]^ { * } { \mathcal{r}}(\beta,\nu;t ) { \mathcal{s}}\!\left(\left(\sqrt{\upsilon_{\alpha}^{(\mu)}}\right)^ { * } l , \sqrt{\upsilon_{\beta}^{(\nu)}}l\right ) \biggr\ } e^{i(e_{\alpha}-e_{\beta})t/\hbar } . \label{functbsurpro1}\end{aligned}\ ] ] r. g. winter , phys . rev . * 123 * , 1503 ( 1961 ) . w. van dijk and y. nogami , phys . lett . * 83 * , 2867 ( 1999 ) ; phys . c * 65 * , 024608 ( 2002 ) . g. garca - caldern , i. maldonado , and j. villavicencio , phys . a * 76 * , 012103 ( 2007 ) . g. casati , g. maspero , and d. l. shepelyansky , phys . e * 56 * , r6233 ( 1997 ) . k. m. frahm , phys . e * 56 * , r6237 ( 1997 ) . d. v. savin and v. v. sokolov , phys . e * 56 * , r4911 ( 1997 ) ; phys . e * 57 * , 7367 ( 1998 ) [ errata ] . f. barra and p. gaspard , phys . e * 65 * , 016205 ( 2001 ) . m. puhlmann , h. schanz , t. kottos and t. geisel , europhys . 69 * , 313 ( 2005 ) . t. taniguchi and s. sawada , phys . a * 84 * , 062707 ( 2011 ) . a. del campo , f. delgado , g. garca - caldern , j. g. muga , and m. g. raizen , phys . a * 74 * , 013605 ( 2006 ) . t. taniguchi and s. sawada , phys . e * 83 * , 026208 ( 2011 ) . a. del campo , phys . a * 84 * , 012113 ( 2011 ) . d. waltner , _ semiclassical approach to mesoscopic systems : classical trajectory correlations and wave interference _ , ( springer - verlag , berlin , 2012 ) . l. a. bunimovich and c. p. dettmann , europhys . lett . * 80 * , 40001 ( 2007 ) . dittes , h. l. harney , and a. mller , phys . rev . a * 45 * , 701 ( 1992 ) ; h. l. harney , f. -m . dittes , and a. mller , ann . * 220 * , 159 ( 1992 ) . m. miyamoto , phys . a * 68 * , 022702 ( 2003 ) . our numerical calculations confirm the condition @xmath257 for the initial conditions ( [ initicondi1 ] ) and ( [ initicondi2 ] ) . this is a supportive evidence that the initial state described by eqs . ( [ initicondi1 ] ) and ( [ initicondi2 ] ) is expanded by the scattering states @xmath258 , namely the condition ( [ wavefunct1 ] ) .

we present an analysis of the stellar populations as a function of mass in a sample of galaxies of all morphological and emission types from the 6df galaxy survey ( 6dfgs ) . we measure velocity dispersions and lick indices from the spectra of the central regions of these galaxies , deriving ages and metallicities from the lick indices using stellar population models . we also derive dynamical masses and dynamical mass - to - light ratios for these galaxies by combining the central velocity dispersions with global photometry in the b , r and k bands from supercosmos and 2mass . together , these data allow us to reduce the degeneracies between age , metallicity and star formation burst - strength that have limited previous studies . we find that old galaxies exhibit a mass - metallicity relation with logarithmic slope d[fe / h]/dlog , while young galaxies show slopes consistent with zero . when we account for the effects of the mass - metallicity relation , we obtain a single , consistent relation between mass - to - light ratio and mass for old galaxies in all passbands , . as we have accounted for stellar population effects , this remaining variation in the mass - to - light with mass ( the residual ` tilt ' of the fundamental plane ) must have a dynamical origin . however , we demonstrate that any simple trend between mass - to - light - ratio and mass or luminosity is inconsistent with the observations , and that a more complex relationship must exist . we find that the central regions of galaxies of all masses often exhibit young stellar populations . however it is only in the lowest - mass galaxies studied ( ) that these populations are evident in the global photometry . in higher - mass galaxies , young central populations have decreasing influence on the global photometry , with there being no discernible impact in galaxies more massive than . we conclude that the young stellar populations detected in spectroscopic studies are generally centrally concentrated , and that there is an upper limit on the mass of star - forming events in massive galaxies . these results have important ramifications for mass - to - light ratios estimated from photometric observations . latexl-.36em.3ex-.15em t-.1667em.7ex-.125emx [ section ] [ firstpage ] galaxies : general , galaxies : stellar content , galaxies : kinematics and dynamics

0803.4032

current models of galaxy formation and assembly can be characterised by two extreme and competing views . one such view is represented by the early `` monolithic collapse '' model , whereby the spheroidal ( bulge ) components of galaxies form by the early collapse of individual gas clouds ( e.g. larson 1974 ; carlberg 1984 ) . in these early collapse models the star formation induced by the collapse is terminated when the source of gas is either consumed or expelled from the galaxy . the bulge then ages passively ( albeit perhaps subject to the later addition of a disk ) . such models are largely motivated by the small scatter in the observed relationships between colour and magnitude ( e.g. baum 1959 ; bower , lucy & ellis 1992a , b ) and velocity dispersion , surface brightness and size ( the fundamental plane ; dressler et al . 1987 ) for early - type galaxies and bulges . to explain the tightness of these relations , early collapse models require that star formation in bulges terminates at early times ( z@xmath52 ; bower , lucy & ellis 1992a , b ) , so that bulge dominated ( e.g. elliptical ) galaxies should be uniformly old . this picture faces challenges on a number of fronts . for instance , the differing slopes found for the fundamental plane in various bands ( e.g. bernardi et al . 2003b ) are difficult to explain in the monolithic collapse scenario . furthermore , spectroscopic studies of the stellar populations in the bulges of galaxies find central ages ranging from @xmath0115 gyr ( e.g. trager et al . 2000 ; proctor & sansom 2002 ; terlevich & forbes 2002 : caldwell , rose & concannon 2003 ; kauffmann et al . 2003a , b ; proctor et al . 2004b ) , indicating evidence for more recent star formation . however , by far the most important challenge to the early collapse picture is the fundamental differences in galaxy formation and assembly histories that are predicted when the growth of structure via gravitational instability is modelled for a dark matter - dominated universe , now characterised extremely well via precision measurements of the cosmological parameters ( e.g. the @xmath6cdm model ; spergel et al . 2003 ) . such `` hierarchical '' models , which represent the other extreme view , predict galaxy - mass objects to be assembled by the successive merging of lower - mass objects via hierarchical merging ( white & rees 1978 ) , usually over an extended period . another important prediction of hierarchical merging models is that the merger rate of galaxies is a function of environment , i.e. on average , galaxies in the dense environs of clusters merge early , while those in the less dense ` field ' merge later ( kauffmann 1996 ) . in the hierarchical merging scenario the properties of galaxies are therefore subject to strong influences from their environment . results from stellar population studies appear to confirm the prediction of hierarchical merging models in that ` field ' galaxies possess slightly younger central ages than their cluster counterparts ( e.g. thomas et al . , 2005 ; bernardi et al . , 2006 ; smith et al . , 2006 ) . in contrast to early collapse models , hierarchical merging models predict galaxy bulges to possess a range of ages , with low - mass galaxies forming earlier , and on shorter timescales , than high - mass galaxies . at first glance , the wide range of ages observed in spectroscopic studies might also appear to confirm the hierarchical merging predictions , however the observed trend of age with mass is for lower - mass systems to be _ younger _ than higher - mass systems . this is inconsistent with hierarchical merging models which predict lower - mass systems to form earlier than high - mass system ( kauffmann 1996 ) . the observed trend is therefore sometimes referred to as ` anti - hierarchical ' . the recent de lucia et al . ( 2006 ) analysis of the millennium simulation provides insights important to the development of a coherent picture of galaxy evolution . the millennium simulation models the evolution of dark matter using an n - body code to which de lucia et al . couple a semi - analytic model of galaxy formation . de lucia et al s analysis clearly demonstrates that a strong distinction must be made between the _ mass assembly _ of galaxies and the _ formation _ of their stellar populations . this can be understood intuitively by the simple realisation that mergers bring old stars into a merger remnant as well as making new stars . merger remnants may therefore be , and often are , still dominated ( in terms of their mass ) by old stellar populations . de lucia et al . conclude that while the mass assembly of galaxies is indeed hierarchical , their star formation histories are nevertheless anti - hierarchical . despite this improvement in our theoretical understanding , the problem remains how to reconcile the broad range of predicted ( and spectroscopically observed ) galaxy ages with the extremely small scatter in some scaling relations . recent surveys of large numbers of nearby galaxies have provided statistically significant results and hence further observational constraints on galaxy formation models . such surveys include the 2df galaxy redshift survey ( 2dfgrs : colless et al . 2001b ) and the sloan digital sky survey ( sdss ; york et al . 2000 ) . from central region spectra and global optical colours the 2dfgrs probed the cosmic star formation history at relatively low redshift ( baldry et al . 2002 ) and its variation with local environmental density ( lewis et al . the sdss also obtained optical colours and central spectra for a large number of low redshift galaxies . this ongoing dataset has been used by several authors to study the star formation history and scaling relations of nearby galaxies . these studies include bernardi et al . ( 2003b ) who investigated scaling relations for @xmath09,000 galaxies , eisenstein et al . ( 2003 ) who combined spectra of 22,000 massive galaxies , gallazzi et al . ( 2005 ) who probed the star formation history of @xmath040,000 galaxies , cid fernandes et al . ( 2005 ) who conducted a spectral synthesis of @xmath050,000 galaxies , and chang et al . ( 2006 ) who combined spectra and optical colours with 2mass near - infrared colours for @xmath03,000 galaxies . such large datasets have allowed these authors to investigate trends as a function of mass , environment , galaxy type etc . already , a galaxy mass - metallicity relation has been detected in the metallicities of both gas ( tremonti et al . 2004 ) and stars ( gallazzi et al . 2005 ) and strong trends with galaxy environment identified ( lewis et al . 2002 ) . however , the above studies have not explicitly explored the link between the _ central _ stellar populations , which they observe , and the _ global _ photometry of the galaxies in their samples . it is this issue that this paper specifically addresses . to this end , we compare the results of stellar population analysis of spectra from the 6df galaxy survey ( 6dfgs ; jones et al . 2004 ) to the global near - infrared photometry of the 2mass extended source catalog ( 2mass xsc ; jarrett et al . 2000a , b ) and optical supercosmos ( hambly , irwin & macgillivray . 2001 ) data for several thousand galaxies . we explore such issues as the fraction of galaxies stellar populations that can have been formed in recent star formation events and the effects of stellar populations on the slopes of colour - magnitude relations and the varying ` tilt ' of the fundamental plane . the paper is laid out as follows . in the next section we outline the photometric and spectroscopic data . in section [ spectroscopicanal ] we detail the spectroscopic data reductions and analysis used to derive stellar population parameters . section [ results ] presents the results of our analysis , which are discussed and summarised in section [ concs ] . the 6dfgs is a spectroscopic survey of the entire southern sky more than 10 degrees from the galactic plane . the survey measured redshifts for more than 124,000 galaxies . the primary survey targets were selected primarily from the 2mass extended source catalog ( xsc ) . all galaxies brighter than k@xmath7=12.75 were included in the sample . secondary samples of 2mass and supercosmos galaxies , complete down to magnitude limits ( h , j , r@xmath8 , b@xmath9 ) = ( 13.00 , 13.75 , 15.60 , 16.75 ) , were also included . the survey was carried out on the uk schmidt telescope using the six - degree field ( 6df ) multi - fibre spectrograph . the 6df instrument simultaneously observed up to 150 spectra over a 5.7@xmath10 field . the science fibres of the instrument were 6.7 arcsec in diameter . the spectrograph is bench mounted in an enclosure inside the telescope dome , which gives it increased instrumental stability compared to a spectrographs mounted on a telescope ( e.g. 2df and the sdss spectrograph ) . this increased instrumental stability maintains consistent instrumental resolution , which is important when working with lick indices . the survey obtained spectra in the visual ( v ) and red ( r ) wavelength ranges providing coverage from @xmath04000 to @xmath08400 . the spectra have 5 - 6 fwhm resolution in v and 9 - 12 resolution in r ( see jones et al . 2004 for more details ) . the main science aims of the survey included : to measure the luminosity function of near - infrared selected galaxies as a function of environment and galaxy type ( jones et al . 2006 ) , to measure the clustering of galaxies , to produce detailed maps of the density and peculiar velocity fields of the nearby universe ( jones et al . 2004 ; erdogdu et al . 2006 ) and to study the properties of the underlying stellar population of the galaxies such as ages and chemical abundances . it is this latter issue to which this paper is addressed . the 6dfgs is particularly suited to the aims of the current study because the selection criteria using 2mass near - infrared photometry provides a sample less biased toward bright , young galaxies than surveys based on optical catalogues . the k band photometry also has the useful property of being relatively insensitive to both metallicity the presence of dust ( unlike optical bands ) , and is therefore an good tracer of mass . it also provides a useful tool in breaking the age metallicity degeneracy . finally , 6dfgs spectra are obtained from region 2 to 3 times larger on the sky than those used in the 2dfgrs and sdss surveys . as a result , at the mean redshifts of the surveys , 2df , sdss and 6dfgs fibres correspond to 2.6 , 3.9 and 4.6 kpc / h respectively . the 6dfgs spectra therefore sample a significantly larger fraction of the total galaxy light than either 2df or sdss spectra . the spectroscopic data presented here are from the 6dfgs first data release ( jones et al . lick index , velocity dispersion and emission line - strength measurements were made on all galaxies with redshift quality @xmath11 ( the ` reliable ' redshifts ; jones et al . this gave a total of 39649 galaxies with 3859 of these observed more than once ( 420 galaxies observed more than twice ) . if a galaxy was observed more than once , only the highest signal - to - noise - ratio spectrum was used for the measurements . however , repeat observations were used to establish the errors ( see section [ error_measurement ] ) . only galaxies with median signal - to - noise ratios @xmath512 @xmath12 were deemed suitable for age and metallicity determinations . such a value lies at the lower limit of signal - to - noise - ratio that can be used to derive ages and metallicities with precision . the above selection criteria resulted in @xmath07000 galaxies suitable for lick index measurements . it should be noted that , as no morphological selection was made , the results presented here include galaxies of the full range of early and late types . + the k band magnitudes used in this work were taken to be the k@xmath7 values estimated from the 2mass data by jones et al . we also use the re - calibrated supercosmos b and r band magnitudes ( hambly et al . 2001 ) . galaxy radii were taken as the 2mass 20th magnitude per arcsec@xmath13 isophotal radius ( r@xmath14 ; jarrett et al . we also use k band half - light ( effective ) radii ( r@xmath15 ) for a spectroscopically selected early - type galaxy subsample , as supplied by lachlan campbell ( private communication ) . absolute magnitudes and physical galaxy radii were calculated using distances based on the measured redshifts and a hubble constant of 70 km s@xmath12 mpc@xmath12 ( our median redshift is 0.035 ) . errors on the distance estimates were assumed to be a combination of recession velocity uncertainty and peculiar motions . in order to be conservative , we assumed the combined uncertainty to be @xmath161000 km s@xmath12 . the distributions of some key parameters in the final sample are shown in fig . [ sample ] . the sample covers of order 5 magnitudes in luminosity ( @xmath17m@xmath18@xmath19 mag and @xmath20m@xmath21@xmath22 mag ) . this corresponds to a dynamical mass range of more than two orders of magnitude ( from @xmath2310@xmath24 to @xmath2510@xmath26 m@xmath27 ; section [ photo ] ) . from the sample of @xmath07000 galaxies suitable for age metallicity determination ( section [ specsample ] ) @xmath06000 also had a full set of 2mass and supercosmos photometry . 7000 galaxies presented here . the signal - to - noise - ratio per is estimated as the median of the 6dfgs spectra . apparent magnitudes are from 2mass and supercosmos data . absolute magnitudes were calculated using redshifts from the 6dfgs spectra and a hubble constant of 70 km s@xmath12mpc@xmath12 . k band 20th magnitude isophotal radius is from 2mass . effective radii are also presented for a ( @xmath025% ) subsample of our 6dfgs galaxies . dashed lines in the radius plots represent the radius of the 6df aperture ( 3.35 arcsec ) . , width=302 ] we measured emission line - strengths for h@xmath28 , h@xmath29 , [ nii]@xmath306548 , [ nii]@xmath306584 , [ oiii]@xmath304959 and [ oiii]@xmath305007 from the 6dfgs spectra . in order to measure the h@xmath31 and [ nii ] emission , a de - blending option in the iraf task _ splot _ this option allows the fitting of gaussians to a series of spectral lines and the measurement of the value of the equivalent widths of each . the strengths of h@xmath31 and the two [ nii ] doublet lines either side of it were measured in this way from continuum divided spectra . this process can lead to false measurements where noise spikes or the continuum level are fit with gaussian profiles rather than real features . by consideration of the position and fwhm of the gaussian fit to the feature , one can reasonably discriminate between real emission features and false detections . if the line centre was more than 7 away from the features expected wavelength ( given the 6dfgs redshift ) , the measurement was discarded . if the fwhm was well below the instrumental resolution of the 6df instrument ( @xmath23 5 ) the measurement was also discarded . finally , extremely large fwhm ( @xmath25 12 ) were discarded . these large fwhm are mostly gaussian fits to features in the continuum , but may also have occasionally included real features of broad - lined active galactic nuclei ( agn ) . the equivalent widths of the two [ oiii ] emission lines at 4959 and 5007 were measured using the procedure and band definitions of gonzalez ( 1993 ) . h@xmath29 emission , on the other hand , was estimated for each galaxy as the difference between the _ observed _ 6dfgs h@xmath29 index value and the value predicted by our best - fit model solutions ( see section [ fitting ] ) . within this work the sample is sometimes subdivided into ` passive ' ( no emission ) and ` emission ' galaxies . passive galaxies are defined as those in which no emission lines are detected within the 6dfgs wavelength range with greater than approximately 1-@xmath32 significance . emission line galaxies are therefore defined as those in which at least one emission line is detected at greater than one sigma confidence . for h@xmath28 this criterion corresponds to emission - line equivalent widths of @xmath01.5 . using this definition the sample contains 4640 passive galaxies and 2270 emission galaxies . unfortunately , a full analysis of morphological types is not available for the sample . however , visual inspection of a sub - sample of passive galaxies suggests that @xmath080% are early types , while amongst the emission galaxies a similar fraction are late types . the method used to measure lick indices from appropriately broadened spectra and to correct the indices for the broadening effect of a galaxy s velocity dispersion was based on that used by stephen moore ( 2001 ) measurements of the velocity dispersion in the 6dfgs galaxy spectra were made using the iraf task _ fxcor _ which carries out a fourier cross correlation between the object spectrum and a zero - redshift , zero - velocity dispersion template spectrum . the task outputs the radial velocity of the object spectrum and the fwhm of the fourier correlation peak between the object and template spectra , which can be converted into a velocity dispersion . the template spectra used were three velocity standard stars observed with the 6df spectrograph in 2003 march by craig harrison ( hr2574 , hr3145 and hr5888 , which are late g and early k giants ) . the raw values of the velocity dispersion measured by _ fxcor _ need to be calibrated . to do this each of the standard star spectra were artificially broadened to a specific velocity dispersion using the iraf task _ gauss_. these artificially broadened spectra were then used to calibrate their ( known ) velocity dispersions against the value measured by _ fxcor_. towards lower velocity dispersions , as one approaches the instrumental velocity resolution ( @xmath33100 km s@xmath12 ) , this calibration becomes less accurate . however , galaxies with such low velocity dispersions require only very small corrections to their measured lick index values . consequently , reliable index determinations can still be obtained in low velocity dispersion galaxies . more importantly , this method of measuring velocity dispersion fails on galaxies that are not well fit by the standard star templates . this mainly occurs in low signal - to - noise ratio galaxy spectra with young ages and/or strong emission lines . a comparison of velocity dispersion values derived from the 6dfgs spectra to values in the literature ( hyperleda ) is shown in fig . [ vdisp_comp ] . the 6dfgs values show good agreement with the literature values with an average offset of 6.1 kms@xmath12 , and an rms scatter of 20 kms@xmath12 . both deviation and scatter are greatest at high velocity dispersion , where we find the hyperlead values to be consistently greater than the 6dfgs values . this is in agreement with bernardi ( 2007 ) who found hyperleda values greater than values derived from the sdss at high velocity dispersion . given the large 6df aperture ( 6.7 arcsec compared to a more typical 2 or 3 arcsec in hyperleda ) it is possible that these deviations are the results of aperture effects . jrgensen et al . ( 1995 ) investigated the effects of varying the aperture size in observations of velocity dispersions in the inner regions ( i.e. r @xmath33 r@xmath34 ) of early - type galaxies . their results showed a weak power law dependence of measured velocity dispersions with the size of aperture . we therefore experimented with correcting velocity dispersions according to jrgensen et al . ( 1995 ) in the sub - sample of 6dfgs galaxies for which hyperlead values were also available . as this subsample of the data is heavily biased to nearby , early - type galaxies , it is well matched to the galaxies used to calibrate the jrgensen et al . ( 1995 ) relation . the results showed that while the offset was reduced to + 0.75 kms@xmath12 , the scatter was increased to 23 kms@xmath12 . we also note that the trend for hyperleda values to be greater than values derived from our data remains unaffected by these corrections . as noted above , the jrgensen et al . ( 1995 ) correction is calibrated for the inner regions of early - type galaxies . however , for our 6dfgs sample as a whole , the _ average _ aperture is @xmath01 r@xmath15 ( see fig . [ sample ] ) . the sample also contains both early- and late - type galaxies . this suggests that the corrections presented by jrgensen et al . ( 1995 ) are inappropriate for the 6df sample as a whole . for such a sample , it is instead informative to consider the study of gebhardt et al . this analysis of the variation in observed velocity dispersions ( including the effects of rotation ) measured through circular apertures of varying radius in a sample of 26 galaxies of mixed hubble types shows that , within 5 r@xmath15 , the sample exhibits little or no systematic variation of the measured velocity dispersion with aperture size . as a result of the considerations above , we have elected to make no corrections for aperture size to the measured velocity dispersion values within this work . + to make lick index measurements suitable for direct comparison with single stellar population ( ssp ) models , it is necessary to broaden the spectra to the instrumental resolution of the lick / ids system . the lick / ids fwhm instrumental resolution as a function of wavelength is described in worthey & ottaviani ( 1997 ) where it is shown to vary from @xmath08.5 at the central wavelength ( @xmath05000 ) to @xmath010 at the spectrum ends ( @xmath04000 and 6000 ) . the calculation of the necessary broadening requires the measurement of the 6df instrumental resolution . this was measured from 6df arc line spectra obtained from the observational time - span of the 6dfgs first data release . the instrumental resolution of the 6df spectra varies not only with wavelength but also with optical fibre number , so measurements were made of arc lines at various wavelengths , _ for all fibres individually_. a bivariate polynomial was fit to the measured fwhm spectral resolution as a function of both wavelength and fibre number . the amount of broadening required to match the lick / ids resolution could then be estimated and each 6df spectrum was broadened to this resolution . the index values given in stellar population models do not include the effect of the velocity dispersion broadening present in galaxy spectra . it is therefore necessary to introduce a correction to the observed galaxy index values to account for this effect . to measure the velocity dispersion correction required for each lick index , we measured the indices of a range of artificially broadened stellar spectra . these values of the lick indices measured in the broadened spectra were then compared to those measured in the un - broadened spectra to determine the index corrections for a range of velocity dispersions . for each index measured in angstroms ( e.g. the fe and ca indices ) a multiplicative correction was defined by : @xmath35 while , for each index measured in magnitudes ( e.g. cn@xmath36 , cn@xmath37 , mg@xmath36 , mg@xmath37 ) , an additive corrections was defined by : @xmath38 where index@xmath39 is the value of the index in the raw stellar spectrum ( i.e. with no broadening ) and index@xmath40 is the index value after broadening of the stellar spectrum to a specific velocity dispersion @xmath41 . for each index , quadratic lines were fitted to the velocity dispersion versus index correction ratio data . the index correction required for a particular galaxy could then be determined from the measured velocity dispersion and the quadratic function . estimating errors for spectroscopic measurements is notoriously difficult , mainly because there are several sources of error that can not be well quantified ( e.g. sky - subtraction , poor flux calibration etc ) . therefore , instead of trying to propagate variance arrays , comparisons of the values from repeated 6dfgs observations of the same galaxy were used to quantify this error . this method was used for calculating the error in lick indices and velocity dispersions , as well as in h@xmath31 and [ nii ] emission equivalent - widths . clipping . the solid line is the least squares fit to the bin values . the dashed line is the relationship between error and signal - to - noise ratio after correcting for the clipping used in forming the rms bins ( see section [ error_measurement ] ) the vertical line marks the lowest signal - to - noise ratio ( 12 @xmath12)used in the present study . * bottom * : same as top plots except the _ fractional _ error is shown.,title="fig:",width=158 ] clipping . the solid line is the least squares fit to the bin values . the dashed line is the relationship between error and signal - to - noise ratio after correcting for the clipping used in forming the rms bins ( see section [ error_measurement ] ) the vertical line marks the lowest signal - to - noise ratio ( 12 @xmath12)used in the present study . * bottom * : same as top plots except the _ fractional _ error is shown.,title="fig:",width=158 ] clipping . the solid line is the least squares fit to the bin values . the dashed line is the relationship between error and signal - to - noise ratio after correcting for the clipping used in forming the rms bins ( see section [ error_measurement ] ) the vertical line marks the lowest signal - to - noise ratio ( 12 @xmath12)used in the present study . * bottom * : same as top plots except the _ fractional _ error is shown.,title="fig:",width=158 ] clipping . the solid line is the least squares fit to the bin values . the dashed line is the relationship between error and signal - to - noise ratio after correcting for the clipping used in forming the rms bins ( see section [ error_measurement ] ) the vertical line marks the lowest signal - to - noise ratio ( 12 @xmath12)used in the present study . * bottom * : same as top plots except the _ fractional _ error is shown.,title="fig:",width=158 ] in the first data release of the 6dfgs , 3859 galaxies were observed more than once . for all repeat observations the differences in values were grouped into bins based on the median signal - to - noise ratio of the galaxy spectra . the signal - to - noise ratio of the lower of the two observations was used for this binning , as this observation would be the cause of most of the error . if a galaxy had been observed more than twice then the two highest signal - to - noise ratio spectra were used . the rms of each signal - to - noise ratio bin was determined and used as the @xmath42 value . to remove the biasing effect of large outliers in the bins only the lowest 95.45% of the values ( the @xmath43 limit ) in each bin was used to calculated the rms value . this requires that the rms be scaled up by 1.164 to take into account this clipping . a function was fitted to the rms of the bins against the signal - to - noise ratio . the fitting function used was a modified rectangular hyperbola of the form @xmath44 ( ` a ' and ` b ' the fitted coefficients ) . the error in any measured quantity for a galaxy could then be estimated using the signal - to - noise ratio of the spectrum and the corresponding error from the fitted function . examples of this procedure can be seen in figure [ error_snr ] . we note that the majority of the sample with repeat measurements were re - sampled due to the low signal - to - noise of the original observation . we used the @xmath45-fitting procedure of proctor & sansom ( 2002 ) ( see also proctor et al . 2004a , b and proctor et al . 2005 ) to measure the derived parameters : log(age ) , [ fe / h ] , [ z / h ] and [ e / fe ] ( a proxy for the ` @xmath28'abundance ratio ; see thomas et al . 2003 for details ) . briefly , the technique for deriving these parameters involves the simultaneous comparison of as many observed indices as possible to models of single stellar populations ( ssps ) . the best fit is found by minimising the square of the deviations between observations and models in terms of the observational errors ( i.e. @xmath45 ) . the rationale behind this approach is that , while all indices show some degeneracy with respect to each of the derived parameters , each index does contain _ some _ information regarding each parameter . in addition , such an approach should be relatively robust with respect to many problems that are commonly experienced in the measurement of spectral indices and their errors . these include poor flux calibration , poor sky subtraction , poorly constrained velocity dispersions , poor calibration to the lick system and emission - line contamination . this robustness is of particular importance in the analysis of large numbers of pipe - line reduced spectra such as those of the 6dfgs which can not be flux calibrated and so are not fully calibrated to the lick system . the method is also relatively robust with respect to the uncertainties in the ssp models used in the interpretation of the measured indices ; e.g. the second parameter effect in horizontal branch morphologies and the uncertainties associated with the asymptotic giant branch . it was shown in proctor et al . ( 2004a ) and proctor et al . ( 2005 ) that the results derived using the @xmath45 technique are , indeed , significantly more reliable than those based on only a few indices . in order to carry out the comparison of observations to ssp models it was first necessary to select and interpolate the models provided in the literature . interpolations are required because models in the literature are presented at only five or six discrete metallicities over the @xmath02.5 dex range . we experimented with a number of different models ( bruzual & charlot 2003 ( hereafter bc03 ) ; thomas , maraston & bender 2003 ; thomas , maraston & korn 2004 and korn , maraston & thomas 2005 ( hereafter kmt05 ) . the results presented here are those from the models of kmt05 , as these provide the necessary coverage in age ( 0.1 to 15 gyr ) and metallicity ( [ z / h ] from 2.3 to + 0.67 ) , as well as including a detailed modelling of the effects on indices of varying ` @xmath28'abundance ratios . however , fits were also obtained to the other model sets for comparison . the process by which the best fits were obtained was iterative . first , fits were obtained for all galaxies using _ all _ the available indices . the pattern of deviations from the fit so obtained was then used to identify indices that matched the models poorly ( see fig . [ chis ] ) . due to the problem of emission - line filling of the h@xmath46 , h@xmath47 and h@xmath29 balmer lines in emission galaxies , we considered ` passive ' and ` emission ' galaxies ( section [ meas_em_lines ] ) separately . in fig . [ chis ] emission galaxies show the expected large deviation between observed and best fit model values of balmer indices and fe5015 ( which is also emission - line affected ) . these indices were therefore omitted from the fitting procedure in emission galaxies . perhaps surprisingly , h@xmath29 and fe5015 also deviate significantly in the ` passive ' galaxies . we take this to indicate that , despite the rather severe definition of emission galaxies outlined in section [ meas_em_lines ] , at least some of the passive galaxies contain low - level emission that has escaped detection during our emission line measurements . indeed , this was the main motivation for the stringent definition of emission - galaxies . to minimise the effects of this low - level emission on our age and metallicity estimates , age sensitive h@xmath29 and metallicity sensitive fe5015 indices were also excluded from the fits to passive galaxy data . in both passive and emission galaxies ca4455 was found to fall below best - fit model values . this effect has been noted in previous studies ( e.g. proctor & sansom 2002 ; clemens et al 2006 ) . the flux - calibration sensitive mg@xmath36 and mg@xmath37 indices were also found to deviate from best fit model values . this was an anticipated effect , as 6dfgs spectra are only approximately flux calibrated and mg@xmath36 and mg@xmath37 have widely space pseudo - continuum bands . ca4455 , mg@xmath36 and mg@xmath37 were therefore excluded from the fitting procedure . fortunately , most of the information lost by the exclusion of the mg@xmath36 and mg@xmath37 indices is captured in the mgb index that is not sensitive to flux - calibration issues . once poorly fitting indices had been identified and omitted , the fitting of the remaining indices was carried out using a clipping procedure in which indices deviating from the model fit by more than 3@xmath32 were excluded , and the fitting procedure performed again . this results in an average of less than one additional index per galaxy being clipped from the fits . most of the clipped indices could be associated with contamination from features such as the 5577 sky - line . the resultant fits are based on between 10 and 15 indices and show good agreement between observed and best - fit values ( fig . [ chis ] ) . indeed , the reduced-@xmath45 values given in fig . [ chis ] ( @xmath48 ) were 1.08 and 1.30 for passive- and emission - galaxies respectively . this clearly indicates that our data are well represented by the models and that our errors reflect well the differences between observations and models . for each galaxy in the sample , errors in log(age ) and [ z / h ] were estimated using 50 monte - carlo realisations of the best - fit model indices , assuming the observational errors for the galaxy in question . we again note that the reduced-@xmath45 values of both passive- and emission - galaxies were close to one , indicating that the assumed errors reflect the average differences between observations and models quite well . the rms scatter in the log(age ) and [ z / h ] values of the 50 realisations was therefore taken as the individual galaxy error . however , in an effort to better understand the detailed behaviour of the errors , deviations from the best - fit log(age ) and [ z / h ] values of all 50 realisations of each galaxy were retained and are presented in section [ azerr ] . for the purposes of the following analysis it is important that the derived ages are well defined . @xmath0 1000 low signal - to - noise galaxies whose error in log(age ) exceeded 0.3 dex were deemed unreliable and were excluded from the analysis . the average errors for the remaining galaxies were @xmath00.15 dex in both log(age ) and [ z / h ] . emission lines can be used to separate star - burst and active galactic nuclei ( agn ) galaxies using the [ oiii]@xmath305007/h@xmath29[nii]/h@xmath28 diagnostic . values of the [ oiii]@xmath305007 , [ nii ] and h@xmath28 emission strengths were estimated from the 6dfgs spectra in a traditional manner ( section [ meas_em_lines ] ) . the strength of h@xmath29 emission , however , was taken to be the deviation between the measured value of the h@xmath29 index and the best - fit ssp model value . due to the relatively small observed dynamical range of h@xmath29 in absorption compared to that observed in emission , this is a fairly robust estimate of h@xmath29 emission . this is evidenced by the classic ` y'-shape evident in the diagnostic plot of [ oiii]@xmath305007/h@xmath29 against [ nii]/h@xmath28 shown in in fig . [ emission ] . the position of the line dividing star - burst galaxies from agn varies between studies in the literature . in fig . [ emission ] we show the lines given by kewley et al . ( 2001 ) ( solid line ) and kauffmann ( 2003a ) ( dashed line ) . in order to ensure reliable identifications of agn we use the definition of kewley et al . in the following . about 2200 of the 2700 emission galaxies in our sample have reliable estimates of all four of these emission lines , 300 of these lying in the region of the diagram associated with agn as defined by kewley et al . + in this section we outline the results of our age and metallicity determinations using lick indices . the distribution of ages and metallicities are presented in fig . the sample has been sub - divided into passive- and emission - galaxies as described in section [ meas_em_lines ] . emission - galaxies classified as agn ( section [ emlingal ] ) are shown as grey symbols , the remainder are plotted in black . before interpreting this plot , we must note the apparent clusterings and ` zones - of - avoidance ' , the clearest examples of which are evident in the emission galaxy plot . these are the inevitable result of linear interpolations in non - rectilinear spaces . returning attention to the actual results of our age and metallicity determinations ; similar trends of increasing [ z / h ] with decreasing log(age ) are evident in both of passive and emission galaxies albeit with emission galaxies tending to younger ages . galaxies classified as possessing agn in section [ emlingal ] also often possess young central ages . we investigate the effect of aperture size on derived parameter using fig . [ aperture ] . in these plots , the galaxies with the largest and smallest sizes ( [email protected] arcsec and [email protected] arcsec ) are shown as red and blue points respectively . the blue points therefore represent the galaxies in which the 6.7 arcsec aaomega aperture encompasses the largest fraction of galaxy light ( on average @xmath03.0 r@xmath15 ) , while the red points represent galaxies in which the smallest fraction of galaxy light ( @xmath00.5 r@xmath15 ) is encompassed by the fibre . the observed age metallicity and age velocity dispersion relations can be seen to be similar in these two extremes , with the most noticeable difference a slightly higher [ z / h ] ( by @xmath00.1 dex ) in large galaxies . we therefore find aperture - size effects to be generally small . due to the large , circular aperture of the 6dfgs , a direct comparison with the literature , which are mainly based on long - slit observations of galaxy centres , is not straight forward . nevertheless , we note that the observed trend is at least qualitatively in agreement with the trends observed in previous studies ( trager et al . 2000 ; proctor & sansom 2002 ; mehlert et al . 2003 ; gallazzi et al . 2005 ; collobert et al . the young central ages of agn are also consistent with studies in the literature showing that star - bursts are often found to accompany agn activity ( e.g. see cid fernandes et al . 2004 and references therein ) . an intriguing feature of fig . [ agez ] is the presence of a few old galaxies with extremely low [ z / h ] values , ( @xmath330.5 dex ) . these galaxies , present in both passive and emission samples , possess a range of velocity dispersions . visual inspection of both spectra and photometry generally reveals nothing exceptional about these galaxies . further analysis of these galaxies are beyond the scope of this paper , but we note that their low numbers result in them having no impact on the results and conclusions of the present paper . as in these literature studies , the slope in the age metallicity relation exhibited by our data is similar to the slope of the age metallicity degeneracy ( worthey 1994 ; ` the 3/2 rule ' ) . before comparing these results with the photometry , it is therefore clearly important to establish that the age metallicity degeneracy has been broken . to this end , in the following section we consider the errors in the derived parameters . as described in section [ errdps ] , errors in age and metallicity were characterised by considering the results of the 50 monte - carlo realisations of each of the 7000 galaxies in our sample . fits using both the index combinations shown in fig . [ chis ] ( with and without balmer lines ) were carried out for the purposes of this analysis ( a total of approximately one million realisations ) . the differences between the input - model values of log(age ) and [ z / h ] and those of each of the realisations were calculated . these were combined in a number of bins depending upon the age metallicity of the input models . the analysis was also carried out using the _ observed _ index values as the inputs to the monte - carlo realisations ( rather than the best - fit model values ) . no significant quantitative or qualitative differences were found . [ error ] shows the results of this analysis as 1-sigma confidence contours for three age metallicity bins . each contour in these plots is based on tens of thousands of individual realisations of the best - fit galaxy data . the contour levels were defined as the iso - densities corresponding to e@xmath12 of the peak values ( as expected for the 1-sigma contour of a two - dimensional gaussian distribution ) . the use of the iso - density peak in estimating the extent of 1-sigma confidence contours results in a small uncertainty in the estimates . however , we estimate these to be of order 2% and therefore insignificant . results of the error analysis for both passive galaxies ( in which balmer indices were included in fits ) and emission galaxies ( in which balmer indices were _ excluded _ from fits ) are shown separately . we recall that h@xmath29 was excluded from all the fits . [ error ] also shows the data divided into two signal - to - noise ratio regimes ( s / n@xmath2316 @xmath12 and s / n@xmath2516 @xmath12 ) . marginal distributions in log(age ) and [ z / h ] are shown at the edges of each main plot . we note that the positions of the contours in fig . [ error ] were chosen for clarity only . to get a full picture one must imagine the whole surface of the log(age)[z / h ] plane populated by such contours . [ error ] shows that , as might be expected , error estimates increase with decreasing signal - to - noise - ratio . for passive galaxies , the analysis reveals log(age ) and [ z / h ] errors of @xmath00.1 dex in galaxies with s / n above @xmath016 . this falls to @xmath00.2 dex in galaxies with s / n below @xmath016 . the errors of the emission - galaxies ( in which all balmer lines are excluded from the fits ) are somewhat larger , but are generally still of order 0.2 dex . it is therefore clear that while the age metallicity degeneracy is still present ( as evidenced by the sloped elliptical error contours in fig . [ error ] ) , the magnitude of the errors is sufficiently small for its effects to be negligible in our results - even in galaxies fit without balmer lines . this is the result , and main advantage , of using large numbers ( @xmath2510 ) indices in the determination of ages and metallicities . the good age resolution achieved is emphasised in the marginal distributions of fig . [ error ] , in which galaxies older than 10 gyr ( 1.0 dex ) , and galaxies younger than 1.5 gyr ( 0.15 dex ) ( delimited by dashed lines in the marginal distributions ) are largely uncontaminated by galaxies with ages of 3 gyr ( 0.5 dex ) . since the following analysis concentrates mainly on the very oldest and very youngest galaxies , these provide confidence that results do not suffer significantly from the effects of the age metallicity degeneracy . the final sample for which velocity dispersions and reliable age / metallicity estimates were measured consists of 4500 passive galaxies and 1000 emission galaxies . + the combination of 6dfgs , 2mass and supercosmos data permits the comparison for spectroscopic age and metallicity determinations to the b , r and k band photometric data for some 6000 galaxies . specifically , by combining the photometry with the velocity dispersion measures from 6dfgs spectroscopy , we are able to investigate trends with age and metallicity in b , r and k band mass - to - light - ratios . this analysis is carried out using a _ dynamical _ mass ( m@xmath51 ) calculated as : @xmath52 where @xmath32 is the central velocity dispersion , r@xmath53 is the half - light radius and the constant c has a value of 5.0 ( cappellari et al . unfortunately half - light radii are only available for 25% of our sample , while k band 20th magnitude isophotal radii ( r@xmath54 ) are available for the entire sample . a calibration was therefore carried out by using equation [ masseq ] to calculate masses for the galaxies in the 25% sub - sample ( which also possess much more accurate , aperture corrected , velocity dispersion estimates ) , and comparing them with mass estimates using r@xmath14 ( instead of r@xmath53 ) in the same equation . a plot of the comparison is shown in fig . [ mass_comp ] . the correlation has equation m@[email protected] with only 0.12 dex of scatter . given error estimates in m@xmath51 and m@xmath56 of 0.06 and 0.10 dex respectively , the correlation is clearly extremely good . masses quoted throughout the remainder of this work are therefore those based on r@xmath14 corrected as detailed . as well as the photometry of 2mass and supercosmos , we also use the photometric results from the ssp models of bruzual & charlot ( 2003 ) . some key values from the bruzual & charlot models are shown in table [ photometry ] . we use these models almost exclusively to estimate differential properties ; e.g. the rate of change of mass - to - light - ratio with log(age ) in populations of given [ z / h ] . it should be noted that the rates of change of the photometric properties are reasonably constant with metallicity and age , as long as age is expressed in logarithmic form . this clearly indicates that a differential approach is a robust use of the model values . we shall flag the one occasion in which the models are used in a non - differential manner . . examples of b , r and k magnitudes from bc03 for single stellar population models of varying age and metallicity are presented . note that these values are used almost exclusively in a differential manner ( see section [ photo ] ) . [ cols="^,^,^,^,^,^ " , ] we now turn our attention to the effects of the young central populations detected in our 6dfgs spectra on the global properties of galaxies as indicated by their photometry . there are 400 ` young ' ( @xmath231.5 gyr ) galaxies ( see section [ eamopp ] ) included in this analysis . plots of mass - to - light - ratio with mass for both young and old galaxies are shown in fig . mass - binned averages of young and old data are shown as blue and red solid lines respectively . also shown in these plots as dashed blue lines are model predictions of the effects of the differing ages and metallicities of the young and old populations in each mass bin . these are based on the comparison of luminosity values from the bc03 ssp models for the differences in age and metallicity of the young and old galaxies in each mass bin . a direct comparison of the data for young galaxies to these lines would therefore implicitly assume that the young central populations detected in the spectroscopy pervade the galaxies populations _ as a whole _ - i.e represents the same fraction of the galaxy population at all radii . young galaxies can be seen to be displaced with respect to the old galaxies in a sense consistent with their young ages . however , the displacements are small compared to the predicted values . they also become smaller towards higher masses . the data therefore show that , on average : ( i ) young populations do _ not _ pervade their host galaxies , but are instead centrally concentrated ; ii ) young populations must constitute relatively small fractions of total galaxy masses ; and ( iii ) the mass fraction in the form of a young stellar population must decrease with increasing galaxy mass . to quantify these conclusions , for each mass bin , the fraction of total galaxy stellar mass and luminosity involved in the recent star - burst ( f@xmath57 and f@xmath58 ) were estimated . these estimates were made by a differential comparison of the bc03 predictions for mass - to - light - ratios to the observed values . in each waveband , the difference between the observed mass - to - light - ratios of the young galaxies from those of the old galaxies in the same mass bin ( @xmath59log[m / l]@xmath60 ) was compared to the difference between model predictions for young and old populations of appropriate metallicity ( @xmath59log[m / l]@xmath61 ) . denoting these mass - to - light - ratio differences as s and r respectively , we derive : @xmath62_{obs}}-1}{10^{-\delta\log[m / l]_{model}}-1 } \ \ or \ \ \frac{10^{-s}-1}{10^{-r}-1}.\ ] ] the derivation of equation 6 is given in appendix a. we estimate the _ luminosity _ fraction in the recent burst by : @xmath63 finally , the total mass ( m@xmath64 , in solar masses ) is estimated by : @xmath65 where m@xmath66 is the total mass of the galaxy . the derivation of these expressions again assumes that the young populations at any given mass are seen against a background old population with a metallicity and mass - to - light - ratio appropriate to that mass ( as given by the mass - binned values for old galaxies in table [ mol ] ) . the young stellar population is nonetheless assumed to _ dominate the central regions _ ( such that issues related to degeneracies with burst - strength do not arise ) . the estimates of mass- and luminosity - fractions and burst - masses are plotted against galaxy mass in fig . [ mass_frac ] . despite the ( expected ) variation in luminosity fractions between wavebands , agreement between the values for the mass fractions derived from the three wavebands is extremely good . this suggests that both our assumption of the same underlying dynamical ` tilt ' in young galaxies as exhibited by old galaxies is correct , and that our age and metallicity corrections are accurate . the trend shows a decreasing mass fraction with increasing mass falling from @xmath010% at 10@xmath67 m@xmath27 to 2% at 10@xmath68 m@xmath27 . this is in agreement with the study of treu et al . ( 2005 ) , who find the stellar fraction formed in recent times varies from 20%-40% below 10@xmath69 m@xmath27 , to below 1% above 10@xmath68 m@xmath27 . also shown in fig . [ mass_frac ] are the estimates of average total mass involved in the recent burst of star formation . agreement between wavebands is again good . however , the absolute values given here must be treated with caution , as they are calculated by the only non - differential application of bc03 models used in this work . nevertheless , in a relative sense , the data exhibit a narrow range of burst - masses and an apparent _ upper limit _ on their size in high mass galaxies . as a final consideration we next investigate the implications of our findings for colour - magnitude relations . note that these relations are independent of dynamical effects . in fig . [ cmr ] we plot the colour - magnitude relations b k and r k against m@xmath70 . galaxies with both old ( grey points ) and young ( black points ) central populations are shown . averages ( binned by luminosity ) of the old and young populations are again shown as solid red and blue lines respectively . slopes in the old galaxies are in good accord with literature values ( thin black lines in each plot ) . as the sample includes late - type galaxies , particular amongst the young galaxies , these must lie close to the ` red sequence ' in our colour - magnitude diagrams . this suggests they therefore correspond to the ` dusty red - sequence ' of wolf , gray & meisenheimer ( 2005 ) . the lack of galaxies in the ` blue - cloud ' is almost certainly the result of a number of selection effects . these include the effects of template mismatches during velocity dispersion measurements and the removal of galaxies with poor fits or large errors during age / metallicity measurement . [ cmr ] shows that , for both colours , young galaxies tend to be bluer than the old galaxies . however , the offsets are extremely small . our results from the previous section indicate that this is largely the result of the low mass fractions formed in the recent star formation events . [ cmr ] therefore shows that galaxies lieing on , or close to , the red sequence may not be ` red and dead ' , but may still be forming modest numbers of new stars . indeed , we note that _ most _ of the galaxies with young central populations also exhibit on - going star formation ( fig . [ agez ] ) . we have measured ages and metallicities from the 6dfgs spectra of @xmath07000 galaxies of all morphological and emission types . we have demonstrated that the age metallicity degeneracy has been broken in this study , particularly with regard to the very young and very old galaxies used in the subsequent analysis . these data for the _ central regions _ are compared with the _ global _ photometry from 2mass and supercosmos for the @xmath06000 galaxies for which such photometry is available . combining spectroscopy and photometry we are able to investigate trends in global b , r and k band mass - to - light - ratios with the central stellar populations of the galaxies . an age - metallicity trend is identified in passive galaxies that is consistent with the results of similar studies in the literature . the stellar populations of galaxies exhibiting emission lines ( arising from either star - bursts or agn ) follow a similar trend , but exhibit younger ages than their passive counterparts . to minimise the effects of complex mixtures of stellar populations , we confine our analysis to the data for very old or very young galaxies . we find a steep mass - metallicity relation in old galaxies ( logarithmic slope of @xmath00.25 ) , while in young galaxies the trend is consistent with zero slope . using only the old galaxies , we find a dynamical trend in mass - to - light - ratios with mass of logarithmic slope @xmath00.15 . this is in good agreement with values reported in the literature . however , there is no agreement in the literature as to the cause of this tilt . padmanabhan et al . ( 2004 ) concluded that most of the tilt is due to variations in dark matter content . from the modelling perspective , robertson et al . ( 2006 ) perform simulations of a series of mergers and investigate the scaling relations of their merger remnants . from mergers of gas - rich disk galaxies they derive a fp with @xmath28 = 1.55 and @xmath29 = 0.82 , in good agreement with our k band values ( see table 3 ) . in their models the tilt away from the virial scaling relation is caused by gas dissipative effects and not dark matter content . trujillo et al . ( 2004 ) on the other hand attribute the tilt to ` non - homology ' , deriving almost identical contributions to the tilt from dynamical and stellar population effects as those we find in this work . our findings , however , also show that whatever the cause of the tilt , it is not a simple in mass - to - light - ratio ( or its scatter ) with mass or luminosity . instead we find a that a more complex description will be required ; this will be a subject of a future paper . considering extremely young galaxies , our results show that the properties of the central stellar populations identified by the spectroscopy do _ not _ generally correlate with the global properties of their host galaxies . indeed , the data indicate that recent bursts of star formation are limited to a mass of @xmath3310@xmath24 m@xmath27 and are strongly centrally concentrated . consequently , in high mass galaxies exhibiting young central populations , recent star forming events have negligible effect on their global photometry . as a consequence we have shown that galaxies sitting on , or near , the red sequence may not be ` red and dead ' , but may still be forming modest numbers of new stars . we therefore conclude that the young ages found in studies of galaxy centres in this and other spectroscopic studies should not be considered representative of the age of their host galaxies as a whole . they may , however , be reasonably interpreted as indicators of recent assembly events . as a result of these considerations , the small scatter in photometric scaling relations , which initially motivated the monolithic collapse models , ceases to be inconsistent with either the hierarchical merging model or the results of the lick index studies indicating young central populations . in future work we will include the results from the completed 6df galaxy survey and report on the full chemical properties ( including @xmath28-abundance ratios ) of the central regions of galaxies . the effective radii , peculiar velocities and environmental parameters of the 6dfgs sample , currently being analysed by the 6dfgs team , will be included . this will allow not only more accurate determination of photometric properties and a more detailed analysis of the fundamental plane , but also the investigation of the strong _ quantitative _ predictions of trends in galaxy formation / assembly epochs and metals - content with galaxy mass and environment . + * acknowledgements * this publication makes use of data products from the two micron all sky survey ( 2mass ) , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology , funded by the national aeronautics and space administration and the national science foundation . the authors acknowledge the data analysis facilities provided by iraf , which is distributed by the national optical astronomy observatories and operated by aura , inc . , under cooperative agreement with the national science foundation . we thank the australian research council for funding that supported this work . the measuring of line values from 6dfgs spectra was begun in philip lah s honours thesis 2003 `` exploring the stellar population of early - type galaxies in the 6df galaxy survey '' which looked at the 6dfgs early data release , ( december 2002 ) . this honours thesis was supervised by matthew colless and heath jones . assistance with the software used in this work was provided by craig harrison . lachlan campbell provided additional assistance with the 6dfgs data particularly with the velocity dispersions and 6df arc spectra . the authors also thank chris blake for valuable input . * references * + baldry i.k . et al . , 2002 , apj , 569 , 582 + baum w.a . , 1959 , iau symp . 10 , the hertzsprung - russell diagram , ed . j. l. greenstein ( paris : iau ) , 23 + bernardi m. et al . , 2003a , aj , 125 , 1882 + bernardi m. et al . , 2003b , aj , 125 , 1866 + bernardi m. et al . , 2006 , aj , 131 , 2018 + bernardi m. , 2007 , aj , 133 , 1954 + blumenthal g.r . , faber s.m . , primack j.r . , rees m.j . , 1984 , nature , 311 , 517 + bower r.g . , lucy j.r . , ellis r.s . , 1992a , mnras , 254 , 589 + bower r.g . , lucy j.r . , ellis r.s . , 1992b , mnras , 254 , 601 + bruzual a.g . , charlot s. , 2003 , mnras , 344 , 1000 + caldwell n. , rose j.a . , concannon k.d . , 2003 , aj , 125 , 2891 + cappellari m. et al . , 2006 , mnras , 366 , 1126 + carlberg r.g . , 1984 , apj , 286 , 403 + chang r. , gallazzi a. , kauffmann g. , charlot s. , ivezi z. , brinchmann j. , heckman t.m . , 2006 , mnras , 366 , 717 + cid fernandes r. , gu q. , melnick j. , terlevich e. , terlevich r. , kunth d. , rodrigues lacerda r. , joguet b. , 2004 , mnras , 355 , 273 + cid fernandes r , mateus , a. , sodr l. , stasinska g. , gomes j.m . , 2005 , mnras , 358 , 363 + clemens m.s . , bressan a. , nikolic b. , alexander p. , annibali f. , rampazzo r. , 2006 , mnras , 370 , 702 + colless m. , saglia r.p . , burstein d. , davies r.l . , mcmahan jr r.k . , wegner g. , 2001a , mnras , 321 , 277 + colless m. et al . , 2001b , mnras , 328 , 1039 + collobert m. , sarzi m. , davies r.l . , kuntschner h. , colless m. , 2006 , mnras , 370 , 1213 + de lucia g. , springel v. , white s.d.m . , croton d. , kauffmann g. , 2006 , mnras , 366 , 499 + dressler a. , lynden - ball d. , burstein d. , faber s.m . , terlevich r. , wegner g. , 1987 , apj , 313 , 42 + eisenstein d.j . et al . , 2003 , apj , 585 , 694 + fernandes r.c . , mateus , a. , sodr l. , stasinska g. , gomes j.m . , 2005 , mnras , 358 , 363 + gallazzi a. , charlot s. , brinchmann j. , white s.d.m . , tremonti c. , 2005 , mnras , 362 , 41 + gebhardt , k. et al . , apj , 539 , l13 + gibbons r.a . , fruchter a.s . , bothun g.d . , 2001 , aj , 121 , 649 + gonzlez j. j. , 1993 , phd thesis , univ . california + hambly n.c . , irwin m.j . , macgillivray h.t . , 2001 , mnras , 326 , 1295 + hudson m.j . , lucey j.r . , smith r.j . , steel j. , 1997 , mnras , 291 , 488 + jarrett t.h . , chester t. , cutri r. , schneider s. , skrutskie m. , huchra j.p . , 2000a , aj , 119 , 2498 + jarrett t.h . , cheste , t. , cutri r. , schneider s. , rosenberg j. , huchra j.p . , mader j. , 2000b , aj , 120 , 298 + jones d.h . et al . , 2004 , mnras , 355,747 + jones d.h . , peterson b.a . , colless m. , saunders w. , 2006 , mnras , 369 , 25 + jrgensen i. , franx m. , kjrgaard p. , 1995 , mnras , 276 , 1341 + jrgensen i. , franx m. , kjrgaard p. , 1996 , mnras , 280 , 167 + kauffmann g. , 1996 , mnras , 281 , 487 + kauffmann g. et al . , 2003 , mnras , 346 , 1055 + kauffmann g. et al . , 2003a , mnras , 341 , 33 + kauffmann g. et al . , 2003b , mnras , 341 , 54 + kewley l.j . , dopita m.a . , sutherland r.s . , heisler c.a . , trevena j. , 2001 , apj , 556 , 121 + korn a.j . , maraston c. , thomas d. , 2005 , a&a , 438 , 685 + larson r.b . , 1974 , mnras , 166 , 585 + lewis i. et al . , 2002 , mnras , 334 , 673 + mehlert d. , thomas d. , saglia r.p , bender r. , wegner g. , 2003 , a&a , 407 , 423 + moore s.a.w . , 2001 , ph.d thesis , univ of durham + padmanabhan n. et al . , 2004 , new astron . , 9 , 329 + pahre m.a . , djorgovski s.g . , de carvalho r.r . , 1998 , aj , 116 , 1591 + proctor r.n . , sansom a.e . , 2002 , mnras , 333 , 517 + proctor r.n . , forbes d.a . , beasley m.a . , 2004a , mnras , 355 , 1327 + proctor r.n . , forbes d.a . , hau g.k.t . , beasley m.a . , de silva g.m . , contreras r. , terlevich a.i . , 2004b , mnras , 349 , 1381 + proctor r.n . , forbes d.a , forestell a. , gebhardt k. , 2005 , mnras , 362 , 857 + robertson b. , cox t.j . , hernquist l. , franx m. , hopkins p.f . , martini p. , springel v. , 2006 , apj , 641 , 21 + smith r.j . , hudson m.j . , lucey , j.r . , nelan , j.e . , wegner , g.a . , 2006 , mnras , 369 , 1419 + spergel d.n . et al . , 2003 , apjs , 148 , 175 + terlevich a.i . , forbes d.a . , 2002 , mnras , 330 , 547 + thomas d. , maraston c. , bender r. , 2003 , mnras , 339 , 897 + thomas d. , maraston c. , bender r. , mendes de oliveira c. , 2005 , apj , 621 , 673 + thomas d. , maraston c. , korn a. , 2004 , mnras , 351 , 19 + trager , s. , faber , s. , worthey , g. , gonzalez , j. , 2000 , aj , 120 , 165 + tremonti c. et al . , apj , 613 , 898 + treu t. , ellis r.s . , liao t.x . , van dokkum p.g . , tozzi p. , coli , a. , newman j. , cooper m.c . , davis m. , 2005 , apj , 633 , 174 + trujjillo i. , burkert a. , bell e.f . , 2004 , aj , 600 , l39 + white s.d.m . , rees m.j . , 1978 , mnras , 183 , 341 + wolf c. , gray m.e . , meisenheimer k. , 2005 , a&a , 443 , 435 + worthey g. , 1994 , apjs , 95,107 + worthey g. , ottaviani d.l . , 1997 , apjs , 111 , 377 + york d.g . et al . , 2000 , aj , 120 , 1579 + [ lastpage ]

we report new hst _ wfpc2 _ and _ nicmos _ observations of the center of the nearest radio galaxy centaurus a ( ngc 5128 ) and discuss their implications for our understanding of the active nucleus and jet . we detect the active nucleus in the near ir ( k and h ) and , for the first time , in the optical ( i and v ) , deriving the spectral energy distribution of the nucleus from the radio to x - rays . the optical and part of the near - ir emission can be explained by the extrapolation of the x - ray power law reddened by , a value consistent with other independent estimates . the 20pc - scale nuclear disk discovered by schreier et al . ( ) is detected in the$}}\lambda 1.64\mu$]m line and presents a morphology similar to that observed in with a / ratio typical of low ionization seyfert galaxies and liners . nicmos 3 pa observations in a 50 circumnuclear region suggest enhanced star formation (m yr ) at the edges of the putative bar seen with iso , perhaps due to shocks driven into the gas . the light profile , reconstructed from v , h and k observations , shows that centaurus a has a core profile with a resolved break at and suggests a black hole mass ofm . a linear blue structure aligned with the radio / x ray jet may indicate a channel of relatively low reddening in which dust has been swept away by the jet . 0.5 cm

astro-ph9907378

centaurus a ( ngc5128 ) , the closest giant elliptical galaxy hosting an active galactic nucleus ( agn ) and a jet , is an ideal benchmark for testing and addressing some open issues in the unified model of agns . its relative proximity ( @xmath9 mpc , hui et al . @xcite ) offers a unique opportunity to investigate the putative supermassive black - hole , the associated accretion disk and jet , i.e. the entities which the unified model considers at the core of agn activity ( see wills @xcite for a recent review ) . the detailed structure of the galaxy and the stellar and globular cluster populations can be linked to study the relationship between star formation and nuclear activity ; it should be possible to assess the role of the merger in the evolution of the giant elliptical galaxy , its multiple populations and the triggering and fueling of the agn itself . however , the study of this nearest giant elliptical at intermediate wavelengths has been severely hindered by the presence of a dust lane which dominates ground - based optical and near - ir observations of the nuclear region . the dust lane , which obscures the inner half kiloparsec of the galaxy , with associated gas , young stars and hii regions , is interpreted as the result of a relatively recent merger event between a giant elliptical galaxy and a small , gas rich , disk galaxy ( baade & minkowski @xcite , graham @xcite , malin , quinn & graham @xcite ) . see israel ( @xcite ) for a recent comprehensive review on centaurus a. the presence of the agn was revealed and studied through its radio and x - ray manifestations . centaurus a is a giant double - lobed radio source , first discovered by bolton , stanley & slee ( @xcite ) and it is considered a prototype low luminosity fanaroff - riley class i radio galaxy . its nucleus was found to be a source of x - ray emission ( bowyer et al . @xcite , kellogg et al . @xcite , grindlay et al . a strong jet was discovered in the x - rays ( schreier et al . @xcite ; feigelson et al . @xcite ) and subsequently studied in the radio with the vla ( schreier , burns & feigelson @xcite ; burns , feigelson & schreier @xcite ; clarke , burns & feigelson @xcite ; clarke , burns & norman @xcite ) and vlbi ( cf . jones et al . @xcite ) ; it is the nearest extragalactic jet . in the outer regions , ignoring the dust lane , centaurus a is a fairly normal giant elliptical galaxy . most of the light comes from an older stellar population , and the light distribution appears to follow a de vaucouleurs law ( cf . graham @xcite ) . the extensive system of shells within the extended elliptical component of the galaxy ( cf . malin , quinn & graham @xcite ) provide further evidence for a merger event occurring @xmath10 yr ago . m & f336w & 1800 u4100102 m & f336w & 600 u4100103 m & f336w & 1800 u4100104 m & f336w & 600 u4100105 m & f555w & 300 u4100106 m & f555w & 1000 u4100107 m & f555w & 1000 u4100108 m & f814w & 300 u4100109 m & f814w & 1000 u410010am & f814w & 1000 u4100201 m & f336w & 1900 u4100202 m & f336w & 600 u4100203 m & f336w & 1900 u4100204 m & f336w & 600 u4100205 m & f555w & 260 u4100206 m & f555w & 1000 u4100207 m & f555w & 1000 u4100208 m & f814w & 260 u4100209 m & f814w & 1000 u410020am & f814w & 1000 n46d02apm & f222 m & 1279 n46d02b6 m & f187n & 2303 n46d02bhm & f190n & 2303 n46d03poq & f160w & 2559 n46d03pxq & f160w & 2559 n46d03q6q & f110w & 2559 n4we01010 & f164n & 2560 n4we01040 & f166n & 2560 n4we01070 & f166n & 2560 n4we010a0 & f164n & 2560 n4k46kxjq & f187n & 768 n4k46kxkq & f160w & 192 n3zb44010 & f160w & 319 ir and co observations of the dust lane can be modeled by a thin warped disk ( quillen et al . @xcite ; quillen , graham & frogel @xcite ) which dominates ground - based near - ir observations along with the extended galaxy emission ( packham et al . @xcite and references therein ) . r - band imaging polarimetry from hst with wfpc ( schreier et al . @xcite ; paper i ) are also consistent with dichroic polarization from such a disk . at large radii the radial light- profile is well fitted with a de vaucouleurs law , but until now the dust lane has prevented secure measurements of the profile within the central arcminute ( @xmath11kpc ) of the galaxy or of the nucleus itself . recent hst nicmos observations in the k - band ( schreier et al . @xcite , hereafter paper ii ) have however revealed the presence of an unresolved ( i.e. fwhm@xmath12 ) source located at the nucleus of centaurus a , superimposed on extended galaxy emission . a @xmath13 1 radius emission line region centered on the nucleus observed in was interpreted as an extended accretion disk around the agn . in this paper , we report new hst observations of the central region of the galaxy : two contiguous wfpc2 fields covering approximately 2by 4 around the nucleus in the f336w , f555w and f814w bands , and nicmos camera 1 , 2 , and 3 observations of the nucleus and inner jet . we discuss the implications of these new high spatial resolution data in u through k on our understanding of the active nucleus , the jet , and the galaxy continuum emission in the innermost regions of centaurus a. these observations are part of an extensive continuing program to study this giant elliptical galaxy the nucleus and jet of the nearest agn , its stellar populations , and the relationship between its merger history and its nuclear activity . the current paper concentrates on the photometric structure of the galaxy interior to the dust lane , the spectral energy distributions of the nucleus , and the search for an optical / ir counterpart to the radio jet . in section [ sec : observ ] we discuss the observations and data reduction . the observational results are described in section [ sec : res ] : section [ sec : agn ] derives the location of the active nucleus and its spectral energy distribution ; section [ sec : feii ] presents new @xmath1$}}\lambda 1.64\mu$]m observations of the nuclear disk discovered in paper ii ; section [ sec : redden ] presents a reddening map and reconstructs the inner galaxy light profile ; section [ sec : nic3 ] presents the pa@xmath2 map which identifies the presence of several star formation regions ; and section [ sec : knota ] summarizes the observations related to the jet `` knot a '' . section [ sec : spnuc ] discusses the implications of the spectral energy distribution of the nucleus . section [ sec : lightprof ] analyses the galaxy light profile and its implications for the mass of the black hole . section [ sec : sbagn ] discusses the implications of these observations on the starburst - agn connection , and on the black hole fueling mechanism . finally , section [ sec : jet ] analyses the effect of the jet on the circumnuclear region . 0.5 cm centaurus a ( ngc 5128 ) was observed with both the wide field and planetary camera 2 ( wfpc2 biretta et al . @xcite ) and the near infrared camera and multi - object spectrograph ( nicmos mackenty et al . @xcite ) on - board hst . the wfpc2 data were taken on august 1 , 1997 and january 10 , 1998 ; the two observations differed in telescope roll angle by approximately 180 optimizing coverage of the nucleus and dust lane region in the wfpc2 field of view . the pointings included the nucleus and knot a in the higher resolution pc chips . the observations are logged in tab . [ tab : logobs ] . multiple exposures were performed in each of the f336w ( u ) , f555w ( v ) and f814w ( i ) filters , and final images were produced by combining long and short exposures in each band , allowing removal of cosmic rays and correction for saturated pixels in the longer exposures . the data were recalibrated with the pipeline software _ calwfp _ in the stsdas / iraf reduction package using up to date reference frames . warm pixels and cosmic rays were removed ( stsdas tasks _ warmpix _ and _ crrej _ ) and mosaiced images with spatial sampling of 01/pix were obtained using the _ wmosaic _ task , which also corrects for the optical distortions in the four wfpc2 chips . the background was estimated from areas with the lowest count rates in large , dust - extincted regions ; flux calibration was performed using the conversion factors given by whitmore ( @xcite ) . nicmos camera 2 ( nic2 ) observations of the nucleus of centaurus a were obtained on august 11 , 1997 in the f222 m ( k ) , f187n ( pa@xmath2 ) and f190n ( pa@xmath2 continuum ) filters ; these data were extensively discussed in paper ii . deep nic2 observations of the x - ray / radio feature `` knot a '' , centered @xmath14 nw of the nucleus , were obtained on september 9 , 1997 in the f160w ( h ) and f110w ( j ) filters . nicmos camera 1 observations of the nucleus of centaurus a were obtained on august , 14 1998 in the narrow band filters f164n and f166n ( and adjacent continuum ) . nicmos camera 3 ( nic3 ) observations with the f160w and f187n filters were carried out during the nic3 campaign in june 1998 as part of a snapshot survey of nearby galaxies ( proposal i d 7919 ) . further nic2 observations of the nuclear region of centaurus a with the f160w filter , obtained as part of a go snapshot survey ( proposal i d 7330 ) , were retrieved from the hst data archive . all the nicmos data were recalibrated using the pipeline software calnica v3.1 ( bushouse et al . @xcite ) and the best reference files in the hubble data archive as of may 1998 . for further discussion of nicmos data reduction see paper ii . the nicmos observations are logged in tab . [ tab : logobs ] . to construct full field images , to compare features seen in different observations taken at different times and with different instruments and , especially , to create color maps , we calculated the relative positions of the different pointings via cross correlation techniques . the wfpc2 observations taken at different dates were initially aligned to one another using image wcss ( world coordinate system , see voit et al . @xcite for a description ) which provide the ra and dec of each pixel , based on the hst guide star catalogue ( gsc ) reference system , accurate up to @xmath151 in absolute position . final alignment was determined by fitting the positions of @xmath1310 stars in the overlap region ; the resulting images are found to be aligned to better than 0.3 wf pixels ( i.e. 003 ) . the nicmos and wfpc2 images were aligned again by first using the image wcss , and then cross - correlating selected regions with well - defined morphologies . the positions of stars observed in both wfpc2 and nicmos images showed that the alignment was accurate to better than 0.4 nic2 pixels ( @xmath16 ) . 0.5 cm figures [ fig : f814w][fig : f336w ] are gray - scale representations of the wfpc2 mosaics in the filters f814w , f555w and f336w , corresponding to the standard i , v and u bands . figure [ fig : truecolor ] is a true color ( red = f814w , green = f555w , blue = f336w ) representation of the above data . the scale is 01/pix . the images reveal significant detail of the filamentary structure of the dust lane as well as the presence of numerous sources over the entire field of view . these comprise both unresolved point sources and many spatially resolved globular clusters and stellar associations . particularly prominent in the true - color image is a cluster of blue stars at the nw upper edge of the dust lane . several clusters of blue stars are also present within the dust lane and at its southern edge , clearly manifesting active star formation ( dufour et al . @xcite , paper i ) . the prominent dark bands at the edges of the dust lane presumably correspond to folds in the dusty molecular disk proposed as responsible for the observed obscuration by quillen , graham & frogel ( @xcite ) . photometry of the stars and stellar associations in the data will be discussed in subsequent papers . 0.5 cm in paper ii , we identified a strong point source observed at 2.2@xmath17 m as the active galactic nucleus of centaurus a , via its spatial coincidence with the radio nucleus ( within the @xmath18absolute positional uncertainty of the hst gsc ) . although the current wfpc2 images in u , v and i do not show a prominent galactic nucleus , even with the superb spatial resolution of hst , the nucleus can be located in the optical data using the accurately aligned ir and optical images ( sec . [ sec : observ ] ) . figure [ fig : nucleus ] shows the @xmath19 region centered around the ir peak . from left to right are shown : the nicmos f222 m data with the prominent nucleus ; the similarly prominent unresolved component in the f160w image ; the optical counterpart of the nucleus seen as a faint point source in the wfpc2 f814w data ; and a marginally detected source in f555w . to our knowledge , these wfpc2 data contain the first unambiguous detection of the nucleus of centaurus a at optical wavelengths . the nucleus is unresolved at all wavelengths , providing upper limits on the fwhm of the nuclear emission region of @xmath20 ( 1.7pc ) at 2.2@xmath21 and @xmath22 ( 0.9pc ) at 8000 respectively . the data allow us to derive the spectral energy distribution of the nucleus from the optical to the near - ir where , at the high spatial resolution of hst , the central source is easily isolated from the galaxy background . the sed is discussed further in section [ sec : spnuc ] below . note that the extreme steepness of the spectrum ( the nucleus is @xmath23 times fainter in v than in k ) has an important impact on the accuracy of the nuclear broad band photometry . the standard procedure used to flux calibrate hst images involves multiplying the count rate by the photflam keyword ( the conversion factor to flux units in erg s@xmath6 @xmath24 @xmath6 ) . however , this conversion factor assumes a flat continuum ( i.e. a constant @xmath25 ) which , clearly , does not apply to the cen a nucleus , whose overall 0.5 2.2 @xmath17 m spectrum can be approximated by a power law with spectral index -6.6 , i.e. @xmath26 . therefore , the flux scale must be re - computed for each filter by convolving the filter transmission curve with the source spectrum . we adopted an iterative procedure : using the photflam keywords we obtained zero order flux estimates , fitted a reddened power law , and re computed the photflam keywords , repeating the process until it converged . the decrease of the photflam keyword was less than 1% in f222 m and f190n , but @xmath27 in f160w , and 40% and 55% for f814w and f555w respectively ( see table [ tab : spnuc ] ) . in conclusion , it is essential to convolve the steep nucleus input spectrum with the instrumental responses . the resulting spectrum derived from our corrected hst photometry is presented in fig . [ fig : spnuc ] ( left panel ) and is discussed in sec . [ sec : spnuc ] . 0.5 cm in paper ii , we presented nicmos 2 observations in of the nuclear region of centaurus a , which we interpreted as an inclined , @xmath28pc diameter ionized disk . in figure [ fig : feiidisk ] we overlay on the image the intensity contours of continuum subtracted @xmath1$}}\lambda 1.64\mu$]m image . morphologically , the emission in and are almost indistinguishable and the / ratio is @xmath29 . the total detected emission of the disk ( i.e. in a @xmath30 aperture ) is @xmath31 erg s@xmath6 @xmath24 which is in excellent agreement with the flux of @xmath32 erg s@xmath6 @xmath24 in a @xmath33 aperture centered on the nucleus found by simpson & meadows ( @xcite ) with ground - based near - ir spectroscopy . the ratio @xmath1$}}/{\mbox{$\rm pa\alpha$}}=1/3 $ ] corresponds to /br@xmath34 ( 6.6 with @xmath35 and 12 with @xmath36 ) . after taking into account extinction the high /br@xmath37 value is typical of low excitation seyfert galaxies and liners ( cf . moorwood & oliva @xcite , alonso - herrero et al . the much larger disks detected by hst at the centers of elliptical galaxies usually have liner spectra , a typical example being the gaseous disk in m87 ( e.g. dopita et al . @xcite ) . 0.5 cm the x - ray spectral low energy cutoff has long provided strong evidence for absorption along the line of sight to the cen a nucleus . it has also long been clear from inspection of the optical images ( cf . [ fig : nucleus ] ) that the dust lane , interpreted as a warped disk of gas and dust crossing the nuclear region , produces significant foreground obscuration . to study the sed of the nucleus , it is crucial to disentangle the foreground ( disk ) component from the intrinsic local absorption near the agn . the extinction due to the dust lane also prevents a direct determination of the radial light profile near the nucleus of the galaxy in the wavelength bands observed . we can derive an average light profile only by combining the available information in the f555w , f160w and f222 m bands after estimating the respective reddening corrections . this is of course possible only assuming that no significant colour gradients are present . we briefly present a simple , zero - order estimate to correct for reddening which , hypothesizing foreground screen extinction , is derived from the color in two bands at @xmath38 and @xmath39 as : @xmath40 where @xmath41 and @xmath42 are the intrinsic and reddened magnitudes at the effective wavelength @xmath38 , @xmath43 is the extinction curve , i.e. @xmath44 , and @xmath45 . @xmath46 can be derived from the color @xmath47 as : @xmath48 where we assume an intrinsic color , @xmath49 , constant over the entire field of view . different point spread functions ( psfs ) are taken into account by convolving each image with the psf of the other . we have adopted the reddening curve by cardelli , clayton and mathis ( @xcite ) according to which @xmath50 , @xmath51 , @xmath52 and @xmath53 . using the above formalism , we construct an e(b - v ) map for the nuclear region by combining the f160w ( h ) and f222 m ( k ) nicmos images . these are more suitable for the reddening determination than the v and i band fluxes , which drop to undetectable levels in many of the heavily reddened regions of the dust lane . we assume an intrinsic colour @xmath54 which is an average value for both spiral bulges and elliptical galaxies with a standard deviation of only 0.1 magnitudes ( e.g. hunt et al . @xcite and references therein ) . note that the color correction due to the non - standard nicmos filters is negligible ( 0.2 in jonhson s system corresponds to 0.24 in nicmos ) . to qualitatively verify the accuracy of the correction procedure , we apply the reddening correction to the f814w wfpc2 image . figure [ fig : reddening ] shows the f222 m image , the e(b - v ) image derived from h - k , the uncorrected and reddening corrected f814w images . the corrected f814w image is seen to be mostly smooth , similar to the f222 m image which is obviously little affected by reddening . the correction fails only in a region of high extinction below the nucleus ( i.e. the fold of the molecular disk , see quillen , graham & frogel @xcite ) where e(b - v ) reaches values higher than @xmath55mag . a small linear structure starts below the nucleus and extends nw across the dust lane . the extinction in the regions immediately surrounding the nucleus is @xmath56mag , in good agreement with the estimate of @xmath57mag given by schreier et al . ( @xcite ) . this value can be considered as a lower limit to the extinction of the nuclear light , caused by foreground extinction . in the rest of the field , apart from patches like a knot north of the nucleus ( @xmath58 ) , the extinction generally decreases from south to north , perpendicularly to the dust lane , reaching values as low as 0.1 . the reddening corrections allow us to combine data taken in different bands and reconstruct the galaxy radial light - profile . we combine the nic2 ( f222 m ) data , with its rather small ( @xmath59 ) field of view , with the larger fov nic3 ( f160w ) data ( reddening correction derived from the i - h color ) and the wfpc2 ( f555w ) data ( reddening correction derived from the v - i color ) . distortions caused by the patchy reddening extinction are well corrected in the azimuthal average , as can be seen in fig . [ fig : profiles ] ( left panel ) , where the h and k band light profiles after reddening correction ( filled symbols ) are smoother than those before correction ( empty symbols ) ; as expected , the improvement is more evident in the h band . the overall radial profile of the nuclear region of is shown in fig . [ fig : profiles ] ( right panel ) . the filled squares represent the points derived from the f555w data , the filled triangles from the f222 m data , and the empty squares from the f160w data . the light profiles from the f160w and the f222 m image are normalized to match that of the f555w image , using scaling factors determined from fitting the light profiles . we use the `` nuker - law , '' an empirical expression introduced by lauer et al . ( @xcite ) , to fit the core properties of elliptical galaxies observed at hst resolution : @xmath60^\frac{\gamma-\beta}{\alpha}\ ] ] where @xmath61 at the center , @xmath62 at larger radii , and the transition occurs at @xmath63 with surface brightness @xmath64 . @xmath2 represents the sharpness of the transition . the fit parameters , including the scaling factors from h to v ( @xmath65 ) and k to v ( @xmath66 ) , are determined via @xmath67 minimization . since it is not possible to correctly normalize the @xmath67 in order to perform a statistical analysis ( see sec . 4.1 of carollo et al . @xcite for an extensive discussion ) we estimated uncertainties by mapping the @xmath67 in the parameter space . for a thousand times , we randomly selected the starting values of the fit parameters and let the minimization algorithm converge to find all local minima of @xmath67 . acceptable fits were then selected by means of visual inspection . the results of the fit are @xmath64=2.55 ( @xmath68 erg @xmath24 s@xmath6 @xmath6 arcsec@xmath69 ) , @xmath70=3.35.6 , @xmath2=0.81.6 , @xmath71=1.62.0 and @xmath37=0.00.3 . the best fit solid line in the figure has @xmath64=3.9 , @xmath72 , @xmath2=1.0 , @xmath73 , @xmath74 and the scaling factors @xmath75 ) . @xmath76 and @xmath77 ( in @xmath75 ) . the relatively large confidence bands for the model parameters arise because of the uncertainties in the scaling factors and the small range of radius available for fitting . never - the - less , the profile appears to fit the data well . we discuss the implications of the radial profile in section [ sec : lightprof ] . 0.5 cm the nic3 f160w image and a continuum subtracted pa@xmath2 ( f187n ) image are presented in figure [ fig : nicmos3 ] , along with a 6 cm radio contour overlay ( discussed further in the next section ) . since no nic3 narrow band continuum image is available , we derive one from the f160w image . to account for reddening , we estimate e(b - v ) by comparing the f160w image with the wfpc2 f814w image ( the f222 m image used earlier has too small a field of view ) and then apply the differential reddening correction to f160w to match that of f187n . the results , as seen in the figure , appear reasonable most of the field is devoid of emission , and the structures due to reddening which appear in a direct line - continuum subtraction have disappeared . moreover , the morphology of the nuclear region , with the disk and bow - shock like features , is the same as seen in the nic2 pa@xmath2 continuum - subtracted data , as presented in paper ii . two chains of emission line knots are observed above and below the nucleus , roughly parallel to the dust lane . these are located in the regions of higher extinction on the folds of the putative warped disk . apart from the pa@xmath2 emission knots close to the nucleus , already extensively discussed in paper ii , the radio overlay shows no obvious correlation with the jet , in contrast with previous hst observations of seyfert or radio galaxies with radio jets ( e.g. capetti et al . @xcite , axon et al . @xcite and references therein ) . the fluxes from the more intense pa@xmath2 knots are of order @xmath78 erg @xmath24 s@xmath6 in a @xmath79 area , corresponding to @xmath80 ph s@xmath6 h - ionizing photons ( using d=3.5 mpc and a@xmath81mag as a reddening estimate in the regions of star formation ) . the total flux associated to these knots is @xmath82 erg @xmath24 s@xmath6 . 0.5 cm there is a remarkable correspondence between the radio jet and the broad - band v - i color map , as seen in fig . [ fig : viradio ] . a relatively blue linear feature coincides with the trajectory of the radio jet from the nucleus to the so - called `` knot a '' , well known since early x - ray / radio observations and stretching from 15 to 30 nw of the nucleus , as indicated in the figure . faint emission in this `` blue channel '' is also apparent in the f336w image ( upper right panel in fig . [ fig : viradio ] ) . the f160w and f110w nic2 images of the region around knot a are displayed in fig . [ fig : knota ] along with the f814w and f555w wfpc2 images ( rescaled and aligned via the cross - correlation technique described in sec . [ sec : observ ] ) . the figure also overlays 6 cm vla radio contours ( e. feigelson , private communication ) on the f110w - f160w and f555w - f814w images . the radio is overlayed on the optical infrared image by aligning the radio and infrared nuclei . although there is no obvious optical emission correlated with the radio knots , there does appear to be a correlation between the radio and the near ir / optical colors : the radio contours lie in a region of bluer continuum emission , similar to that observed in the linear region discussed above . particularly intriguing is the presence of redder spots , apparently coincident with bends in the radio jet . these red spots might represent dense , dusty clouds which deflect the radio jet . it is worth noting that we do not detect any line or continuum emission associated with this putative interaction region . we have investigated the possibility that continuum emission from the jet might be present but `` masked '' by galaxy emission and the patchy extinction . using as a template the spectrum of a blue , bright region far from knot a , we find that within statistical errors , knot a has the same spectral shape , but with different scaling factors and e(b - v ) , implying no significant emission from the jet itself . note also that an extrapolation of the radio synchrotron spectrum to the optical would not predict any detectable flux . 0.5 cm lcccr + nicmos / f222 m & 2.218 & 2.38e-15 & 5% & 0.99@xmath3photflam + nicmos / f190n & 1.900 & 1.62e-15 & 5% & 1.00@xmath3photflam + nicmos / f160w & 1.610 & 5.31e-16 & 5% & 0.93@xmath3photflam + wfpc2/f814w & 0.803 & 3.15e-18 & 12% & 0.60@xmath3photflam + wfpc2/f555w & 0.547 & 6.34e-20 & 33% & 0.45@xmath3photflam + wfpc2/f336w & 0.337 & @xmath836.8e-19 & & 1.00@xmath3photflam + + isocam / lw2 & 7 & 0.6 & & + isocam / lw3 & 15 & 1.2 & & + scuba/450 & 450 & 13 & & + scuba/850 & 850 & 19 & & + + iras & 12 & 11.2 & + iras & 25 & 20.1 & + iras & 60 & 145 & + iras & 100 & 217 & + + rxte / pca / hexte & 210 & 3.40e-10 & 2% & phot . index @xmath84 + + vla & 20 & 3584 & 3% + vla & 6 & 6984 & 3% + 0.5 cm fig . [ fig : spnuc ] ( right panel ) compares our photometry of the nucleus of centaurus a with selected measurements from the literature ( see also table [ tab : spnuc ] ) . in x - rays , the 2.5240 kev spectrum observed by rxte is well fitted by a power law ( plus a fe k shell line ) with photon index @xmath131.9 , absorbed by a column density of @xmath85 @xmath24 ( rothschild et al . @xcite ) . in the mid - ir submm range , we consider the recent iso and scuba observations of mirabel et al . ( @xcite ) . we also plot the fluxes reported by iras although , with its relatively poor angular resolution , these points are dominated by the extended galaxy emission ; as shown by iso , the nucleus contributes at most 10% of the total mid - ir flux ( mirabel et al . @xcite ) . indeed , the iso points themselves overestimate the contribution from nucleus , since even the smallest 7 aperture includes significant extended galaxy emission . at radio wavelengths we employ the vla fluxes of the nucleus of burns , feigelson and schreier ( @xcite ) . the absorption toward the nucleus of centaurus a is inferred both from x - rays and infrared observations . in x - rays the observed absorbing column density corresponds to a visual extinction of @xmath86 mag ( assuming a standard gas - to - dust ratio @xmath87 ) . we estimate @xmath88 mag applying the extinction curve of draine and lee ( @xcite ) to the the 10@xmath21 silicate absorption feature in the iso spectrum ( mirabel et al . the two values closely follow the relationship observed in seyfert galaxies which , on average , have @xmath89 , where @xmath90 is the extinction determined from ir spectra and @xmath91 is determined from x - ray absorbing column densities , with the assumption of galactic dust - to - gas ratio ( see sec . 3 of granato , danese & franceschini @xcite ) . we note that in section [ sec : redden ] above we have estimated a lower limit for the extinction toward the nucleus of @xmath92 . the presence of a non - thermal source in cen a is well established from both the x - ray and radio observations and for this reason , a power law fit of the sed of the nucleus is a viable starting hypothesis . indeed , packham et al . ( @xcite ) described the ir nuclear spectrum of cen a as a power law reddened by 16 mag . however , an extrapolation of their fit underestimates our v band flux by more than a factor of 1000 while , at the same time , over - predicting the 10 @xmath17 m iso flux by a factor of 20 . this drawback is common to any reddened power law reddening can not reproduce the sharp break at @xmath93 m unless the absorption is as large as e(b - v ) @xmath94 , making the optical emission unobservable . an intrinsically steep spectrum is also unacceptable since the observed spectrum at long wavelengths , where reddening is negligible , is essentially flat . on the other hand , an extrapolation to the optical of the observed x - ray power law , reddened by @xmath95mag ( consistent with the extinction derived from the silicate feature and from the x - ray extinction ) , accounts for the v and i measurements without exceeding the ir points ( dotted line in fig . [ fig : spnuc ] ) which , therefore , must be accounted for by another component . the infrared emission of agns at wavelengths longer than 2@xmath17 m is usually interpreted as thermal emission from hot dust close to the active nucleus ( e.g. , rieke & lebofsky @xcite ) . the observed sed of the cen a nucleus in the mid - infrared is also qualitatively consistent with the presence of warm dust , while the sharp drop of the nuclear intensity longward of 3 @xmath17 m is reminiscent of dips seen in quasar spectra , usually ascribed to the cutoff in the dust temperature distribution due to the sublimation of grains ( sanders et al . we note that silicates sublime at @xmath96 1400 k and graphite at @xmath96 1750 k. we test quantitatively if dust emission is consistent with the observed sed by estimating the dust temperature , @xmath97 , which corresponds to the colors measured in adjacent bands : @xmath98 where @xmath99 is the dereddened flux , @xmath100 is the blackbody function and the dust emissivity is @xmath101 . @xmath71 is usually in the range 1.5 - 2 and here we assume @xmath102 . the derived temperature increases from @xmath103 obtained from @xmath104 , to @xmath105k for i - h , and to @xmath106 for v i . this result is consistent with the idea that we are seeing dust with a broad temperature distribution , and that at shorter wavelengths we are sampling regions of higher temperature . however , as we have shown above , the nucleus of cen a suffers a foreground extinction of at least e(b - v ) @xmath107 and the reddening corrected v - i slope becomes incompatible with thermal emission since the corresponding temperature @xmath108 exceeds the dust sublimation temperature . the spectrum drawn with a thick solid line in fig . [ fig : spnuc ] ( left and right panel ) is the combination of two components : the extrapolation of the x - ray power law reddened by @xmath95 mag and dust emission at 700k reddened by @xmath109 mag , i.e. @xmath110 the minimum foreground extinction to the nucleus . dust emission is described by @xmath111 . this composite spectrum well reproduces the observed points : the emission in v and i is completely `` non - thermal '' , that in k is mostly ( @xmath112 ) thermal , while in the h band the two emission processes are comparable . of course these results are only indicative and the emission in the near infrared could be completely dust dominated . note that even if the 700k component fits the observed hst data points , lower temperature components are required in order to fit the iso and scuba measurements . however , a complete analysis of the infrared emission spectrum is beyond the aims of this paper . therefore our data points on the nucleus are consistent with a scenario in which dust dominates emission in the near infrared but non - thermal emission accounts for the v and i bands and contributes to the near - ir emission . this is in agreement with previous suggestions that centaurus a hosts a misaligned bl lac nucleus ( e.g. bailey et al . @xcite , morganti et al . @xcite , steinle et al . @xcite and references therein ) . near - ir polarimetric observations at hst resolution can further test this scenario . turner et al . ( @xcite ) reported that the nucleus of centaurus a at 3.5 @xmath17 m decreased its flux by a factor 2.5 in less than 5 years from 1987 to 1992 ( comparing their data with those of lpine et al . @xcite ) . comparing the surface brightness profiles and performing aperture photometry in our f160w and f187n images , obtained over 9 and 10 month baselines , respectively , we find no evidence for variability of the nucleus above the 10% level . we note that even at the highest spatial resolution 01 , corresponding to 1.7pc or @xmath113 years the variability reported by turner et al . ( @xcite ) is not necessarily indicative of non - thermal emission , since the timescales are also consistent with variability in dust emission . using the formula by barvainis ( @xcite ) the dust sublimation radius is @xmath114pc ( with @xmath115 ) therefore only variability on time scales shorter than @xmath116 months can unambiguously prove the presence of a non thermal component in the near - ir . for completeness , even if the proposed scenario accounts for many of the properties of the centaurus a nucleus , we note that emission in the v band could also be accounted for by a central star cluster . 0.5 cm according to the classification by faber et al . ( @xcite ) , the brightness distribution of centaurus a derived earlier is consistent with a `` core '' profile , with @xmath117 , and with a break at a radius of @xmath118 . in general , galaxies with core profiles have @xmath119 , while those with power law profiles have @xmath120 , and both types can exist at intermediate values . fitting a de vaucouleurs law to the v - band isophotes of the galaxy outside the dust lane , dufour et al . ( @xcite ) found @xmath121 and estimated the total v magnitude for the elliptical component of centaurus a as 6.20 mag which , when corrected for e(b - v)=0.1 of foreground extinction , becomes v@xmath122 mag . from our profile fitting of the inner region , we estimate the contribution of the central cusp , and find that the total flux is increased by @xmath123 with respect to a simple de vaucouleurs law extrapolated in to the nucleus . this corresponds to -0.15 in magnitude i.e. v@xmath124 mag . using a cen a distance modulus of @xmath125 ( hui et al . @xcite ) , we find @xmath126 ( i.e. log l = 10.8 in solar units ) , consistent with faber s core profile classification . van der marel ( @xcite ) has recently shown that the faber classification is consistent with the hypothesis that ( i ) all early - type galaxies have central bhs that grew adiabatically in homogeneous isothermal cores and ( ii ) these `` progenitor '' cores follow scaling relations similar to those of the fundamental plane . this model suggests a roughly linear correlation between the mass of the central black hole and the v - band galaxy luminosity @xmath127 in solar units ( with an rms scatter of 0.33 dex , or a factor two in the black hole mass ) . this relationship is in agreement with the average relationship for nearby galaxies with kinematically determined black hole masses ( e.g. magorrian et al . @xcite ) . applying the above hypotheses to the case of centaurus a , we estimate the black hole mass to be @xmath128 m@xmath5 with a factor @xmath132 rms uncertainty . given the large uncertainties , this value is well in agreement with a dynamical mass of @xmath129 m@xmath130 within 40pc , derived from h@xmath131 spectroscopy ( israel @xcite ) . 0.5 cm the pa@xmath2 emission we observe in the dust lane strongly suggests that active star formation is taking place in the circumnuclear region . we can estimate the star formation rate from in the inner @xmath132 ( i.e. nic 3 field of view ) . as shown in section [ sec : nic3 ] , the total observed flux is @xmath133 erg @xmath24 s@xmath6 , which , assuming a mean extinction of @xmath35 , corresponds to an ionizing photon rate of @xmath134 ph s@xmath6 . according to the models by leitherer & heckman ( @xcite ) , the inferred star formation rate ( sfr ) depends mostly on the assumed imf ( initial mass function ) slope and upper mass cutoff and just slightly on metallicity . with a salpeter imf and 30m@xmath5 upper mass cutoff , the continuous sfr is 0.3m@xmath5 yr@xmath6 . this further decrease to 0.1m@xmath5 yr@xmath6 if the upper mass cutoff is 100m@xmath5 , but increases to 1m@xmath5 yr@xmath6 with the same upper mass cutoff and a steeper imf . in any case we find a moderately high sfr in a small ( @xmath135pc ) region . star formation in the nuclear environment of agns is not uncommon . in particular local seyfert 2 galaxies , i.e. obscured agns , appear characterized by enhanced circumnuclear star formation . observational evidence comes from the strength of the far - ir continuum ( rodriguez - espinoza et al . @xcite ) , from mid - ir 10@xmath17 m data ( maiolino et al . @xcite ) , from the nuclear mass - to - light ratio ( oliva et al . @xcite ) , from narrow band optical images ( gonzales - delgado et al . @xcite ) and from the detection of uv spectral features typical of o stars in a sample of seyfert 2 galaxies ( heckman et al . @xcite , gonzales - delgado et al . @xcite ) . a likely explanation of this starburst - agn connection is that , in order to feed the agn , large amounts of gas must be dumped in the inner kpc of an active galaxy . the processes which can accomplish this ( galaxy interactions , bars etc . , see below ) also trigger star formation in the dense , concentrated gas . a spectacular example of this process is given by maiolino et al . ( @xcite ) who recently discovered a nuclear gaseous bar in the circinus galaxy , a well known seyfert 2 . the observed detailed morphology in suggests two parallel chains of star formation located in regions of high extinction n and s of the nucleus . we can relate this morphology both to the warping of a molecular disk comprising the dust lane , and also to the bar - like structure recently observed by iso . quillen et al . ( @xcite ) suggested that the regions of high extinction above and below the nucleus are folds in the thin , warped molecular disk evidenced by co emission . the extinction is higher at the folds because the disk is nearly edge - on . if star formation is taking place in this molecular disk , then the pa@xmath2 morphology is also easily explained , since a higher number of hii regions are projected along our line of sight . however , the pa@xmath2 emission is also seen to be correlated with the edges of the putative bar observed with iso at 7 and 15 @xmath17 m ( mirabel et al . @xcite ) , as shown in figure [ fig : isopa ] . this correspondence is not unexpected , since the 7@xmath17 m emission is presumably dominated by pah re - radiation , which could be excited by the hot young stars seen in pa@xmath2 . note that the more diffuse ir emission can be attributed to older ( b or a ) stars which still emit enough uv photons to excite the pahs , but not enough to create the giant hii regions observed . if the ir morphology is indeed indicative of a bar , as suggested by mirabel et al . ( @xcite ) , the observed morphology is consistent with the overall scenario which relates the onset of activity in galactic nuclei to interactions between galaxies ( eg . , baade & minkowski @xcite ; toomre & toomre @xcite ; heckman et al . @xcite ; see also balick & heckman @xcite , and barnes & hernquist @xcite for reviews ) . specifically , `` minor mergers '' between a host galaxy and a small gas - rich companion or dwarf galaxy are regarded as an important means of fueling active nuclei ( eg . , mihos & hernquist @xcite ; hernquist & mihos @xcite ; walker et al . @xcite ; bahcall et al . @xcite ; de robertis , yee & hayhoe @xcite ) , in which case @xmath136 @xmath137 of stars , gas and dust can be dumped in toward the central regions , providing a fuel supply for the agn . such a scenario has long been invoked for cen a ( cf . discussion in paper ii ) . galaxy interactions are also known to be appealing mechanisms for triggering star formation via shocks induced in molecular clouds . detailed n - body / hydrodynamic models suggest that interactions between ellipticals and small gas - rich galaxies result in the accretion of gas into the central regions of the elliptical ( eg . weil & hernquist @xcite ; mihos & hernquist @xcite ) , and that star formation can then be triggered via accretion shocks in molecular clouds . the gas is driven toward the center of the galaxy in response to tidal forces ( binney & tremaine @xcite ; barnes & hernquist @xcite ) . in centaurus a , we see star formation regions both within the dust lane and at the edges of the putative bar . models predict shocks predominantly at the leading edges of such bars ( cf . athanassoula @xcite , @xcite ) . the shocks are likely to trigger star formation , explaining the observed morphology , wherein the hii region are located at the edges of the bar and along the `` spiral arms '' . at smaller scales , the pa@xmath2 `` disk '' around the nucleus , discussed in paper ii , could then be supplied by the gas streaming in from the bar , ionized by the agn . note that the orientation of the pa@xmath2 feature is perpendicular to the edges of the bar . in summary , the feeding of the agn requires gas to be driven in toward the nucleus , and the same galaxy interaction / merger which enriches the elliptical galaxy with gas , can trigger star formation via shocks induced in molecular clouds . 0.5 cm the data presented in sec . [ sec : knota ] provide no evidence for synchrotron emission from the jet , either in the channel leading to knot a or from knot a itself . we note that an extrapolation of a simple synchrotron spectrum from the radio would not produce detectable optical synchrotron emission ( see e.g. paper i ) . in this regard , note also that we see no pa@xmath2 related to the jet , except possibly within a few arcseconds of the nucleus ( see paper ii ) . we thus exclude emission from young stars whose formation might have been triggered by the interaction between the jet and the ism . we also note that we can not exclude free - free emission from hot shocked gas . we suggest that the `` blue channel '' corresponds to a region of gas `` evacuated '' by the jet , causing lower extinction than the surroundings . we can investigate the feasibility of this hypothesis the evacuation by the jet of a region of gas corresponding to the observed extent of the blue feature given what is known about the lifetime of the source and the properties of the jet . the detailed radio studies by schreier , burns & feigelson ( @xcite ) , burns , feigelson & schreier ( @xcite ) , clarke , burns & feigelson ( @xcite ) and clarke , burns & norman ( @xcite ) , suggest that the bulk kinetic power of the northern jet is @xmath138 erg s@xmath6 , while the minimum age estimate of the inner lobes is @xmath139 yr . a simple , order - of - magnitude estimate of the amount of energy required to evacuate material from this region assumes it is described by a cylindrical geometry with a diameter @xmath140 ( @xmath141pc ) and length @xmath142 ( @xmath143 pc ) , initially filled uniformly by an ism with density @xmath144 @xmath145 . to displace this amount of material by a distance approximately equal to its length , over the lifetime of the radio source ( assuming @xmath146 yr as a lower limit ) , would require it to be accelerated to a velocity @xmath147 km s@xmath6 over this timescale . assuming a constant kinetic energy input , the required kinetic energy flux into the ism is therefore @xmath148 erg s@xmath6 . thus , evacuating the required volume of gas over the source lifetime would require a conversion efficiency of order @xmath149% of the total jet kinetic power , well within the bounds of feasibility . the lack of strong line emission from the jet could suggest a scenario in which the surrounding ism is relatively dense . if the interaction with the jet only results in slow shocks into the gas ( @xmath150 km s@xmath6 ) , then little or no optical line emission is produced . however , the lack of pa@xmath2 or h2 suggesting no obvious star formation argues against this . we conclude that we may indeed be seeing the channel evacuated by the jet between the agn and the first bright x - ray / radio knot , despite the lack of detected emission from the knot itself . 0.5 cm our hst _ wide field and planetary camera 2 _ and _ near infrared and multi object spectrograph _ observations of centaurus a ( ngc 5128 ) have provided several new insights about the active galactic nucleus . detection of unresolved emission in both visible light and near - ir suggests that two different emission mechanisms are required : a non - thermal component which represents an extrapolation of the x - ray power law , reddened by @xmath0 ( consistent with mid - ir and x - ray estimates ) , and a thermal component probably caused by emission from hot dust within 2pc of the nucleus . these results are in agreement with previous suggestions that centaurus a harbors a misaligned bl lac nucleus . we have shown that only variability on time scales shorter than @xmath116 months can unambiguously prove the existence of a non - thermal component in the near - ir . we have detected the 20pc - scale nuclear disk in the @xmath1$}}\lambda 1.64\mu$]m line which shows a morphology similar to that observed in with an / ratio typical of low ionization seyfert galaxies and liners . we do not see evidence for optical emission from , or star formation associated with , the radio / x ray jet . we do detect a blue linear structure , aligned with jet and extending from the nucleus to knot a , which we interpret as a region of relatively low reddening . this feature may represent a channel in which gas and dust have been swept away by the jet , consistent with simple energetic arguments . the data allow us to derive a map of dust extinction , e(b - v ) , in a 20@xmath320 circumnuclear region and reveal a several arcsecond long dust feature near but below the nucleus , oriented in a direction transverse to the large dust lane . this structure may be related to the bar observed with iso and scuba , as reported by mirabel et al . ( @xcite ) . we find rows of pa@xmath2 emission knots along the top and bottom edges of the bar , which we interpret as star formation regions , possibly caused by shocks driven into the gas . the inferred star formation rates are moderately high ( @xmath151m@xmath5 yr@xmath6 ) . if we hypothesize , with mirabel , that the bar represents a mechanism for transferring gas in to the center of the galaxy , then the large dust lane across the galaxy , the bar , the knots , and the inner pa@xmath2 disk previously reported in paper ii , all represent aspects of the feeding of the agn . gas and dust are supplied by a recent galaxy merger ; a several arcminute - scale bar allows the dissipation of angular momentum and infall of gas toward the center of the galaxy ; subsequent shocks trigger star formation ; and the gas eventually accretes onto the agn via the 20pc disk . reconstructing the radial light profile of the galaxy to within 01 of the nucleus by combining v , h and k measurements , corrected for reddening , shows that centaurus a has a core profile , according to the classification of faber et al . ( @xcite ) . using the models of van der marel ( @xcite ) , we estimate a black hole mass of @xmath8m@xmath5 , consistent with ground based kinematical measurements ( israel @xcite ) . this is consistent with the presence of a strong agn , as suggested by the large radio lobes , the jet , and the strong x - ray emission . it further suggests that the ionized gas disk seen in pa@xmath2 ( paper ii ) would have relatively high rotational velocities , of order 800 km / sec . planned high spatial resolution near - ir spectroscopy should be able to accurately determine the mass of the super - massive black hole in this nearest agn . a.m. acknowledge the partial support of the italian space agency ( asi ) through the grant ars98116/22 and of the italian ministry for university and research ( murst ) under grant cofin98 - 02 - 32 . a.m. and a.c . acknowledge support from the stsci visitor program . acknowledge support through go grants o0570/go-7267 and o0568/go-6578 from space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 526555 . we thank nino panagia , tino oliva , marco salvati , roberto maiolino and lucia pozzetti for helpful discussions which greatly improved this paper . we also thank felix mirabel and dave saunders for useful discussions and pre - publication access to the iso data . alonso - herrero a. , rieke m.j . , rieke g.h . , ruiz m. , 1997 apj , 482 , 747 athanassoula e. , 1992a mnras , 259 , 328 athanassoula e. , 1992b mnras , 259 , 345 axon d.j . , marconi a. , capetti a. , macchetto f.d . , schreier e.j . , robinson a. , 1998 , apj , 496 , l75 baade w. , minkowski r. , 1954 apj , 119 , 215 bahcall j.n . , kirhakos s. , saxe d.h . , schneider d.p . , 1997 apj , 479 , 642 bailey j. , sparks w.b . , hough j.h . , axon d.j . , 1986 nature , 32 , 150 balick b. , heckman t.m . , 1982 ara&a , 20 , 431 barnes j.e . , hernquist l. , 1992 ara&a , 30 , 705 barvainis r. , 1987 apj , 320 , 537 binney j. , tremaine s. , 1987 , `` galactic dynamics '' , princeton university press biretta j.a , et al . , 1996 , _ wfpc2 instrument handbook _ , version 4.0 ( baltimore : stsci ) bolton j.g . , stanley g.j . , slee o.b . , 1949 nature , 164 , 101 bowyer c.s . , lampton m. , mack j. , 1971 , apj , 161 , l1 burns j.o . , feigelson e.d . , schreier e.j . , 1983 apj , 273 , 128 bushouse h. , skinner c.j . , mackenty j.w . , 1997 , _ nicmos instrument sci . _ 97 - 28 ( baltimore : stsci ) capetti a. , axon d.j . , macchetto f.d . , 1997 apj , 487 , 560 cardelli j.a . , clayton g.c . , mathis j.s . , 1989 apj , 345 , 245 carollo c.m . , et al . , 1997 , apj , 481 , 710 clarke d.a . , burns j.o . , feigelson e.d . , 1986 apj , 300 , l41 clarke d.a . , burns j.o . , norman m.l . , 1992 apj , 395 , 444 de robertis m.m . , yee h.k.c . , hayhoe k. , 1998 apj , 496 , 93 dopita m.a . , et al . , 1997 , apj , 490 , 202 draine b.t . , lee h.m . , 1984 apj , 285 , 89 dufour r.j . , van den bergh s. , harvel c.a . , et al . , 1979 aj , 84 , 284 faber s.m . , tremaine s. , ajhar e.a . , et al . , 1997 aj , 114 , 1771 feigelson e.d . , schreier e.j . , delvaille j.p . , et al . , 1981 apj , 251 , 31 granato g.l . , danese l. , franceschini a. , 1997 apj , 486 , 147 gonzales delgado r.m . , leitherer c. , heckman t.m . , cervino m. , 1997 , apj , 483 , 705 gonzales delgado r.m . , heckman t.m . , leitherer c. , et al . , 1998 , apj , 505 , 174 graham j.a . , 1979 apj , 232 , 60 grindlay j.e . , schnopper h. , schreier e.j . , gursky h. , parsignault d.r . , 1975 , apj , 201 , l133 heckman t.m . , smith e.p . , baum s.a . , et al . , 1986 apj , 311 , 526 heckman t.m . , gonzales delgado r.m . , leitherer c. , et al . , 1997 , apj , 482 , 114 hernquist l. , mihos j.c . , 1995 apj , 448 , 41 hui x. , ford h.c . , ciardullo r. , jacoby g.h . , 1993 apj , 414 , 463 hunt l.k . , malkan m.a . , salvati m. , et al . , 1997 apjs , 108 , 229 israel f.p . , 1998 a&a rev . , 8 , 237 jones d.l . , tingay s.j . , murphy d.w . , et al . , 1996 apj , 466 , l63 kellogg e. , gursky h. , leong c. , schreier e. , tananbaum h. , giacconi r. , 1971 , apj , 165 , l49 lauer t.r . , ajhar e.a . , byun y .- i . , et al . , 1995 aj , 110 , 2622 leitherer c. , heckman t.m . , 1995 apjs , 96 , 9 lpine j.r.d . , braz m.a . , epchtein n. , 1984 a&a , 131 , 72 mackenty j.w . , et al . , 1997 , _ nicmos instrument handbook _ , version 2.0 ( baltimore : stsci ) magorrian j. , et al . , 1998 , aj , 115 , 2285 maiolino r. , ruiz m. , rieke g.h . , keller l.d . , 1995 apj , 446 , 561 maiolino r. , alonso - herrero a. , anders s. , et al . , 1999 apj , submitted malin d.f . , quinn p.j . , graham j.a . , 1983 apj , 272 , l5 mihos j.c . , hernquist l. , 1994 apj , 425 , l13 mihos j.c . , hernquist l. , 1996 apj , 464 , 641 mirabel i.f . , laurent o. , sanders d.b . , et al . , 1999 a&a , 341 , 667 moorwood a.f.m . , oliva e. , 1988 a&a , 203 , 278 morganti r. , robinson a. , fosbury r.a.e . , di serego alighieri s. , tadhunter c.n . , malin d.f . , 1991 mnras , 249 , 91 oliva e. , origlia l. , maiolino r. , moorwood a.f.m . , 1999 a&a , submitted packham c. , hough j.h . , young s. , et al . , 1996 mnras , 278 , 406 quillen a.c . , de zeeuw p.t . , phinney e.s . , phillips t.g . , 1992 apj , 391 , 121 quillen a.c . , graham j.r . , frogel j.a . , 1993 apj , 412 , 550 rieke g.h . , lebofsky m.j . , 1981 apj , 250 , 87 rodriguez - espinoza j.m . , rudy r.j . , jones b. , 1986 , apj , 309 , 76 rothschild r.e . , band d.l . , blanco p.r . , et al . , 1999 apj , 510 , 651 sanders d.b . , phinney e.s . , neugebauer g. , et al . , 1989 apj , 347 , 29 schreier e.j . , feigelson e. , delvaille j. , et al . , 1979 apj , 234 , l39 schreier e.j . , burns j.o . , feigelson e.d . , 1981 apj , 251 , 523 schreier e.j . , capetti a. , macchetto f. , sparks w.b . , ford h.j . , 1996 apj , 459 , 535 ( paper i ) schreier e.j . , marconi a. , axon d.j . , caon n. , macchetto d. , capetti a. , hough j.h . , young s. , packham c. , 1998 , apj , 499 , l143 ( paper ii ) simpson c. , meadows v. , 1998 apj , 505 , l99 steinle h. , bennett k. , bloemen h. , et al . , 1998 a&a , 330 , 97 toomre a. , toomre j. , 1972 apj , 178 , 623 turner p.c . , forrest w.j . , pipher j.l . , shure m.a . , 1992 apj , 393 , 648 van der marel r.p . , 1999 aj , 117 , 744 voit , m. , et al . , 1997 , _ hst data handbook - vol . i _ , version 3.0 ( baltimore : stsci ) walker i.r . , mihos j.c . , hernquist l. , 1996 apj , 460 , 121 weil m.l . , hernquist l. , 1993 apj , 405 , 142 whitmore b. , 1995 , in _ calibrating hst : post servicing mission _ , a. koratkar and c. leitherer ( eds . ) , baltimore stsci , p.269 wills b.j . , in `` quasars and cosmology '' , a.s.p . conference series ( 1999 ) , eds . g.ferland , j.baldwin ( san francisco : asp ) , in press ( astro - ph/9905093 )

collective behaviour is a widespread phenomenon in biology , cutting through a huge span of scales , from cell colonies up to bird flocks and fish schools . the most prominent trait of collective behaviour is the emergence of global order : individuals synchronize their states , giving the stunning impression that the group behaves as one . in many biological systems , though , it is unclear whether global order is present . a paradigmatic case is that of insect swarms , whose erratic movements seem to suggest that group formation is a mere epiphenomenon of the independent interaction of each individual with an external landmark . in these cases , whether or not the group behaves truly collectively is debated . here , we experimentally study swarms of midges in the field and measure how much the change of direction of one midge affects that of other individuals . we discover that , despite the lack of collective order , swarms display very strong correlations , totally incompatible with models of noninteracting particles . we find that correlation increases sharply with the swarm s density , indicating that the interaction between midges is based on a metric perception mechanism . by means of numerical simulations we demonstrate that such growing correlation is typical of a system close to an ordering transition . our findings suggest that correlation , rather than order , is the true hallmark of collective behaviour in biological systems . * author summary : * _ our perception of collective behaviour in biological systems is closely associated to the emergence of order on a group scale . for example , birds within a flock align their directions of motion , giving the stunning impression that the group is just one organism . large swarms of midges , mosquitoes and flies , however , look quite chaotic and do not exhibit any group ordering . it is therefore unclear whether these systems are true instances of collective behaviour . here we perform the three dimensional tracking of large swarms of midges in the field and find that swarms display strong collective behaviour despite the absence of collective order . in fact , we discover that the capability of swarms to collectively respond to perturbations is surprisingly large , comparable to that of highly ordered groups of vertebrates . _

1307.5631

intuition tells us that a system displays collective behaviour when all individuals spontaneously do the same thing , whatever this thing may be . we surely detect collective behaviour when all birds in a flock fly in the same direction and turn at the same time @xcite , as well as when all spins in a magnet align , giving rise to a macroscopic magnetization @xcite . on the other hand , we would not say that there is any collective behaviour going on in a gas , despite the large number of molecules . the concept of collective behaviour seems therefore closely linked to that of emergent collective order , or synchronization . indeed , explaining how order spontaneously arises from local inter - individual interactions has been one of the major issues in the field @xcite . the case of insect swarms is tricky in this respect . several species in the vast taxonomic order diptera ( flies , mosquitoes , midges ) form big swarms consisting largely of males , whose purpose is to attract females @xcite . swarming therefore has a key reproductive function and , in some cases , relevant health implications , the obvious , but not unique , example being that of the malaria mosquito , _ anopheles gambiae _ @xcite . it is well - known that swarms form in proximity of some visual marker , like a water puddle , or a street lamp @xcite . swarming insects seem to fly independently around the marker , without paying much attention to each other ( see video s1 ) . for this reason , the question of whether swarms behave as truly collective systems is debated @xcite . in fact , it has even been suggested that in diptera there is no interaction between individuals within the swarm and therefore no collective behaviour at all @xcite . although other studies observed local coordination between nearest neighbors @xcite , it remains controversial whether and to what extent collective patterns emerge over the scale of the whole group . clarifying this issue is a central goal in swarms containment @xcite . in absence of quantitative evidence telling the contrary , the hypothesis that external factors , as the marker , are the sole cause of swarming and that no genuine collective behaviour is present , is by far the simplest explanation . physics , however , warns us to be careful in identifying collective behaviour with collective order . there are systems displaying important collective effects both in their ordered _ and _ in their disordered phase . the most important example is that of a ferromagnet : the collective response to an external perturbation is as strong in the disordered phase slightly above the critical temperature @xmath0 , i.e. the temperature below which a spontaneous magnetization emerges , as it is in the ordered phase slightly below @xmath0 . in fact , once below the critical temperature , increasing the amount of order _ lowers _ the collective response @xcite . hence , although certainly one of its most visible manifestations , emergent order is not necessarily the best probe of collective behaviour . the crucial task for living groups is not simply to achieve an ordered state , but to respond collectively to the environmental stimuli . for this to happen , correlation must be large , namely individuals must be able to influence each other s behavioural changes on a group scale . the question then arises of whether correlation in biological systems is a consequence of collective order or whether it can be sustained even in absence of order , as it happens in some physical systems . the question is relevant because the way individuals in a group synchronize their behavioural fluctuations ( correlation ) is possibly a more general mechanism than the synchronization of behaviour itself ( order ) . all experimental studies performed up to now , however , concerned highly synchronized groups ( as bird flocks , fish shoals and marching locusts @xcite ) , which displayed both order and correlation . hence , the question of whether or not order and correlations are two sides of the same phenomenon remained open until now . here , we try and give an answer to this question by experimentally studying large swarms of insects in the field . as we will show , despite the lack of collective order , we do find strong correlations , indicating that even in biological systems collective behaviour and group - level coordination do not require order to be sustained . * experiments and tracking . * we perform an experimental study of swarms of wild midges in the field . midges are small non - biting flies belonging to the order diptera , suborder nematocera ( diptera : chironomidae and diptera : ceratopogonidae - see methods ) . the body length of the species we study is in the range @xmath1@xmath2 mm . swarms are found at sunset , in the urban parks of rome , typically near stagnant water . as noted before @xcite , we find that swarms form above natural or artificial landmarks . moving the landmark leads to an overall displacement of the swarm . the swarms we studied range in size between @xmath3 and @xmath4 individuals ( see table s1 ) . to reconstruct the @xmath5 trajectories of individual insects we use three synchronized cameras shooting at @xmath6 frames - per - seconds ( trifocal technique - fig.[fig : experiment]e and methods ) . our apparatus does not perturb the swarms in any way . the technique is similar to the one we used for starling flocks @xcite , with one notable difference . to reach the desired experimental accuracy we need to know the mutual geometric relations between the three cameras very accurately . in the case of flocks , this could be achieved only by an _ a priori _ alignment of the cameras . in the case of swarms , though , we proceed differently . after each swarm acquisition , we pin down the geometry of the camera system by taking multiple images of a calibrated target ( fig.[fig : experiment]f ) . this procedure is so accurate that the error in the @xmath5 reconstruction is dominated by the image segmentation error due to the pixel resolution . if we assume this to be equal to @xmath7 pixel ( typically it is smaller than that because midges occupy many pixels ) , we make an error of @xmath8 cm in the determination of the distance between two points @xmath9 cm apart from each other ( a reference value for nearest neighbour distance ) . the absolute error is the same for more distant points , making the relative precision of our apparatus even higher . this accuracy makes the determination of the correlation functions we study here very reliable . the @xmath5-tracking of each midge is performed by using the technique described in @xcite . sample @xmath5 reconstructions are shown in fig.[fig : experiment]b and in video s1 . compared to previous field @xcite and lab @xcite studies , the present work is the most extensive experimental study of swarming insects in three dimensions to date . $ ] . in red we report the reference value of the polarization in a flock of starlings . * middle : * rotational order parameter , @xmath10 $ ] . * bottom : * dilatational order parameter @xmath11 $ ] . ] * lack of collective order . * swarms are in a disordered phase . the standard order parameter normally used in collective behaviour is the polarization , @xmath12 , where @xmath13 is the number of midges in the swarm and @xmath14 is the velocity of insect @xmath15 . the polarization measures the degree of alignment of the directions of motion ; it is a positive quantity and its maximum value is @xmath7 . the average polarization over all swarms is quite small , @xmath16 ( see fig.[fig : order ] and table s1 ) . as a reference , in starling flocks we find @xmath17 , on average @xcite . the probability distributions of the polarization fully confirms the swarms lack of translational order and the stark difference with flocks ( fig.[fig : order ] ) . clearly , swarms are not in a polarized state . translation is not the only possible collective mode , though . for example , it is well - known that fish school can produce rotating ( milling ) configurations . moreover , a group can expand / contract , giving rise to dilatational ( or pulsive ) collective modes . for this reason we have defined and measured also a rotational and a dilatational order parameter ( see methods ) . we find , however , that these quantities too have very small values ( fig.[fig : order ] ) . the time series , on the other hand , show that the order parameters can have rare , but strong fluctuations , during which their value may become significantly larger than that of an uncorrelated system ( fig.[fig : order ] ) . these large fluctuations are a first hint that nontrivial correlations are present . * correlation . * the connected correlation function measures to what extent the change in behaviour of individual @xmath15 is correlated to that of individual @xmath18 , at distance @xmath19 . correlation is the most accessible sign of the presence of interaction between the members of a group . the absence of interaction implies the absence of correlation . conversely , the presence of correlation implies the presence of an effective interaction ( see supporting information , section i ) . correlation can be measured for different quantities , but in the case of midges , as with birds and other moving animals , the principal quantity of interest is the direction of moti to compute the connected correlation we first need to introduce the velocity fluctuations , namely the individual velocity subtracted of the overall motion of the group , @xmath20 ( see methods for the detailed definition of @xmath21 ) . this fluctuation is a dimensional quantity , hence it is unsuitable to compare the correlation in natural vs numerical systems , as we shall do later on . we therefore introduce the dimensionless velocity fluctuation , @xmath22 the connected correlation function is then given by , @xmath23 where @xmath24 if @xmath25 and zero otherwise , and @xmath26 is the space binning factor . the form of @xmath27 in natural swarms is reported in fig.[fig : main ] : at short distances there is strong positive correlation , indicating that midges tend to align their velocity fluctuations to that of their neighbours . after some negative correlation at intermediate distances , @xmath27 relaxes to no correlation for large distances . this qualitative form is quite typical of all species analyzed ( see fig.[fig : main ] ) . the smallest value of the distance where @xmath27 crosses zero is the correlation length , @xmath28 , that is an estimate of the length scale over which the velocity fluctuations are correlated @xcite . the average value of this correlation length over all analyzed swarms is , @xmath29 m . this value is about @xmath30 times larger than the nearest neighbours distance , whose average over all swarms is , @xmath31 m ( see fig.[fig : main ] and table s1 ) . previous works noticed the existence of pairing maneuvers and flight - path coordination between nearest neighbours insects @xcite . our results , however , indicate that midges within a natural swarm influence each other s motion far beyond their nearest neighbours . * susceptibility . * the collective response of the swarm depends crucially on two factors : how distant in space the behavioural change of one insect affects that of another insect ( spatial span of the correlation ) and how strong this effect is ( intensity of the correlation ) . to combine these two factors in one single observable we calculate the cumulative correlation up to scale @xmath19 , @xmath32 it can be shown ( see si - ii ) that this dimensionless function is related to the space integral of the correlation function @xmath27 hence , @xmath33 reaches a maximum where @xmath27 vanishes , i.e. for @xmath34 ( see fig.[fig : main ] ) . this maximum , @xmath35 , is a measure of the total amount of correlation present in the system . in statistical physics @xmath36 is exactly equal to the _ susceptibility _ , namely the response of the system to an external perturbation @xcite . in collective animal behaviour , we do not have a quantitative link between integrated correlation and response , so that calling @xmath36 susceptibility is not strictly correct . however , if the probability distribution of the velocities is stationary , we can follow a maximum entropy approach @xcite and still find that the total amount of correlation in the system , @xmath36 , is related to the way the group responds collectively to a perturbation ( see si - ii ) . the value of @xmath36 for midges swarms is reported in fig.[fig : main ] . * noninteracting swarm . * in order to judge how significant is the correlation function @xmath27 and how large is the susceptibility @xmath36 in natural swarms , we need an effective zero for these quantities , i.e. some null hypothesis baseline . as we have seen in the introduction , the minimal assumption is that all individuals in the swarm interact with an external landmark independently from each other . following okubo @xcite ( but see also @xcite and @xcite ) , we therefore simulate a ` swarm ' of noninteracting particles performing a random walk in a three - dimensional harmonic potential ( see methods ) . visually , the group behaviour of this noninteracting harmonic swarm ( nhs ) looks remarkably similar to that of a real swarm ( see video s2 and s3 ) : all ` midges ' fly around the marker and the group lacks collective order . this similarity , however , is deceptive . in the nhs , the correlation function @xmath27 simply fluctuates around zero , with no spatial span , nor structure ( fig.[fig : main ] ) . moreover , the susceptibility in the nhs is extremely small , @xmath37 , whereas the susceptibility in natural swarms is up to @xmath3 times larger than this non - interacting benchmark ( fig.[fig : main ] ) . we conclude that swarming behaviour is _ not _ the mere epiphenomenon of the independent response of each insect with the marker . despite the lack of collective order , natural swarms are strongly correlated on large length scales . there are big clusters of midges that move coherently , contributing to the ` dancing ' visual effect of the swarm . the only way this can happen is that midges interact . what kind of interaction is that ? as a function of the distance at one instant of time . the dashed vertical line marks the average nearest neighbour distance , @xmath38 , for that swarm . the correlation length , @xmath28 , is the first zero of the correlation function . red : correlation function in the nhs case . the value of @xmath19 for the nhs has been rescaled to appear on the same scale as natural distances . each natural swarm is compared to a nhs with the same number of particles . * middle : * comulative correlation , @xmath33 . this function reaches a maximum @xmath34 . the value of the integrated correlation at its maximum , @xmath39 , is the susceptibility @xmath36 . * bottom : * numerical values of the susceptibility , @xmath36 in all analyzed swarms . for each swarm the value of @xmath36 is a time average over the whole acquisition ; error bars are standard deviations . red : the average susceptibility @xmath40 in the noninteracting case . ] as a function of the rescaled nearest neighbour distance , @xmath41 , where @xmath42 is the body length . each point represents a single time frame of a swarming event , and all events are reported on the same plot ( symbols are the same for all species ) . the full line is the best fit to equation ( [ umma ] ) . * right : * logarithm of the average susceptibility as a function of @xmath41 . _ dasyhelea flavifrons _ - blue squares ; _ corynoneura scutellata _ - green circles ; _ cladotanytarsus atridorsum _ - red triangles . the full line represents the best fit to equation ( [ umma ] ) . each data point represents the time average over the entire acquisition of one swarming event . error bars indicate standard deviations . ] * metric interaction . * to answer this question we study the susceptibility across swarms of different densities . interestingly , we find that @xmath36 increases when the average nearest neighbour distance , @xmath38 , decreases ( fig.[fig : chi ] ) . denser swarms are more correlated than sparser ones . this result indicates that midges interact through a metric perceptive apparatus : the strength of the perception decreases with the distance , so that when midges are further apart from each other ( larger @xmath38 ) the interaction is weaker and the susceptibility @xmath36 is lower . this is at variance with what happens in bird flocks : birds interact with a fixed number of neighbours , irrespective of their nearest neighbour distance @xmath38 @xcite ; such kind of topological interaction does not depend on the group density , hence the susceptibility does not depend on the nearest neighbour distance . figure [ fig : chi ] , on the other hand , shows that midges interact metrically , namely with all neighbours within a fixed metric range , @xmath43 . hence , in swarms the number of interacting neighbours increases with decreasing @xmath38 ( increasing density ) , and as a consequence of this increased amount interaction , the system becomes also more correlated . in a system ruled by metric interaction we expect all lengths to be measured in units of the perception range , @xmath43 . this implies that the natural variable for the susceptibility is the rescaled nearest neighbour distance , @xmath44 . the problem is that we are considering different species of midges , likely to have different metric perception ranges . the simplest hypothesis we can make is that @xmath43 is proportional to the insect body length @xmath42 ( which we can measure ) , so that @xmath45 . this hypothesis is confirmed by the data : the susceptibility is significantly more correlated to the variable @xmath41 ( p - value @xmath46 ) than to @xmath38 ( p - value @xmath47 ) . the fact that the natural variable is @xmath41 is a further indication that the interaction in swarms is based on a metric mechanism . the difference in the nature of the interaction between flocking birds and swarming midges ( topological vs. metric ) is possibly due to the significant differences between vertebrates and arthropods . topological interaction , namely tracking a fixed number of neighbours irrespective of their distance , requires a level of cognitive elaboration of the information @xcite more sophisticated than a metric interaction , where the decay of the effective force is merely ruled by the physical attenuation of the signal with increasing metric distance . in other words , within a metric mechanism the range of the interaction is fixed by a perceptive cut - off , rather than a cognitive one . metric interaction is known to be more fragile than topological one against external perturbations @xcite , and indeed it is far more likely to observe the dispersion of a swarm in the field than that of a flock . this may be the reason why the presence of an external marker is crucial for the swarming behaviour of midges @xcite . * correlation without order . * the experimental observations of a notrivial connected correlation and of a large susceptibility indicate that midges are effectively interacting with each other by acting on their directions of motion . this does not exclude , of course , that other types of interaction are present . first of all , the empirical observation that the swarm uses a visual marker as a reference for maintaining its mean spatial position , strongly suggests that each individual interacts with the marker . besides , it is certainly possible that effective positional attraction - repulsion forces between midges , as those described in @xcite , exist . however , the directional correlations indicate that insects are _ also _ effectively interacting by adjusting their velocities . moreover , the fact that these correlations are positive for short distances means that midges tend to _ align _ their direction of motion . this fact may seem surprising , because alignment interactions typically lead to the formation of ordered ( polarized ) groups , which is clearly not the case for midges . swarms are disordered , and yet interacting and highly correlated systems . is this a paradox ? in fact , it is not . an alignment interaction does not _ per se _ lead to global order in the group . in all models where imitation of the neighbours is present , the onset of long - range order depends on the value of some key tuning parameter . in a ferromagnet , this parameter is the temperature @xmath48 , namely the amount of noise affecting the interaction between the neighbouring spins . at high temperature the system is in a disordered state , whereas by lowering @xmath48 one reaches a critical temperature below which an ordering transition occurs . in models of active matter there is another parameter tuning the transition between disorder and order , that is density or , equivalently , nearest neighbour distance : the system gets ordered once the nearest neighbour distance falls below some transition value . the crucial point is that , in general , the correlation of the system tends to be very large _ around _ the transition point , irrespective of whether the system is in the ordered or in the disordered phase . hence , even a disordered system can display large correlations , provided that it is not too far from an ordering transition . in what follows , we want to show that this is indeed what happens with midge swarms . * vicsek model . * the simplest model based on alignment interaction that predicts an order - disorder transition on changing the density is the vicsek model of collective motion @xcite . in this model each individual tends to align its direction of motion to that of the neighbours within a _ metric _ perception range , @xmath43 . the rescaled nearest neighbour distance , @xmath49 , is the control parameter : for low noise , the model predicts a transition from a disordered phase ( low polarization ) at high values of @xmath50 ( low density ) , to an ordered phase ( large polarization ) at low values of @xmath50 ( high density ) @xcite . we numerically study the vicsek model in three dimensions . as we have seen , real swarms hold their average position with respect to a marker ; to reproduce this behavioural trait we introduce an harmonic attraction force that each individual experiences towards the origin ( see methods ) . also in central potential the model displays an ordering transition : at large density , for @xmath51 the system is ordered and it has large polarization ( video xxxx ) . on the other hand , the polarization is low in the disordered phase , @xmath52 ( fig.[fig : vicsek ] ) . however , the correlation function is nontrivial when @xmath50 is sufficiently close to @xmath53 ( fig.[fig : vicsek ] ) , indicating the existence of large clusters of correlated individuals , which can be clearly detected in video xxxx . we calculate the susceptibility @xmath36 , in the same manner as we did for natural swarms , in the disordered phase , @xmath52 and find a clear increase of @xmath36 on lowering @xmath50 ( fig.[fig : vicsek ] ) . this increase of the susceptibility is coherent with the existence of an ordering transition at @xmath53 . in the thermodynamic limit , the transition in the metric vicsek model is first order @xcite . however , it has been shown that , unless @xmath13 is much larger than the values analyzed here , the transition is hardly distinguishable from a second order one @xcite . as a consequence , the susceptibility is expected to become very large approaching @xmath53 and to follow the usual scaling relation of critical phenomena @xcite , @xmath54 a fit to equation ( [ umma ] ) of the @xmath5-vicsek data is reported in fig.[fig : vicsek ] , giving @xmath55 and a transition point , @xmath56 . the reason for the growth of @xmath36 approaching @xmath53 in the vicsek model is quite intuitive . the model is metric , so that at large @xmath50 , namely when the nearest neighbour distance @xmath38 is much larger than the interaction range @xmath43 , very few individuals interact with each other , and coordination is small . the smaller @xmath50 becomes , the larger the number of particles within the mutual interaction range , thus promoting the correlation of larger and larger clusters of particles . for this reason the correlation length and the susceptibility grow when the nearest neighbour distance decreases . when @xmath50 approaches its critical value , the coordinated clusters become as large as the whole system , so that the groups orders below @xmath53 . in the disordered phase , but close to the ordering transition . the dashed line is the nearest neighbour distance . inset : polarization as a function of time . for this value of @xmath50 the system is disordered . * right : * logarithm of the susceptibility as a function of the rescaled nearest neighbour distance , @xmath57 , where @xmath43 is the metric interaction range . the full line represents the best fit to equation ( [ umma ] ) . ] the low order parameter , the nontrivial correlation function , and especially the increase of @xmath36 on decreasing the nearest neighbour distance , are phenomenological traits that the metric visek model shares with natural swarms . we conclude that a system based solely on alignment can be in its disordered phase and yet display large correlations , as midge swarms do . it is interesting to note that by approaching the ordering transition a compound amplification of the coorelation occurs : when the nearest neighbour distance , @xmath38 , decreases , the spatial span of the correlation , @xmath28 , increases , so that the effective perception range in units of nearest neighbour distance , @xmath58 is boosted . we emphasize that we are not quantitatively fitting vicsek model to our data . our only aim is to demonstrate a general concept : large correlation and lack of global order can coexist even in the simplest model of nearest neighbours alignment , provided that the system is sufficiently close to an ordering transition . * estimating the interaction range . * the consistency between our experimental data and the vicsek model suggests that an underlying ordering transition could be present in swarms as well . an ordering transition as a function of the density has been indeed observed in laboratory experiments on locusts @xcite , fish @xcite and in observations of oceanic fish shoals @xcite . in these cases , both sides ( low and high density ) of the ordering transition . however , midge swarms in the field are always disordered , living in the low - density / high-@xmath50 side of the transition . locating a transition point having data on just one side of it , is a risky business . the reason why we want to do this here is because it will allow us to give a rough estimate of the metric range of interaction in midges , which can be compared with other experiments . if a vicsek - like ordering transition exist , we can use equation ( [ umma ] ) to fit the swarms data for @xmath36 ( fig.[fig : chi ] ) . as we already mentioned , we do not know the value of the metric perception range , @xmath43 , in swarms . therefore , we use as scaling variable @xmath41 , where @xmath42 is the body length . although the fit works reasonably well ( fig.[fig : chi ] ) , the scatter in the data is quite large ; hence , given the nonlinear nature of the fit , it would be unwise to pin down just one value for the parameters , and we rather report confidence intervals . the fit gives a transition point in the range , @xmath59 $ ] , with an exponent in the range , @xmath60 $ ] ( larger exponents correspond to lower transition points ) . interestingly , there is an alternative way to locate the ordering transition that does not rely on the fit of @xmath36 . let us establish a link between pairs of insects closer than the perception range @xmath43 and calculate the size of the biggest connected cluster in the network . given a swarm with nearest neighbour distance @xmath38 , the larger @xmath43 , the larger this cluster . when @xmath43 exceeds the percolation threshold , @xmath61 , a giant cluster of the same order as the group size appears @xcite . we calculate the percolation threshold in swarms ( fig.[fig : percolation ] and methods ) and find @xmath62 . the crucial point is that varying the perception range @xmath43 at fixed nearest neighbour distance @xmath38 , is equivalent to varying @xmath38 at fixed @xmath43 . hence , at fixed @xmath43 , there is an equivalent percolation threshold of the nearest neighbour distance , @xmath63 , such that for @xmath64 a giant cluster appears . clearly , @xmath65 . it is reasonable to hypothesize that the critical nearest neighbour distance is close to the maximal distance compatible with a connected network , given @xmath43 . a sparser network would cause the swarm to lose bulk connectivity . therefore , given a certain perception range @xmath43 , the ordering transition occurs at values of the nearest neighbour distance @xmath38 close to its percolation threshold , @xmath66 . at this point we have two independent ( and possibly equally unreliable ) estimates of the transition point in natural swarms of midges : the first one in units of body - lenghts , @xmath59 $ ] ; the second one in units of interaction range , @xmath67 . putting the two together we finally obtain an estimate of the metric interaction range in units of body - lengths , @xmath68 . the body length of the species under consideration is in the range , @xmath69mm@xmath70 mm . this implies a perception range of a few centimeters , @xmath71 cm , depending on the species . this crude estimate of the midge interaction range is compatible with the hypothesis that midges interact acoustically . in @xcite the male - to - male auditory response in _ chironomus annularius _ ( diptera : chironomidae ) was studied and it was found that the range of the response was about @xmath72 cm , not too far from our estimate . similar measurements in mosquitoes ( diptera : culicidae ) show that the auditory perception range is about @xmath73 cm @xcite , which is again compatible with our determination of the interaction range in midge swarms . we have shown that natural swarms of midges lack collective order and yet display strong correlations . such correlations extends spatially much beyond the inter - individual distance , indicating the presence of significant cluster of coordinated individuals . this phenomenology is incompatible with a system of noninteracting particles whose swarming behaviour is solely due to the attraction to an external landmark . we conclude that genuine collective behaviour is present in swarms . by using vicsek model as a simple conceptual framework , we have showed that this coexistence of disorder and correlation is a general feature of systems with alignment interaction close to their ordering transition . we stress that the existence of correlation , and therefore of inter - individual interaction , is not in contradiction with the fact that a swarm almost invariably forms in proximity of a marker . the effect of the marker ( external force ) is merely to keep the swarm at a stationary position with respect to the environment . however , as we have shown in the case of the noninteracting swarm , this stationarity ( which superficially would seems the only visible trait of swarming ) , can not by itself produce the observed strong correlations . we believe that correlation , rather than order , is the most significant experimental signature of collective behaviour . correlation is a measure of how much and how far the behavioural change of one individual affects that of other individuals not directly interacting with it . our data show that in swarms correlations are so strong that the effective perception range of each midge is much larger than the actual interaction range . if the change of behaviour is due to some environmental perturbations , such large correlation guarantees that the stimulus is perceived at a collective level . a notion of collective behaviour based on correlation is more general and unifying than one based on order . for example , bird flocks and insect swarms look like completely different systems as long as we stick to collective order . however , once we switch to correlation , we understand that this big difference may be deceptive : both flocks and swarms are strongly correlated systems , in which the effective perception range , or correlation length , is far larger than the interaction range @xcite . in this perspective , the striking difference in emergent order between the two systems , namely the fact that flocks move around the sky , whereas swarms do not , may be related to different ecological factors , rather than to any fundamental qualitative difference in the way these systems interact . strong correlations similar to those found in bird flocks and midge swarms have also been experimentally measured in neural assemblies @xcite . this huge diversity - birds , insects , neurons - is bewildering but fascinating , and it suggests that correlation may be a universal condition for collective behaviour , bridging the gap between vastly different biological systems . 0.3 truecm * acknowledgments . * we thank william bialek , enzo branchini , massimo cencini , francesco ginelli and dov levine for discussions . we also acknowledge the help of tomas s. grigera in running numerical tests on the single sample susceptibility in the ising model . this work was supported by grants iit seed artswarm , erc stg n.257126 and us - afosr - fa95501010250 ( through the university of maryland ) . 0.3 truecm * authors contribution . * a.c . and i.g designed the study . coordinated the experiment . l.p . , e.shen and m.v . set up and calibrated the 3d system . l.d.c . , o.p . and e.shen performed the experiment . b.r . performed the species recognition and monitored all entomological aspects of the study . , l.d.c . , s.m . , l.p . , e.shen , e.s . and m.v . performed the tracking and produced the @xmath5 data . , s.m . , l.p . and m.v . analyzed the data . e.s . performed the numerical simulations . a.c . wrote the paper . correspondence and requests for materials should be addressed to s.m . ( [email protected] ) or a.c . ( [email protected] ) . * experiments . * data were collected in the field ( urban parks of rome ) , between may and october , in @xmath74 and in @xmath75 . we acquired video sequences using a multi - camera system of three synchronized cameras ( idt - m5 ) shooting at @xmath6 fps . two cameras ( the stereometric pair ) were at a distance between @xmath76 and @xmath77 depending on the swarm and on the environmental constraints . a third camera , placed at a distance of @xmath78 from the first camera was used to solve tracking ambiguities . we used schneider xenoplan @xmath79 @xmath80 lenses . typical exposure parameters : aperture @xmath81 , exposure time @xmath82ms . recorded events have a time duration between @xmath83 and @xmath84 seconds . no artificial light was used . to reconstruct the @xmath5 positions and velocities of individual midges we used the techniques developed in @xcite . wind speed was recorded . after each acquisition we captured several midges in the recorded swarm for lab analysis . a summary of all swarms data can be found in table s1 . * midge identification . * we recorded swarms of midges belonging to the family diptera : ceratopogonidae ( _ dasyhelea flavifrons _ ) and diptera : chironomidae ( _ corynoneura scutellata _ and _ cladotanytarsus atridorsum _ ) . midges belonging to the family chironomidae were identified to species according to @xcite , the ones belonging to the family ceratopogonidae were identified according to @xcite and @xcite . specimens used for identification were captured with a hand net and fixed in 70@xmath85 alcohol , cleared and prepared according to @xcite . permanent slides were mounted in canada balsam and dissected according to @xcite . species identification was based on morphology of the adult male , considering different characters , as wing venation , antennal ratio ( length of apical flagellomere divided by the combined length of the more basal flagellomeres ) and genitalia , which in diptera are named hypopygium ( a modified ninth abdominal segment together with the copulatory apparatus - see fig.[fig : experiment ] ) . * definition of the velocity fluctuations . * let @xmath86 and @xmath87 be respectively the sets of coordinates of the objects in the system at time @xmath88 and @xmath89 . for the sake of simplicity in the notation , assume @xmath90 . we define the velocity vector of each object @xmath15 as @xmath91 . to compute the correlation function we need to subtract the contributions due to the collective motion from the velocity . we identify three collective modes : translation , rotation and dilatation ( expansion / contraction ) . let @xmath92 be the position of the center of mass at time @xmath88 and let @xmath93 denote the position of the @xmath15-th object in the center of mass reference frame . we define the optimal translation as the displacement @xmath94 . the velocity fluctuation subtracting the overall translation is then @xmath95 . the optimal rotation about the origin is defined as the @xmath96 orthogonal matrix @xmath97 which minimizes the quantity @xmath98 ^ 2 $ ] , see @xcite . the velocity fluctuations subtracting the optimal translation and rotation is @xmath99 . finally the optimal dilatation is defined as the scalar @xmath100 that minimizes the quantity @xmath101 ^ 2 $ ] , see @xcite . therefore the velocity fluctuation subtracting the optimal translation , rotation and dilatation is : @xmath102 where @xmath103 is the contribution to the individual velocity due to the three collective modes of the system . * rotation and dilatation order parameters . * the rotational order parameter is defined as , @xmath104 where @xmath105 is the projection of @xmath106 on the plane orthogonal to the axis of rotation , the cross indicates a vectorial product , and @xmath107 is a unit vector in the direction of the axis of rotation . in ( [ rot_param ] ) , @xmath108 is the angular momentum of midge @xmath15 with respect to the axis @xmath107 . in a perfectly coherent rotation all individuals would have angular momenta parallel to the axis , so that @xmath109 . in a noncoherent system , some of the projections of the angular momentum on @xmath107 would be positive and some negative , so @xmath110 . note that @xmath107 is the axis of the rotation defined in the previous section , computed using kabsch algorithm @xcite . the dilatational order parameter is defined as , @xmath111 \cdot [ \vec y_i(t+\delta t)-r \ , \vec y_i(t ) ] } { |r \,\vec y_i(t ) |\ ; |\vec y_i(t+\delta t)-r\,\vec y_i(t)| } \ .\ ] ] @xmath112 $ ] and it measures the degree of coherent expansion ( positive @xmath113 ) and contraction ( negative @xmath113 ) of the swarm . in a perfectly coherent expansion / contraction @xmath114 would be parallel to @xmath115 and so the scalar product in equation [ eqn::lambda ] will be @xmath7 for an expansion and @xmath116 for a contraction . * normalization of the correlation function*. in the study of flocks @xcite , we normalized @xmath27 by its limiting value for @xmath117 , which is equivalent to divide it by the value in the first bin . in that way the normalized correlation function tends to @xmath7 for @xmath117 , so that its value is amplified . in the study of flocks we were only looking at the correlation length , which is not altered by such a normalization however , here we will be interested in both the range and the intensity of the correlation , so we must not amplify artificially the correlation signal . normalizing the fluctuations as in ( [ pongo ] ) is equivalent to normalize the correlation function by its value at _ exactly _ @xmath118 , i.e. for @xmath119 , which is different from its limit for @xmath120 . * noninteracting harmonic swarm . * the nhs is an elementary model of non interacting particles performing a random walk in a three - dimensional harmonic potential . the dynamics of each particle is defined by the langevin equation , @xmath121 where @xmath122 is the position of the @xmath15-th particle at time @xmath88 , @xmath123 is the mass , @xmath124 the friction coefficient , @xmath125 the harmonic constant and @xmath126 is a random vector with zero mean and unit variance , @xmath127 , with @xmath128 . clearly , in this model there is no interaction between particles . the parameter @xmath129 tunes the strength of the noise . the equation of motion are integrated with the euler method @xcite . we simulated the nhs in the critically damped regime ( @xmath130 ) , which gives the best similarity to natural swarms . the number of particles @xmath13 is set equal to that of the natural swarm we want to compare with . parameters have been tuned to have a ratio between the distance traveled by a particle in one time step ( frame ) and the nearest neighbor distance comparable to natural swarms , @xmath131 : @xmath132 . * vicsek model . * we performed numerical simulations of the vicsek model in @xmath5 @xcite . the direction of particle @xmath15 at time @xmath133 is the average direction of all particles within a sphere of radius @xmath43 around @xmath15 ( including @xmath15 itself ) . the parameter @xmath43 is the metric radius of interaction . the resulting direction of motion is then perturbed with a random rotation ( noise ) . natural swarms are known to form close to a marker and to keep a stationary position with respect to it @xcite . to mimic this behaviour we modified the vicsek model by adding an external harmonic force equal for all particles . this potential also grants cohesion , without the need to introduce an inter - individual attraction force @xcite . the update equation for velocities is therefore given by , @xmath134 \ , \ ] ] where @xmath135 is the spherical neighborhood of radius @xmath43 centered around @xmath15 , @xmath136 is the normalization operator , @xmath137 , and @xmath138 performs a random rotation uniformly distributed around the argument vector with maximum amplitude of @xmath139 . the term @xmath140 is the harmonic force directed towards the origin . for @xmath141 we recover the standard vicsek model . the update equation for the positions is , @xmath142 . thanks to the central force we can use open boundary conditions . particles have all fixed velocity modulus @xmath143 . each simulation has a duration of @xmath144 time steps , with initial conditions consisting in uniformly distributed positions in a sphere and uniformly distributed directions in the @xmath145 solid angle . after a transient of @xmath146 time steps , we saved 500 configurations at intervals of 1000 time steps in order to have configurations with velocity fluctuations uncorrelated in time . the control parameter of interest is @xmath49 , where @xmath38 is the nearest neighbour distance , which is tuned by @xmath147 . the model displays a transition to an ordered phase when @xmath51 . we studied the susceptibility @xmath148 for different values of @xmath149 $ ] . to observe the power - law behaviour of @xmath148 predicted by the model we performed standard finite - size scaling @xcite : at each fixed value of the system s size @xmath150 $ ] we calculated @xmath151 and worked out the maximum of the susceptibility @xmath152 and its position @xmath153 ; we finally plotted @xmath154 vs. @xmath155 parametrically in @xmath13 , to obtain the function @xmath148 in fig.[fig : vicsek ] . the noise , @xmath129 , affects the position of the transition point point @xmath53 @xcite , but this is irrelevant for us , because we do not use any quantitative result from the model to infer the biological parameters of real swarms . the data reported in figs.[fig : vicsek ] have @xmath156 . * percolation threshold . * for each frame we run a clustering algorithm with scale @xmath43 @xcite : two points are connected when their distance is lower than @xmath43 . for each value of @xmath43 we compute the ratio @xmath157 between the number of objects in the largest cluster and the total number of objects in the swarm ( fig.[fig : percolation ] ) . the percolation threshold , @xmath158 , is defined as the point where a giant cluster , i.e. a cluster with size of the same order as the entire system , forms @xcite . we define @xmath158 as the point where @xmath159 . the percolation threshold scales with the nearest neighbour distance , @xmath160 ( fig.[fig : percolation ] ) . strictly speaking , the percolation argument only holds at equilibrium , because in a system where particles are self - propelled there may be order even at low density @xcite . however , at low values of the noise , we still expect the percolation argument to give a reasonable , albeit crude , estimate of the perception range . 99 krause , j. & ruxton , g. d. ( 2002 ) _ living in groups _ ( oxford university press , oxford ) . cardy , j. ( 1996 ) _ scaling and renormalization in statistical physics _ ( cambridge lecture notes in physics , cambridge university press , cambridge ) parisi , g. ( 1998 ) _ statistical field theory _ ( westview press ) okubo , a. ( 1986 ) dynamical aspects of animal grouping : swarms , schools , flocks , and herds _ adv . biophys . _ * 22 * : 1 - 94 . camazine , s. , deneubourg , j .- l . , franks , n. r. , sneyd , j. , theraulaz , g. , & bonabeau , e. ( 2003 ) _ self - organization in biological systems _ ( princeton university press , princeton ) sumpter , d.j.t . ( 2010 ) _ collective animal behavior _ ( princeton university press , princeton ) downes , j. a. ( 1969 ) the swarming and mating flight of diptera._ann . entomol . _ * 14 * : 271 - 298 . sullivan rt ( 1981 ) insect swarming and mating , _ florida entomol . _ * 64 * : 4465 . nielsen , e.t . & haeger , j.s . ( 1960 ) swarming and mating in mosquitoes . _ miscell . am . _ * 1 * : 7295 charlwood , j.d . & jones , m.d.r . ( 1980 ) mating in the mosquito anopheles gambiae sl . 2 . swarming behavior . _ physiol . entomol . _ * 5 * : 315320 . manoukis , n. c. , diabate , a. , abdoulaye , a. , diallo , m. & dao ( 2009 ) a. structure and dynamics of male swarms of anopheles gambiae . entomol . _ * 46 * : 227235 . okubo , a. & grunbaum , d. , & edelstein - keshet , l. ( 2001 ) . the dynamics of animal grouping . _ interdiscip . appl . math _ * 14*:197 - 237 . downes , j. a. ( 1955 ) observations on the swarming flight and mating of culicoides ( diptera : ceratopogonidae ) 1 . _ r. entomol . soc . london_*106 * : 213 - 236 . blackwell , a. , mordue ( luntz ) , a. j. & mordue , w. ( 1992 ) the swarming behaviour of the scottish biting midge , culicoides impunctatus ( diptera : ceratopogonidae ) _ ecol . entomol . _ * 17 * : 319325 . okubo , a. & chiang , h.c . ( 1974 ) an analysis of the kinematics of swarming of anarete prithcardi kim ( diptera : cecidomyiidae ) , _ res . * 16 * : 142 butail , s. , manoukis , n. c. , diallo , m. , ribeiro , j. m. , & paley , d. a. ( 2013 ) the dance of male anopheles gambiae in wild mating swarms . entomol . _ * 50 * : 552 - 559 . ali , a. ( 1980 ) nuisance chironomids and their control : a review . _ bulletin of the esa _ * 26 * : 3 . rose , r. i. ( 2001 ) pesticides and public health : integrated methods of mosquito management . _ emerging infectious diseases _ * 7 * : 17 . cavagna , a. , cimarelli , a. , giardina , i. , parisi , g. , santagati , r. , stefanini , f. , & viale , m. ( 2010 ) scale - free correlations in starling flocks _ proc . usa _ * 107*:1186511870 . tunstrm , k. , katz , y. , ioannou , c. , huepe , c. , lutz , m. j. , & couzin , i. d. ( 2013 ) collective states , multistability and transitional behavior in schooling fish . _ plos comput . * 9 * : 1 - 11 . buhl , j. , sumpter , d. j. t. , couzin , i. d. , hale , j. j. , despland , e. , miller , e. r. , & simpson , s. j. ( 2006 ) from disorder to order in marching locusts . _ science _ * 312 * : 1402 - 1406 . cavagna , a. , giardina , i. , orlandi , a. , parisi , g. , procaccini , a. , viale , m. & zdravkovic , v. the starflag handbook on collective animal behaviour : 1 . empirical methods _ anim . * 76 * , 217236 ( 2008 ) . attanasi , a. , cavagna , a. , del castello , l. , giardina , i. , jelic , a. , melillo , s. , parisi , l. , shen , e. , silvetsri , e. & viale , m. ( 2013 ) . tracking in three dimensions via multi - path branching . _ preprint _ arxiv:1305.1495 . shinn , e. a. & long , g. e. ( 1986 ) technique for 3-d analysis of cheumatopsyche pettiti ( trichoptera : hydropsychidae ) swarms . _ environ . entomol . _ * 15 * : 355 - 359 . kelley , d. h. , & ouellette , n. t. ( 2013 ) emergent dynamics of laboratory insect swarms . _ scientific reports _ * 3 * : 1073 . huang , k. ( 1990 ) _ statistical mechanics_. ( wiley , john & sons , new york ) binney , j. j. , dowrick , n. j. , fisher , a. j. , & newman , m. ( 1992 ) _ the theory of critical phenomena : an introduction to the renormalization group_. ( oxford university press , oxford ) bialek , w. , cavagna , a. , giardina , i. , mora , t. , silvestri , e. , viale , m. , & walczak , a. m. statistical mechanics for natural flocks of birds . natl . acad . usa _ * 109 * , 4786 - 4791 ( 2012 ) . ballerini , m. , cabibbo , n. , candelier , r. , cavagna , a. , cisbani , e. , giardina , i. , lecomte , v. , orlandi , a. , parisi , g. , procaccini , a. , et al . ( 1998 ) interaction ruling animal collective behavior depends on topological rather than metric distance : evidence from a field study _ proc . * 105 * : 12321237 . couzin , i. d. , krause , j. , james , r. , ruxton , g. d. , & franks , n. r. ( 2002 ) collective memory and spatial sorting in animal groups _ j. theor . biol . _ * 218 * : 111 . vicsek , t. , czirk , a. , ben - jacob , e. , cohen , i. & shochet , o. ( 1995 ) novel type of phase transition in a system of self - driven particles _ phys . _ * 75*:12261229 . gonci , b. , nagy , m. , & vicsek , t. ( 2008 ) phase transition in the scalar noise model of collective motion in three dimensions . . j. special topics _ * 157 * : 53 - 59 . chat , h. , ginelli , f. , grgoire , g. , & raynaud , f. ( 2008 ) collective motion of self - propelled particles interacting without cohesion . e _ * 77*:046113 . baglietto , g. & albano , e.v . ( 2008 ) finite - size scaling analysis and dynamic study of the critical behavior of a model for the collective displacement of self - driven individuals _ e _ * 78 * : 021125 . becco , c. , vandewalle , n. , delcourt , j. & poncin , p. ( 2002 ) experimental evidences of a structural and dynamical transition in fish school , _ physica a _ * 367 * : 487493 . makris , n. c. , ratilal , p. , symonds , d.t . , jagannathan , s. , lee , s. & nero , rw . ( 2006 ) critical population density triggers rapid formation of vast oceanic fish shoals . _ science _ * 311 * : 660 - 663 . kertesz , j. , stauffer , d. , & coniglio , a. in : deutcher , g.d . zallen , j. , adles , j. eds . ( 1983 ) _ percolation structures and processes _ ( adam hilger , bristol ) fyodorova , m. v. , & azovsky , a. i. ( 2003 ) interactions between swarming chironomus annularius ( diptera : chironomidae ) males : role of acoustic behavior . _ j. insect behav . _ * 16 * : 295 - 306 . pennetier , c. , warren , b. , dabir , k. r. , russell , i. j. , & gibson , g. ( 2010 ) singing on the wing as a mechanism for species recognition in the malarial mosquito _ anopheles gambiae _ _ current biology _ * 20 * : 131 - 136 . schneidman , e. , berry ii , m. j. , segev , r. , & bialek , w. ( 2006 ) weak pairwise correlations imply strongly correlated network states in a neural population . _ nature _ * 440 * : 1007 - 1012 . h. & pinder l. c. v. ( 2007 ) keys to the adult male chironomidae of britain and ireland vol i , vol ii , ( sp 64 , the freshwater biological association , ambleside ) . kieffer , j. j. ( 1925 ) _ faune de france , vol . 11 : diptres nmatocres piqueurs : chironomidae , ceratopogoninae _ ( lechevalier , paris ) . dominiak , p. ( 2012 ) biting midges of the genus dasyhelea kieffer ( diptera : ceratopogonidae ) _ poland polish journal of entomology _ * 81 * : 211 - 304 . wirth , w.w . , & marston , n. ( 1968 ) a method for mounting small insects on microscope slides in canada balsam . _ annals of the entomological society of america _ * 61 * : 783 - 784 . wiederholm , t. ( 1989 ) chironomidae of the holarctic region . keys and diagnoses . adult males . _ entomologica scadinavica suppl . _ * 34 * : 1 - 532 . kabsch , w. ( 1976 ) a solution for the best rotation to relate two sets of vectors . _ acta crystallographica _ * 32 * : 922923 . butcher , j.c . ( 2003 ) _ numerical methods for ordinary differential equations _ ( john wiley & sons , new york ) czirok , a. , vicsek , m. & vicsek , t. collective motion of organisms in three dimensions . _ physica a _ * 264 * , 299 - 304 ( 1999 ) . lu , s. y. & fu , k. s. ( 1978 ) . a sentence - to - sentence clustering procedure for pattern analysis . _ ieee trans . syst . man cybern._*8 * : 381389 . the most basic kind of correlation one can measure in a system is the scalar product of the velocities of two different individuals , @xmath161 . this quantity is large if velocities are pointing in the same direction and low if they are uncorrelated . this is what is called _ non - connected _ correlation , and it has a problem : its value is trivially dominated by the _ mean _ motion of the system . imagine that a gust of wind shifts the entire swarm , so that each midges s velocity is dominated by the wind speed . as an effect of the wind , midges @xmath15 and @xmath18 would be moving nearly parallel to each other , so that the non - connected correlation would be high . this , however , is simply an effect of the wind , and it is not a genuine sign of correlation , nor of interaction between the individuals . the same thing would happen in a system of uncorrelated particles put in rotational motion around an axis : velocities of nearby particle are mostly parallel as a mere effect of the overall rotation . these examples show that , in order to get information about the _ bona fide _ interaction between individuals , we need to compute the correlation between the _ fluctuations _ around the mean motion of the system . in other words , what we want to detect is to what extent the individual changes of behaviour with respect to the global behaviour of the system are correlated . this is what the _ connected _ correlation does and it is the only reliable measure of correlation in a system . the presence of a non - connected correlation is not in general proof of anything at the level of the interaction , as the wind example clearly shows . on the other hand , the presence of non - zero connected correlation in a system is unambiguous proof that there is interaction , and strong enough to produce collective effects . to compute the connected correlation function we therefore need to identify the collective modes of motion of the system and subtract them from the individual motion ( see methods ) . in this way we obtain the velocity fluctuation , namely the velocity of midge @xmath15 in a reference frame that not only is co - moving with the centre of mass , but also rotating and expanding / contracting as the whole swarm . therefore , what is left is the deviation of @xmath15 from the mean group motion , which is the only quantity that is safe to correlate . it is very important to realize that an error or an artifact in computing the fluctuations can lead to spurious values of the correlation . as an example , consider two different and unrelated swarms moving in opposite directions , because of some weird fluctuation of the wind . if we fail to notice that these are _ two _ systems and analyze our data as if they were _ one _ , we get a zero net motion of the centre of mass . hence , the velocity fluctuations are equal to the full velocities , and we are effectively computing a non - connected correlation , rather than a connected one , giving the delusion of very large correlation . in the main text we show that swarms are mostly disordered . however , the fact that order parameters are low on average , does not mean that we can use the full velocities to compute the correlation function . as we have already said , a brief gust of wind can push the non - connected correlation function , to very high values . in this study , we are not investigating the origin of the order parameters fluctuations , but we focus on correlations . hence , we have to be sure that correlation is computed in a way to avoid any spurious signal from the collective modes . in a stationary system , it can be proved @xcite that the susceptibility is equal to the collective response of the system to uniform external perturbations . maximum entropy calculations @xcite show that the stationary probability distribution of the velocities in systems where there is an alignment interaction is given by , @xmath162 where @xmath163 is the strength of the interaction ( depending on distance @xmath19 in a metric system ) and @xmath164 is a normalizing factor ( the partition function ) , @xmath165 if an external perturbation ( or field ) @xmath166 couples uniformly to all velocities , this distribution gets modified as , @xmath167 now we ask what is the collective response @xmath36 of the system to a small variation of the perturbation @xmath166 . to answer this question we calculate the variation of the global order parameter , i.e. of the space average of the velocity , under a small variation of @xmath166 . we have , @xmath168 where we have disregarded the vectorial nature of the quantities not to burden the notation and where we have defined , @xmath169 apart from the missing normalization , needed to make @xmath36 dimensionless , the quantity in ( [ monster ] ) is exactly the susceptibility defined in the main text , equation @xmath170 . let us now analyze in detail the relation between correlation function and susceptibility in a finite size system , where instead of the ensemble averages , @xmath171 , we can only perform space averages . from equations @xmath172 and @xmath170 in main text , we obtain : @xmath173 if we make the hypothesis that mass fluctuations are not strong , we can write , @xmath174 where @xmath175 is the density . hence , we get , @xmath176 where we have used the simple relationship between density and nearest neighbours distance , @xmath177 . in an infinitely large system , the bulk susceptibility is simply , @xmath178 , that is the full volume integral of the connected correlation function . in a finite size system , however , due to the constraint , @xmath179 , we must have , @xmath180 for all systems , be they natural or synthetic , irrespective of the amount of real correlation . this relation is simply the mathematical consequence of the way velocity fluctuations are defined . therefore , in a finite systems the susceptibility can be estimated as the maximum value reached by @xmath33 ( this maximum is a lower bound for the bulk susceptibility ) . we know that , @xmath181 , so that the function @xmath33 reaches its maximum at @xmath34 . hence the finite size susceptibility is given by , @xmath182 0.1 in [ cols="<,^,^,^,^,^,^,^,^,^",options="header " , ] 99 cavagna , a. , cimarelli , a. , giardina , i. , parisi , g. , santagati , r. , stefanini , f. , & viale , m. scale - free correlations in starling flocks _ proc . usa _ * 107 * , 1186511870 ( 2010 ) binney , j. j. , dowrick , n. j. , fisher , a. j. , & newman , m. _ the theory of critical phenomena : an introduction to the renormalization group_. ( oxford university press , oxford , 1992 ) bialek , w. , cavagna , a. , giardina , i. , mora , t. , silvestri , e. , viale , m. , & walczak , a. m. statistical mechanics for natural flocks of birds . usa _ * 109 * , 4786 - 4791 ( 2012 ) . cardy , j. _ scaling and renormalization in statistical physics _ ( cambridge lecture notes in physics , cambridge university press , cambridge , 1996 ) toner , j. & tu , y. flocks , herds , and schools : a quantitative theory of flocking _ phys . e _ * 58 * , 48284858 ( 1998 ) cavagna , a. , giardina , i. , ginelli , f. boundary information inflow enhances correlation in flocking . * 110 * , 168107 . * video s1 : * wild swarm of roughly @xmath3 midges in the field ( diptera : ceratopogonidae ) . the swarm has been video recorded at 170 frames per seconds , with a resolution of 4mpix , by a idt - m5 camera . three dimensional visualization of the same natural swarm as in video s1 . this 3d reconstruction has been obtained through the dynamical tracking algorithm based on our trifocal experimental technique . three dimensional visualization of a numerically simulated swarm of non - interacting particles in a harmonic potential ( nhs - noninteracting harmonic swarm ) . the number of ` midges ' in the nhs is the same as in video s2 . three dimensional visualization of a numerically simulated swarm obtained using vicsek model with the addition of an harmonic attraction force towards the origin . the video refers to a swarm in the ordered phase , with the polarization equal to @xmath183 . the number of `` midges '' in the swarm is @xmath184 , the harmonic constant @xmath147 is equal to @xmath185 while the simulation noise @xmath129 is equal to @xmath186 . * video s5 : * three dimensional visualization of a numerically simulated swarm obtained using vicsek model with the addition of an harmonic attraction force towards the origin . the video refers to a swarm in the disordered phase , with the polarization equal to @xmath187 . the number of `` midges '' in the swarm is @xmath184 , the harmonic constant @xmath147 is equal to @xmath188 while the simulation noise @xmath129 is equal to @xmath189 .

control over electron - spin states , such as coherent manipulation , filtering and measurement promises access to new technologies in conventional as well as in quantum computation and quantum communication . we review our proposal of using electron spins in quantum confined structures as qubits and discuss the requirements for implementing a quantum computer . we describe several realizations of one- and two - qubit gates and of the read - in and read - out tasks . we discuss recently proposed schemes for using a single quantum dot as spin - filter and spin - memory device . considering electronic epr pairs needed for quantum communication we show that their spin entanglement can be detected in mesoscopic transport measurements using metallic as well as superconducting leads attached to the dots . 2

cond-mat0004182

theoretical research on electronic properties in mesoscopic condensed matter systems has focussed primarily on the charge degrees of freedom of the electron , while its spin degrees of freedom have not yet received the same attention . however , an increasing number of spin - related experiments @xcite show that the spin of the electron offers unique possibilities for finding novel mechanisms for information processing and information transmission most notably in quantum - confined nanostructures with unusually long spin dephasing times @xcite approaching microseconds , as well as long distances of up to @xmath0 @xcite over which spins can be transported phase - coherently . besides the intrinsic interest in spin - related phenomena , there are two main areas which hold promises for future applications : spin - based devices in conventional @xcite as well as in quantum computer hardware @xcite . in conventional computers , the electron spin can be expected to enhance the performance of quantum electronic devices , examples being spin - transistors ( based on spin - currents and spin injection ) , non - volatile memories , single spin as the ultimate limit of information storage etc . @xcite . on the one hand , none of these devices exist yet , and experimental progress as well as theoretical investigations are needed to provide guidance and support in the search for realizable implementations . on the other hand , the emerging field of quantum computing @xcite and quantum communication @xcite requires a radically new approach to the design of the necessary hardware . as first pointed out in ref . , the spin of the electron is a most natural candidate for the qubit the fundamental unit of quantum information . we have shown @xcite that these spin qubits , when located in quantum - confined structures such as semiconductor quantum dots or atoms or molecules , satisfy all requirements needed for a scalable quantum computer . moreover , such spin - qubits being attached to an electron with orbital degrees of freedom can be transported along conducting wires between different subunits in a quantum network @xcite . in particular , spin - entangled electrons can be created in coupled quantum dots and as mobile einstein - podolsky - rosen ( epr ) pairs @xcite provide then the necessary resources for quantum communication . for both spin - related areas conventional computers and quantum computers similar and sometimes identical physical concepts and tools are needed , the common short - term goal being to find ways to control the coherent dynamics of electron spins in quantum - confined nanostructures . it is this common goal that makes research on the electron spin in nanostructures spintronics a highly attractive area . while we advance our basic knowledge about spin physics in many - body systems , we gain insights that promise to be useful for future technologies . we have remarked earlier @xcite that there have been almost as many proposals for solid state implementations of quantum computers as all the other proposals put together . a clear reason for this is that solid state physics is a most versatile branch of physics , in that almost any phenomenon possible in physics can be embodied in an appropriately designed condensed matter system . a related reason is that solid state physics , being so closely allied with computer technology , has exhibited great versatility in the creation of artificial structures and devices . this has been exploited to produce ever more capable computational devices . it appears natural to expect that this versatility will extend to the creation of solid state quantum computers as well ; the plethora of proposals would indicate that this is indeed true , although only time can tell whether any of these proposals will actually provide a successful route to a quantum computer . in the following we will review the current status of our theoretical efforts towards the goal of implementing quantum computation and quantum communication with electron spins in quantum - confined nanostructures . most of the results presented here have been discussed at various places in the literature to which we refer the interested reader for more details . the long - term goal of our investigations is quantum information processing with electron spins . thus , a brief description of this emerging research field and its goals are in order . quantum computing has attracted much interest recently as it opens up the possibility of outperforming classical computation through new and more powerful quantum algorithms such as the ones discovered by shor @xcite and by grover @xcite . there is now a growing list of quantum tasks @xcite such as cryptography , error correcting schemes , quantum teleportation , etc . that have indicated even more the desirability of experimental implementations of quantum computing . in a quantum computer each quantum bit ( qubit ) is allowed to be in any state of a quantum two - level system . all quantum algorithms can be implemented by concatenating one- and two - qubit gates . there is a growing number of proposed physical implementations of qubits and quantum gates . a few examples are : trapped ions @xcite , cavity qed @xcite , nuclear spins @xcite , superconducting devices @xcite , and our qubit proposal @xcite based on the spin of the electron in quantum - confined nanostructures . coupled quantum dots provide a powerful source of deterministic entanglement between qubits of localized but also of delocalized electrons @xcite . e.g. , with such quantum gates it is possible to create a singlet state out of two electrons and subsequently separate ( by electronic transport ) the two electrons spatially with the spins of the two electrons still being entangled the prototype of an epr pair . this opens up the possibility to study a new class of quantum phenomena in electronic nanostructures @xcite such as the entanglement and non - locality of electronic epr pairs , tests of bell inequalities , quantum teleportation @xcite , and quantum cryptography @xcite which promises secure information transmission . in the present work , quantum dots play a central role and thus we shall make some general remarks about these systems here . semiconductor quantum dots are structures where charge carriers are confined in all three spatial dimensions , the dot size being of the order of the fermi wavelength in the host material , typically between @xmath1 and @xmath2 @xcite . the confinement is usually achieved by electrical gating of a two - dimensional electron gas ( 2deg ) , possibly combined with etching techniques , see fig . [ figarray ] . precise control of the number of electrons in the 2 conduction band of a quantum dot ( starting from zero ) has been achieved in gaas heterostructures @xcite . the electronic spectrum of typical quantum dots can vary strongly when an external magnetic field is applied @xcite , since the magnetic length corresponding to typical laboratory fields @xmath3 is comparable to typical dot sizes . in coupled quantum dots coulomb blockade effects @xcite , tunneling between neighboring dots @xcite , and magnetization @xcite have been observed as well as the formation of a delocalized single - particle state @xcite . a fundamental problem in quantum physics is the issue of the decoherence of quantum systems and the transition between quantum and classical behavior . of course , a lot of attention has been devoted in fundamental mesoscopic research to characterizing and understanding the decoherence of electrons in small structures . we remind the reader , however , that most of what has been probed ( say in weak localization studies or the aharonov - bohm effect ) is the _ orbital _ coherence of electron states , that is , the preservation of the relative phase of superpositions of spatial states of the electron ( e.g. , in the upper and lower arm of an aharonov - bohm ring ) . the coherence times seen in these investigations are almost completely irrelevant to the _ spin _ coherence times which are important in our quantum computer proposal . there is some relation between the two if there are strong spin - orbit effects , but our intention is that conditions and materials should be chosen such that these effects are weak . under these circumstances the spin coherence times ( the time over which the phase of a superposition of spin - up and spin - down states is well - defined ) can be completely different from the charge coherence times ( a few nanoseconds ) , and in fact it is known that they can be orders of magnitude longer ( see below ) . this was actually one of our prime motivations for proposing spin @xcite rather than charge as the qubit in these structures . the experimental measurement of this kind of coherence ( i.e. for spins ) is not so familiar in mesoscopic physics , and thus it is worth describing it briefly here . in recent magneto - optical experiments , based on time - resolved faraday rotation measurements , long spin coherence times were found in doped gaas in the bulk and a 2deg@xcite . at vanishing magnetic field and @xmath4 , a transverse spin lifetime ( decoherence time ) @xmath5 exceeding @xmath6 was measured , with experimental indications that this time is a single - spin effect@xcite . since this number still includes inhomogeneous effects g - factor variations in the material , leading to spins rotating with slightly different frequencies and thus reducing the total magnetization it represents only a lower bound of the transverse lifetime of a _ single _ spin , @xmath7 , which is relevant for using spins as qubits . using the same pump - probe technique , spin lifetimes in semiconductor quantum dots have been measured @xcite , with at most one spin per dot . the relatively small @xmath5 decoherence times ( a few ns at vanishing magnetic field ) , which have been seen in these experiments , probably originate from a large inhomogeneous broadening due to a strong variation of g - factors @xcite . nevertheless , the fact that many coherent oscillations were observed provides strong experimental support to the idea of using electron spin as qubits . since none of the experiments have been done on an actual quantum computing structure as we envision it ( see below ) , the existing results can not be viewed as conclusive . because of sensitivity to details , theory can only give general guidance about the mechanisms and dependencies to be looked for , but can not make reliable _ a priori _ predictions of the decoherence times . in fact there are further complications @xcite : we know theoretically that decoherence is not actually fully characterized by a single rate ; in fact , a whole set of numbers is needed to fully characterize the decoherence process ( 12 in principle for individual qubits ) , and no experiment has been set up yet to completely measure this set of parameters , although the theory of these measurements is available . even worse , decoherence effects will in principle be modified by the act of performing quantum computation ( during gate operation , decoherence is occurring in a coupled qubit system@xcite ) . we believe that the full characterization of decoherence will involve ongoing iteration between theory and experiment , and will thus be inseparable from the act of building a reliable quantum computer . still , we should mention that recent calculations @xcite including spin - orbit interaction lead to unusually low phonon - assisted spin - flip rates in quantum dots , which indicates long spin - decoherence times . we will discuss spin - qubit errors due to nuclear spins @xcite below in sec . [ ssecnuc ] . for the implementation of realistic calculations on a quantum computer , a large number of qubits will be necessary ( on the order of @xmath8 ) . for this it is essential that the underlying concept can be scaled up to a large number of qubits , which then can be operated in parallel ( parallelism is required in known error correction schemes , see sec . [ ssecerror ] ) . this scaling requirement is well achievable with spin - based qubits confined in quantum dots , since producing arrays of quantum dots @xcite is feasible with state - of - the - art techniques of defining nanostructures in semiconductors . of course , the actual implementation of such arrays including all the needed circuits poses experimental challenges , but at least we are not aware of physical restrictions which would exclude such an upscaling for spin - qubits . as we discuss in sec . [ sec2bit ] and [ secsspinrot ] , quantum gate operations will be controlled through an effective hamiltonian @xmath9 which is switched via external control fields @xmath10 . note that in the following the exchange coupling is local , i.e. @xmath11 is finite only for neighboring qubits . however , in cavity - qed systems , there is also a long - range coupling of qubits as some of us have described in ref . but even if the exchange coupling is only local , operations on non - neighboring qubits can still be performed . since one can swap the state of two qubits with the help of the exchange interaction only , as we will show in sec . [ sec2bit ] , the qubits can be moved around in an array of quantum dots . thus , a qubit can be transported to a place where it can be coupled with a desired second qubit , where single - qubit operations can be performed , or where it can be measured . the gating mechanisms described in sec . [ sec2bit ] and [ secsspinrot ] do not depend on the shape of @xmath12 , where @xmath13 stands for the exchange coupling @xmath14 or the zeeman interaction . only the time integral @xmath15 needs to assume a certain value ( modulo @xmath16 ) . the exchange interaction @xmath17 should be switched adiabatically , i.e. such that @xmath18 , where @xmath19 is the energy scale on which excitations may occur . here , @xmath19 should be taken as the energy - level separation of a single dot ( if spin is conserved ) . a rectangular pulse leads to excitation of higher levels , whereas an adiabatic pulse with amplitude @xmath20 is e.g. given by @xmath21 where @xmath22 controls the width of the pulse . we need to use a switching time @xmath23 , such that @xmath24 becomes vanishingly small . we then have @xmath25 , so we need @xmath26 for adiabatic switching . the fourier transform @xmath27 has the same shape as @xmath10 but width @xmath28 . in particular , @xmath29 decays exponentially in the frequency @xmath30 , whereas it decays only with @xmath31 for a rectangular pulse . single qubit operations can be performed for example in g - factor - modulated materials , as proposed in sec . [ secsspinrot ] . a spin can be rotated by a relative angle of @xmath32 through changing the effective g - factor by @xmath33 for a time @xmath34 . thus , a typical switching time for an angle @xmath35 , a field @xmath36 , and @xmath37 is @xmath38 . if slower operations are required , they are easily implemented by choosing a smaller @xmath33 , reducing the magnitude of the field @xmath39 , or by replacing @xmath40 by @xmath41 with integer @xmath42 , thus `` overrotating '' the spin . next we consider two exchange - coupled spins , which perform a square - root - of - swap gate for the integrated pulse @xmath43 , as described in sec . [ sec2bit ] . we apply a pulse ( see sec . [ ssecpulse ] ) @xmath44 with @xmath45 and @xmath46 . again , we calculate a switching time @xmath47 , while the adiabaticity criterion is @xmath48 . once more , the switching time can be easily increased by adding @xmath49 with integer @xmath42 to the integrated pulse @xmath50 , i.e. by `` overswapping '' the two spins . this increased switching time allows a slower switching of @xmath17 if required . further , we note that the total time consumed by an algorithm can be optimized considerably by simultaneously switching different parameters of the hamiltonian , i.e. producing parallel instead of serial pulses . as an example , we have shown that for an error - correcting algorithm using only three qubits , a speed - up of a factor of two can be achieved @xcite . for algorithms handling a larger number of qubits , a more drastic optimization can be expected . one of the main goals in quantum computation is the realization of a reliable error - correction scheme @xcite , which requires gate operations with an error rate not larger than one part in @xmath51 . taking the ratio of the dephasing time from sec . [ sseccoherence ] , @xmath52 , and the switching times from sec . [ ssecgatetime ] , @xmath53 , we see that the targeted error rate seems not to be out of reach in the near future . from there on , an arbitrary upscaling of a quantum computer becomes feasible and is no further limited by decoherence and lacking gate precision , at least when systems with a scalable number of qubits are considered . we note that a larger number of qubits also requires a larger total number of gate operations to be performed , in order to implement the error - correction schemes . therefore it is inevitable to perform these operations in parallel ; otherwise the pursued gain in computational power is used up for error correction . hence , one favors concepts where a localized control of the gates can be realized such that operations can be performed in parallel . however , since there are still many milestones to reach before sophisticated error - correction schemes can be applied , one should by no means disregard setups where gate operations are performed in a serial way . quantum computation is not only spoiled by decoherence , but also by a limited precision of the gates , i.e. by the limited precision of the hamiltonian . in order for error correcting schemes to work , the ( time integrated ) exchange and zeeman interaction need to be controlled again in about one part in @xmath51 . while this requirement is present in all quantum computer proposals , it emphasizes the importance of gates with fine control . after a gate operation was performed on two qubits , one should be able to turn off the coupling between these qubits very efficiently , e.g. exponentially in the external fields , such that errors resulting from the remaining coupling can be reduced efficiently ( if there is still a remaining coupling this can easily result in correlated errors ; however , such correlated errors would pose new problems since known error correction schemes explicitly exclude them ) . the exchange coupling between two quantum dots can be indeed suppressed exponentially , as we will describe below in sec . [ sec2bit ] . a further possible source of errors are fluctuating charges in the environment ( e.g. moving charges in the leads attached to the gates ) since they can lead to unknown shifts of the electrostatic potentials raised and lowered for switching . however , it is known from experiments on single quantum dots that such charge fluctuations can be controlled on the scale of hours @xcite which is sufficiently long on the time - scale set by the spin decoherence time which can be on the order of @xmath54 secs . still , the ability to suppress 1/f noise will be very important for well - controlled switching in quantum computation . finally , we note that uncontrolled charge switching is not nearly so great a problem for spin qubits as for charge qubits , since this switching does not couple directly to the spin degree of freedom . it turns out that a serious source of possible qubit errors using semiconductors such as gaas is the hyperfine coupling between electron spin ( qubit ) and nuclear spins in the quantum dot @xcite . in gaas semiconductors , both ga and as possess a nuclear spin @xmath55 , and no ga / as isotopes are available with zero nuclear spin this is in contrast to si - based structures which would be more advantageous from this aspect . however , in si the control over nanostructures such as quantum dots is not as advanced as in gaas yet , but this might be just a question of time . anyway , we shall now see that such decoherence effects can also be controlled by creating an overhauser field @xcite . the hyperfine coupling between the electron spin @xmath56 and the nuclear spins @xmath57 , is given by @xmath58 , where @xmath59 is the hyperfine coupling constant . due to this coupling , a flip of the electron spin with a concomitant change of one nuclear spin may occur , causing an error in the quantum computation . we have analyzed this error in the presence of a magnetic field @xmath60 @xcite , and find in time - dependent perturbation theory that the total probability for a flip of the electron spin oscillates in time . the amplitude of these oscillations is @xmath61 where @xmath39 is defined below and @xmath62 is the maximal magnitude of the effective nuclear field ( overhauser field ) . in typical quantum dots we have @xmath63 . if @xmath64 and with a polarization @xmath65 , @xmath66 of the nuclear spins , an effective nuclear field @xmath67 is produced and the transition probability becomes suppressed with @xmath68 . such a polarization @xmath69 can be established by dynamically spin - polarizing the nuclear spins , e.g. by optical pumping @xcite or by spin - polarized currents at the edge of a 2deg @xcite . for these methods , nuclear overhauser fields are reported as large as @xmath70 in gaas ( corresponding to @xmath71 ) @xcite and which can have a lifetime on the order of minutes @xcite . alternatively , for unpolarized nuclei , the amplitude of @xmath72 can be suppressed by an external field @xmath73 [ eq . ( [ eqnprobflip ] ) ] . thus , the decoherence of an electron spin due to hyperfine interaction can be suppressed drastically , either by dynamically polarizing the nuclear spins in the host material or by applying an external magnetic field . it would be highly desirable to test this prediction by measuring the electron - spin @xmath74 time with and without overhauser field . at the beginning of most algorithms for quantum computers as well as an input for error correcting schemes , initialized qubits are required , i.e. qubits in a well defined state such as spin up , @xmath75 . single spins can be polarized by exposing them to a large magnetic field @xmath76 and letting them relax to the ground state . such a magnetic field could be applied locally or realized by forcing the electrons ( via external gates ) into a magnetized layer , into a layer with a different effective g - factor @xcite or into a layer with polarized nuclear spins ( overhauser effect ) @xcite etc . , see also fig . [ figarray ] and sec . [ secsspinrot ] . if a spin - polarized current can be produced , such as by spin - polarizing materials @xcite or by spin - filtering with the help of another dot @xcite ( see sec . [ ssecspinfilter ] ) , polarized electrons can be injected into an empty quantum dot , i.e. the dot is filled with an already initialized spin . for some algorithms , it is favorable to start with a given initial state , such as @xmath77 , instead of a ground state @xmath78 . this can be readily implemented with spins as qubits using standard electron spin resonance ( esr ) techniques @xcite : we start with a ground state @xmath78 . then we produce a zeeman splitting by applying a static local magnetic field for these spins , which should be initialized into state @xmath79 . an ac magnetic field is then applied perpendicularly to the first field with a resonant frequency that matches the larmor frequency @xmath80 . due to paramagnetic resonance @xcite , this causes spin - flips in the quantum dots with the corresponding zeeman splitting , thus producing the desired state . we note that since we do not want to affect the other spins ( having a different zeeman splitting ) the amplitude of the ac field must be switched adiabatically , see sec . [ ssecpulse ] . of course , spin precession can also be used to perform single - spin rotations ( see sec . [ secsspinrot ] ) . the main component for every computer concept is a multi-(qu)bit gate , which eventually allows calculations through combination of several ( qu)bits . since two - qubit gates are ( in combination with single - qubit operations ) sufficient for quantum computation @xcite they form a universal set we now focus on a mechanism that couples pairs of spin - qubits . such a mechanism exists in coupled quantum dots , resulting from the combined action of the coulomb interaction and the pauli exclusion principle . two coupled electrons in absence of a magnetic field have a spin - singlet ground state , while the first excited state in the presence of strong coulomb repulsion is a spin triplet . higher excited states are separated from these two lowest states by an energy gap , given either by the coulomb repulsion or the single - particle confinement . the low - energy dynamics of such a system can be described by the effective heisenberg spin hamiltonian @xmath81 where @xmath17 denotes the exchange coupling between the two spins @xmath82 and @xmath83 , i.e. the energy difference between the triplet and the singlet . after a pulse of @xmath17 with @xmath84 ( mod @xmath16 ) , the time evolution @xmath85 corresponds to the `` swap '' operator @xmath86 , whose application leads to an interchange of the states in qubit 1 and 2 @xcite . while @xmath86 is not sufficient for quantum computation , any of its square roots @xmath87 , say @xmath88 , turns out to be a _ universal _ quantum gate . thus , it can be used , together with single - qubit rotations , to assemble any quantum algorithm . this is shown by constructing the known universal gate xor @xcite , through combination of @xmath87 and single - qubit operations @xmath89 , applied in the sequence @xcite , @xmath90 with these universal gates at hand , we can reduce the study of general quantum computation to the study of single - spin rotations ( see sec . [ secsspinrot ] ) and the _ exchange mechanism _ , in particular how @xmath17 can be controlled experimentally . the central idea is that @xmath17 can be switched by raising or lowering the tunneling barrier between the dots . in the following , we shall review our detailed calculations to describe such a mechanism . we note that the same principles can also be applied to other spin systems in quantum - confined structures , such as coupled atoms in a crystal , supramolecular structures , and overlapping shallow donors in semiconductors @xcite etc . , using similar methods as explained below . we point out that , beyond the mechanisms described in sec . [ ssecldots ] and sec . [ ssecvdots ] , spins in quantum dots can also be coupled on a long distance scale by using a cavity - qed scheme @xcite or by using superconducting leads to which the quantum dots are attached @xcite , see sec . [ ssecsclead ] . we consider a system of two coupled quantum dots in a two - dimensional electron gas ( 2deg ) , containing one ( excess ) electron each , as described in sec . [ ssecqd ] . the dots are arranged in a plane , at a sufficiently small distance @xmath91 , such that the electrons can tunnel between the dots ( for a lowered barrier ) and an exchange interaction @xmath14 between the two spins is produced . we model this system of coupled dots with the hamiltonian @xmath92 , where the single - electron dynamics in the 2deg ( @xmath93-plane ) is described through @xmath94 with @xmath95 being the effective mass and @xmath96 the confinement potential as given below . a magnetic field @xmath97 is applied along the @xmath98-axis , which couples to the electron spin through the zeeman interaction @xmath99 and to the charge through the vector potential @xmath100 . in almost depleted regions , like few - electron quantum dots , the screening length @xmath101 can be expected to be much larger than the screening length in bulk 2deg regions ( where it is @xmath102 for gaas ) . thus , for small quantum dots , say @xmath103 , we need to consider the bare coulomb interaction @xmath104 , where @xmath105 is the static dielectric constant . the confinement and tunnel - coupling in eq . ( [ eqnhonee ] ) for laterally aligned dots is modeled by the quartic potential @xmath106,\ ] ] with the inter - dot distance @xmath91 and @xmath107 the effective bohr radius of the dot . separated dots ( @xmath108 ) are thus modeled as two harmonic wells with frequency @xmath109 . this is motivated by the experimental evidence that the low - energy spectrum of single dots is well described by a parabolic confinement potential @xcite . now we consider only the two lowest orbital eigenstates of @xmath110 , leaving us with one symmetric ( spin - singlet ) and one antisymmetric ( spin - triplet ) orbital state . the spin state for the singlet is @xmath111 , while the triplet spin states are @xmath112 , @xmath113=@xmath114 , and @xmath115=@xmath116 . for temperatures with @xmath117 , higher - lying states are frozen out and @xmath110 can be replaced by the effective heisenberg spin hamiltonian [ eq . ( [ heisenberg ] ) ] . the exchange energy @xmath118 is given as the difference between the triplet and singlet energy . for calculating these energies , we use the analogy between atoms and quantum dots and make use of variational methods similar to the ones in molecular physics . using the heitler - london ansatz with ground - state single - dot orbitals , we find @xcite , @xmath119 \bigg\},\end{aligned}\ ] ] where we have introduced the dimensionless distance @xmath120 between the dots and the magnetic compression factor @xmath121 with the larmor frequency @xmath122 . the zeroth order bessel function is denoted by @xmath123 . in eq . ( [ j ] ) , the first term comes from the confinement potential , while the terms proportional to the parameter @xmath124 result from the coulomb interaction @xmath125 ; the exchange term is recognized by its negative sign . we are mainly interested in the weak coupling limit @xmath126 , where the ground - state heitler - london ansatz is self - consistent . we plot @xmath14 [ eq . ( [ j ] ) ] in fig . [ jplots ] as a function of @xmath39 and @xmath127 . we note that @xmath128 , which is generally true for a two - particle system with time - reversal invariance . we observe that over a wide range of the parameters @xmath129 and @xmath130 , the sign of @xmath131 changes from positive to negative at a finite value of @xmath39 ( for the parameters chosen in fig . [ jplots](a ) at @xmath132 ) . @xmath14 is suppressed exponentially either by compression of the electron orbitals through large magnetic fields ( @xmath133 ) , or by large distances between the dots ( @xmath134 ) , where in both cases the orbital overlap of the two dots is reduced . this exponential suppression , contained in the @xmath135 prefactor in eq . ( [ j ] ) , is partly compensated by the exponentially growing exchange term @xmath136 . in total , @xmath14 decays exponentially as @xmath137 for large @xmath138 or @xmath127 . since the sign reversal of @xmath14signalling a singlet - triplet crossing results from the long - range coulomb interaction , it is not contained in the standard hubbard model which takes only short - range interaction into account . in this latter model one finds @xmath139 in the limit @xmath140 ( see fig . [ jplots ] ) . the heitler - london result [ eq . ( [ j ] ) ] was refined by taking higher levels and double occupancy of the dots into account ( implemented in a hund - mullikan approach ) , which leads to qualitatively similar results @xcite , in particular concerning the singlet - triplet crossing . we remark again that the exponential suppression of @xmath14 is very desirable for minimizing gate errors , see sec . [ ssecprec ] . in the absence of tunneling between the dots we still might have direct coulomb interaction left between the electrons . however , this has no effect on the spins ( qubit ) provided the spin - orbit coupling is sufficiently small , which is the case for s - wave electrons in gaas structures with unbroken inversion symmetry ( this would not be so for hole - doped systems since the hole has a much stronger spin - orbit coupling due to its p - wave character ) . finally , the vanishing of @xmath14 can be exploited for switching by applying a constant homogeneous magnetic field to an array of quantum dots to tune @xmath14 to zero ( or close to some other desirable value ) . then , for switching @xmath14 on and off , only a small gate pulse or a small local magnetic field is needed . we have also investigated the case of vertically tunnel - coupled quantum dots @xcite . such a setup of the dots has been produced in multilayer self - assembled quantum dots ( sad ) @xcite as well as in etched mesa heterostructures @xcite . we apply the same methods as described in sec . [ ssecldots ] for laterally coupled dots , but now we extend the hamiltonian eq . ( [ eqnhonee ] ) from two to three dimensions and take a three - dimensional confinement @xmath141 . we implement the vertical confinement @xmath142 as a quartic potential similar to eq . ( [ eqnv ] ) , with curvature @xmath143 at @xmath144 [ see fig . [ vdots](b ) ] , implying an effective bohr radius @xmath145 and a dimensionless distance @xmath146 . we have modeled a harmonic potential for the lateral confinement , while we have allowed different sizes of the two dots @xmath147 . this allows additional switching mechanisms as it is explained in the next paragraph . since we are considering a three - dimensional setup , the exchange interaction is not only sensitive to the magnitude of the applied fields , but also to their direction . we now give a brief overview of our results @xcite for in - plane ( @xmath148 , @xmath149 ) and perpendicular ( @xmath150 , @xmath151 ) fields ; this setup is illustrated in fig . [ vdots](a ) : ( 1 ) an in - plane magnetic field @xmath148 suppresses @xmath14 exponentially ; a perpendicular field in laterally coupled dots has the same effect ( sec . [ ssecldots ] ) . ( 2 ) a perpendicular magnetic fields @xmath150 reduces on the one hand the exchange coupling between identically sized dots @xmath152 only slightly . on the other hand , for different dot sizes @xmath153 , the behavior of @xmath154 is no longer monotonic : increasing @xmath150 from zero amplifies the exchange coupling @xmath14 until both electronic orbitals are magnetically compressed to approximately the same size , i.e. @xmath155 . from this point , @xmath14 decreases weakly , as for identically sized dots . ( 3 ) a perpendicular electric field @xmath151 detunes the single - dot levels , and thus reduces the exchange coupling ; the very same finding was made for for laterally coupled dots and an in - plane electric field @xcite . ( 4 ) an in - plane electric field @xmath149 and different dot sizes provide another switching mechanism for @xmath14 . the dots are shifted parallel to the field by @xmath156 , where @xmath157 . thus , the larger dot is shifted a greater distance @xmath158 and so the mean distance between the electrons grows as @xmath159 , taking @xmath160 . since the exchange coupling @xmath14 is exponentially sensitive to the inter - dot distance @xmath161 , it is suppressed exponentially when an in - plane electric field is applied , @xmath162 $ ] , which is illustrated in fig . [ vdots](c ) . thereby we have given an exponential switching mechanism for quantum gate operation relying only on a tunable electrical field , in addition to the magnetically driven switching discussed above . an operation which encodes a single spin 1/2 state @xmath163 into a singlet or triplet state can be used for measuring the state of the qubit represented by @xmath163 , when a measurement device capable of distinguishing singlet / triplet states is available ( see e.g. sec . [ ssecddot ] ) . further , such an operation acts as an entangler " for electron pairs used in quantum communication ( see sec . [ secqcom ] ) . indeed , we can construct such a two - qubit operation explicitly . while quantum dot 1 is in state @xmath163 , we prepare the state of the quantum dot 2 to @xmath75 , perform a @xmath87 gate and finally apply a local zeeman term , generating the time evolution @xmath164 , thus @xmath165 in other words , this operation maps the triplet @xmath114 ( and @xmath116 ) into itself , while the state @xmath166 is mapped into the singlet ( and @xmath167 into the triplet @xmath168 ) , up to phase factors . a requirement for quantum computing is the possibility to perform one - qubit operations , which translates in the context of spins into single - spin rotations . so it must be possible to expose a specific qubit to a time - varying zeeman coupling @xmath169 @xcite , which can be controlled through both the magnetic field @xmath170 and/or the g - factor @xmath171 . since only phases have a relevance , it is sufficient to rotate all spins of the system at once ( e.g. by an external field @xmath39 ) , but with a different larmor frequency . we have proposed a number of possible implementations @xcite for spin - rotations : the equilibrium position of the electron can be moved around through electrical gating . thus , if the electron wave function is pushed into a region with a different magnetic field strength or ( effective ) g - factor , one produces a relative rotation around the direction of @xmath170 by an angle of @xmath172 , see fig . [ figarray ] . regions with an increased magnetic field can be provided by a magnetic ( dot ) material while an effective magnetic field can be produced e.g. with dynamically polarized nuclear spins ( overhauser effect ) @xcite . we shall now explain a concept for using g - factor - modulated materials @xcite . in bulk semiconductors the free - electron value of the land g - factor @xmath173 is modified by spin - orbit coupling . similarly , the g - factor can be drastically enhanced by doping the semiconductor with magnetic impurities @xcite . in confined structures such as quantum wells , wires , and dots , the g - factor is further modified and becomes sensitive to an external bias voltage @xcite . we have numerically analyzed a system with a layered structure ( algaas - gaas - inalgaas - algaas ) , in which the effective g - factor of electrons is varied by shifting their equilibrium position from one layer to another by electrical gating@xcite . we have found that in this structure the effective g - factor can be changed by about @xmath37 @xcite . alternatively , one can use esr techniques for switching ( as already explained in sec . [ secinitialization ] ) . furthermore , localized magnetic fields can be generated with the magnetic tip of a scanning force microscope , a magnetic disk writing head , by placing the dots above a grid of current - carrying wires , or by placing a small wire coil above the dot etc . one scheme for reading out the spin of an electron on a quantum dot is implemented by tunneling of this electron into a supercooled paramagnetic dot @xcite . there the spin induces a magnetization nucleation from the paramagnetic metastable phase into a ferromagnetic domain , whose magnetization direction @xmath174 is along the measured spin direction and which can be measured by conventional means . since this direction is continuous rather than only one of two values , we describe this generalized measurement in the formalism of positive - operator - valued ( pov ) measurements @xcite as projection into the overcomplete set of spin-@xmath175 coherent states @xmath176 . thus if we interpret a magnetization direction in the upper hemisphere as @xmath75 , we have a @xmath177%-reliable measurement , since @xmath178 , using the normalization constant @xmath16 for the coherent spin states . while spins have the intrinsic advantage of long decoherence times , it is very hard to measure a single spin directly via its magnetic moment . however , measuring the charge of single electrons is state of the art . thus it is desirable to have a mechanism for detecting the spin of an electron via measuring charge , i.e. voltage or current @xcite . a straightforward concept yielding a potentially 100% reliable measurement requires a switchable `` spin - filter '' tunnel barrier which allows only , say , spin - up but no spin - down electrons to tunnel . when the measurement of a spin in a quantum dot is to be performed , tunneling between this dot and a second dot , connected to an electrometer , is switched on , but only spin - up electrons are allowed to pass ( spin - filtering ) . thus if the spin had been up , a charge would be detected in the second dot by the electrometer @xcite , and no charge otherwise . again , this is a pov type of measurement ( see above ) . it is known how to build electrometers with single - charge detection capabilities ; resolutions down to @xmath179 of one electron charge have been reported @xcite . spin filtering and also spin - state measurements can be achieved by tunneling through a quantum dot @xcite as we shall discuss next . we discuss now a setup quantum dot attached to in- and outgoing current leads @xmath180which can be operated as a spin filter , or as a read - out device , or as a spin - memory where a single spin stores the information @xcite . a new feature of this proposal is that the spin - degeneracy is lifted with _ different _ zeeman splittings in the dot and in the leads , e.g. by using materials with different effective g - factors for leads and dot @xcite . this results in coulomb blockade peaks and spin - polarized currents which are uniquely associated with the spin state on the dot . the setup is described by a standard tunneling hamiltonian @xmath181 @xcite , where @xmath182 describes the leads and the dot . @xmath183 includes the charging and interaction energies of the electrons in the dot as well as their zeeman energy @xmath184 in an external magnetic field @xmath170 . the tunneling between leads and the dot is described by @xmath185 , where @xmath186 annihilates electrons with spin @xmath187 and momentum @xmath188 in lead @xmath189 and @xmath190 annihilates electrons in the dot . we consider the coulomb blockade regime @xcite where the charge on the dot is quantized . then we apply a standard master - equation approach @xcite with a reduced density matrix of the dot and calculate the transition rates in a `` golden - rule '' approach up to 2nd order in @xmath191 . the first - order contribution to the current is the sequential tunneling current @xmath192 @xcite , where the number of electrons on the dot fluctuates and thus the processes of an electron tunneling from the lead onto the dot and vice versa are allowed by energy conservation . the second - order contribution is the cotunneling current @xmath193 @xcite , involving a virtual intermediate state with a different number of electrons on the dot ( see also sec . [ ssecddot ] ) . we now consider a system , where the zeeman splitting in the leads is negligible ( i.e. much smaller than the fermi energy ) while on the dot it is given as @xmath194 . we assume a small bias @xmath195 between the leads at chemical potential @xmath196 and low temperatures so that @xmath197 , where @xmath198 is the characteristic energy - level distance on the dot . first we consider a quantum dot in the ground state , filled with an odd number of electrons with total spin @xmath175 , which we assume to be @xmath75 and to have energy @xmath199 . if an electron tunnels from the lead onto the dot , a spin singlet is formed with energy @xmath200 , while the spin triplets are ( usually ) excited states with energies @xmath201 and @xmath202 . at the sequential tunneling resonance , @xmath203 , where the number of electrons on the dot fluctuates between @xmath204 and @xmath205 , and in the regime @xmath206 , energy conservation only allows ground state transitions . thus , spin - up electrons are not allowed to tunnel from lead @xmath207 via the dot into lead @xmath208 , since this would involve virtual states @xmath209 and @xmath210 , and so we have @xmath211 for sequential tunneling . however , spin down electrons may pass through the dot in the process ( 7,9 ) ( 3,3.6)(0,0)@xmath212 ( 13,10 ) ( 6,3.6 ) ( 6,3.6)(0,0)@xmath213 @xmath214 @xmath215 ( 13,10 ) ( 6,3.6 ) ( 6,3.6)(0,0)@xmath216 @xmath217 , followed by ( 13,10 ) ( 6,3.6 ) ( 6,3.6)(0,0)@xmath216 @xmath214 @xmath215 ( 13,10 ) ( 6,3.6 ) ( 6,3.6)(0,0)@xmath213 ( 7,9 ) ( 3,3.6)(0,0)@xmath212 @xmath218 . here the state of the quantum dot is drawn inside the circle , while the states in the leads are drawn to the left and right , _ resp . _ , of the circle . this leads to a _ spin - polarized _ sequential tunneling current @xmath219 , which we have calculated as @xcite @xmath220 , \quad k_bt > \delta\mu , \label{eqnlarget}\end{aligned}\ ] ] where @xmath221 and @xmath222 . here @xmath223 is the tunneling rate between lead @xmath189 and the dot and we have introduced the matrix elements @xmath224 . similarly , for @xmath204 even we find @xmath225 while for @xmath226 a similar result holds @xcite as in eqs . ( [ eqnsmallt ] ) , ( [ eqnlarget ] ) . even though @xmath192 is completely spin - polarized , a leakage of current with opposite polarization arises through cotunneling processes @xcite ; still the leakage is small , and the efficiency for @xmath227 for spin filtering in the sequential regime becomes @xcite @xmath228 and equivalently for @xmath229 at the even - to - odd transition . in the sequential regime we have @xmath230 , thus , for @xmath231 , we see that the spin - filtering is very efficient . we discuss now the opposite case where the leads are fully spin polarized with a much smaller zeeman splitting on the dot @xcite . such a situation can be realized with magnetic semiconductors ( with effective g - factors reaching 100 @xcite ) where spin - injection into gaas has recently been demonstrated for the first time@xcite . another possibility would be to work in the quantum hall regime where spin - polarized edge states are coupled to a quantum dot@xcite . in this setup the device can be used as read - out for the spin state on the dot . assume now that the spin polarization in both leads is up , and the ground state of the dot contains an odd number of electrons with total spin @xmath175 . now the leads can provide and absorb only spin - up electrons . thus , a sequential tunneling current will only be possible if the dot state is @xmath210 ( to form a singlet with the incoming electron , whereas the triplet is excluded by energy conservation ) . hence , the current is much larger for the spin on the dot being in @xmath210 than it is for @xmath75 . again , there is a small cotunneling leakage current for the dot - state @xmath75 , with a ratio of the two currents given by eq . ( [ efficiencyst ] ) . thus , we can probe ( read out ) the spin - state on the quantum dot by measuring the current which passes through the dot . given that the sequential tunneling current is typically on the order of @xmath232 na @xcite , we can estimate the read - out frequency @xmath233 to be on the order of @xmath232 ghz . combining this with the initialization and read - in techniques from sec . [ secinitialization ] , i.e. esr pulses to switch the spin state , we have a _ spin memory _ at the ultimate single - spin limit , whose relaxation time is just the spin relaxation time . this relaxation time can be expected to be on the order of @xmath234 s of nanoseconds @xcite , and can be directly measured via the currents when they switch from high to low due to a spin flip on the dot @xcite . measurements of the faraday rotation @xcite originating from a pair of coupled electrons would allow us to distinguish between spin singlet and triplet @xcite : in the singlet state ( @xmath235 , no magnetic moment ) there is no faraday rotation , whereas in the triplet state ( @xmath236 ) the polarization of linearly polarized light is rotated slightly due to the presence of the magnetic moment . a single spin @xmath163 can be measured either directly via faraday rotation or by first entangling it with another spin @xmath75 and then applying the singlet / triplet - measurement . this entanglement is achieved by applying the gate defined in sec . [ ssecegate ] , resulting in either a triplet or singlet , depending on whether @xmath163 was @xmath75 or @xmath210 . however , much more work is required to analyze the faraday rotation ( in particular to calculate the oscillator strength for such processes ) in order to assess its efficiency for spin measurements . a ( pure ) state of two particles ( qubits ) is called entangled , if it can not be expressed as a tensor product of two single - particle states . many tasks in quantum communication require maximally entangled states of two qubits ( epr pairs ) such as the spin singlet @xcite . note that also the triplet @xmath237 is an entangled state , while the other two triplets @xmath238 are not . the quantum gate mechanism described in sec . [ ssecegate ] is one possibility for producing such entangled states ( we call in general such a device an _ entangler _ , for which a number of realizations are conceivable ) . here we discuss three experimental setups by which the entanglement of electrons can be detected via their charge in transport and noise measurements in mesoscopic nanostructures @xcite . this investigation touches on fundamental issues such as the non - locality of quantum mechanics , especially for massive particles , and genuine two - particle aharonov - bohm effects which are fascinating topics in their own right . the main idea here is to exploit the unique relation between the symmetry of the orbital state and the spin state ( for two electrons ) which makes it possible to detect the spin state again via the charge ( orbital ) degrees of freedom of the electrons . we should emphasize here that entanglement _ per se _ is rather the rule than the exception in condensed matter systems . for instance every ground state of a many - electron system is entangled simply by the antisymmetry requirement for the wave function . however , the key here is to have separate control over each specified particle which belongs to an entangled many - particle state . in quantum optics , violations of bell inequalities and quantum teleportation with photons have been investigated @xcite , while so far no corresponding experiments for electrons in a solid - state environment are reported . when we consider the injection of entangled electrons into a fermi sea , we must keep in mind that there is always coulomb interaction present with all the other electrons in the leads . so we need to analyze its effect on the entanglement @xcite . specifically , when we add an electron in state @xmath239 to a fermi sea ( lead ) , the quasiparticle weight of that state will be renormalized by @xmath240 ( see below ) , i.e. some weight @xmath241 to find the electron in the original state @xmath239 will be distributed among all the other electrons @xcite . this rearrangement of the fermi system due to the coulomb interaction happens very quickly , on a timescale given by the inverse plasmon frequency . so , the question now is : how big is this renormalization ? more precisely , when a triplet / singlet electron pair ( @xmath242 and @xmath187 for short ) is injected from an entangler into two leads @xmath207 and @xmath208 , we obtain the state @xmath243 with the filled fermi sea @xmath244 , @xmath245 , @xmath246 the momentum of an electron , and @xmath189 the lead number . the operator @xmath247 creates an electron in state @xmath248 with spin @xmath249 . the propagation of the triplet or singlet , interacting with all other electrons in the fermi sea , can be described by the 2-particle green s function @xmath250 . if we prepare a triplet ( singlet ) , @xmath251 is the amplitude of finding a triplet ( singlet ) after time @xmath242 . assuming sufficiently separated leads with negligible mutual interaction , we find@xcite @xmath252 . for a spin - independent hamiltonian with bare coulomb interaction only and within rpa @xcite , the quasiparticle weight for a 2deg is given by @xcite @xmath253 , in leading order of the interaction parameter @xmath254 , where @xmath255 is the bohr radius and @xmath256 the fermi wavevector . in a gaas 2deg we have @xmath257 nm and @xmath258 , and thus we obtain @xmath259 . therefore , we conclude that the entanglement of a pair of electrons injected into a fermi liquid will be reduced but there is still a finite probability left to preserve the entangled state . this holds provided the spin - scattering effects are small . that this is indeed the case in gaas 2degs is supported by experiments @xcite where the electron spin has been transported phase - coherently over distances of up to 100 @xmath260 @xcite . it has been known @xcite for quite some time that bosons such as photons show `` bunching '' behavior when measuring the correlations between particles ( noise " ) in an incoming particle current . more recently , the opposite behavior for fermions , `` antibunching '' , was expected theoretically @xcite and found experimentally @xcite , in particular for electrons . however , as we have pointed out recently @xcite the noise of electrons in current - carrying wires is not sensitive to the symmetry of the total wave function but only to the symmetry of the _ orbital _ part of it , at least if no spin - scattering processes are present . thus , if we now consider a two - electron state , we expect antibunching for the triplet states , since they have an antisymmetric orbital wave function , whereas the orbital wave function associated with the spin singlet state is symmetric , and so we expect a bunching behavior . this leads to an observable decrease or increase in noise for electrons , depending on their common spin state , as we shall discuss next @xcite . we assume that an entangler generates pairs of entangled electrons which are then injected into lead 1 and 2 , one electron each , as shown in fig . [ figentangler ] . a beam splitter is inserted in order to create two - particle interference effects in the sense that there is an equal probability amplitude for incoming electrons ( from lead 1 or 2 ) to leave into lead 3 or 4 ( note that the electrons in a fermi liquid wire hardly interact which each other ; the role of the beam splitter is thus to simulate direct and exchange coulomb processes ) . the quantity of interest is then the noise , i.e. the current - current correlations , measured in leads @xmath261 and/or @xmath262 . the amplitude of recovering a singlet or triplet state after injecting it into an interacting fermi sea is reduced by a factor of @xmath263 ( see sec . [ ssecfs ] ) . except for this renormalization , the entanglement of the singlet or triplet state is not affected by the interacting electrons in the filled fermi sea . thus we can now calculate transport quantities using the standard scattering theory for non - interacting quasiparticles in a fermi liquid . we consider the entangled incident states @xmath264 with one electron per lead and the quantum numbers @xmath265 , where @xmath266 is the energy of the electron . considering a multiterminal conductor with density of states @xmath267 , we assume that the leads consist of only one quantum channel ; the generalization to several channels is straightforward . the ( unpolarized ) current operator for lead @xmath268 can be written as @xcite @xmath269 e^ { i(\!\varepsilon-\varepsilon ' \!)t/\hbar } \ ! , \label{current_def}\ ] ] where @xmath270 creates an incoming electron with spin @xmath249 and energy @xmath271 in the lead @xmath268 . the operators @xmath272 for the outgoing electrons are given by @xmath273 with the scattering matrix @xmath274 , which is assumed to be spin- and energy - independent . the average currents in the leads , @xmath275 , are not sensitive to the orbital symmetry of the wavefunction . the spectral densities of the fluctuations @xmath276 between the leads @xmath268 and @xmath277 are @xmath278 which are now evaluated with the scattering matrix for the beamsplitter ( fig . [ figentangler ] ) with the reflection and transmission amplitudes @xmath279 and @xmath242 , thus @xmath280 , and @xmath281 and no backscattering , so @xmath282 . we obtain for the noise at zero frequency @xcite @xmath283 here , the minus ( plus ) sign refers to the spin triplet ( singlet ) and @xmath284 is the transmission coefficient of the beam splitter . if two electrons with the same energies , @xmath285 , in the singlet state are injected into the leads @xmath207 and @xmath208 , the shot noise is enhanced by a factor of two compared to the value for uncorrelated particles @xcite , @xmath286 . this amplification of the noise arises from _ bunching _ of the electrons due to their symmetric orbital wavefunction , such that the electrons preferably appear in the same outgoing leads . if the electron pairs are injected as a triplet , an _ antibunching _ effect appears , completely suppressing the noise , i.e. @xmath287 . we stress that the sign of cross - correlations does not carry any signature of statistics , e.g. here the different signs of @xmath288 and @xmath289 [ eq . ( [ eqnnoise ] ) ] merely reflect current conservation and absence of backscattering . since the bunching effect appears only for a state with a symmetric orbital wave function , which is not the case for unentangled electron states , measuring noise enhancement in the outgoing arms of the beamsplitter provides unique evidence for entanglement @xcite . we turn now to a setup by which the entanglement of two electrons in a double - dot can be measured through current and noise @xcite . for this we consider a double - dot which is weakly coupled , with tunneling amplitude @xmath290 , to in - and outgoing leads at chemical potentials @xmath196 . as shown in fig . [ figddot ] , the dots are put in parallel in contrast to the standard series connection . we work in the coulomb blockade regime @xcite where the charge on the dots is quantized and in the cotunneling regime @xcite , with @xmath291 , where @xmath292 is the single - dot charging energy , @xmath267 the lead density of states , and @xmath14 the exchange coupling ( see sec . [ sec2bit ] ) . the cotunneling current involves a coherent virtual process where an electron tunnels from a dot to , say , lead 2 and then a second electron tunnels from lead 1 to this dot . assuming @xmath293 , elastic as well as inelastic cotunneling occurs . further , @xmath290 is assumed to be sufficiently weak so that the double - dot will return to its equilibrium state before the next electron passes through . since an electron can either pass through the upper or lower dot , a closed loop is formed by these two paths , and in the presence of a magnetic flux the upper and the lower paths collect a phase difference given by the aharonov - bohm phase @xmath294 ( with @xmath59 being the loop area ) , thus leading to interference effects . if the two electrons on the double - dot are in the _ singlet state _ , then the tunneling current acquires an additional phase of @xmath295 ( see below and fig . [ figddot ] ) leading to a sign reversal of the coherent contribution compared to that for triplets . explicitly , we find for the cotunneling current @xcite @xmath296 and for the shot noise power @xmath297 , where the upper sign refers to the triplet states in the double - dot and the lower sign to the singlet state . ( [ eqnidd ] ) can be reproduced , up to a prefactor , by the following heuristic argument . consider the two spins on the double dot to be in the singlet state @xmath298 or in a triplet state , say , @xmath299 . these superpositions are illustrated in fig . [ figddot ] by drawing the first term in black in the left part of the dots and the second term in gray on the right . we consider the contribution @xmath300 to the current , where we start with one spin - up electron in the left lead and end with a spin - down electron in the right lead and the triplet state @xmath209 on the double dot ( see inset of fig . [ figddot ] ) . for this process , either a spin - down electron tunnels first @xmath301 from the lower dot into the right lead and then @xmath302 the spin - up electron from the left lead tunnels into the lower dot . or the upper dot participates via @xmath303 and @xmath304 , but now the state @xmath166 is involved , thus if the initial state on the double dot is a singlet , the transition amplitudes for upper and lower path acquire opposite signs , whereas there is no sign change if we started out from a triplet ( as shown for @xmath237 in fig . [ figddot ] ) . therefore , we can write the transition amplitudes @xmath305 for the lower path and @xmath306 for the upper path , where the upper / lower sign stands for a triplet / singlet initial state on the double - dot . this leads to a total transition amplitude of @xmath307 , and a current @xmath308 . note that the transition @xmath309 is inelastic whereas @xmath310 is not . for an initial singlet state on the double - dot , the other inelastic processes @xmath311 also yield a current proportional to @xmath312 , while the current from the elastic process @xmath313 is proportional to @xmath314 . similarly , starting with a triplet , the sign of the @xmath315 term is negative for an inelastic process , while it is positive for an elastic one . note that there is only one inelastic process @xmath316 , whereas there are more elastic processes allowed for @xmath317 . the total current is obtained by summing over all terms , yielding @xmath318 , where the upper sign stands for an initial triplet state and the lower sign for a singlet , in agreement with eq . ( [ eqnidd ] ) . we finally emphasize that for the singlet @xmath319 and for the triplet @xmath237 the double - dot state is entangled , i.e. a correlated two - particle state , and thus the proposed setup probes a genuine two - particle interference effect via the aharonov - bohm oscillations in the current ( noise ) . note also that we can continuously transmute the statistics from fermionic to bosonic ( like for anyons ) : the symmetric orbital part of @xmath319 goes into an antisymmetric one at half a flux quantum , and vice versa for @xmath237 . we have evaluated the noise also for finite frequencies @xcite , and found that again @xmath320 , and , moreover , that the odd part of @xmath321 leads to slowly decaying oscillations of the noise in real time , @xmath322 , @xmath323 , which can be ascribed to a charge imbalance on the double dot during an uncertainty time @xmath324 . we finally note that the three triplets can be further distinguished by an orientationally inhomogeneous magnetic field which results in a spin - berry phase @xcite that leads to left , right or no phase - shift in the aharonov - bohm oscillations of the current ( noise ) . we have considered a further scenario of double - dots @xcite , where the dots are aligned in parallel between the leads , as in sec . [ ssecddot ] , but now no direct coupling is assumed between them . however , they are coupled with a tunneling amplitude @xmath290 to two superconducting leads . the s - wave superconductor favors an entangled singlet - state on the dots ( like in a cooper pair ) and further provides a mechanism for detecting the spin state via the josephson current . it turns out that in leading order @xmath325 the spin coupling is again described by a heisenberg hamiltonian @xcite @xmath326 where @xmath327 , and the energy of the dot is @xmath328 below the lead fermi energy . here , @xmath329 is the average phase difference across the superconductor double - dot superconductor ( s - dd - s ) junction . we can modify the exchange coupling between the spins by tuning the external control parameters @xmath290 and @xmath329 . thus , we have presented here another implementation of a two - qubit quantum gate ( see sec . [ sec2bit ] ) or an entangler " for epr transport ( see sec . [ ssecenoise ] ) . furthermore , the spin state on the dot can be probed if the superconducting leads are joined with one additional ( ordinary ) josephson junction with coupling @xmath330 and phase difference @xmath331 into a squid - ring . the supercurrent @xmath332 through this ring is given by @xcite @xmath333 where @xmath334 . measurement of the spin- and flux - dependent critical current @xmath335 probes the spin state of the double dot . this is realized by biasing the system with a dc current @xmath336 until a finite voltage @xmath337 appears for @xmath338 @xcite . we have described a concept for a quantum computer based on electron spins in quantum - confined nanostructures , in particular quantum dots , and presented theoretical proposals for manipulation , coupling and detection of spins in such structures . we have discussed the requirements for initialization , read - in , gate operations , read - out , coherence , switching times and precision and their actual realization . by putting it all together , we have illustrated how a scalable , all - electronically controlled quantum computer can be envisioned . we have shown that there is a fruitful link between mesoscopic transport phenomena and quantum communication that is based on production , detection and transport of electronic epr pairs . we have proposed and analyzed a variety of experimental setups which would probe novel spin - based phenomena in open and closed mesoscopic nanostructures . the involved physics , which is based on strong correlations and spin phase - coherence of electrons , is of fundamental interest in its own right quite apart from future applications . finally , by implementing the ideas proposed here , experimental evidence could be gained to demonstrate controlled entanglement and coherence of electron spins in nanostructures . this would be a first step in showing that the proposed scheme of spin - based qubits is indeed suitable for quantum computing and quantum communication . kikkawa , i.p . smorchkova , n. samarth , and d.d . awschalom , science * 277 * , 1284 ( 1997 ) ; j.m . kikkawa and d.d . awschalom , phys . * 80 * , 4313 ( 1998 ) ; d.d . awschalom and j.m . kikkawa , phys . today * 52*(6 ) , 33 ( 1999 ) . d.p . divincenzo , g. burkard , d. loss , and e. sukhorukov , in _ quantum mesoscopic phenomena and mesoscopic devices in microelectronics _ , kulik and r. ellialtoglu ( nato asi , turkey , june 13 - 25 , 1999 ) ; see cond - mat/99112445 . cirac and p. zoller , phys . 74 * , 4091 ( 1995 ) ; c. monroe _ et al . _ , _ ibid . _ * 75 * , 4714 ( 1995 ) . q.a . turchette _ _ , phys . rev . lett . * 75 * , 4710 ( 1995 ) . d. cory , a. fahmy , and t. havel , proc . * 94 * , 1634 ( 1997 ) ; n. a. gershenfeld and i. l. chuang , science * 275 * , 350 ( 1997 ) . b. kane , nature * 393 * , 133 ( 1998 ) . a. shnirman , g. schn , and z. hermon , phys . lett . * 79 * , 2371 ( 1997 ) . et al . _ , * 80 * , 4032 ( 1998 ) ; _ ibid . _ * 81 * , 689 ( 1998 ) . et al . _ , nature * 395 * , 873 ( 1998 ) ; i.j . maasilta and v.j . goldman , phys . * 84 * , 1776 ( 2000 ) . shor , phys . rev . a * 52 * , r2493 ( 1995 ) ; a.m. steane , phys 77 * , 793 ( 1996 ) ; d.p . divincenzo and p.w . shor , _ ibid . _ * 77 * , 3260 ( 1996 ) ; e. knill and r. laflamme , phys . a * 55 * , 900 ( 1997 ) ; d. gottesman , _ ibid . _ * 54 * , 1862 ( 1996 ) ; e. dennis , quant - ph/9905027 .

we present a method for reconstructing the phonon relaxation time distribution ( including polarization ) in a material from thermal spectroscopy data . the distinguishing feature of this approach is that it does not make use of the effective thermal conductivity concept and associated approximations . the reconstruction is posed as an optimization problem in which the relaxation times are determined by minimizing the discrepancy between the experimental relaxation traces and solutions of the boltzmann transport equation ( bte ) for the same problem . the latter may be analytical , in which case the procedure is very efficient , or numerical . the proposed method is illustrated using monte carlo solutions of thermal grating relaxation as synthetic experimental data . the reconstruction is shown to agree very well with the relaxation times used to generate the synthetic monte carlo data and remains robust in the presence of uncertainty ( noise ) .

1609.05864

by probing the thermal response at timescales and lengthscales on the order of the phonon relaxation times and free paths , respectively , thermal spectroscopy provides a potentially powerful means of extracting information about these intrinsic transport properties of real materials @xcite . as a result , thermal spectroscopy has recently received a lot of attention for applications related to the development of thermoelectric materials @xcite . the success of this technique relies on reliably extracting the information embedded in the material response , an inverse problem of considerable complexity that currently remains open . in order to use precise language , in this paper we focus on time - domain thermoreflectance . we believe that the methodology proposed in this paper can be straightforwardly extended to complementary approaches based on frequency - domain thermoreflectance @xcite . traditional approaches to time - domain thermoreflectance data analysis start by matching ( in a least - squares - fit sense ) the experimental response to numerical solutions of the heat conduction equation with the thermal conductivity treated as an adjustable , effective " , quantity @xcite . the free path distribution is subsequently extracted by assuming that the fitted effective thermal conductivity is given by the convolution of the differential free path distribution in the material with a suppression function that is assumed to be geometry dependent @xcite . an approximate suppression function has been developed for the thermal grating geometry @xcite ; suppression functions for other , more complex , geometries have yet to be developed . robust optimization frameworks for inverting the resulting integral equations have also been developed @xcite . however , since , by design , the material response in the experiment is not in the traditional fourier regime defined by characteristic timescales and lengthscales being much longer than the phonon mean free time and mean free path , respectively approaches based on fitting the material response using fourier theory can only be accepted as approximate . as shown in section [ discussion ] , the requirement of long times is particularly difficult to satisfy even if lengthscale corrections are applied . in general , in situations where the requirements for fourier behavior are not carefully met , the effective thermal conductivity obtained by such fits will depend on the fitting time window , the measurement location @xcite , and the experimental geometry , highlighting the fact that it is not a material property and casting doubt over the reliability of the inverse calculation . to avoid these conceptual but also practical problems , in the present work , we propose a methodology for reconstructing the relaxation time distribution in the material from thermal spectroscopy experimental results using an approach that at no time makes an assumption of an underlying fourier behavior . in our work , the reconstruction problem is formulated as a non - convex optimization problem whose goal is to minimize the difference between the experimental material response and the response as calculated by a bte - based model of the experimental process . here , we note the recent paper @xcite in which reconstruction is based on a fourier - space solution of a bte model of the two - dimensional transient thermal grating experiment . despite emphasizing the bte more than in previous approaches , the work in @xcite , ultimately , still reverts to a fourier - based formulation by fitting effective thermal diffusivities and introducing related approximations ( e.g. late times compared to the mean phonon relaxation time ) . a further distinguishing feature of our work is the use of deviational monte carlo ( mc ) simulation @xcite as a method to solve the bte which expands the domain of applicability of the approach , since it does not rely on the simplicity of the experimental setup or the degree to which it is amenable to analytical treatment . the remainder of the paper is organized as follows . in section [ formulation ] , we formulate the reconstruction as an optimization problem requiring only solutions of the bte and present the optimization framework used for the reconstruction . in section [ application / validation ] , we discuss the implementation of the proposed method in the context of an archetypal experimental setup , namely that of a transient thermal grating @xcite . in the same section , we validate our proposed method using synthetic data generated through monte carlo ( mc ) simulation . in section [ discussion ] we provide some comparison to effective thermal conductivity approaches and a discussion on the validity of the latter . in section [ conclusion ] we provide a summary of our work and suggestions for future improvement . given the small temperature differences usually associated with thermal spectroscopy experiments , here we consider the linearized bte @xmath1 where @xmath2 is the deviational energy distribution , @xmath3 is the phonon frequency , @xmath4 is the phonon traveling direction , @xmath5 is the occupation number of phonon modes , @xmath6 is the phonon group velocity , @xmath0 is the frequency - dependent relaxation time and @xmath7 is the reduced planck constant . here and in what follows , we use @xmath3 to denote the dependence on both frequency and polarization . the above equation is linearized about the equilibrium temperature @xmath8 , to be understood here as the experimental reference temperature . in general , @xmath9 ; however , as a result of the linearization , @xmath10 ; in other words , the solutions ( and associated reconstruction ) are valid for the experiment baseline temperature @xmath8 . also , @xmath11 and @xmath12 is the bose - einstein distribution with temperature parameter @xmath13 , given by @xmath14 where @xmath15 is boltzmann s constant . finally , @xmath16 is referred to as the deviational pseudo - temperature ( @xmath17 is the pseudo - temperature ) . note that the deviational pseudo - temperature , which is different from the deviational temperature defined below , is defined by the energy conservation statement @xcite @xmath18 d \omega d \mathbf{\omega}= 0 , \label{pseudotemperature}\ ] ] in which @xmath19 is the density of states , @xmath20 is the frequency - dependent volumetric heat capacity , and @xmath21 represents the differential solid angle element such that @xmath22 and @xmath23 refer to the polar and azimuthal angles in the spherical coordinate system , respectively . the temperature @xmath24 is computed from @xmath25 d \omega d \mathbf{\omega}= 0 , \label{temperature}\ ] ] where @xmath26 is the deviational temperature . the frequency - dependent free path is given by @xmath27 where @xmath28 is the group velocity magnitude . our goal is to obtain an accurate and reliable approximation to the function @xmath29 from the experimental measurements , with the latter typically in the form of a temperature ( relaxation ) profile . reconstruction of the free path distribution follows from relationship ; in other words , the group velocities @xmath30 are assumed known , since they can be reliably calculated either experimentally @xcite by means of raman spectroscopy , x - ray scattering , etc , or theoretically @xcite using methods such as density functional theory ( dft ) . we propose the use of an optimization framework in which the experimental measurements provide targets that need to be reproduced by the bte solution ; in other words , @xmath29 is determined as the function that optimizes ( minimizes ) the discrepancy between the experimental result and the bte prediction for the same quantity . one important consideration is use of a suitable discrete representation for @xmath29 . in our formulation any number of longitudinal / transverse acoustic / optical branches @xmath31 may exist and are solved for explicitly , where the superscript @xmath32 denotes the branch ; the total scattering rate @xmath33 is obtained using the matthiessen s rule . in the examples following below we have taken @xmath34 where @xmath35 denotes the longitudinal acoustic branch , while @xmath36 and @xmath37 represent the two transverse acoustic branches . we have not considered optical phonons in this work because acoustic modes account for most of the heat conduction @xcite , but also because the proposed methodology can be straightforwardly extended to include them . in order to account for the most general situation where no a - priori information is available on the functional form of the @xmath38 relation , we have used a piecewise linear relation between @xmath39 and @xmath40 . the intersections between the piecewise linear segments are smoothed via a third order polynomial function ( of @xmath40 ) that guarantees continuity of the relaxation time function and its first derivative @xcite ( the continuity requirement may be easily removed , allowing jumps between different segments ) . note that our parametrization is equivalent to a ( piecewise ) relationship of the form of @xmath41 for some real numbers @xmath42 and @xmath43 , which is related to relations commonly used in the literature @xcite . our approach , however , does not constrain the value of @xmath43 to any particular value or even an integer value . the actual functional form is given by @xmath44 \mathds{1 } _ { \omega \in [ x^{s}_{2j } , x^{s}_{2j+1 } ] } + \\ \sum_{j=1}^{m-1 } \left [ a^{s}_{j } \left [ \log(\omega ) \right]^3 + b^{s}_{j } \left [ \log(\omega ) \right]^2 + c^{s}_{j } \log(\omega)+ d^{s}_{j } \right ] \mathds{1}_{\omega \in [ x^{s}_{2j-1 } , x^{s}_{2j } ] } , \label{parametrization_2ta } \end{gathered}\ ] ] where @xmath45 determines the number of segments . we note that since the minimum and maximum frequencies for each branch ( @xmath46 and @xmath47 ) are known ( input ) , there are @xmath48 unknowns in the model for each branch @xmath49 , consisting of @xmath50 , ... , @xmath51 , and @xmath52 , ... , @xmath53 , which are the inputs to the optimization algorithm , leading to a total of @xmath54 unknowns ( see section [ optimalgorithm ] for implementation details ; also note the input unknowns are denoted @xmath55 in that section ) . also , @xmath56 denotes the indicator function whose value is @xmath57 if @xmath58 , and @xmath59 if @xmath60 . once @xmath61s and @xmath62s are known , the coefficients @xmath63 , @xmath64 , @xmath65 , @xmath66 , and the intervals in which each line or polynomial function is active inside , @xmath67s , can be computed . the relationships for calculating these parameters are provided in appendix a. in the interest of compactness , we will use the vectorial notations @xmath68 and @xmath69 to represent the unknown parameters . moreover , we will use the symbol @xmath70 to denote the set of all unknown vectors . the reconstruction proceeds by minimizing the objective function @xmath71 , \label{objectivefunction_2ta}\ ] ] where @xmath72 denotes the experimentally measured temperature , @xmath73 is the temperature computed from solution of the bte , and @xmath74 is the total number of ( independent ) measurements available . as indicated above , @xmath75 is in general a function of space , time , but also the characteristic lengthscale of the thermal relaxation problem , @xmath76 . as a result , the optimization process can be based on data for @xmath73 and @xmath75 at various time instances , spatial locations , and for different characteristic system parameters , with @xmath77 . we also point out that @xmath73 may be obtained by any method that can provide accurate solutions of as applied to the experimental setup ; here , we consider both semi - analytical solution ( using fourier transform techniques ) and mc simulation . the second term in exploits the fact that the bulk value of the thermal conductivity , @xmath78 , is usually known , to enhance the importance of the low frequency modes in the optimization . although this term is optional " , we have found that including this term with a weight of @xmath79 improves the reconstruction quality considerably in the low frequency regime , because the low density of states associated with those frequencies prevents them from influencing the optimization process if the objective function only includes a comparison between @xmath73 and @xmath75 ( the first term in ) . determination of @xmath70 ( the vectors @xmath80 , @xmath81 , @xmath82 , @xmath83 , @xmath84 , and @xmath85 ) which minimize the objective function is achieved using the nelder - mead ( nm ) algorithm @xcite , a simplex - based search method that is free from gradient computation . although nm can neither be categorized as a local optimizer or a global optimizer , it performs significantly better than common local minimizers ( which fail to converge to the correct solution in the presence of highly non - convex objective functions such as ) @xcite , while it is significantly less costly than common global optimizers like genetic algorithms @xcite or simulated annealing algorithms @xcite . in the nm algorithm , the optimization process proceeds as follows ( adopted from @xcite ) : 1 . start with @xmath86 initial points @xmath87 , @xmath88 , in the @xmath89-dimensional space defined by the @xmath89 unknown parameters ( here @xmath90 ) . a possible choice of initial simplex ( the set of @xmath86 initial points ) is @xmath91 and @xmath92 for scalars @xmath93 , @xmath94 ; here @xmath91 is an arbitrary point in the @xmath89-dimensional space of unknown parameters and @xmath95 is the unit vector in the @xmath96-th direction @xcite . 2 . order the points @xmath87 , @xmath88 , in an ascending order based on their objective function values . assign new superscripts such that @xmath97 . 3 . calculate the centroid of the @xmath89 points with lowest objective function values , @xmath98 . 4 . calculate the reflected point , @xmath99 , where @xmath100 . 5 . if @xmath101 , replace @xmath102 with @xmath103 and start the next iteration ; go to step 2 . if @xmath104 , compute @xmath105 , the expanded point , where @xmath106 . 1 . if @xmath107 , replace @xmath102 with @xmath108 and start the next iteration ; go to step 2 . 2 . if @xmath109 , replace @xmath102 with @xmath103 and start the next iteration ; go to step 2 . 7 . if @xmath110 1 . if @xmath111 , compute @xmath112 , the contracted point , where @xmath113 . if @xmath114 , replace @xmath102 with @xmath115 and start the next iteration ; go to step 2 otherwise , go to step 8 . if @xmath116 , compute @xmath117 , the new contracted point . if @xmath118 , replace @xmath102 with @xmath119 and start the next iteration ; go to step 2 . otherwise , go to step 8 . 8 . for @xmath94 , compute @xmath120 where @xmath121 , replace @xmath87 with @xmath122 , and start the next iteration ( go to step 2 ) . in words , in each iteration , the objective is to eliminate the worst point ( with the largest value of objective function ) among the current @xmath86 points of the simplex via reflection ( step 5 or 6-b ) , expansion ( step 6-a ) , contraction ( step 7 ) or shrinkage ( step 8) . figure [ nm ] illustrates this algorithm for the case @xmath123 . in the present work , the parameters @xmath124 , and @xmath125 are taken to be 1 , 1 , and 0.5 , respectively , as proposed in @xcite . we have also chosen the shrinkage coefficient to be @xmath126 as recommended in @xcite for robustness in the presence of noise . [ h ] .5 ) is shown . ( a)- point @xmath127 of the previous simplex ( which consists of points @xmath91 , @xmath128 , and @xmath127 ) is replaced with a new point from the set @xmath129 . the new simplex consists of @xmath91 , @xmath128 , and the replacement of @xmath127 . ( b)- if none of the candidate points in figure [ nm1 ] have an objective function smaller than @xmath130 , then the simplex shrinks . the new simplex is @xmath91 , @xmath131 , and @xmath132 . , title="fig : " ] .5 ) is shown . ( a)- point @xmath127 of the previous simplex ( which consists of points @xmath91 , @xmath128 , and @xmath127 ) is replaced with a new point from the set @xmath129 . the new simplex consists of @xmath91 , @xmath128 , and the replacement of @xmath127 . ( b)- if none of the candidate points in figure [ nm1 ] have an objective function smaller than @xmath130 , then the simplex shrinks . the new simplex is @xmath91 , @xmath131 , and @xmath132 . , title="fig : " ] the most expensive step in this process is the calculation of the objective function which requires evaluation of @xmath73 . the nm algorithm typically requires one or two evaluations of at each iteration , except when a shrinkage step takes place where @xmath133 evaluations of are required . single function evaluation occurs as a result of successful initial reflection corresponding to step 5 of the algorithm . two evaluations of the objective function result from either expansion at step 6-a , reflection at step 6-b , outside contraction at step 7-a or inside contraction at step 7-b . the simplex typically starts to shrink near the convergence point where it can not find any new direction that yields lower values of the objective function and consequently instead searches for such point inside the previous simplex by shrinking it toward its best vertex @xcite . in this section , we consider the one - dimensional transient thermal grating ( ttg ) experiment as an example for illustrating and validating the methodology presented in section [ formulation ] . the mathematical model of the ttg experiment will be briefly discussed in section [ ttg ] . using this mathematical formulation , we generate synthetic experimental data ; this process is described in section [ synthetic ] . we then use these synthetic data to reconstruct @xmath29 for the material model used in the mc simulations ; the details of the reconstruction process are discussed in section [ example : reconstruction ] . we consider reconstruction both via semi - analytical solution of the bte and mc simulation ; the advantages of each approach are discussed and contrasted . in section [ validation ] , we compare the reconstructed relaxation times and free path distribution with those of the original data used in the mc simulations for generating the synthetic data . ttg experiment has been widely used to study transient thermal transport . in this experiment , a sinusoidal profile with unit amplitude is generated by crossing two short pump pulses @xcite . by assuming a one - dimensional heat transport along the @xmath134 direction , the bte from equation can be written in the form @xmath135 in which the last term is due to the sinusoidal initial temperature profile . here , @xmath76 is the grating wavelength ( see discussion in section [ inverseproblem ] ) , and @xmath136 is the imaginary unit . this equation is sufficiently tractable to allow analytical solutions in fourier / reciprocal space @xcite . we follow a similar procedure ( see appendix b ) to obtain a fourier space solution for @xmath73 ; the final result is @xmath137 d \omega } \right ] , \label{analytical}\ ] ] where @xmath138 is the volumetric heat capacity and @xmath139 denotes the fourier transform of @xmath140 with respect to the time variable . also , @xmath141 is given by @xmath142 we note here that the temperature profile required for evaluating is given by @xmath143 . instead of using experimental data ( for @xmath75 ) to validate the proposed methodology , we use synthetic data generated from mc simulation of . this approach enables considerably more precise validation because it sidesteps issues of experimental error , which is hard to quantify , but also modeling error ( how accurately does model the ttg experiment ? ) , which is even harder to quantify . in other words , reconstruction from @xmath75 measurements " obtained from mc simulations should be able to reproduce the function @xmath29 used to generate the synthetic data exactly , making any discrepancies directly attributable to numerical / methodological error . in the interest of simplicity , the material was assumed to be silicon . however , in order to fully explore the ability of the reconstruction process to capture arbitrary functional relations @xmath0 , we consider both a simple analytical model and an ab initio model of this material . the simple model considered here is described in @xcite ( thermal conductivity @xmath144 ) and will be referred to as the _ holland _ model . the second model considered ( thermal conductivity @xmath145 ) is described in @xcite and will be referred to as the _ ab initio _ model throughout this paper . mirroring the experimental procedure as closely as possible , we obtain transient temperature relaxation data ( solutions of ) at one spatial location for a number of characteristic ( grating ) lengths , namely @xmath146 ( @xmath147 for example ) . these solutions were obtained using the adjoint monte carlo method recently proposed @xcite that is particularly efficient if solutions at particular spatial locations are of interest . the mc simulations used a sufficiently high number of particles ( @xmath148 ) to ensure that the synthetic data was essentially noise free . as discussed in section [ validation ] , we have also produced noisy synthetic data using significantly fewer particles in order to investigate the robustness of the optimization method to noise . relaxation data was recorded " in the time period @xmath149 , where @xmath150 ; here , @xmath151 denotes the time taken for the response to decay to 1% of its original amplitude for each @xmath76 . during this time period , 100 discrete @xmath75 measurements were obtained . eight different wavelengths were simulated , namely , @xmath152 , @xmath153 , @xmath154 , @xmath155 , @xmath156 , @xmath157 , @xmath158 , and @xmath159 . as a result , 800 total @xmath75 measurements were available for reconstruction ( @xmath160 ) . reconstruction proceeds by comparing solutions of 1-d ttg experiment , equation , in the form of @xmath161 to the counterpart @xmath74 measurements of @xmath75 as a means of finding the optimum unknown vectors . here , solutions of were obtained using inverse fast fourier transform ( ifft ) of as explained in @xcite , or by using adjoint mc simulations @xcite . although mc simulations are more expensive , they are investigated here because they make the proposed method significantly more general , since they are not limited to problems which lend themselves to analytical or semi - analytical solutions . as discussed before , phonon group velocities were assumed known , while @xmath29 was described by the model given in with @xmath162 ( a piecewise linear function with three segments ) . as shown in section [ validation ] , this approximation level gives very good results . to generate the initial starting points ( @xmath163 vertices of the simplex ) , we have used @xmath164 as recommended in @xcite , where @xmath165 represents the @xmath96-th component of vector @xmath91 as discussed in the first step of the nm algorithm in section [ optimalgorithm ] . larger values of @xmath93 ( we have tested @xmath166 ) may lead to large steps during the initial iterations of the optimization , which consequently can move the simplex toward a wrong local minimum ; on the other hand , very small values of @xmath93 ( we have tested @xmath167 ) lead to smaller steps throughout the optimization process , which require a ( significantly ) larger number of iterations in order to guarantee that the parameter space is adequately sampled . we have obtained similar performances to @xmath164 using @xmath168 and @xmath169 . we perform the optimization in four stages . in the first stage , we solve for one segment ( @xmath170 ) which is the same for all branches , leading to only two unknown parameters . due to the small number of unknowns , this step is very cheap but very valuable since it results in a significant reduction in the value of objective function in only 10 - 20 iterations . in the second stage , @xmath35 and @xmath171 modes are still assumed to be the same but the intended number of segments is used , increasing the number of unknowns to @xmath48 ( six in the present case ) . the initial condition for this stage is taken to be the same as the optimized value of the previous stage ( or a slightly perturbed version ) . in the third stage , we repeat the optimization process , now for @xmath172 unknowns ( the two @xmath171 branches are assumed to be the same , @xmath173 ) , starting from the optimized parameters of the second stage . finally , we perform the optimization for all @xmath54 unknowns starting from the optimized parameters of the previous stage ( repeated for each of the three branches independently ) . this procedure is more robust than direct optimization for @xmath54 unknowns , which depending on the initial condition may be trapped in a local minimum of characterized by significantly different relaxation times for the @xmath35 and @xmath171 branches . to reduce the probability of the reconstruction being trapped in a local minimum , in our approach , the first and second stages were repeated starting from 5 distinct initial conditions . the result at each stage with the lowest value of @xmath174 was used as the initial condition for the next stage . although this consideration may not be necessary , it is expected to improve the quality of the optimization process ( by providing a smaller final value of @xmath175 ) due to the non - convexity of the objective function . although the number of iterations varies depending on the initial condition , we have observed that for the value of @xmath45 used in our work ( corresponding to @xmath176 unknowns ) , the third and fourth stages require on the order of 150 - 200 iterations for convergence ( each ) . when using analytical or semi - analytical ( ifft ) solutions for @xmath73 , the computational cost associated with each evaluation is very small and thus cost is not a consideration . using mc simulation to obtain solutions for @xmath73 is more costly and is discussed below . to reduce cost , mc simulations during the first two stages of the optimization process used @xmath177 particles for calculation of @xmath73 . during the third and fourth stages of the process , the number of particles is increased to @xmath178 . figure [ obj_func_particle_num ] investigates the sensitivity of the final value of the objective function @xmath175 in the third stage as a function of the number of particles used ( in this stage of the optimization ) for the ab initio material model . specifically , the figure compares the mean and variance of the objective function compared to the third - stage final value obtained from ifft of ( which is free from noise and independent of number of particles ) . the mc result appears to converge ( asymptotically ) to the deterministic result with increasing number of computational particles , @xmath179 . although this convergence appears to be slow , it is encouraging that the additional error ( compared to deterministic reconstruction ) is small and thus the shallow slope of the convergence implies that the reconstruction is quite robust to noise in the range @xmath180 . [ h ] at the third stage on @xmath179 . increasing the number of particles decreases the mean and variance of @xmath175.,title="fig : " ] since the computational time for each adjoint mc simulation is proportional to @xmath179 @xcite , given our choices for @xmath179 ( @xmath177 for stages 1 and 2 and @xmath178 for stages 3 and 4 ) , the cost of the first two stages of the optimization is negligible compared to the cost of the last two stages . here we note that no attempt has been made to optimize this process ; it is therefore possible that considerably more computational benefits may be possible if a more structured approach is used @xcite that takes into account information such as that contained in figure [ obj_func_particle_num ] . we also note that given the small cost of the first two optimization stages , repeating them a number of times starting from different initial conditions does not increase the cost of the reconstruction process significantly . for an average of 150 - 200 iterations , the cost of each of the third or fourth stages corresponds to 300 - 600 mc simulations , considering that on average 2 - 3 evaluations of at each iteration are required . this number of iterations can be performed efficiently using the adjoint method proposed recently in @xcite . note that a significant fraction of the computational effort is spent on iterations close to the final solution , which is primarily due to the high computational cost of the shrinkage step near convergence as was discussed previously in section [ optimalgorithm ] ( compare @xmath181 function evaluations associated with the shrinkage step , with one or two function evaluations associated with reflection , expansion or contraction ) . in contrast to the present case where validation required very small error tolerances , in actual practice , tolerances will be set by experimental considerations and are expected to be larger , reducing the number of expensive shrinkage processes and making the cost of reconstruction from experimental data smaller . in this section , we discuss reconstruction results with particular emphasis on their ability to reproduce the original material properties ( @xmath0 ) used to generate the synthetic measurements @xmath75 , referred to as true " . although our scheme returns @xmath29 as output , as is typical in the literature , we also provide comparison between the true " and reconstructed " cumulative distribution function ( cdf ) of free paths defined as @xmath182 , where @xmath183 is the set of modes such that @xmath184 . in this section , we present results for the holland material model . for this model , we assumed that the two @xmath171 branches are the same and as a result , the optimization process consists of three stages with a total of @xmath172 unknowns . figure [ compwithanalytical ] shows a comparison between the true material parameters and the reconstructed ones using the semi - analytical solution for @xmath73 . the agreement is excellent . [ h ] .5 .5 figure [ compwithmc ] shows a similar comparison for the case where mc simulations were used for obtaining @xmath73 . we observe that the error in reconstructed relaxation times is more significant when using mc simulations . we attribute this primarily to the noise associated with mc simulation and its interaction with the nm algorithm . [ h ] .5 obtained from adjoint mc simulation using the holland material model.,title="fig : " ] .5 obtained from adjoint mc simulation using the holland material model.,title="fig : " ] the synthetic data used to obtain the results in figures [ compwithanalytical ] and [ compwithmc ] were generated by mc simulations using @xmath148 computational particles . this high number of computational particles leads to a very small variance , corresponding to a very accurate experimental measurement " . in order to evaluate the performance of the reconstruction method in the presence of noise in the experimental measurement , we have also performed reconstruction calculations with noisy synthetic data . for that purpose , we have used only @xmath185 computational particles in the mc simulations that generated the synthetic data , leading to a standard deviation of @xmath186 . this number of particles makes the uncertainty in @xmath75 significantly larger than the noise in common experimental data ( e.g. compare the noisy temperature profile of figure [ noisydata ] with figure 2c of @xcite ) . figure [ noisydata ] shows a comparison between the synthetic data obtained with @xmath148 and @xmath185 particles . [ h ] .5 and @xmath185 computational particles based on the holland material model . two different grating wavelengths are shown . , title="fig : " ] .5 and @xmath185 computational particles based on the holland material model . two different grating wavelengths are shown . , title="fig : " ] the reconstructed relaxation times and free path distribution corresponding to the noisy synthetic data ( @xmath187 ) using semi - analytical and mc solutions with the holland material model are provided in figures [ noisy - analytical ] and [ noisy - mc ] , respectively . in both figures we observe that even in the presence of the considerable noise in the measurement , the algorithm is able to infer the relaxation times and the free path distribution with reasonable accuracy . [ h ] .5 .5 [ h ] .5 .5 in this section , we repeat the validation process , namely using ifft of and adjoint mc simulation in the presence and absence of noise in the synthetic data , for the ab initio material model . the reconstructed material properties in the absence of noise ( @xmath188 ) are provided in figures [ compwithanalytical_2ta ] and [ compwithmc_2ta ] . the reconstructed material properties in the presence of noise ( @xmath187 ) using ifft of and mc simulation are provided in figures [ noisy - analytical_2ta ] and [ noisy - mc_2ta ] , respectively . we observe that the general level of agreement is very good , but not at the level of the holland model discussed in section [ holland ] . this is clearly to be expected given the relative complexity of the two models and the fact that the parametrization is related to the holland model in limiting cases . we also note that although the discrepancy is particularly noticeable for low frequencies , this is primarily an artifact of the low density of states of modes in these frequency ranges . specifically , the density of states of the @xmath35 modes for @xmath189 is zero ( i.e. none of the two terms in influences @xmath190 for @xmath191 ) . the role of density of states can be further verified by noting that the reconstructed free path distribution is in all cases in better agreement with the input ( true " ) data compared to the corresponding relaxation times . we also observe that , as in the case of the holland model , the noise in the synthetic data has only a small effect on the reconstruction quality . [ h ] .5 .5 [ h ] .5 obtained from adjoint mc simulation of the ab initio material model . , title="fig : " ] .5 obtained from adjoint mc simulation of the ab initio material model . , title="fig : " ] [ h ] .5 .5 [ h ] .5 .5 the accuracy of the reconstruction process shown in figures [ compwithanalytical ] , [ compwithmc ] , [ noisy - analytical]-[noisy - mc_2ta ] is very encouraging and , generally speaking , superior to that obtained by effective - thermal - conductivity - based approaches ( see , for example , figure 2 in @xcite , or figure [ lessdata ] of present manuscript ) . in @xcite , it is pointed out that the effective - thermal - conductivity - based approach is less accurate outside the range of lengthscales over which input data is available and thus the discrepancies observed in @xcite may be attributed to the lack of data in the range @xmath192 and @xmath193 . on the other hand , as shown in figure [ lessdata ] , reconstruction using the method proposed here with @xmath194 and @xmath195 , reveals that this method is minimally sensitive to the range in which data exist . [ h ] .5 and @xmath195 ) . relaxation data was recorded " in the time period @xmath149 , where @xmath196 . the reconstructed cdf , @xmath197 " curve corresponds to data transcribed @xcite from figure 2 of @xcite , which uses the effective thermal conductivity approach with 12 relaxation profiles with wavelengths in the range @xmath198 . , title="fig : " ] .5 and @xmath195 ) . relaxation data was recorded " in the time period @xmath149 , where @xmath196 . the reconstructed cdf , @xmath197 " curve corresponds to data transcribed @xcite from figure 2 of @xcite , which uses the effective thermal conductivity approach with 12 relaxation profiles with wavelengths in the range @xmath198 . , title="fig : " ] sensitivity to the range over which input data exists is very important , not only because some lengthscales may not be experimentally accessible @xcite , but also because the effective thermal conductivity concept can not be used to analyze small lengthscale or early - time data recall that fourier - based descriptions are only accurate in the limit of large lengthscales ( compared to the mean free path ) and long timescales ( compared to the mean time between scattering events ) . below , we theoretically analyze the limitations arising from these requirements by studying in detail the response associated with the ttg setup . another problem associated with the effective thermal conductivity approach is that , other than in the thermal grating geometry , the associated suppression functions are unknown ( and depend not only on the experimental setup , but also on the material parameters that are being calculated ) . using gray mc simulations as a substitute for the suppression function introduces approximations ( see @xcite for a discussion ) that are not always acceptable and more importantly not _ a priori _ controllable . clearly , in order to extract information from the experiment correctly , a model needs to be able to fit the functional form of the experimental response accurately . we now examine , both theoretically and via simulation , if this requirement is satisfied by the effective thermal conductivity formalism in the case of the ttg problem . we first recall that the effective thermal conductivity concept applies to late times , as defined by @xmath199 @xcite ( @xmath200 is the fourier transform variable with respect to time , see discussion in section [ ttg ] ) . following previous work @xcite , we simplify in the limit of @xmath199 to obtain ( see appendix c ) @xmath201 . \label{s_omega_quasi}\ ] ] here , @xmath202 . substituting equation in and performing the inverse fourier transform in time leads to the following temperature response in the _ late time limit _ @xmath204 where @xmath205 \right\ } d \omega \label{k_quasi}\ ] ] is the effective thermal conductivity as found before @xcite and @xmath206 ^ 2 } { \left [ \int_{\omega } \frac{c_{\omega } } { k\!n_{\omega}^2 + 1 } d \omega \right]^3 } \label{a}\ ] ] is the response amplitude , and @xmath207 is an effective heat capacity . in other words , although the response is of exponential form , it is not a solution of the fourier heat equation with an effective thermal conductivity @xmath208 , as usually assumed . response is a diffusive type of solution which differs from the classical fourier solution by featuring * an amplitude , @xmath209 , different from the original temperature perturbation in the ttg experiment ( taken to be @xmath210 here ) * a thermal conductivity given by @xmath208 * a heat capacity given by @xmath211 moreover , we always need to keep in mind that the above is only true for @xmath199 , which turns out to have significant implications . using the fact that @xmath212 we can write in the form @xmath213 where @xmath214 is the ( average ) knudsen number , @xmath215 is the mean free path and @xmath216 is the mean free time . this leads us to conclude that the requirement @xmath217 ( @xmath218 ) is incompatible with a solution of this form ( @xmath219 ) unless @xmath220 . mathematical " incompatibility is a simple statement of the physical fact that relaxation times scale with the system size and thus for a small system not satisfying @xmath221 the response timescale will be on the order of the relaxation time or smaller ( no experimentally measurable signal will be available at times much longer than the relaxation time)in other words , no fully diffusive relaxation regime is possible for systems not satisfying @xmath220 . the above discussion is validated in figures [ temp_comparison_2ta ] and [ temp_comparison ] ; the figures compare various fourier - based relaxation profiles to the bte solution of the grating relaxation problem for various values of the wavelength @xmath76 . in addition to solution , we also consider the traditional effective thermal conductivity approach given by @xcite , @xmath222 the figures also show the solution of the same problem obtained via ifft of using the _ reconstructed _ @xmath29 data ( as shown in figure [ compwithanalytical ] for the holland model and figure [ compwithanalytical_2ta ] for the ab initio model ) . figure [ temp_comparison_2ta ] shows results for the ab initio material model ( @xmath223 ) and figure [ temp_comparison ] for the holland model ( @xmath224 ) . here , @xmath215 was calculated using the procedure outlined in @xcite . in both figures , we observe that the bte solution based on reconstructed relaxation times , referred to as present " , is the only solution that is able to predict the correct temperature profiles ( ifft of , denoted by bte " ) in all @xmath225-regimes , for all times . referring to figure [ temp_comparison_2ta ] , at @xmath226 , the ballistic solution of the bte ( provided in appendix d ) is also able to predict the temperature profile , although for @xmath227 some discrepancies are observable ( scattering is no longer completely negligible ) ; fourier - based approximations are inaccurate . for @xmath228 ( @xmath229 ) , none of the fourier - based approximations are able to predict the correct temperature profile ; although solution provides a significantly better approximation compared to expression at late times , the bte solution is clearly not of exponential form at early and moderate times ( in other words , although appears to be close to the bte solution , this is partly because both go to 0 at late times ) . at @xmath230 ( @xmath231 ) the agreement between and the bte solution is good ; equation remains inaccurate . the comparison at @xmath232 ( @xmath233 ) shows good agreement between equation and the bte solution ; equation also becomes a reasonable approximation . further comparison ( not shown here ) shows that at @xmath234 ( @xmath235 ) , both fourier - based solutions are able to reproduce the boltzmann solution . a comparison for the holland model can be found in figure [ temp_comparison ] . the results are similar , but generally reveal worse performance by equation in the critical @xmath236 range . also , the ballistic behavior is observed at smaller wavelengths ( approximately @xmath237not shown here ) . [ h ] .5 . solutions based on equations and are denoted equation " and equation " , respectively.,title="fig : " ] .5 . solutions based on equations and are denoted equation " and equation " , respectively.,title="fig : " ] .5 . solutions based on equations and are denoted equation " and equation " , respectively.,title="fig : " ] .5 . solutions based on equations and are denoted equation " and equation " , respectively.,title="fig : " ] [ h ] .5 . solutions based on equations and are denoted equation " and equation " , respectively.,title="fig : " ] .5 . solutions based on equations and are denoted equation " and equation " , respectively.,title="fig : " ] .5 . solutions based on equations and are denoted equation " and equation " , respectively.,title="fig : " ] .5 . solutions based on equations and are denoted equation " and equation " , respectively.,title="fig : " ] in summary , although relation shows that a regime featuring exponential relaxation is possible , this is only physically realizable ( and measurable ) for @xmath220 . this is clearly more favorable than the typical requirement ( @xmath238 for homogeneous materials , @xmath239 in the presence of boundaries @xcite ) for validity of fourier s law uncorrected ; the improved range of validity is due to the corrections - , which are available , in part , due to the simplicity of the ttg problem and geometry . although extensions of fourier approaches are sometimes possible ( e.g. see @xcite for the role of boundaries ) , these are not always guaranteed to exist or be tractable and their range of validity can not be expected to extend to @xmath240 , except fortuitously . with the above considered , for the ttg case , the effective thermal conductivity construction and associated solution is only strictly theoretically justified in the limit @xmath220 . in some cases this coincides with the range in which experiments operate @xcite , in which case reconstruction using a diffusive approximation may be acceptable provided : * the correct asymptotic relation ( i.e. equation ) is used * some loss of accuracy outside the range over which experimental data is available is acceptable we also point out that the validation in @xcite featured wavelengths satisfying @xmath220 . in general , however , the requirements @xmath220 and @xmath203 are very restrictive , especially when considering that typical pump - probe response time is on the order of a few nanoseconds @xcite . in contrast , the method proposed here introduces no lengthscale or timescale restriction ; moreover , if input data is available over a limited range of wavelengths , the quality of the reconstruction does not suffer significantly . by analyzing the ttg response in detail we have shown that , as expected , reconstruction using the effective thermal conductivity concept is limited to late ( compared to the mean free time ) times and large ( compared to the mean free path ) lengthscales ( @xmath220 for the ttg in particular ) . using input data that does not satisfy these requirements will lead to reconstruction error ; on the other hand , using data in a limited range of characteristic lengthscales also leads to reconstruction error @xcite . in this paper we have proposed an alternative approach to the reconstruction problem that avoids the limitations associated with the effective thermal conductivity approach . the proposed method assumes knowledge of the phonon group velocities and poses the reconstruction as an optimization problem seeking the function @xmath29 that minimizes the discrepancy between the observed experimental data and the solution of the bte as applied to the experimental process . although much research and improvements remain to be made , one of the main contributions of the present paper is to show that reconstruction based on a more rigorous foundation ( which does not make use of the assumption of fourier behavior ) is possible . the proposed formulation is sufficiently general to accept solutions of the bte obtained by any means . in fact , we envision applications to problems including interfaces between materials @xcite ( and where the reconstruction will include inference of interface properties ) as direct extensions of the present work . the optimization process is achieved by the nelder - mead algorithm that does not require gradient calculation and is thus robust to noise . our results verify that reconstruction is robust both in the presence of noise in the input ( experimental ) data and the bte solutions when the latter are obtained by mc simulation . although mc simulations are considerably more expensive , they are , in fact , the more relevant to applications of practical interest ( e.g. @xcite ) which are in general high - dimensional and not expected to lend themselves to analytical solutions . the cost of mc simulations is partially mitigated by their excellent parallel efficiency . moreover , in cases where the experimental measurement is limited to one ( or a few ) discrete spatial locations , the mc simulation cost can be made independent of dimensionality using techniques such as adjoint formulations . our validation tests started from multiple distinct initial conditions thus subjecting the optimization process to a worst - case scenario , since in most cases some a - priori information on the nature of the solution is expected to exist . in the present case , the lack of prior information was overcome by starting the optimization process , as a first stage , from a sweep of different initial conditions and choosing the result with the lowest objective function value as the initial condition for a second optimization stage . in the case of mc simulations , this approach was possible due to the robustness of the nm algorithm to noise , which enabled us to perform the first two stages of calculations using very cheap ( noisy ) simulations . in general , a - priori information could be very useful in ensuring convergence to the correct solution , which due to the highly non - convex nature of the minimization problem is not guaranteed . future research will focus on formulations which exploit a - priori information on the nature of the solution as well as alternative optimization algorithms to improve convergence rates and accuracy . our results also show that the reconstruction process is very robust with respect to the time window over which the material response is observed , in contrast to fourier - based approaches which assume a late - time response . more specifically , by matching the material response to bte solutions , the reconstruction is successful with response data limited to very early times ( e.g. on the order of 5 nanoseconds ) that are typical in pump - probe experiments @xcite but do not satisfy assumptions required for diffusive behavior to set in . validation of the proposed methodology using experimental data of thermal relaxation processes will be the subject of a future publication . the authors would like to thank v. chiloyan , s. huberman , a. maznev and l. zeng for many useful comments and discussions . this work was supported by the solid - state solar - thermal energy conversion center ( s3tec ) , an energy frontier research center funded by the u.s . department of energy , office of science , basic energy sciences under award # de - sc0001299 and de - fg02 - 09er46577 . h , _ superior thermal conductivity of single - layer graphene _ , nano letters 8 , 3 , 902 - 907 ( 2008 ) . , _ frequency dependence of the thermal conductivity of semiconductor alloys _ , physical review b 76 , 075207 ( 2007 ) . , _ heating - frequency - dependent thermal conductivity : an analytical solution from diffusive to ballistic regime and its relevance to phonon scattering measurements _ , physical review b 91 , 165311 ( 2015 ) . , _ analytical interpretation of nondiffusive phonon transport in thermoreflectance thermal conductivity measurements _ , physical review b 90 , 064302 ( 2014 ) . , _ length - dependent thermal conductivity in suspended single - layer graphene _ , nature communications 5 , 3689 ( 2014 ) . , _ onset of nondiffusive phonon transport in transient thermal grating decay _ , physical review b 84 , 195206 ( 2011 ) . , _ thermal conductivity spectroscopy technique to measure phonon mean free paths _ , physical review letters 107 , 095901 ( 2011 ) . , _ determining phonon mean free paths from observations of quasiballistic thermal transport _ , physical review letters 109 , 205901 ( 2012 ) . , _ spectral mapping of thermal conductivity through nanoscale ballistic transport _ , nature nanotechnology 10 , 701 - 706 ( 2015 ) . , _ variational approach to extracting the phonon mean free path distribution from the spectral boltzmann transport equation _ , physical review b 93 , 155201 ( 2016 ) . , _ transport regimes in quasiballistic heat conduction _ , physical review b 89 , 094302 ( 2014 ) . , _ disparate quasiballistic heat conduction regimes from periodic heat sources on a substrate _ , journal of applied physics 116 , 064307 ( 2014 ) . , _ multidimensional quasiballistic thermal transport in transient grating spectroscopy _ , physical review b 92 , 085203 ( 2015 ) . , _ efficient simulation of multidimensional phonon transport using energy - based variance - reduced monte carlo formulations _ , physical review b 84 , 205331 ( 2011 ) . , _ an alternative approach to efficient simulation of micro / nanoscale phonon transport _ , applied physics letters 101 , 153114 ( 2012 ) . , _ monte carlo methods for solving the boltzmann transport equation _ , annual review of heat transfer 17 , 205 - 265 ( 2014 ) . , _ adjoint - based deviational monte carlo methods for phonon transport calculations _ , physical review b 91 , 235321 ( 2015 ) . , _ non - diffusive relaxation of a transient thermal grating analyzed with the boltzmann transport equation _ , journal of applied physics 114 , 104302 ( 2013 ) . , _ probing the experimental phonon dispersion of graphene using @xmath241c and @xmath242c isotopes _ , physical review b 86 , 085409 ( 2012 ) . , _ raman study of isotope effects and phonon eigenvectors in sic _ , physical review letters 82 , 3089 ( 1999 ) . , _ determination of the phonon dispersion of zinc blende ( 3c ) silicon carbide by inelastic x - ray scattering _ , applied physics letters 80 , 4360 ( 2002 ) . , _ phonon dispersion of carbon nanotubes _ , solid state communications 121 , 471 - 474 ( 2002 ) . , _ phonon dispersion relation in rhodium : ab initio calculations and neutron - scattering investigations _ , physical review b 57 , 324 ( 1998 ) . , _ gallium arsenide thermal conductivity and optical phonon relaxation times from first - principles calculations _ , europhysics letters 101 , 16001 ( 2013 ) . , _ the phonon dispersion of graphite revisited _ , solid state communications 131 , 3 , 141 - 152 ( 2004 ) . , _ ab initio calculations of phonon dispersion relations in aluminum _ , acta physica polonica a 97 , 2 , 349 - 354 ( 2000 ) . , _ spectral phonon transport properties of silicon based on molecular dynamics simulations and lattice dynamics _ , journal of computational and theoretical nanoscience 5 , 1 - 12 ( 2008 ) . , _ on smoothing of non - smooth functions _ , math track 2 , 17 - 21 , ( 2006 ) . , _ analysis of lattice thermal conductivity _ , physical review 132 , 6 , ( 1963 ) . , _ a simplex method for function minimization _ , the computer journal 7 , 308 - 313 , ( 1965 ) . , _ nelder - mead simplex optimization routine for large - scale problems : a distributed memory implementation _ , computational economics 43 , 447 - 461 ( 2014 ) . , _ an introduction to genetic algorithms _ , mit press ( 1998 ) . , _ optimization by simulated annealing _ , science 220 , 4598 , 671 - 680 ( 1983 ) . , _ convergence properties of the nelder - mead simplex method in low dimensions _ , siam journal on optimization 9 , 1 , 112 - 147 ( 1998 ) . , _ numerical recipes in c : the art of scientific computing _ , cambridge university press , cambridge ( 1992 ) . , _ modifications of nelder - mead simplex method for stochastic simulation response optimization _ , proceedings of the winter simulation conference ( 1991 ) . , _ elastic modulus and in - plane thermal diffusivity measurements in thin polyimide films using symmetry - selective real - time impulsive stimulated thermal scattering _ , applied physics a 58 , 523 - 534 ( 1994 ) . , _ exploring electron and phonon transport at the nanoscale for thermoelectric energy conversion _ , phd thesis , massachusetts institute of technology , cambridge , ma ( 2011 ) . , _ the fast fourier transform _ , prentice - hall , new jersey ( 1988 ) . , _ hybrid atomistic - continuum formulations and the moving contact - line problem _ , journal of computational physics 154 , 245 - 265 ( 1999 ) . due to a small discrepancy between the true data of @xcite and the present work ( likely related to discretization ) , the reconstructed cdf , @xmath197 " data was slightly scaled ; at the same time , care was taken to ensure that the discrepancy with the present work true data is the same or smaller than that shown in figure 2 of @xcite . , _ efficient multiscale methods for micro / nanoscale solid state heat transfer _ , phd thesis , massachusetts institute of technology , cambridge , ma ( 2015 ) . , _ extending the range of validity of fourier s law into the kinetic transport regime via asymptotic solution of the phonon boltzmann transport equation _ , physical review b 93 , 045424 ( 2016 ) . , _ phonon transport at crystalline si / ge interfaces : the role of interfacial modes of vibration _ , scientific reports 6 , 23139 ( 2016 ) . , _ table of integrals , series , and products _ , 7th edition , academic press , burlington , ma , ( 2007 ) . the ranges at which each line or polynomial in relationship is active , @xmath67s , are calculated via the following equations @xmath243 where @xmath244 , @xmath245 , and the parameter @xmath246 is the range of frequency at which each polynomial is active which is taken to be @xmath247 in this work . the coefficients of polynomials , @xmath63 , @xmath64 , @xmath65 , and @xmath66 are calculated from the following equations @xmath248 ^ 3 } \left [ \frac{\log \left ( \frac{\tau^{s}_{\omega_{j+1}}}{\tau^{s}_{\omega_{j } } } \right ) } { \log \left ( \frac{\omega^{s}_{j+1}}{\omega^{s}_{j } } \right ) } - \frac{\log \left ( \frac{\tau^{s}_{\omega_{j } } } { \tau^{s}_{\omega_{j-1 } } } \right ) } { \log \left ( \frac{\omega^{s}_{j}}{\omega^{s}_{j-1 } } \right ) } \right ] , \nonumber \\ b^{s}_{j } & = 0.5 \left\ { \frac{1 } { \log \left ( \frac{\omega^{s}_{j}+ \delta } { \omega^{s}_{j}- \delta } \right ) } \left [ \frac{\log \left ( \frac{\tau^{s}_{\omega_{j+1 } } } { \tau^{s}_{\omega_{j } } } \right ) } { \log \left ( \frac{\omega^{s}_{j+1 } } { \omega^{s}_{j } } \right ) } - \frac{\log \left ( \frac{\tau^{s}_{\omega_{j } } } { \tau^{s}_{\omega_{j-1 } } } \right ) } { \log \left ( \frac{\omega^{s}_{j } } { \omega^{s}_{j-1 } } \right ) } \right]- 3 a^{s}_{j } \log \left ( \left ( \omega^{s}_{j } \right)^2- \delta^2 \right ) \right\ } , \nonumber \\ c^{s}_{j } & = \frac{\log \left ( \frac{\tau^{s}_{\omega_{j } } } { \tau^{s}_{\omega_{j-1 } } } \right ) } { \log \left ( \frac{\omega^{s}_{j } } { \omega^{s}_{j-1 } } \right ) } - 3 a^{s}_{j } \left [ \log \left ( \omega^{s}_{j}- \delta \right ) \right]^2- 2 b^{s}_{j } \log \left ( \omega^{s}_{j}- \delta \right ) , \nonumber \\ d^{s}_{j } & = \log ( \tau^{s}_{\omega_{j-1}})- \log ( \omega^{s}_{j-1 } ) \frac{\log \left ( \frac{\tau^{s}_{\omega_{j } } } { \tau^{s}_{\omega_{j-1 } } } \right ) } { \log \left ( \frac{\omega^{s}_{j } } { \omega^{s}_{j-1 } } \right ) } + \nonumber \\ & \hspace{80pt } 0.5 \frac { \left [ \log \left ( \omega^{s}_{j}- \delta \right ) \right]^2 } { \log \left ( \frac{\omega^{s}_{j}+ \delta } { \omega^{s}_{j}- \delta } \right ) } \left [ \frac{\log \left ( \frac{\tau^{s}_{\omega_{j+1 } } } { \tau^{s}_{\omega_{j } } } \right ) } { \log \left ( \frac{\omega^{s}_{j+1 } } { \omega^{s}_{j } } \right ) } - \frac{\log \left ( \frac{\tau^{s}_{\omega_{j } } } { \tau^{s}_{\omega_{j-1 } } } \right ) } { \log \left ( \frac { \omega^{s}_{j}}{\omega^{s}_{j-1 } } \right ) } \right]- \nonumber \\ & \hspace{80pt } 0.5 a^{s}_{j } \left [ \log \left ( \omega^{s}_{j}- \delta \right ) \right]^2 \left [ 3 \log ( \omega^{s}_{j}+ \delta)- \log ( \omega^{s}_{j}- \delta ) \right ] , \label{optim_parameter_2}\end{aligned}\ ] ] where @xmath245 and @xmath49 . here , we derive the semi - analytical frequency - domain solution used in section [ ttg ] using fourier transform in the frequency domain ( with respect to the time variable ) and wavenumber domain ( with respect to the space variable ) . we will use the subscript @xmath200 to denote the former and the subscript @xmath249 to denote the latter . by applying a fourier transform to in both variables we obtain @xmath250 which provides the following relationship for @xmath251 @xmath252 } { i \zeta+ i \eta v_{\omega } \cos(\theta)+ \tau_{\omega}^{-1 } } .\ ] ] applying fourier transform in both variables to equation and replacing @xmath251 with the above result , we obtain the following expression for @xmath253 @xmath254 } { i \tau_{\omega } \zeta+ i \eta v_{\omega } \tau_{\omega } \cos(\theta ) + 1 } d_{\omega } \right ] d \omega d \mathbf{\omega}= 0 , \label{pseudo_fourier}\ ] ] which after integrating over @xmath255 and @xmath256 and algebraic simplification leads to the following relationship @xmath257 equation can be further simplified by applying the inverse fourier transform in wavenumber domain @xmath249 . noting that equation is in the form of @xmath258 , we use the identity @xmath259 to obtain @xmath260 d \omega } , \label{pseudo_final}\ ] ] where @xmath261 is given in . equation is similar to the result obtained in @xcite . to obtain an expression for the temperature , we insert into and apply the inverse fourier transform in wavenumber domain using the identity to obtain @xmath262 .\ ] ] this relation is then substituted in @xmath263 d \omega d \mathbf{\omega}= 0 , \label{temperature_fourier}\ ] ] obtained by applying fourier transform in the frequency domain to . solving the resulting equation for @xmath264 leads to the expression provided in . introducing the definition @xmath265 in equation , provides the following relationship for @xmath261 @xmath266 using a taylor expansion for @xmath199 of the logarithmic term in we obtain @xmath267 , \label{s_omega_new_taylor}\ ] ] which is the same expression provided in equation . if we substitute @xmath261 in with and neglect terms of @xmath268 we obtain @xmath269 d \omega+ \\ \frac { \left [ \int_{\omega } c_{\omega } \left ( \frac{\tan^{-1 } \left ( k\!n_{\omega } \right ) } { k\!n_{\omega } } - \frac{\tau_{\omega } i \zeta } { k\!n_{\omega}^2 + 1 } \right ) d \omega \right]^2 } { \int_{\omega } \frac{c_{\omega } } { \tau_{\omega } } \left [ 1- \frac{\tan^{-1 } \left ( k\!n_{\omega } \right ) } { k\!n_{\omega}}+ \frac{\tau_{\omega } i \zeta } { k\!n_{\omega}^2 + 1 } \right ] d \omega } \bigg\ } . \label{analytical_quasi}\end{gathered}\ ] ] equation is in the general form of @xmath270 where @xmath271 , @xmath272 , ... , @xmath273 are independent of fourier transform variable @xmath200 and depend only on the geometry and the material properties . equation can be further simplified in the following form @xmath274 applying inverse fourier transform to leads to the following result @xmath275 where @xmath276 and @xmath277 are the delta dirac function and its first derivative , respectively , and @xmath278 is the heaviside function . at late times ( more precisely , @xmath279 ) , only the last term of is nonzero . substituting constants @xmath271 through @xmath273 in with their equivalents from equation leads to the result provided in . the temperature relaxation profile in the ballistic limit can be obtained by setting @xmath280 in equation and . using this substitution , for @xmath261 we have @xmath281 substituting this expression in , we observe that the second term of vanishes in this limit ( the numerator is @xmath282 , while the denominator is @xmath283 ) . therefore , we have @xmath284 the inverse fourier transform of can be calculated analytically @xcite and gives @xmath285

the neutrino mass hierarchy can be determined , in principle , by measuring a phase in the disappearance oscillation probability in vacuum , without relying on the matter effect , using a single channel . this phase is not the same for the normal and inverted neutrino mass spectra . in this paper , we give a complete description and physics understanding of the method . the key feature of the method is to detect advancement ( normal ) or retardation ( inverted ) of the phase of the atmospheric - scale oscillation relative to the solar - scale oscillation . we then show that this method can be realized with the recently proposed resonant absorption reaction enhanced by mssbauer effect . the unique feature of this setup is the ultra - monochromaticity of the observed s . because of this feature the phase advancement or retardation of atmospheric - scale neutrino oscillation is detectable after 20 or more oscillations if the source and the target are made sufficiently compact in size . a quantitative estimate of the sensitivity to mass hierarchy resolution is given . we have also examined how a possible continuation of such experiment can be carried out in order also to achieve high precision ( few % ) determination of the solar - scale oscillation parameters and . 1.6 cm

hep-ph0701151

among the remaining unknowns in neutrino physics , resolving the neutrino mass hierarchy may be one of the key importance to our understanding of physics of neutrino masses and the lepton flavor mixing @xcite . given the significance of the problem , it is natural that many methods are proposed for determining the mass hierarchy . the most conventional one would be to do neutrino and antineutrino comparison in long - baseline ( lbl ) accelerator based oscillation experiments @xcite . because of the difference in how the matter effect interferes with the vacuum oscillation effect it can signal the sign of @xmath3 , and hence the mass hierarchy . other less conventional ideas include the methods using atmospheric neutrinos @xcite , @xmath4 and @xmath5 disappearance channels @xcite , supernova neutrinos @xcite , and neutrino - less double beta decay @xcite . these lists of references can not be complete , and some of the other relevant references are quoted therein . in this paper , we discuss yet another way of determining the neutrino mass hierarchy by a neutrino oscillation disappearance experiment using _ a single neutrino flavor _ in vacuum whereas in previous works @xcite the method required the use of two different disappearance channels close to the first oscillation maximum . the possibility that the neutrino survival probability in vacuum can be sensitive to the neutrino mass hierarchy was first suggested in @xcite using reactor neutrinos in the context of high-@xmath6 large mixing angle ( lma ) solar neutrino solution . it was also pursued recently in the hanohano project using a fourier transform technique @xcite . we present here a complete formulation and physics understanding of the new method to determine the neutrino mass hierarchy . this method utilizes the feature that the atmospheric-@xmath7 scale oscillation `` interferes '' with the solar-@xmath7 scale oscillation in a different way depending upon the mass hierarchy . we first carefully re - formulate the difference as an advancement or retardation of the phase of the atmospheric - scale neutrino oscillation relative to the solar - scale oscillation ; for the @xmath8 and @xmath9 disappearance probabilities , the atmospheric oscillation is continuously advanced ( retarded ) for the normal ( inverted ) hierarchy . since the core of the problem exists in extracting long wavelength solar oscillation out of the two short wavelength atmospheric scale oscillations due to @xmath10 and @xmath3 , it is the critical matter to identify which quantity to be held fixed in defining the mass hierarchy reversal from the normal to the inverted , or vice versa . to our understanding , the only known appropriate quantity to hold is the effective neutrino mass squared difference @xmath11 ( see eq.([eqn : dmsqee ] ) ) , the one which is actually observed in @xmath4 ( or @xmath0 ) disappearance measurement at the first few oscillation maxima and/or minima @xcite . we find it important to have the proper formalism because otherwise the sensitivity to the mass hierarchy resolution can be significantly overestimated . then , we point out that an experiment using the resonant @xmath0 absorption reaction @xcite enhanced by mssbauer effect recently proposed by raghavan @xcite gives the first concrete realization of the idea . ( see @xcite for earlier proposals of mssbauer enhanced experiments . ) because of the monochromaticity of the @xmath0 beam from the bound state beta decay @xcite , the phase advancement or retardation of atmospheric - scale neutrino oscillation is detectable after 20 or more oscillations if the source and the target are made sufficiently compact in size . the unique feature of this method is the use of this ultra - monochromatic neutrino source thereby sidestepping the very high energy resolution requirement requested with reactor or accelerator neutrinos . a concrete plan for the experiment may be as follows ; in the first stage the atmospheric @xmath11 must be accurately measured in a way described in @xcite by a measurement at @xmath1210 m baseline . as we will see , the precision of the measurement must reach sub - percent level in order for this method to work . then , in the second stage relative phase advancement or retardation of the atmospheric - scale neutrino oscillation must be detected by moving the detector to @xmath12350 m , this is just beyond the solar - scale oscillation distance for 18.6 kev neutrinos . based on this plan we give a detailed estimate of the sensitivity to the resolution of the mass hierarchy . upon finishing the above measurement , it is natural to think about the continuation of the experiment by moving the detector to distances tuned for precision measurement of the oscillation parameters relevant for solar neutrinos , @xmath1 and @xmath2 . we also give a detailed estimate of the accuracy for the determination of these parameters . it is notable that a resonant @xmath13 absorption experiment could allow to make precise measurements of _ all mixing parameters _ that appear in the @xmath14 oscillation channel : @xmath15 , @xmath2 , @xmath1 , the effective atmospheric @xmath16 and the neutrino mass hierarchy . in sec . [ sec : how ] , we first describe the essence of the method in an explanatory way . we then define our theoretical machinery using the effective @xmath11 . we also clarify the basic requirement for the measurement . in sec . [ sec : exp ] , we discuss the concrete set up of the experiment and describe our statistical method for the analysis . we then give the analysis results on sensitivity to the mass hierarchy resolution . in sec . [ sec : solar ] , we describe the analysis for the determination of the solar oscillation parameters . in sec . [ sec : conc ] , we finish by giving our concluding remarks . in appendix [ appendixa ] , we give a different interpretation of the phenomena in terms of the effective mass squared differences whereas in appendix [ appendixb ] , we discuss how this method can be applied to the @xmath17 disappearance channel . this possibility has been discussed in @xcite however the baseline required is @xmath18 km . the vacuum @xmath8 survival probability using @xmath19 ( @xmath20 ) as shorthand for the kinematical phase , can be written without any approximation as @xmath21 - p_{\odot } , \label{pee}\end{aligned}\ ] ] where the usual parameterization of the mixing matrix @xcite is used with the abbreviations @xmath22 and @xmath23 . here @xmath24 is the neutrino energy and @xmath25 is the baseline distance . @xmath26 , as defined by , @xmath27 is the oscillation probability associated with the solar @xmath7 scale . in this paper , we assume the standard three neutrino framework with cpt conservation ( except for where we discuss possible cpt violation ) , which implies that eq . ( [ pee ] ) is also valid for @xmath28 . 0.2 cm is plotted as a function of @xmath25 . the blue and the red curves are for the normal and the inverted mass hierarchies , respectively . in the right panel , @xmath29 , which is defined in the text , is shown . here the neutrino energy was fixed to 18.6 kev . this choice comes from the experiment we will discuss later in this paper.,title="fig:",scaledwidth=49.0% ] is plotted as a function of @xmath25 . the blue and the red curves are for the normal and the inverted mass hierarchies , respectively . in the right panel , @xmath29 , which is defined in the text , is shown . here the neutrino energy was fixed to 18.6 kev . this choice comes from the experiment we will discuss later in this paper.,title="fig:",scaledwidth=49.0% ] -0.8 cm the neutrino mass hierarchy can be determined if one knows which kinematical phase , @xmath30 or @xmath31 , is larger . namely , if @xmath32 the normal hierarchy is the case , and if @xmath33 the inverted one . how we can know which is larger ? when two waves with slight different frequencies travel together there arises the phenomenon of beating . the amplitude of superposed waves beats with the frequency corresponding to the difference of two frequencies . for @xmath8 disappearance probability , eq . ( [ pee ] ) , we have beating because the frequency of the two waves in ( [ pee ] ) with @xmath30 and @xmath31 differ by the frequency associated with the @xmath34 characterized by the solar @xmath1 scale , which is smaller by a factor of @xmath35 compared to the atmospheric @xmath36 scale . one can decompose the superposed waves into the beating low frequency wave and the high frequency wiggles within the beat . here , unlike the case addressed in most textbooks of basic physics , the amplitude of the two frequencies is different . therefore , to tell which of the high frequencies has a larger amplitude , one has to determine whether the phase of the high frequency wiggles advances or retards as the wave evolves . this phase advancement or retardation can tell us whether the neutrino mass hierarchy is normal or inverted using only the disappearance measurement in a single channel , @xmath37 in vacuum . to demonstrate that it is possible , we plot in the left panel of fig . [ rio - sp ] the oscillation probability @xmath38 as a function of the distance @xmath25 from a source . as one can clearly see in the figure , the case of normal hierarchy ( @xmath39 , blue curve ) starts to be distinguished from the case of inverted hierarchy ( @xmath40 , red curve ) at around @xmath41 m because of the advancement of the normal hierarchy s high frequency wiggles and the retardation of the inverted hierarchy s high frequency wiggles , i.e. the phase of the high frequency wiggles . what we mean by `` beating '' may be clearer in the left panel of fig . [ fig : lovere ] where the solar term is switched off . to see the difference between the oscillation probabilities of the normal and the inverted hierarchies more clearly , we define the quantity @xmath42 where the hierarchy flip is defined so that @xmath43 is minimized for short baseline measurements . ( see sec . [ subsec : theo ] for the precise definition of the hierarchy flip . ) we normalize @xmath44 by dividing by @xmath45 since @xmath46 is the `` background '' for seeing @xmath15 effects in @xmath8 disappearance . the quantity @xmath47 is plotted in the right panel of fig . [ rio - sp ] . the envelope of @xmath48 slowly rises from 0 to a plateau value of @xmath49 at a distance about 25% further from the source than the first solar oscillation maximum ( @xmath50 occurs when @xmath51 m ) and remains reasonably constant at this value beyond the first solar oscillation minimum ( @xmath52 ) . clearly , even though the difference between the hierarchies is very small at short distances it becomes as large as theoretically possible , @xmath49 ( see below ) , at large distances . to decompose the superposed waves into the beat and wiggle , one has to correctly define the high frequency mode . we choose to do this by defining a short distance effective @xmath7 in the @xmath8 disappearance channel , see @xcite . it is given by @xmath53 this is the only linear combination of @xmath54 and @xmath55 that can be determined by a short distance ( over the first few atmospheric oscillations ) @xmath8 disappearance experiment . @xmath11 is defined to be positive for both hierarchies . the choice of the atmospheric @xmath7 is crucial ; extraction of the low frequency mode can be done correctly only if the high frequency mode is identified in an appropriate way . see sec . [ reversal ] for more about it . now , we rewrite @xmath56 without any approximation , using the positive and measurable variables @xmath57 and @xmath58 and the phase @xmath59 ( which will be given below ) , as follows : @xmath60 - p_{\odot } , \label{eqn : pdis}\end{aligned}\ ] ] where the positive ( negative ) _ sign _ in front of the phase @xmath59 corresponds to the normal ( inverted ) mass hierarchy ; if @xmath61 is precisely determined , the only effect of the hierarchy manifest itself as this _ sign_. clearly , the amplitude of the @xmath49 oscillations is modulated by @xmath62 which varies from 1 to @xmath63 as the solar kinematic phase , @xmath58 , changes from @xmath64 to @xmath65 . this amplitude modulation is the beat discussed in the previous section . the phase @xmath59 in eq . ( [ eqn : pdis ] ) depends only on solar parameters , @xmath58 and @xmath2 , and given as , @xmath66 or , alternatively @xmath67 however , eq . ( [ eqn : phi ] ) contains quadrant information not available in eq . ( [ eqn : phi2 ] ) . in fig . [ fig : phi ] , we show @xmath59 as a function of @xmath58 . note that it is an odd , monotonically increasing function of @xmath58 . -0.5 cm as function of @xmath58 . @xmath68 when @xmath69.(b ) @xmath59 as function of @xmath25 for various choices of the solar parameters within the current allowed region . @xmath70 at about 350 m somewhat independent of the precise values for the solar parameters.,title="fig:",scaledwidth=46.0% ] as function of @xmath58 . @xmath68 when @xmath69.(b ) @xmath59 as function of @xmath25 for various choices of the solar parameters within the current allowed region . @xmath70 at about 350 m somewhat independent of the precise values for the solar parameters.,title="fig:",scaledwidth=46.0% ] -1.0 cm due to the monotonically increasing nature of @xmath59 and the @xmath71 sign in front of @xmath59 , in eq . ( [ eqn : pdis ] ) , for the normal hierarchy the high frequency wiggles undergo a phase advancement as the wave evolves , whereas for the inverted hierarchy there is a phase retardation . this interpretation is explored in appendix a. ] . it is easy to show that @xmath72 i.e. the phase advancement ( normal ) or retardation ( inverted ) is @xmath73 for every @xmath74 increase of @xmath58 . ( [ eqn : pdis ] ) and ( [ eqn : phi ] ) are the foundations of our investigation . some important observations are worth emphasizing immediately : * only @xmath75 varies at the atmospheric scale . everything else , including phase @xmath59 , varies at the solar scale . this is a useful distinction because these two scales differ by a factor of @xmath76 . * the difference between probabilities corresponding to the two hierarchies ( [ deltap ] ) is given by @xmath77 to leading order in @xmath15 . @xmath78 becomes visible when the phase @xmath59 develops to order unity . from fig . [ fig : phi ] this occurs at around the first solar oscillation maximum , ( @xmath79 ) , in agreement with the features exhibited in fig . [ rio - sp ] . from ( [ eq : pdif ] ) we also see that the normal and the inverted hierarchies are maximally out of phase when @xmath80 . this occurs when @xmath81 the approximate solution to this equation , for @xmath2 around 30 degrees , is @xmath82 . for @xmath83 , eq . ( [ eqn : phieqpiover2 ] ) is satisfied when @xmath84 , i.e. , just beyond the first solar oscillation maximum . * we notice that by doing transformation @xmath85 ( or @xmath86 ) in ( [ eqn : phi ] ) flips the sign of @xmath87 . it means that the mass hierarchy is confused by the presence of the particular type of cpt violation , the subtle one discussed in @xcite . it occurs because we use the antineutrino channel ; if we are able to use the neutrino channel , the problem of confusion does not arise because we know that @xmath88 from the solar neutrino observation . we now describe our experimental strategy for determining the hierarchy . first one needs to make a very precise measurement of @xmath89 at short distances from the source , at around half the atmospheric oscillation length , @xmath90 where @xmath91 \left [ \frac { \delta m^2_{\text{ee } } } { 2.5 \times 10^{-3 } \text{ev}^2 } \right]^{-1 } \text{m}\ , .\end{aligned}\ ] ] then , one must determine the sign in front of @xmath59 where the phase difference between normal and inverted hierarchy is maximal , i.e. @xmath92 as can be seen in fig . [ fig : phi ] . with @xmath93 , it can be translated into the distance @xmath94 l_0 \simeq 20 \ , l_0 $ ] away from the source . how well do we need to know @xmath11 ? if we do not know the value of @xmath11 with sufficient precision , one can adjust @xmath11 , depending upon the normal ( nh ) and the inverted ( ih ) hierarchies , so that the relation @xmath95 holds at some @xmath96 . then , from eq . ( [ eqn : pdis ] ) the hierarchy is completely confused as we will explicitly see in sec . [ setup ] . it can be translated into @xmath97 which leads to the requirement for the accuracy of measurement of @xmath11 as @xmath98% . this is the critical value at which the hierarchy is completely confused , and if we want to do better , e.g. , at least a factor of 3 times better than 3% , the first requirement is @xmath991% measurement of @xmath11 . at this point the astute reader will be concerned that we are attempting to determine a oscillation phase after 20 or so oscillations . are nt the atmospheric oscillations completely washed out long before 20 @xmath100 ? to observe atmospheric oscillations 20 or so oscillations out , assuming coherence , demands that we can determine @xmath96 per event to high precision . figure [ fig : lovere ] shows how the oscillations are reduced as the resolution on @xmath96 is increased and that a precision better than 1% is required to see oscillation at @xmath12 20 oscillation lengths from the source . -0.5 cm versus distance measured in number of oscillation lengths , @xmath100 . to help with visualization @xmath101 has been set to zero for these figures . ( a ) 0.1% gaussian resolution on @xmath96 . ( b ) 0.5% resolution on @xmath96 . ( c ) 2% resolution on @xmath96 . in ( b ) the oscillations are reduced at 20 or so oscillations out , but still observable . whereas in ( c ) the oscillations are averaged out by 15 @xmath100 . , title="fig:",scaledwidth=33.0% ] versus distance measured in number of oscillation lengths , @xmath100 . to help with visualization @xmath101 has been set to zero for these figures . ( a ) 0.1% gaussian resolution on @xmath96 . ( b ) 0.5% resolution on @xmath96 . ( c ) 2% resolution on @xmath96 . in ( b ) the oscillations are reduced at 20 or so oscillations out , but still observable . whereas in ( c ) the oscillations are averaged out by 15 @xmath100 . , title="fig:",scaledwidth=33.0% ] versus distance measured in number of oscillation lengths , @xmath100 . to help with visualization @xmath101 has been set to zero for these figures . ( a ) 0.1% gaussian resolution on @xmath96 . ( b ) 0.5% resolution on @xmath96 . ( c ) 2% resolution on @xmath96 . in ( b ) the oscillations are reduced at 20 or so oscillations out , but still observable . whereas in ( c ) the oscillations are averaged out by 15 @xmath100 . , title="fig:",scaledwidth=33.0% ] -0.5 cm in summary , to measure the sign of the phase and determine the hierarchy we have _ two _ important initial requirements : * the resolution on @xmath102 measured in a short baseline experiment ( @xmath103 from the source ) must be @xmath991% or less . * the resolution on @xmath96 for the long baseline experiment ( @xmath12 20 @xmath100 from the source ) must be @xmath991% or less . usually , the neutrino energy can not be determined experimentally with this accuracy , therefore , these preconditions are very difficult to meet , if not impossible . we have emphasized in the foregoing discussions the importance of identifying the quantity to be held fixed to define the mass hierarchy reversal , and proposed @xmath11 as the solution . in this subsection we want to elaborate this point and clarify how the difference between the normal and the inverted hierarchies can be made artificially larger by choosing different variables to define the hierarchy reversal . in fig . [ fig : mh - reversal ] , presented is the survival probability of electron anti - neutrinos at a baseline of 50 km from a source which is averaged over the uncertainty of energy 3%/@xmath104 as a function of neutrino energy @xmath24 . it is shown in the left panel of fig . [ fig : mh - reversal ] that if we hold @xmath105 fixed in reversing the hierarchy from the normal to the inverted ( as was done in @xcite ) the difference between the normal and the inverted hierarchies is clearly visible . however , the obvious distinction seen in the left panel of fig . [ fig : mh - reversal ] disappears if we use @xmath89 to hold when switching between the hierarchies as shown in the middle panel of fig . [ fig : mh - reversal ] . a simple explanation for such a marked difference is that by holding @xmath106 fixed in reversing the hierarchy from the normal to the inverted one maps the neutrino mass spectrum into a significantly different one , @xmath107 . ( see appendix [ appendixa ] for more about it . ) this artificially enhances the difference between the hierarchies , as demonstrated in fig . [ fig : mh - reversal ] . -0.5 cm disappearance probability at 50 km from a reactor source as a function of the neutrino energy . the resolution on the energy is assumed to be 3%/@xmath108 . the blue ( light gray ) and the red ( dark gray ) curves are for the normal and the inverted mass hierarchies , respectively . in the left panel @xmath109 is held fixed when the hierarchy is flipped , whereas in the middle panel @xmath11 is held fixed . in the right panel @xmath11 for the inverted hierarchy is 0.6% larger then for the normal hierarchy . , title="fig:",scaledwidth=33.0% ] disappearance probability at 50 km from a reactor source as a function of the neutrino energy . the resolution on the energy is assumed to be 3%/@xmath108 . the blue ( light gray ) and the red ( dark gray ) curves are for the normal and the inverted mass hierarchies , respectively . in the left panel @xmath109 is held fixed when the hierarchy is flipped , whereas in the middle panel @xmath11 is held fixed . in the right panel @xmath11 for the inverted hierarchy is 0.6% larger then for the normal hierarchy . , title="fig:",scaledwidth=33.0% ] disappearance probability at 50 km from a reactor source as a function of the neutrino energy . the resolution on the energy is assumed to be 3%/@xmath108 . the blue ( light gray ) and the red ( dark gray ) curves are for the normal and the inverted mass hierarchies , respectively . in the left panel @xmath109 is held fixed when the hierarchy is flipped , whereas in the middle panel @xmath11 is held fixed . in the right panel @xmath11 for the inverted hierarchy is 0.6% larger then for the normal hierarchy . , title="fig:",scaledwidth=33.0% ] -0.5 cm despite that there exist some discrepancies at energies around 4 mev , a peak energy of signal events in reactor experiments , the two disappearance probabilities can be made very close to each other over the energy range of 4 to 6 mev if we allow the value of @xmath89 for the inverted hierarchy to increase by 0.6% relative to the value for the normal hierarchy , as can be seen in the right panel of fig . [ fig : mh - reversal ] . the lesson we have learned from this exercise is that unless we know @xmath11 to a sub - percent level , the hierarchy can be easily confused by a tiny shift of the atmospheric @xmath7 . one might want to use neutrinos from reactors to determine the mass hierarchy by the present method , as suggested by the authors of @xcite . however , the experimental requirements are very severe in order for this method to work . we believe that the three points discussed above deserve attention if one want to make a reliable estimate of sensitivity to the mass hierarchy resolution with reactor neutrinos ; ( 1 ) definition of hierarchy reversal , ( 2 ) implementation of realistic energy resolution , and ( 3 ) robustness of the sensitivity to mass hierarchy resolution against the uncertainty of @xmath110 . the recent hanohano proposal , @xcite , sidesteps some of these requirements by using a sophisticated fourier transform technique . detailed analysis around the atmospheric peak in fourier space can tells us whether @xmath111 is greater or smaller than@xmath112 , thus determining the hierarchy . there is no need to explicitly deal with the definition of hierarchy reversal . in this way , the two curves shown in the two right panels of fig . [ fig : mh - reversal ] can , in principle , be distinguished . if this can be achieved in a realistic setting of the experiment , it represents a remarkable advance for reactor neutrino experiments , demonstrating the power of the fourier transform technique . most probably , the unique possibility to satisfy both requirements of the previous section is to make use of monochromatic beam from the bound - state beta decay @xmath113 and detect it by the time - reversed resonant absorption reaction @xmath114 , the idea recently revived by raghavan @xcite . for the mssbauer enhancement to work the method requires @xmath115 better than @xmath116 , a truly monochromatic neutrino beam of 18.6 kev . thus , the energy spread of the beam in such an experiment is automatically much narrower than what is required . having monochromaticity satisfied , the remaining requirement is on the baseline @xmath25 . the short and long baseline measurement are to be done at the distances of @xmath1210 m and @xmath117 m from the source , respectively . then , the requirement on @xmath118 of , say , 0.3% implies that the size of the source and the target must be limited respectively to @xmath123 cm and @xmath121 m in short and long baseline measurement . the former places severe limits on the sizes of the source and the detector . assuming this requirement can be met , it has been shown by analyzes with several concrete settings that a sensitivity to @xmath11 of @xmath119% at 1@xmath120 cl is possible @xcite . therefore , the first requirement listed in sec . [ subsec : req ] can be met if @xmath121 0.05 . we consider a possible experimental set up for determination of the neutrino mass hierarchy based on a mssbauer enhanced @xmath28 absorption experiment @xcite . by using the revised estimate of the cross section @xmath122 @xcite the event rate is given by @xmath123 therefore , one obtains about one million ( 2500 ) events per 3 days ( 10 days ) for 1 mci source and 100 g @xmath124he target at a baseline distance @xmath125 ( 350 ) m. the experiment should be done in two stages ; in the first stage , @xmath11 and @xmath49 should be determined very precisely by a measurement at baseline around @xmath1210 m as demonstrated in @xcite . we thus do not discuss it further by just assuming the setting with 10 location measurement called the run iib in @xcite . for the second stage , we examine the setting with detectors placed at five locations @xmath126 m from the source , which covers a complete atmospheric oscillation length . the results of these two stages should be combined as we will discuss bellow . -0.20 cm -0.5 cm ev@xmath127 and @xmath128 for the normal hierarchy , indicated by solid circles with error bars . here @xmath129 is included ( solar parameters are fixed to their best fit values ) . expected number of events in the absence of oscillation is assumed to be 2000 at each detector position . the blue ( light gray ) solid curve which passes through the data points indicates a theoretical expectation assuming the normal hierarchy with @xmath130 ev@xmath127 and @xmath128 whereas the red ( dark gray ) curve off the data points is the one of inverted hierarchy with same values of @xmath11 and @xmath15 . assuming that @xmath11 is known to @xmath131% precision , the thick - dotted ( thin - dotted ) lines above and below the solid lines are the expectations with @xmath11 0.5% above ( below ) that value . , title="fig:",scaledwidth=56.0% ] -0.6 cm in fig . [ 5point ] we present the expected number of events for @xmath130 ev@xmath127 and @xmath132 at 5 locations mentioned above for the normal hierarchy , for 2000 events at each detector location in the absence of oscillation , which correspond to an exposure of about 20 mci@xmath133g@xmath133day . , one may think that measurement at least 3 different baseline distances would be enough to detect relative phases of the beat and the wiggle around it . though it is true , it turned out that the five point measurement can achieve better accuracy than the three point measurement for an equal total number of events . ] throughout our analysis except for that in sec . [ sec : solar ] , we fix the solar parameters to their best fit values as @xmath134 and @xmath135 ev@xmath127 and do not try to vary them . @xmath11 is taken with a tentative expectations for its accuracy , @xmath136 % ) ev@xmath127 . the case of the normal and the inverted hierarchy are plotted by the blue ( light gray ) and the red ( dark gray ) curves , respectively . it appears that , if enough statistics is accumulated and @xmath11 is separately measured accurately , the five point measurement , can discriminate between the normal and the inverted hierarchies . the details of the statistical procedure is described in the next subsection . we show in three panels in fig . [ allowed - region ] the allowed regions in the @xmath137 space obtained by the first stage alone ( upper panels ) , by the second stage alone ( middle panels ) , and the one obtained when the first and the second stages are combined ( lower panels ) . the input parameters are taken as @xmath128 , @xmath138 ev@xmath127 , and the mass hierarchy is assumed to be the normal one . in the first - stage measurement to be done at around the distance @xmath139 m from the source , @xmath49 as well as @xmath11 can be determined rather precisely . but , there is no sensitivity to the mass hierarchy because the baseline is too short , as is the case of reactor @xmath15 experiment . -0.5 cm plane obtained by the first stage alone ( upper panels ) , the second stage alone ( middle panels ) and the combined result of these two stages ( lower panels ) for the case where the input parameters are @xmath128 , @xmath140 ev@xmath127 with the normal hierarchy , based on the @xmath141 analysis described in sec . [ method ] . the solar parameters are fixed to their current best fit values . the first stage is assumed to be performed at 10 positions at around @xmath142 m following the run iib in @xcite whereas the second stage is assumed to be performed at 5 locations at @xmath143 m from the source . the three closed curves ( red , blue , green ) from inner to outer denote the ones obtained at 90% cl , 95% cl , and 99 % cl . , title="fig:",scaledwidth=70.0% ] -0.5 cm the feature of the second - stage experiment presented in the middle panels of fig . [ allowed - region ] requires some explanations . first of all , there exist many isolated island - like allowed regions . suppose that @xmath15 is given from some other experiments . then , due to the cosine term in eq . ( [ eqn : pdis ] ) the measurement will give a periodic solution @xmath144 for both of the hierarchies . if we take approximation @xmath145 , there is a shift @xmath146 between the adjacent @xmath11 solutions for the normal and the inverted hierarchies . hence , the solutions of @xmath89 are alternating in the normal and the inverted hierarchies with a constant shift @xmath146 between the adjacent solutions , the feature clearly visible in the middle panels of fig . [ allowed - region ] . the contours have finite widths , though prolonged , along the @xmath49 direction because the second stage measurement alone can determine @xmath15 but only with a limited accuracy . it s clear that the second stage alone can not determine neither the value of @xmath11 nor the mass hierarchy . however , once the results of the first and the second stages are combined all the fake solutions are eliminated , as demonstrated in the lower panels of fig . [ allowed - region ] . it is evident that @xmath89 must be determined with high accuracy in the first stage measurement . now we give details of our analysis method . to calculate the sensitivity of the determination of the neutrino mass hierarchy by the first and second stages of the experiment combined , we compute @xmath147 where @xmath148 is the minimum of @xmath149 with the first stage @xmath141 given by @xmath150\ , ( v^{-1}_{\text{1}})_{ij}\ , \left [ \frac{n_j^{\text{1-obs}}-n_{j}^{\text{1-theo}}}{n_j^{\text{1-theo}}}\right ] \ , , \ ] ] and the second stage @xmath141 given by @xmath151\ , ( v^{-1}_{\text{2-obs}})_{ij}\ , \left [ \frac{(n_j^{\text{2}}-n_{j}^{\text{2-theo}})}{n_j^{\text{2-theo}}}\right].\ ] ] @xmath152 is the number of events at baseline @xmath153 for a given input hierarchy and values of the mixing parameters ( observed events ) , while @xmath154 is the theoretically expected number of events at baseline @xmath153 for a given set of oscillation parameters . we use , as in ref . @xcite , the correlation matrices @xmath155},\ ] ] where @xmath156 . the estimation of the uncorrelated systematic uncertainty is the key issue in our sensitivity analysis . in @xcite two cases of the error , @xmath157 % and 1% , are examined . the former corresponds to the case of movable detector , assuming that a direct extraction or real - time detection of produced @xmath124h atom is possible . the latter is a tentative estimate for the case of multiple identical detectors . we use the correlated systematic uncertainty @xmath158 % , though the results are insensitive to @xmath159 . we adopt the same numbers as these for the uncertainties in our analysis for both the first and the second stages of the experiment . in our @xmath141 analysis , for simplicity , the solar mixing parameters are fixed to the current best fit values , ignoring their uncertainties . therefore , the fitting parameters are : @xmath11 , @xmath49 and the sign of @xmath3 . note , however , that @xmath160 for the first stage has practically no dependence on solar parameters as well as the mass hierarchy due to too short baseline . the current uncertainty on the values of the solar parameter are not so relevant for the determination for the hierarchy because @xmath59 does not depend strongly on them ( see fig . [ fig : phi]b ) as pointed out previously . -0.2 cm event number ( per detector ) space . the black solid , the red dashed , and the blue dotted curves denote the region boundary at 90% , 95% , and 99% cl , respectively . the left and right panels are for the cases of uncorrelated systematic uncertainties of 0.2% and 1% , respectively . , title="fig:",scaledwidth=52.0% ] -0.8 cm event number ( per detector ) space . the black solid , the red dashed , and the blue dotted curves denote the region boundary at 90% , 95% , and 99% cl , respectively . the left and right panels are for the cases of uncorrelated systematic uncertainties of 0.2% and 1% , respectively . , title="fig:",scaledwidth=52.0% ] -0.6 cm now , we present our estimation of the sensitivity of the two - stage mssbauer @xmath13 experiment to the determination of the neutrino mass hierarchy . in fig . [ fig : sensitivity ] we present the region of sensitivity to resolving the mass hierarchy at 90% , 95% , and 99% cl , denoted by the black solid , the red dashed , and the blue dotted curves , respectively . the abscissa is @xmath49 and the ordinate is the number of events per detector in long baseline experiment . here , 90% , 95% , and 99% cl correspond to @xmath161 = 2.71 , 3.84 and 6.63 , respectively . left and right panels are for the cases where the uncorrelated systematic uncertainty is 0.2% and 1% , respectively . assuming @xmath157 % , we can see that the hierarchy can be determined at 90 ( 99 ) % cl for @xmath49 down to 0.035 ( 0.06 ) for 1000 events and down to 0.025 ( 0.04 ) for 2000 events . for a larger error @xmath162% , the critical value of @xmath49 will be @xmath12 0.08 ( 0.11 ) at 90 ( 99 ) % cl approximately independent of the number of events larger than @xmath12 1000 . this is caused by the fact that the precision on the determination of @xmath11 at the first stage is highly dependent on this systematic uncertainty @xcite . in the second stage of the mssbauer experiment , the detectors are placed just after the first solar oscillation maximum in order to determine the mass hierarchy , as we discussed in the previous section . in a possible third stage of the experiment , one could envisage moving the detectors somewhat closer to the source in order to cover the region around this maximum . it will allow us a precise determination of the solar - scale oscillation parameters , @xmath1 and @xmath2 . in order to optimize the determination of these parameters , we assume that the measurements will be performed at the following 10 different detector locations , @xmath163 \ ; { \rm m } , \label{location - solar1 } \\ l'_n & = & l_n - l_0/2,\quad \quad n=1,2, .. ,5 . \label{location - solar2}\end{aligned}\ ] ] first , in order to observe the oscillation driven by the solar @xmath164 , we consider the configuration of 5 detector locations ( @xmath165 ) separated by 50 m , ranging from 200 m to 400 m as in eq . ( [ location - solar1 ] ) . in this way , we can cover the whole range of solar - scale oscillation before and after the dip due to the oscillation maximum ( see fig . [ rio - sp ] ) . second , in order to minimize the unwanted effect due to the rapid oscillations driven by the atmospheric @xmath166 , we have to place , for each location @xmath165 , another detector ( or move detector to another location ) at @xmath167 separated from @xmath165 by half the atmospheric oscillation length , @xmath90 . ( [ location - solar2 ] ) . the setting of five pairs of detector locations works because the average of the oscillation probabilities corresponding to these two positions @xmath165 and @xmath167 is , at first approximation , independent of @xmath15 and @xmath166 and thus is free from any ambiguity associated with the atmospheric oscillations . in this calculation we assume that the exposure is such that the same number of events ( 1000 ) in the absence of oscillation are collected at each detector location . we take the pessimistic value of the systematic uncertainty @xmath168 1% instead of 0.2 % because the results do not differ much ; the sensitivity is still dominated by the statistics in this regime . as in the previous sections , the input values of the other oscillation parameters are fixed as @xmath169 ev@xmath127 , @xmath134 and @xmath135 ev@xmath127 . -0.3 cm for 1000 events taken at each detector position at 68.27% ( inner curves ) , 95.0% ( middle curves ) and 99.73% ( outer curves ) cl for 2 dof . we assumed the measurement will be done at 10 different detector locations , @xmath170 $ ] m and @xmath171 with @xmath172 . the input values used are : @xmath173 , @xmath174 ev@xmath127 , @xmath169 ev@xmath127 . the green and black curves are for the input values of @xmath175 and @xmath176 , respectively . the other curves are also for @xmath177 but fitted with : the wrong hierarchy ( red ) , @xmath11 shifted by 4% ( magenta ) and the wrong hierarchy and @xmath178 shifted by 4% ( cyan ) . here we have supposed @xmath179 1% . , title="fig:",scaledwidth=60.0% ] -0.4 cm in fig . [ fig : solar ] we show how well @xmath1 and @xmath180 can be determined by this experiment . the obtained sensitivity contours with @xmath181 are presented by the green lines ( which practically overlapped with the black ones ) in fig . [ fig : solar ] . from this result , we conclude that the precision one can achieve for @xmath180 is about 1.6 ( 4.0 ) % at 68.27 ( 99.73 ) % cl for 2 dof whereas that for @xmath1 is about 0.9 ( 2.2 ) % at 68.27 ( 99.73 ) % cl for 2 dof . comparing this result with the sensitivity which can be achieved by the dedicated reactor experiment for the similar number of events studied in ref . @xcite , the precision we obtained for @xmath180 and @xmath1 in this work are comparable or better , in particular , for @xmath1 ( see e.g. fig . 5 of the first article in @xcite ) . -0.5 cm ( upper panel ) and @xmath1 ( lower panel ) as a function of detector exposure at 68.27% and 99.73% cl for 1 dof . here the dashed lines show the input values . , title="fig:",scaledwidth=66.0% ] -0.8 cm in the above argument , we have assumed that @xmath100 ( or @xmath166 ) can be determined very precisely in the first stage of the experiment allowing us to place the detector pair separated exactly by @xmath90 . in reality , however , the determination of @xmath100 will always carry some uncertainties . moreover , if @xmath49 is too small or the uncorrelated systematic uncertainty ( @xmath183 ) is not small enough , the neutrino mass hierarchy will not be determined in the second stage . then , the question is ; what is the additional ambiguities that occur in determination of the solar oscillation parameters in the presence of these uncertainties ? fortunately , we can show that our choice of detector locations defined in eqs . ( [ location - solar1 ] ) and ( [ location - solar2 ] ) guarantees that these uncertainties will not deteriorate the precision in the determination of the solar parameters . in order to check the stability of the results against the uncertainties mentioned above , we performed a fit by assuming the incorrect mass hierarchy and/or slightly incorrect value of @xmath11 by shifting its value by 4% . the results are shown by the blue , the magenta and the light blue curves in fig . [ fig : solar ] where we can see that for each cl , all the curves are very close to each other , indicating that our results are quite stable against these uncertainties . we also examine the stability of the sensitivities to possible nonzero values of @xmath15 . in fig . [ fig : solar ] we show the sensitivity contours for the input values of @xmath184 ( black curve ) in comparison to that with @xmath185 ( green curve ) . we note that these 2 curves are very close to each other implying that the effect of nonzero @xmath15 is quite small . we emphasize that this feature is in a marked contrast with any of the alternative methods for measuring @xmath2 , including the reactor @xcite and the solar neutrino methods @xcite . in these methods there is no way , unless determined by some other experiments , to control the errors which comes from nonzero @xmath15 @xcite , whereas in our present method , there is a built - in mechanism which eliminates the uncertainties that come from the atmospheric oscillation parameters , in particular from @xmath15 . in fig . [ fig : dms12 ] we show how the precision for 1 dof of each of the solar parameters improves with statistics , that is , as a function of the number of events that would be collected at each detector in the absence of oscillation . as in the previous plot , we have assumed that @xmath168 1% . we observe that until @xmath186 events , the precision improves rapidly as the number of events increase and then after , the rapidness of the improvement become smaller . the ultimate precision which can be achieved in the limit of very large statistics with @xmath187 events ( per detector ) for @xmath168 1% would be @xmath12 0.4 ( 1.2 ) % for 68.27% ( 99.73% ) cl for @xmath188 and @xmath12 0.2 ( 0.7 ) % for 68.27% ( 99.73% ) cl for @xmath1 . notice that measurement of @xmath7 to the accuracy of sub % level is usually not possible because the absolute energy scale of leptons is hard to be determined at such accuracy , and it acts as a limiting factor for reducing the @xmath7 error @xcite . in our case , monochromaticity of the @xmath0 beam with known energy removes the necessity for such calibration for determining absolute energy scale . in this paper , we have shown that the neutrino mass hierarchy can be determined only by _ a single experiment _ based on resonant @xmath28 absorption reaction enhanced by the mssbauer effect . in its first stage , @xmath11 must be accurately determined by @xmath18910 m measurements . in the second stage of the experiment , the relative phase of the atmospheric oscillation ( wiggles ) between the normal vs. inverted mass hierarchies , phase advancement or retardation must be discriminated by measurements at @xmath190350 m. it can be executed thanks to the ultra monochromatic nature of neutrino beam which would allow us to detect relative phase of atmospheric oscillation even after 20 oscillations . we stress that the monochromaticity is special to the resonant @xmath28 absorption reaction , most probably the unique experimental way to realize this principle . we concluded that assuming @xmath157 % , the hierarchy can be determined at 90 ( 99 ) % cl for @xmath49 down to 0.035 ( 0.06 ) for 1000 events collected at five different positions and down to 0.025 ( 0.04 ) for 2000 events . for a larger error @xmath162% , the critical value of @xmath49 will be @xmath12 0.08 ( 0.11 ) at 90 ( 99 ) % cl approximately independent of the number of events for more than @xmath12 1000 events . we have also shown that in a third stage of the experiment , by positioning detectors in different locations , it is possible to determine the solar - scale parameters too . the precision one can achieve with 1000 events in each detector for @xmath191 is about 1.6 ( 4.0 ) % at 68.27 ( 99.73 ) % cl whereas that for @xmath1 is about 0.9 ( 2.2 ) % at 68.27 ( 99.73 ) % cl , all for 2 dof . the accuracies improve as data taking proceeds and the errors would reduce by a factor of @xmath192 when an order of magnitude larger number of events is taken . it is in fact remarkable that the experimental set up considered in this paper allows us to make a precise measurement of all the mixing parameters appear in the oscillation probability for @xmath14 channel , @xmath2 , @xmath15 , @xmath1 , @xmath11 and the neutrino mass hierarchy only by a single experiment , independent of any other experiments . the only parameters inaccessible by the method are @xmath193 and the cp phase @xmath194 , but precision measurement of the other mixing parameters will certainly help in determining these parameters by other experiments . some remarks are in order : * as noted in sec . [ subsec : theo ] , our method of determining the mass hierarchy can be confused by the cpt violation of the type with @xmath2 in different octants in the neutrino and the antineutrino sectors . it in turn implies that once the mass hierarchy is determined by some other means , the measurement we propose in this paper can be regarded as testing that particular type of cpt violation . * due to the capability of resolving atmospheric wiggles even after 20 atmospheric oscillations away from the source , this experiment could be sensitive to other dynamics @xcite which differ from the standard three flavor mass induced oscillation one . in particular , one can perhaps probe mechanisms such as : neutrino decay @xcite , decoherence @xcite , mixing with sterile neutrinos @xcite , violation of lorentz invariance or the equivalence principle @xcite , long - range leptonic forces @xcite , extra - dimensions @xcite . finally , we emphasize that realizing the proposed mssbauer @xmath28 experiment is crucial in order to make all the discussions in this paper meaningful . the principle of the experiment is well formulated by having the decay producing source @xmath28 and the absorption reaction in the target as t - conjugate channels with each other , allowing mutual cancellation of large part of the atomic shift @xcite . nonetheless , it still remains unclear to what extent the @xmath124he atom in the target experiences exactly the same environment as the @xmath124h atom in the source @xcite . this point has to be ( and will be ) investigated in a systematic way . if the experiment is indeed realizable it will certainly open a new window for neutrino experiments which can explore the neutrino properties in otherwise inaccessible manner . one of the authors ( h.m . ) thanks the abdus salam international center for theoretical physics , where part of this work was done , for the hospitality . two of us ( h.n . and r.z.f ) are grateful for the hospitality of the theory group of the fermi national accelerator laboratory during the summer of 2006 . this work was supported in part by the grant - in - aid for scientific research , nos . 16340078 and 19340062 , japan society for the promotion of science , by fundao de amparo pesquisa do estado de so paulo ( fapesp ) , fundao de amparo pesquisa do estado de rio de janeiro ( faperj ) and conselho nacional de cincia e tecnologia ( cnpq ) . fermilab is operated under doe contract no . de - ac02 - 76ch03000 . the advancement or retardation of the phase , @xmath59 , could be interpreted as a change in the effective atmospheric @xmath7 . such an effective atmospheric @xmath7 , at @xmath96 , is naturally defined by @xmath195 & = & \left(\frac{c^2_{12 } -\frac{1}{2}\sin^2 2 \theta_{12}\sin^2 \delta_{21 } } { 1- \sin^2 2 \theta_{12 } \sin^2 \delta_{21 } } \right ) \vert \delta m^2_{31 } \vert + \left(\frac{s^2_{12 } -\frac{1}{2}\sin^2 2 \theta_{12}\sin^2 \delta_{21 } } { 1- \sin^2 2 \theta_{12 } \sin^2 \delta_{21 } } \right)\vert \delta m^2_{32 } \vert . \nonumber\end{aligned}\ ] ] where the positive ( negative ) sign , in eq . ( [ eqn : dmsq_eff ] ) , is for the normal ( inverted ) hierarchy . for @xmath196 we recover @xmath11 , since this is in essence its definition , see ref . @xcite for details . thus eq . ( [ eqn : dmsq_eff ] ) is the generalization of @xmath11 to arbitrary @xmath96 . in fig . [ fig : dmsq ] we show the variation of @xmath197 as a function of baseline for both the normal and inverted hierarchies . the extrema of the difference between @xmath198 and @xmath102 are given by @xmath199 as function of baseline at 18.6 kev neutrino energy for both the normal and inverted hierarchies . the dashed line is @xmath102 which has been assumed to be the same for the normal and inverted hierarchies for this plot . in general , unless @xmath102 is determined precisely it need not take exactly the same value for both hierarchies . the maximum difference between @xmath197 and @xmath200 occurs at the solar oscillation maxima , i.e. when @xmath201 . , scaledwidth=55.0% ] as the oscillations go through the solar maxima , the fractional increase ( decrease ) in the effective atmospheric @xmath7 is @xmath202 for the normal ( inverted ) hierarchy . it is the accumulated effect of this small difference in the effective atmospheric @xmath7 for the normal and inverted hierarchies that leads to the observable phase difference shown in fig . [ rio - sp ] . since @xmath203 as it must . for @xmath17 disappearance in vacuum , most of the results of this paper can be applied , using the following translation @xmath204 with this translation the atmospheric @xmath205 and phase , @xmath206 , are given by @xmath207 then @xmath208 . \label{eqn : pdismu}\end{aligned}\ ] ] since numerically @xmath209 the phase for @xmath17 disappearance is approximately minus the phase for @xmath8 disappearance , i.e. @xmath210 therefore the normal hierarchy is retarded while the inverted hierarchy is advanced . this is the opposite sign to what happens for @xmath8 disappearance but the rate of change is approximately the same . the effective atmospheric @xmath7 for @xmath17 disappearance at any @xmath96 is given by @xmath211\vert \delta m^2_{31 } \vert \nonumber \\[0.1 cm ] & & \quad + \left [ \frac{\vert u_{\mu2 } \vert^2(\vert u_{\mu1 } \vert^2 + \vert u_{\mu2 } \vert^2 ) -2\vert u_{\mu1 } \vert^2\vert u_{\mu2 } \vert^2 \sin^2 \delta_{21 } } { ( \vert u_{\mu1 } \vert^2+\vert u_{\mu2 } \vert^2)^2 - 4\vert u_{\mu1 } \vert^2\vert u_{\mu2 } \vert^2 \sin^2 \delta_{21 } } \right ] \vert \delta m^2_{32 } \vert.\end{aligned}\ ] ] again for small @xmath58 this reduces to @xmath212 given by eq . ( [ eqn : dmsqmumu ] ) . v. d. barger , s. geer , r. raja and k. whisnant , phys . b * 485 * , 379 ( 2000 ) [ arxiv : hep - ph/0004208 ] ; a. cervera , a. donini , m. b. gavela , j. j. gomez cadenas , p. hernandez , o. mena and s. rigolin , nucl . b * 579 * , 17 ( 2000 ) [ erratum - ibid . b * 593 * , 731 ( 2001 ) ] [ arxiv : hep - ph/0002108 ] ; h. minakata and h. nunokawa , jhep * 0110 * , 001 ( 2001 ) [ arxiv : hep - ph/0108085 ] . p. huber , m. lindner and w. winter , nucl . b * 645 * , 3 ( 2002 ) [ arxiv : hep - ph/0204352 ] ; v. barger , d. marfatia and k. whisnant , phys . d * 65 * , 073023 ( 2002 ) [ arxiv : hep - ph/0112119 ] ; h. minakata , h. nunokawa and s. j. parke , phys . d * 68 * , 013010 ( 2003 ) [ arxiv : hep - ph/0301210 ] ; o. mena and s. j. parke , phys . rev . d * 70 * , 093011 ( 2004 ) [ arxiv : hep - ph/0408070 ] ; o. mena requejo , s. palomares - ruiz and s. pascoli , phys . d * 72 * , 053002 ( 2005 ) [ arxiv : hep - ph/0504015 ] . m. ishitsuka , t. kajita , h. minakata and h. nunokawa , phys . d * 72 * , 033003 ( 2005 ) [ arxiv : hep - ph/0504026 ] . d. beavis _ et al . _ , arxiv : hep - ex/0205040 ; m. v. diwan _ et al . _ , phys . d * 68 * , 012002 ( 2003 ) [ arxiv : hep - ph/0303081 ] ; v. barger , m. dierckxsens , m. diwan , p. huber , c. lewis , d. marfatia and b. viren , phys . d * 74 * , 073004 ( 2006 ) [ arxiv : hep - ph/0607177 ] ; r. gandhi , p. ghoshal , s. goswami , p. mehta and s. uma sankar , phys . d * 73 * , 053001 ( 2006 ) [ arxiv : hep - ph/0411252 ] . t. kajita , talk given at next generation of nucleon decay and neutrino detectors ( nnn05 ) , aussois , savoie , france , april 7 - 9 , 2005 . http://nnn05.in2p3.fr/ ; s. palomares - ruiz and s. t. petcov , nucl . b * 712 * , 392 ( 2005 ) [ arxiv : hep - ph/0406096 ] ; s. t. petcov and t. schwetz , nucl . b * 740 * , 1 ( 2006 ) [ arxiv : hep - ph/0511277 ] ; r. gandhi , p. ghoshal , s. goswami , p. mehta and s. uma sankar , arxiv : hep - ph/0506145 . a. s. dighe and a. y. smirnov , phys . d * 62 * , 033007 ( 2000 ) [ arxiv : hep - ph/9907423 ] ; h. minakata and h. nunokawa , phys . b * 504 * , 301 ( 2001 ) [ arxiv : hep - ph/0010240 ] ; c. lunardini and a. y. smirnov , jcap * 0306 * , 009 ( 2003 ) [ arxiv : hep - ph/0302033 ] ; v. barger , p. huber and d. marfatia , phys . lett . b * 617 * , 167 ( 2005 ) [ arxiv : hep - ph/0501184 ] . a. s. dighe , m. t. keil and g. g. raffelt , jcap * 0306 * , 005 ( 2003 ) [ arxiv : hep - ph/0303210 ] ; _ ibid . _ * 0306 * , 006 ( 2003 ) [ arxiv : hep - ph/0304150 ] ; r. tomas , m. kachelriess , g. raffelt , a. dighe , h. t. janka and l. scheck , jcap * 0409 * , 015 ( 2004 ) [ arxiv : astro - ph/0407132 ] . w. m. yao _ et al . _ [ particle data group ] , j. phys . g * 33*,1 ( 2006 ) . a. de gouvea and c. pena - garay , phys . d * 71 * , 093002 ( 2005 ) [ arxiv : hep - ph/0406301 ] . h. minakata , h. nunokawa , w. j. c. teves and r. zukanovich funchal , phys . d * 71 * , 013005 ( 2005 ) [ arxiv : hep - ph/0407326 ] ; nucl . suppl . * 145 * , 45 ( 2005 ) [ arxiv : hep - ph/0501250 ] ; a. bandyopadhyay , s. choubey , s. goswami and s. t. petcov , phys . rev . d * 72 * , 033013 ( 2005 ) [ arxiv : hep - ph/0410283 ] ; j. f. kopp , m. lindner , a. merle and m. rolinec , arxiv : hep - ph/0606151 ; s. t. petcov and t. schwetz , arxiv : hep - ph/0607155 . m. nakahata , talk at the 5th workshop on `` neutrino oscillations and their origin '' ( noon2004 ) , february 11 - 15 , 2004 , odaiba , tokyo , japan . http://www-sk.icrr.u-tokyo.ac.jp/noon2004/ m. c. gonzalez - garcia and c. pea - garay , phys . b * 527 * , 199 ( 2002 ) [ arxiv : hep - ph/0111432 ] . m. c. gonzalez - garcia and m. maltoni , phys . d * 70 * , 033010 ( 2004 ) [ arxiv : hep - ph/0404085 ] ; r. foot , c. n. leung and o. yasuda , phys . b * 443 * , 185 ( 1998 ) [ arxiv : hep - ph/9809458 ] . j. f. beacom and n. f. bell , phys . d * 65 * , 113009 ( 2002 ) [ arxiv : hep - ph/0204111 ] . f. benatti and r. floreanini , jhep * 0002 * , 032 ( 2000 ) [ arxiv : hep - ph/0002221 ] ; f. benatti and r. floreanini , phys . d * 64 * , 085015 ( 2001 ) [ arxiv : hep - ph/0105303 ] ; a. m. gago , e. m. santos , w. j. c. teves and r. zukanovich funchal , phys . d * 63 * , 073001 ( 2001 ) [ arxiv : hep - ph/0009222 ] ; _ ibid _ , arxiv : hep - ph/0208166 . m. cirelli , g. marandella , a. strumia and f. vissani , nucl . b * 708 * , 215 ( 2005 ) [ arxiv : hep - ph/0403158 ] ; a. y. smirnov and r. zukanovich funchal , phys . d * 74 * , 013001 ( 2006)[arxiv : hep - ph/0603009 ] . m. gasperini , phys . d * 38 * ( 1988 ) 2635 ; m. gasperini , phys . d * 39 * , 3606 ( 1989 ) ; s. l. glashow , a. halprin , p. i. krastev , c. n. leung and j. t. pantaleone , phys . d * 56 * , 2433 ( 1997 ) [ arxiv : hep - ph/9703454 ] ; s. r. coleman and s. l. glashow , phys . rev . d * 59 * , 116008 ( 1999 ) [ arxiv : hep - ph/9812418 ] ; h. minakata and h. nunokawa , phys . d * 51 * , 6625 ( 1995 ) [ arxiv : hep - ph/9405239 ] ; g. l. fogli , e. lisi , a. marrone and g. scioscia , phys . rev . d * 60 * , 053006 ( 1999 ) [ arxiv : hep - ph/9904248 ] . l. okun , sov . j. nucl . * 10 * , 206 ( 1969 ) , _ ibid _ , phys . b * 382 * , 389 ( 1996 ) ; j. a. grifols and e. masso , phys . b * 396 * , 201 ( 1997 ) [ arxiv : astro - ph/9610205 ] ; j. a. grifols and e. masso , phys . b * 579 * , 123 ( 2004 ) [ arxiv : hep - ph/0311141 ] ; m. c. gonzalez - garcia , p. c. de holanda , e. masso and r. zukanovich funchal , jcap * 0701 * , 005 ( 2007 ) [ arxiv : hep - ph/0609094 ] . g. r. dvali and a. y. smirnov , nucl . b * 563 * , 63 ( 1999 ) [ arxiv : hep - ph/9904211 ] . r. s. raghavan , talk given at neutrino oscillation workshop ( now2006 ) , conca specchiulla , otranto , lecce , italy , september 9 - 16 , 2006 , avilable at http://www.ba.infn.it/@xmath12now/now2006/

reaction - diffusion systems can describe a wide class of rhythmic spatiotemporal patterns observed in chemical and biological systems , such as circulating pulses on a ring , oscillating spots , target waves , and rotating spirals . these rhythmic dynamics can be considered limit cycles of reaction - diffusion systems . however , the conventional phase - reduction theory , which provides a simple unified framework for analyzing synchronization properties of limit - cycle oscillators subjected to weak forcing , has mostly been restricted to low - dimensional dynamical systems . here , we develop a phase - reduction theory for stable limit - cycle solutions of infinite - dimensional reaction - diffusion systems . by generalizing the notion of isochrons to functional space , the phase sensitivity function a fundamental quantity for phase reduction is derived . for illustration , several rhythmic dynamics of the fitzhugh - nagumo model of excitable media are considered . nontrivial phase response properties and synchronization dynamics are revealed , reflecting their complex spatiotemporal organization . our theory will provide a general basis for the analysis and control of spatiotemporal rhythms in various reaction - diffusion systems .

1406.0274

the phase - reduction theory provides a general framework to simplify multidimensional ordinary differential equations ( odes ) describing weakly perturbed limit - cycle oscillators to one - dimensional approximate phase equations @xcite . it has drastically facilitated theoretical and experimental analysis of the synchronization properties of weakly interacting nonlinear oscillators such as chemical oscillators and spiking neurons @xcite . methods for controlling limit - cycle oscillators have also been developed on the basis of the phase reduction theory @xcite . in real - world systems , rhythmic dynamics often arise collectively from a number of spatially distributed interacting elements , rather than from a single isolated oscillator , e.g. , heartbeats generated by an ensemble of pulsating cardiac cells @xcite . such systems are often modeled by reaction - diffusion ( rd ) systems , and the collective spatiotemporal rhythms are described by stable limit - cycle solutions of the rd systems @xcite . synchronization of collective spatiotemporal rhythms has been investigated experimentally in chemical systems @xcite and may be of significant practical importance , e.g. , in biomedical engineering @xcite . in order to analyze and control the dynamics of collective spatiotemporal rhythms , it is desirable to develop a phase - reduction theory for the rd systems . various types of low - dimensional phase equations have been derived for rd systems , in particular for traveling pulses @xcite and for rotating spirals @xcite . in most cases , however , it is assumed that the system is symmetric with respect to continuous spatial translation or rotation and the spatial structure is rigidly translating or rotating , so that their location or angle is simply identified as the phase . however , such assumptions exclude various intriguing rhythmic dynamics of rd systems that lack continuous spatial symmetry . since limit cycles are essentially associated with _ temporal _ translational symmetry , we should be able to derive phase equations from general rd systems without recourse to spatial symmetry . our goal in the present study is to develop , without assuming any spatial symmetry or rigidity , a phase - reduction theory for weakly perturbed rd systems exhibiting stable rhythmic dynamics . we solve this problem by generalizing the conventional phase - reduction theory for odes to rd systems . our theory gives a systematic method to approximate rhythmic dynamics of infinite - dimensional rd systems by one - dimensional phase equations , thereby facilitating detailed analysis of the synchronization dynamics of rhythmic spatiotemporal patterns . as a simple example , we analyze mutual synchronization between two interacting layers of rd systems exhibiting rhythmic dynamics . the proposed theory provides a simple , unified description of rhythmic spatiotemporal patterns and will be the basis for developing methods to control and design rhythmic spatiotemporal patterns in rd systems . in this section , we summarize the essential results of the proposed phase reduction theory for rd systems and apply it to mutual synchronization of a pair of coupled rd systems . full derivation of the theory will be given in appendix b. see also appendix a for a review of the phase reduction theory for ordinary limit - cycle oscillators described by odes . we consider weakly perturbed rd systems exhibiting stable rhythmic dynamics , described by @xmath0 here , the vector field @xmath1 represents the state ( e.g. , concentrations of chemical species ) of the rd medium at point @xmath2 at time @xmath3 , @xmath4 represents the local reaction dynamics at @xmath2 , @xmath5 represents the diffusion of @xmath6 over the medium with a matrix @xmath7 of diffusion constants , and @xmath8 represents weak spatiotemporal perturbations . explicit dependence of @xmath9 on @xmath2 , such as medium heterogeneity , may exist . we assume that the rd system ( [ eq : perturbedrd ] ) without perturbation ( @xmath10 ) exhibits a stable rhythmic dynamics , i.e. , it possesses a stable limit - cycle solution @xmath11 of period @xmath12 , where @xmath13 denotes frequency , and that this solution persists and deforms only slightly even if the system is weakly perturbed ( @xmath14 ) . such a limit cycle includes the circulating pulses on a ring , oscillating spots , target waves , and rotating spirals that we will analyze in section iii ( see figs . [ fig : a ] , [ fig : b ] , [ fig : c ] , and [ fig : d ] ) . the purpose of the phase reduction theory is to derive a simple closed equation for the phase @xmath15 approximately describing limit - cycle oscillations of eq . ( [ eq : perturbedrd ] ) under weak perturbation ( @xmath14 ) . as in the ode case ( see appendix a ) , we first introduce a phase @xmath16 to a system state @xmath17 on the limit cycle @xmath18 so that @xmath19 constantly holds , and denote the system state as @xmath20 using the phase @xmath15 . to perform phase reduction , we also need to assign a phase to a system state @xmath21 that is not on the limit cycle @xmath18 but eventually converges to @xmath18 , because the system state can deviate from @xmath18 due to perturbations . specifically , we need a functional @xmath22 that maps @xmath21 in the basin of @xmath18 to a scalar phase @xmath15 such that @xmath19 constantly holds . this leads to the notion of _ isochrons _ @xcite , i.e. , equal - phase contours of the system state around @xmath18 . the notion of the isochrons is at the core of the conventional phase reduction theory for odes and should be generalized to rd systems . it is , however , generally impossible to obtain such a functional explicitly . to proceed , we use the assumption that the perturbation is weak and focus on the vicinity of @xmath18 . we make an ansatz that the phase @xmath23 of a system state @xmath24 near @xmath20 can be linearly approximated , using a certain function @xmath25 , as @xmath26 \label{eq : approxphs}\end{aligned}\ ] ] around @xmath18 , where @xmath27 = \int { \bf a}({\bf r } ) \cdot { \bf b}({\bf r } ) d{\bf r}$ ] is the inner product between two functions . for a system state @xmath28 on @xmath18 with phase @xmath15 , an identity @xmath29 should hold by the above definition of the phase . moreover , for any system state @xmath21 near @xmath18 evolving under eq . ( [ eq : perturbedrd ] ) with @xmath10 , we require that @xmath30 satisfies @xmath19 constantly within linear approximation . as we will derive in appendix b , if @xmath25 is a periodic solution to a generalized _ adjoint equation _ @xmath31 with a normalization condition @xmath32 = 1 , \label{eq : normalization}\end{aligned}\ ] ] the functional @xmath33 assumed in eq . ( [ eq : approxphs ] ) satisfies the above requirements for the phase , and that such @xmath25 plays the role of the _ phase sensitivity function _ @xcite for the rd system . here , @xmath34 is a jacobi matrix of @xmath9 estimated at @xmath35 on @xmath18 . namely , we can show that the phase @xmath30 of the _ infinite - dimensional _ rd system ( [ eq : perturbedrd ] ) approximately obeys a simple _ one - dimensional _ phase equation @xmath36 , \label{eq : phaseeq}\end{aligned}\ ] ] which is correct up to the lowest order of the perturbation @xmath8 . thus , once @xmath37 is obtained from eqs . ( [ eq : adjointrd ] ) and ( [ eq : normalization ] ) , rhythmic dynamics of the rd system subjected to weak spatiotemporal perturbations , eq . ( [ eq : perturbedrd ] ) , can easily be analyzed using eq . ( [ eq : phaseeq ] ) . this is the main result of the present study . the phase equation ( [ eq : phaseeq ] ) also shows that , if a system state @xmath20 with phase @xmath15 on @xmath18 is instantaneously perturbed by a weak spatial stimulus @xmath38 , the response of the system phase after relaxation , namely , the _ phase response curve _ ( prc ) @xcite , is given by @xmath39\end{aligned}\ ] ] within linear approximation . this prc @xmath40 can be directly measured in numerical simulations by applying impulsive perturbations to the rd system as we will illustrate in section iii . as a simple example of the phase reduction approach , let us consider synchronization of a pair of weakly coupled rd systems exhibiting rhythmic dynamics , @xmath41 where @xmath42 represent the system states . we assume local and linear mutual coupling , @xmath43 , with a diagonal matrix @xmath44 representing the intensity of the weak mutual coupling . experimental systems like eq . ( [ eq : coupledrd ] ) have been realized by coupling a pair of photosensitive belousov - zhabotinsky chemical reactions via video cameras and projectors @xcite , and by coupling a pair of electrochemical oscillators via electrodes @xcite . denoting the phase variables of the two systems as @xmath45 and considering the coupling term @xmath46 as weak perturbations , we can approximate eq . ( [ eq : coupledrd ] ) by a pair of coupled phase equations , @xmath47 , \cr \dot{\theta}_{2}(t ) & = \omega + [ { \bf q}({\bf r } ; \theta_{2 } ) , { \bf g}\ { { \bf x}_{0}({\bf r } ; \theta_{2 } ) , { \bf x}_{0}({\bf r } ; \theta_{1})\ } ] , \label{eq : coupledphs}\end{aligned}\ ] ] where @xmath42 in @xmath46 are approximated by @xmath48 as the lowest - order approximation @xcite . note that the _ two infinite - dimensional _ rd systems are reduced to just _ two one - dimensional _ phase equations . the coupled phase equations ( [ eq : coupledphs ] ) can be analyzed in the same way as those for ordinary limit cycles @xcite . since the coupling term @xmath46 is small , we can apply the averaging method to eqs . ( [ eq : coupledphs ] ) , which yields @xmath49 where the phase - coupling function @xmath50 is given by @xmath51 \ d\theta.\end{aligned}\ ] ] by subtraction , the phase difference @xmath52 obeys @xmath53 where @xmath54 is a @xmath55-periodic , anti - symmetric function . thus , the phase difference @xmath52 between the two rd systems approximately obeys a quite simple one - dimensional equation . by examining the zeros of @xmath54 and their stability , we can predict the stable phase differences at which phase synchronization occurs between the two limit - cycle solutions of the coupled rd systems under the phase - reduction approximation . since we are considering symmetrically coupled identical rd systems , @xmath54 always vanishes at @xmath56 and at @xmath57 , so that the existence of in - phase ( @xmath58 ) and anti - phase ( @xmath59 ) synchronized states is assured . their stability is determined by the slope of @xmath54 . with periodic boundary conditions . the system parameters are @xmath60 , @xmath61 , @xmath62 , @xmath63 , and @xmath64 . with these values , the pulse exhibits a wavy tail @xcite . the oscillation period ( time needed for the pulse to go around the ring ) is @xmath65 . ( a ) snapshots of the stable circulating pulse with a wavy tail , @xmath66 , and the corresponding phase sensitivity function , @xmath67 . ( b ) phase response curves @xmath40 of the circulating pulse normalized by the stimulus intensity @xmath68 . either bell - shaped [ @xmath69 or cosine [ @xmath70 perturbation is given to the activator ( @xmath71 ) component . ( c ) evolution of phase differences between two systems coupled through the @xmath71 component with the coupling intensity matrix @xmath72 and the anti - symmetric part of the phase - coupling function @xmath54 ( rescaled by the coupling intensity @xmath73 ) . the two pulses show multimodal phase synchronization . ( d ) snapshots of phase - locked pulses with four different stable phase phase differences . graphs ( a d ) correspond to the stable phase differences shown in ( c ) . ] in this section , we illustrate the phase reduction theory for rd systems by numerical simulations . we analyze phase response properties and synchronization dynamics of circulating pulses on a ring , oscillating spots , target waves , and rotating spirals of the fitzhugh - nagumo model of excitable media ( figs . [ fig : a]-[fig : d ] ) . among these rhythmic patterns , the circulating pulses and rotating spirals are rigid and spatially symmetric , so that they may in principle be analyzed using the conventional methods @xcite . in contrast , the oscillating spots and target waves are not rigid and lack translational or rotational symmetry ; therefore , they can not be treated by the conventional methods that rely on such assumptions . in any case , the phase reduction can provide a simple , unified approach to the synchronization properties of spatiotemporal rhythms . as we will see , complex spatiotemporal profiles of the rhythmic patterns can lead to interesting synchronization dynamics . the fitzhugh - nagumo ( fhn ) reaction - diffusion model is a classical model of neural spike transmission , whose dynamics is described by @xmath74 where @xmath75 and @xmath76 are activator and inhibitor variables , respectively . by appropriately choosing the parameters @xmath77 , @xmath78 , @xmath79 , and the diffusion constants @xmath80 and @xmath81 , the fhn model can exhibit various types of rhythmic spatiotemporal dynamics @xcite , such as the circulating pulses on a ring ( fig . [ fig : a ] ) , oscillating spots ( fig . [ fig : b ] ) , target waves ( fig . [ fig : c ] ) , and rotating spirals ( fig . [ fig : d ] ) . in numerical simulations , the size of the system is @xmath82 for 1d cases and discretized using @xmath83 spatial grids . for 2d cases , the system size is @xmath84 , and discretized with @xmath85 spatial grids . the explicit euler method with a time step @xmath86 is used for numerical simulations of the rd system . to numerically obtain the phase sensitivity function @xmath25 , the adjoint equation ( [ eq : adjointrd ] ) is integrated backward in time @xcite . namely , one period of the limit - cycle oscillation is recorded by integrating the original rd system forward with sufficiently small time grids ; then , the adjoint equation is spatially discretized and numerically integrated backward using the recorded time sequence of the limit cycle , with occasional normalization of the solution so that eq . ( [ eq : normalization ] ) is satisfied . owing to the assumed stability of the limit - cycle solution , all modes other than the zero mode corresponding to temporal translational invariance eventually decay ( the floquet theorem ) , and the resultant solution gives the phase sensitivity function . or @xmath87 with no - flux boundary conditions . the parameter @xmath77 is space - dependent , i.e. , @xmath88 with @xmath89 and @xmath90 , so that @xmath77 is the largest at the center ( @xmath91 ) and the smallest at the boundaries ( @xmath92 ) . other parameter values are @xmath93 , @xmath94 , @xmath63 , @xmath95 . with these conditions , an oscillating spot constrained at the center can be generated . the oscillation period is @xmath96 ( @xmath97 ) or @xmath98 ( @xmath87 ) . ( a ) snapshots of the oscillating spot solution @xmath99 and the corresponding phase sensitivity function @xmath100 for @xmath101 . ( b ) evolution of @xmath102 and @xmath103 during @xmath104 . ( c ) phase response curves @xmath40 of the oscillating spot normalized by the stimulus intensity @xmath68 . perturbation @xmath105 is either bell - shaped [ @xmath106 or sinusoidal [ @xmath107 and is given to the activator ( @xmath71 ) component . results obtained by direct numerical simulations are compared with the theory , @xmath108 $ ] , where the inner product is taken over the 1d interval ( @xmath109 ) . ( d ) evolution of phase differences between two systems coupled through the @xmath71 component with the intensity matrix @xmath110 , showing in - phase synchronization for @xmath97 ( a ) and anti - phase synchronization for @xmath87 ( b ) . the anti - symmetric part of the phase - coupling function @xmath54 ( rescaled by the coupling intensity @xmath111 ) is shown for comparison . ( e ) snapshots of activator patterns @xmath112 of both systems in the in - phase ( a ) and anti - phase ( b ) synchronized states . ] our first example is a circulating - pulse solution of the fhn model with a wavy tail on a 1d ring of length @xmath113 . since the pattern is rigid and the system is translationally symmetric , the phase @xmath15 can simply be identified as the pulse location in this case . figure [ fig : a](a ) shows snapshots of the limit - cycle solution @xmath114 and the corresponding phase sensitivity function @xmath115 for @xmath116 , both propagating to the right . results for other values of @xmath15 can simply be obtained by translating fig . [ fig : a](a ) in the @xmath117 direction . it is observed that @xmath115 is _ localized _ near the pulse , indicating that perturbations given only in this region can affect the phase of the pulse . it is also seen that @xmath115 has a _ wavy front _ , reflecting the wavy tail of the pulse . this counterintuitive result can be explained as follows . the system exhibits localized damped oscillations when it is perturbed at some spatial point . if the pulse propagates into such a region , the pulse location ( i.e. , the system phase ) is either advanced or retarded depending on the timing of the collision , yielding the wavy front of @xmath115 . figure [ fig : a](b ) compares the prcs @xmath40 of the system to the weak spatial stimulus @xmath118 obtained by direct numerical simulations ( dns ) with the theoretical results , @xmath119 $ ] , where @xmath115 is obtained from the adjoint equation . the stimulus is either a bell shape localized at the center or a cosine curve , and it is given only to the activator component for the sake of simplicity . when the intensity @xmath68 of the stimulus is sufficiently small , good agreement is obtained . figure [ fig : a](c ) shows the synchronization dynamics of the two rd systems , i.e. , the evolution of the phase difference @xmath52 from various initial conditions obtained by the dns , and compares them with the theoretical function @xmath54 . reflecting the wavy shapes of @xmath120 and @xmath121 , @xmath54 is also wavy with many zeros , which implies the coexistence of multiple stable phase - locking points for the two - coupled circulating pulses . this is confirmed by dns , which shows that the final phase differences are in good agreement with the zero - crossing points of @xmath54 with negative @xmath122 . figure [ fig : a](d ) shows several pairs of stably phase - locked pulses obtained by evolving the system from four different initial conditions . the two pulses synchronize where their wavy tails match , yielding multiple stable phase differences as predicted by the phase - reduction analysis . similar multi - modal phase locking is also observed in complex oscillations of delay - differential systems @xcite . our second example is an oscillating spot solution of the 1d fhn system of length @xmath113 with no - flux boundaries @xcite . to pin the spot at the center , the parameter @xmath77 of the model is assumed to be spatially heterogeneous , namely , the excitability of the system is the largest at the center and the smallest at the boundaries . note that the pattern is not rigid and the system lacks spatial symmetry . figure [ fig : b](a ) shows snapshots of the limit - cycle solution @xmath114 and the phase sensitivity function @xmath115 for @xmath101 . the activator component of @xmath115 is sharply localized at both fronts of the spot ; namely , the phase @xmath15 of the system is sensitive only to perturbations near the fronts . figure [ fig : b](b ) shows @xmath114 and corresponding @xmath115 for one oscillation period ( @xmath104 ) . perturbations given to the pulse fronts result in an advance or a delay in phase , depending on the timing , i.e. , whether the spot is expanding or shrinking . the inhibitor component of @xmath115 also reflects the oscillation of the spot . figure [ fig : b](c ) shows the prcs @xmath40 to the weak stimulus @xmath118 , which is either a bell shape or a cosine curve and is only applied to the activator . there is good agreement between the results of numerical simulations and theory . the synchronization properties of a pair of oscillating spots coupled through the activator component are shown in fig . [ fig : b](d ) , where the function @xmath54 and evolution of the phase difference @xmath123 are plotted . for comparison , two different system sizes , @xmath97 and @xmath87 , are used . since the parameter @xmath77 is spatially heterogeneous , the shape and oscillation period of the spot vary with @xmath113 . when @xmath97 , in - phase synchronization ( @xmath56 ) is linearly stable because @xmath124 . in contrast , when @xmath87 , in - phase synchronization is unstable and anti - phase synchronization ( @xmath57 ) becomes stable . this prediction is confirmed by numerical simulations with various initial phase differences . typical snapshots of the synchronized patterns are shown in fig . [ fig : b](e ) . with no - flux boundary conditions . to create a pacemaker region , the parameter @xmath77 is assumed to possess localized circular heterogeneity , i.e. , @xmath125 , where @xmath126^{1/2}$ ] is the distance from the pacemaker center at @xmath127 and @xmath128 is the radius of the pacemaker region , so that @xmath129 as @xmath130 and @xmath131 as @xmath132 . the parameters are @xmath133 , @xmath134 , @xmath135 , and @xmath136 . with these values , the system is self - oscillatory near the pacemaker center , and is excitable otherwise . other parameters are @xmath137 , @xmath138 , @xmath139 , and @xmath140 . the temporal oscillation period is @xmath141 . ( a ) target wave solution @xmath142 and the corresponding phase sensitivity function @xmath143 at @xmath101 . ( b ) phase response curves @xmath40 of the target wave normalized by the stimulus intensity @xmath68 . sinusoidal perturbation @xmath144 is given either to activator ( @xmath71 ) or inhibitor ( @xmath145 ) component . results obtained by direct numerical simulations are compared with the theory , @xmath146 $ ] , where the inner product is now taken over the 2d square ( @xmath147 ) . ( c ) evolution of phase differences between two systems coupled through the @xmath71 component with the intensity matrix @xmath148 , compared with the anti - symmetric part of the phase - coupling function @xmath54 ( rescaled by the coupling intensity @xmath149 ) . both in - phase ( a ) and anti - phase ( b ) synchronization can occur depending on initial conditions . ( d ) evolution of the activator @xmath71 measured at the center @xmath150 $ ] of systems 1 and 2 in the in - phase ( a ) and anti - phase ( b ) synchronized states . solid line corresponds to system 1 , and the dashed line corresponds to system 2 . ( e ) snapshots of the activator patterns @xmath151 of both systems in the in - phase ( a ) and anti - phase ( b ) synchronized states . ] with no - flux boundary conditions . to pin the spiral at the center , a localized circular heterogeneity of radius @xmath152 is introduced to the parameter @xmath153 as @xmath154 , where @xmath155^{1/2}$ ] is a distance from center of system . we assume @xmath156 and @xmath157 , so that excitability is the highest at center . other parameters are fixed at @xmath137 , @xmath138 , @xmath139 , and @xmath140 . with these parameters , the oscillation period of the spiral is @xmath158 . ( a ) spiral solution @xmath159 and the corresponding phase sensitivity functions @xmath160 at @xmath101 . ( b ) phase response curves @xmath40 of the spiral normalized by the stimulus intensity @xmath68 . checkerboard - like spatial perturbation is given either to the activator ( @xmath71 ) or inhibitor ( @xmath145 ) component , where @xmath161 for @xmath162 or @xmath163 , and @xmath164 otherwise . ( c ) evolution of phase differences between two systems coupled through the @xmath71 component with the coupling intensity matrix @xmath165 , compared with the anti - symmetric part of the phase - coupling function @xmath54 ( rescaled by the coupling intensity @xmath166 ) . both in - phase synchronization ( a ) and anti - phase synchronization ( b ) can occur depending on initial conditions . solid line corresponds to system 1 , and the dashed line corresponds to system 2 . ( d ) snapshots of the in - phase and anti - phase synchronized states . ] as the third example , we consider a target wave solution @xcite of the 2d fhn model on a square of side @xmath113 with no - flux boundaries . a circular pacemaker region is created by assuming the parameter @xmath77 to be heterogeneous . the system is rotationally symmetric around the pacemaker region in this case , but the target pattern is not rigid ; the phase should be associated with the temporal dynamics of the pattern . figure [ fig : c](a ) shows the limit - cycle solution @xmath167 and the corresponding phase sensitivity function @xmath168 for @xmath101 . as @xmath15 increases , @xmath167 undergoes oscillations corresponding to the emission of concentric target waves from the pacemaker , and @xmath168 oscillates accordingly . reflecting that the pacemaker dominates overall rhythms of the system , @xmath168 is localized at the pacemaker . figure [ fig : c](b ) shows the prcs @xmath40 obtained by applying weak cosine spatial stimulus @xmath169 to either the activator or inhibitor component . the numerical results are in good agreement with the theory . figure [ fig : c](c ) shows synchronization between two target waves . here , we consider counter - propagating target waves , i.e. , one of the rd systems is inverted in the @xmath117 direction as shown in fig . [ fig : c](e ) . the function @xmath54 has five zeros , with the in - phase ( @xmath58 ) and anti - phase ( @xmath59 ) synchronized states both being stable . therefore , depending on initial conditions , the two target waves can exhibit both types of synchronization , as confirmed by numerical simulations . figure [ fig : c](d ) shows time sequences of the activator at the center of the two systems corresponding to the in - phase and anti - phase synchronized states , and fig . [ fig : c](e ) shows corresponding snapshots . our final example is a rotating - spiral solution @xcite of the fhn model on a 2d square of side @xmath113 with no - flux boundaries . synchronization between a pair of rotating spirals was experimentally studied in @xcite . here , to pin the core of the spiral at the center of the system , circular heterogeneity in the parameter @xmath77 is introduced . the spiral rigidly rotates around this pinning region without changing its shape . figure [ fig : d](a ) shows snapshots of the spiral solution @xmath167 and the corresponding phase sensitivity function @xmath168 at @xmath116 . both rotate in the clockwise direction as @xmath15 increases . since the system is symmetric with respect to spatial rotation around the center and the pattern is rigid , the phase @xmath15 simply corresponds to the rotation angle and the results for other values of @xmath15 can be obtained by rotating fig . [ fig : d](a ) . as in the other cases , @xmath170 is strongly localized near the core of the spiral , indicating that the spiral tip dominates the overall phase of the system ; perturbations given only to this region can affect the overall system phase . figure [ fig : d](b ) compares the prcs obtained by dns with the theory , @xmath171 $ ] , to a checkerboard - like stimulus @xmath172 applied either to the activator or inhibitor , showing good agreement . figure [ fig : d](c ) shows the synchronization process between two spirals . as expected from the function @xmath54 with five zeros , the two spirals can exhibit either in - phase ( @xmath58 ) or anti - phase ( @xmath59 ) synchronization as determined by the initial conditions . typical time sequences of the activator component measured at @xmath173 in the in - phase and anti - phase synchronized states are shown in fig . [ fig : d](d ) , and typical snapshots of the synchronized spirals are shown in fig . [ fig : d](e ) . we developed a phase - reduction theory for limit - cycle solutions of infinite - dimensional rd systems and illustrated its validity by analyzing mutual synchronization of a pair of rd systems exhibiting rhythmic dynamics . our theory does not assume rigidity and spatial symmetry ; therefore , it is generally applicable to a wide class of rhythmic spatiotemporal dynamics in rd systems . the theory can readily be applied , for example , to the analysis of phase locking to periodic external stimulus and noise - induced synchronization @xcite of spatiotemporal rhythms , and will be a basis for controlling and designing spatiotemporal rhythmics in various systems . 10 a. t. winfree , _ the geometry of biological time _ ( springer , 1980 ) . y. kuramoto , _ chemical oscillations , waves , and turbulence _ ( springer , 1984 ) . f. c. hoppensteadt and e. m. izhikevich , _ weakly connected neural networks _ ( springer , 1997 ) . e. brown , j. moehlis , and p. holmes , _ on the phase reduction and response dynamics of neural oscillator populations . _ neural computation * 16 * , 673 - 715 ( 2004 ) . g. b. ermentrout and d. h. terman , _ mathematical foundations of neuroscience _ ( springer , 2010 ) . a. pikovsky , m. rosenblum , and j. kurths , _ synchronization : a universal concept in nonlinear sciences _ ( cambridge university press , 2001 ) . i. z. kiss , y. zhai , and j. l. hudson , _ emerging coherence in a population of chemical oscillators _ , science * 296 * , 1676 - 1678 ( 2002 ) . a. f. taylor , m. r. tinsley , f. wang , z. huang , and k. showalter , _ dynamical quorum sensing and synchronization in large populations of chemical oscillators _ , science * 323 * , 614 - 617 ( 2009 ) . m. r. tinsley , s. nkomo , and k. showalter , _ chimera and phase - cluster states in populations of coupled chemical oscillators _ , nature physics * 8 * , 662 - 665 ( 2012 ) . i. z. kiss , c. g. rusin , h. kori , and j. l. hudson , _ engineering complex dynamical structures : sequential patterns and desynchronization _ , science * 316 * , 1886 - 1889 ( 2007 ) . j. moehlis , e. shea - brown , and h. rabitz , _ optimal inputs for phase models of spiking neurons _ , journal of computational and nonlinear dynamics * 1 * , 358 - 367 ( 2006 ) . t. harada , h. a. tanaka , m. j. hankins , and i. z. kiss , _ optimal waveform for the entrainment of a weakly forced oscillator _ , physical review letters * 105 * , 088301 ( 2010 ) . a. zlotnik , y. chen , i. z. kiss , h. a. tanaka , and j. s. li , _ optimal waveform for fast entrainment of weakly forced nonlinear oscillators _ , physical review letters * 111 * , 024102 ( 2013 ) . l. glass , _ synchronization and rhythmic processes in physiology _ , nature * 410 * , 277 - 284 ( 2001 ) . l. glass , l , _ multistable spatiotemporal patterns of cardiac activity _ , pnas * 102 * , 10409 - 10410 ( 2005 ) . g. bub , a. shrier , and l. glass , _ spiral wave generation in heterogeneous excitable media _ , physical review letters * 88 * , 058101 ( 2002 ) . m. i. rabinovich , a. b. ezersky , and p. d. weidman , _ the dynamics of patterns _ ( world scientific , singapore , 2000 ) . a. s. mikhailov and k. showalter , _ control of waves , patterns and turbulence in chemical systems _ , physics reports * 425 * , 79 - 194 ( 2006 ) . s. p. hastings , _ on the existence of homoclinic and periodic orbits for the fitzhugh - nagumo equations _ , the quarterly journal of mathematics * 27 * , 123 - 134 ( 1976 ) . a. hagberg and e. meron , _ pattern formation in non - gradient reaction - diffusion systems : the effects of front bifurcations _ , nonlinearity * 7 * , 805 ( 1994 ) . t. nomura and l. glass , _ entrainment and termination of reentrant wave propagation in a periodically stimulated ring of excitable media _ , physical review e * 53 * , 6353 ( 1996 ) . t. yanagita , h. suetani , and k. aihara , _ bifurcation analysis of solitary and synchronized pulses and formation of reentrant waves in laterally coupled excitable fibers _ , physical review e * 78 * , 056208 ( 2008 ) . a. s. mikhailov and g. ertl ( eds . ) , _ engineering of chemical complexity ( world scientific lecture notes in complex systems vol . _ ( world scientific publishing , 2013 ) . j. j. tyson and j. p. keener , _ singular perturbation theory of traveling waves in excitable media ( a review ) _ , physica d * 32 * , 327 - 361 ( 1988 ) . p. manneville , _ dissipative structures and weak turbulence _ , ( academic press , 1990 ) . m. hildebrand , j. cui , e. mihaliuk , j. wang , and k. showalter , _ synchronization of spatiotemporal patterns in locally coupled excitable media _ , physical review e * 68 * , 026205 ( 2003 ) . s. fukushima , s. nakanishi , k. fukami , s. i. sakai , t. nagai , t. tada , and y. nakato , _ observation of synchronized spatiotemporal reaction waves in coupled electrochemical oscillations of an ndr type _ , electrochemistry communications * 7 * , 411 - 415 ( 2005 ) . s. i. ei , _ the motion of weakly interacting pulses in reaction - diffusion systems _ , journal of dynamics and differential equations * 14 * , 85 - 137 ( 2002 ) . m. c. cross and p. c. hohenberg , _ pattern formation outside of equilibrium _ , reviews of modern physics * 65 * , 851 - 1112 ( 1993 ) . h. mori and y. kuramoto , _ dissipative structures and chaos _ , ( springer , 1997 ) . t. ohta , _ pulse dynamics in a reaction - diffusion system _ , physica d * 151 * , 61 - 72 ( 2001 ) . g. b. ermentrout , j. z. jalics , and j. e. rubin , _ stimulus - driven traveling solutions in continuum neuronal models with a general smooth firing rate function _ , siam journal on applied mathematics * 70 * , 3039 - 3064 ( 2010 ) . j. lber , m. br , and h. engel , _ front propagation in one - dimensional spatially periodic bistable media _ , physical review e * 86 * , 066210 ( 2012 ) . b. sandstede , a. scheel , and c. wulff , _ dynamics of spiral waves on unbounded domains using center - manifold reductions _ , journal of differential equations * 141 * , 122 - 149 ( 1997 ) . v. n. biktashev , d. barkley , and i. v. biktasheva , _ orbital motion of spiral waves in excitable media _ , physical review letters * 104 * , 058302 ( 2010 ) . a. t. winfree , _ biological rhythms and the behavior of populations of coupled oscillators _ , journal of theoretical biology * 16 * , 15 - 42 ( 1967 ) . j. guckenheimer , _ isochrons and phaseless sets _ , journal of mathematical biology * 1 * , 259 - 273 ( 1975 ) . j. n. teramae and d. tanaka , _ robustness of the noise - induced phase synchronization in a general class of limit cycle oscillators _ , physical review letters * 93 * , 204103 ( 2004 ) . h. nakao , k. arai , and y. kawamura , _ noise - induced synchronization and clustering in ensembles of uncoupled limit - cycle oscillators _ , physical review letters * 98 * , 184101 ( 2007 ) . y. kawamura , h. nakao , k. arai , h. kori , and y. kuramoto , _ collective phase sensitivity _ , physical review letters * 101 * , 024101 ( 2008 ) . k. kotani , i. yamaguchi , y. ogawa , y. jimbo , h. nakao , and g. b. ermentrout , _ adjoint method provides phase response functions for delay - induced oscillations _ , physical review letters * 109 * , 044101 ( 2012 ) . h. nakao , t. yanagita , and k. kawamura , _ phase description of stable limit - cycle solutions in reaction - diffusion systems _ , procedia iutam * 5 * , 227 - 233 ( 2012 ) ( not refereed ) . in this section , we review the classical phase reduction theory for ordinary limit - cycle oscillators described by finite - dimensional odes . see refs . @xcite for details of the theory and applications to various synchronization phenomena in systems of coupled oscillators . we consider a limit - cycle oscillator described by the ode @xmath174 where @xmath175 is a @xmath176 dimensional vector representing the oscillator state at time @xmath3 and @xmath9 determines its dynamics . suppose that eq . ( [ seq1 ] ) has a stable limit - cycle solution of period @xmath177 , @xmath178 which is denoted as @xmath18 . we introduce a phase @xmath179 to the state @xmath180 on @xmath18 in such a way that @xmath181 increases with a constant frequency @xmath182 as @xmath180 evolves along @xmath18 under eq . ( [ seq1 ] ) . this can be performed by choosing a certain state @xmath183 on @xmath18 as the origin of the phase , i.e. , @xmath116 , and assigning a phase value @xmath184 to the oscillator state @xmath180 on @xmath18 ( @xmath185 ) evolving under eq . ( [ seq1 ] ) from the phase origin @xmath183 . namely , we identify the oscillator phase with the time multiplied by the frequency . we will denote the oscillator state on @xmath18 with the phase value @xmath15 as @xmath186 henceforth . the above definition of the phase on @xmath18 can be extended to the whole basin of @xmath18 by assigning the same phase value @xmath181 to the set of oscillator states @xmath187 that asymptotically approach the oscillator state @xmath188 on @xmath18 under eq . ( [ seq1 ] ) , i.e. , @xmath189 where @xmath190 represents the ordinary vector norm . this defines a phase function @xmath191 that maps a given oscillator state @xmath6 in the basin of @xmath18 to a scalar phase @xmath15 . it is clear that the phase @xmath192 of the state @xmath175 evolving under eq . ( [ seq1 ] ) obeys a simple phase equation , @xmath193 not only on the limit - cycle solution @xmath18 but also in the whole basin of @xmath18 . using the chain rule for the derivatives , it can be shown that @xmath194 where @xmath195 is the gradient of the phase function @xmath196 at @xmath197 . thus , the phase function @xmath196 should satisfy @xmath198 in the basin of @xmath18 . the set of oscillator states sharing the same phase value is called the _ isochron _ @xcite and is the fundamental concept in the analysis of limit - cycle oscillators @xcite . the whole basin of @xmath18 is foliated by such isochrons . see fig . [ figs0ode](a ) for a schematic illustration of the isochrons . + now we consider the case that the limit - cycle oscillator is weakly perturbed as @xmath199 where the perturbation @xmath200 , generally a function of the oscillator state @xmath6 and time @xmath3 , is assumed to be sufficiently weak so that the original limit - cycle solution @xmath18 is only slightly deformed . from eq . ( [ eq : s7 ] ) , the phase @xmath192 of the perturbed oscillator obeys @xmath201 however , this is not a closed equation for @xmath181 because the gradient @xmath195 and the perturbation @xmath202 still depend on @xmath175 . to obtain a closed equation for @xmath181 , @xmath175 in these terms are replaced by @xmath188 at the lowest order approximation , assuming that the perturbation @xmath202 is sufficiently weak so that @xmath175 does not significantly deviate from @xmath188 on @xmath18 . this yields an approximate closed phase equation for @xmath181 , @xmath203 which is correct up to @xmath204 . thus , by denoting @xmath205 and introducing a function @xmath206 the @xmath207-dimensional ode ( [ seq7 ] ) describing a perturbed limit cycle can be reduced to a simple one - dimensional phase equation @xmath208 at the lowest order in the perturbation . the key quantity for this approximation is the _ phase sensitivity function _ @xmath209 defined in eq . ( [ seq8 ] ) , which is the gradient of the isochron estimated at @xmath210 on the limit - cycle solution @xmath18 . the function @xmath209 quantifies linear phase response property of the oscillator state @xmath186 with phase @xmath15 on @xmath18 to infinitesimal perturbations . if @xmath186 is instantaneously perturbed by a weak stimulus @xmath211 , the resulting phase response is given by @xmath212 under linear approximation . this @xmath40 is called the _ phase response curve _ ( prc ) of the limit - cycle oscillator described by eq . ( [ seq1 ] ) . the prc can be obtained by applying impulsive perturbations to a limit - cycle oscillator and has been measured in various experimental systems @xcite . though we have developed a formal geometric theory by assuming the existence of the phase function @xmath196 , it is generally impossible to obtain @xmath196 explicitly , except for a few simple models of limit - cycle oscillators . however , to obtain the lowest - order phase equation ( [ seq8 ] ) for weak perturbations , only the phase sensitivity function @xmath209 is actually necessary . as we show below , the function @xmath209 can be obtained as the @xmath55-periodic solution to the following _ adjoint equation _ @xcite : @xmath213 with the constraint @xmath214 , or equivalently , @xmath215 for @xmath104 , where @xmath216 is the jacobi matrix of @xmath217 at @xmath210 on @xmath18 and @xmath218 indicates matrix transpose . the adjoint equation ( [ seq12 ] ) and the normalization condition ( [ seq13 ] ) can be derived in several different ways . we here use a simple argument as in @xcite with an emphasis on the linear approximation of the isochrons near the limit cycle ( see eq . ( [ seqy1 ] ) below ) . we use the same idea to develop a phase reduction theory for the limit - cycle solutions of reaction - diffusion systems in appendix b. see fig . [ figs0ode](b ) for a schematic illustration . we first note that , when @xmath219 is sufficiently small , the phase function @xmath196 for the ode can be expanded in a taylor series around @xmath186 as @xmath220 using the phase sensitivity function @xmath209 in eq . ( [ seq8 ] ) . therefore , if the oscillator state @xmath6 is close to the oscillator state @xmath186 with phase @xmath15 on the limit - cycle solution @xmath18 , @xmath196 can be linearly approximated as @xmath221 suppose a initial state @xmath222 on @xmath18 , and a slightly perturbed initial state @xmath223 near @xmath224 , where @xmath225 is a small perturbation given to @xmath224 . we evolve these two states without applying further perturbations . then , from eq . ( [ seq1 ] ) , the linearized equation for the small perturbation @xmath226 is given by @xmath227 where @xmath228 is the jacobi matrix of @xmath217 at @xmath210 on @xmath18 . from eq . ( [ seqy1 ] ) , the phase of the unperturbed state is given by @xmath229 , and that of the perturbed state can be expressed as @xmath230 under the linear approximation . note here that @xmath231 should also increase with a constant frequency @xmath13 , i.e. , @xmath232 within the linear approximation , because no perturbation is given after @xmath233 . thus , the following equation should hold for @xmath234 evolving from arbitrary @xmath225 : @xmath235 & = \frac{d { \bf z}(\theta(t ) ) } { dt } \cdot { \bf y}(t ) + { \bf z}(\theta(t ) ) \cdot { \rm j}(\theta(t ) ) { \bf y}(t ) = \left\ { \frac{d}{dt } { \bf z}(\theta(t ) ) + { \rm j}(\theta(t))^{\dag } { \bf z}(\theta(t ) ) \right\ } \cdot { \bf y}(t).\end{aligned}\ ] ] therefore , @xmath236 should satisfy the following adjoint equation : @xmath237 which is equivalent to eq . ( [ seq12 ] ) by the relation @xmath238 ( note that @xmath239 ) . to obtain the normalization condition eq . ( [ seq13 ] ) , we differentiate the identity @xmath229 by @xmath3 , which yields @xmath240 this gives the normalization condition eq . ( [ seq13 ] ) , again by the relation @xmath238 . thus , the function @xmath209 can be obtained by solving the adjoint equation ( [ seq12 ] ) under the normalization condition eq . ( [ seq13 ] ) , and the phase function near @xmath18 is given by eq . ( [ seqy1 ] ) within linear approximation . it can also be shown that @xmath209 is the unique solution to eq . ( [ seq12 ] ) by using the floquet theorem characterizing the linear stability of the limit cycle , since @xmath209 is closely related to the floquet eigenvector with the zero floquet exponent @xcite . in actual numerical calculations , it is useful to integrate eq . ( [ seq12 ] ) backward in time to avoid numerical overflow , with occasional normalization using eq . ( [ seq13 ] ) @xcite . then , by virtue of the floquet theorem , only the functional component corresponding to @xmath209 remains numerically . once we obtain the frequency @xmath13 and the phase sensitivity function @xmath209 , we can write down the approximate phase equation ( [ seq9 ] ) for a weakly perturbed limit - cycle oscillator described by eq . ( [ seq7 ] ) . this approximation , called the _ phase reduction _ , greatly simplifies theoretical analysis of weakly perturbed limit cycles and has been extensively used for analyzing synchronization dynamics of weakly interacting nonlinear oscillators @xcite . in this section , we give a full derivation of the phase reduction theory for rd systems that we briefly summarized in section ii a. our aim is to derive a simple one - dimensional phase equation for rhythmic spatiotemporal patterns described as limit - cycle solutions of rd systems without recourse to spatial symmetry of the patterns . we do not require the patterns to be rigidly translating or rotating in the rd medium without changing their spatial profiles , as typically assumed in the conventional derivation of the phase equations for rd systems . such rhythmic patterns can thus include oscillating spots and target waves , which vary their spatial profiles periodically . rigidly circulating waves or rotating spirals with spatial translational or rotational symmetry are also limit - cycle solutions of rd systems , and thus they can also be treated in the same framework as we showed in section iii . our strategy is to generalize the conventional phase reduction theory for ordinary limit cycles described by odes ( see appendix a for comparison ) , which assumes only temporal translational symmetry of the oscillator dynamics , to limit - cycle solutions of infinite - dimensional rd systems , thereby avoiding the assumptions on spatial symmetry . we can develop the theory almost in parallel with the ode case by noticing that the finite - dimensional vector @xmath175 is replaced by a vector field @xmath1 , and correspondingly the ordinary dot product of two vectors is replaced by the inner product of two vector fields . we consider a rd equation of the form @xmath241 where the @xmath207-dimensional vector @xmath1 represents the state of the rd medium at point @xmath2 in the @xmath242-dimensional space at time @xmath3 , @xmath4 specifies local reaction dynamics at point @xmath2 , and @xmath243 represents diffusion of @xmath6 over the medium with a constant diffusion matrix @xmath7 . explicit dependence of @xmath9 on @xmath2 , such as heterogeneity of the medium , may exist . appropriate boundary conditions ( e.g. , periodic or no - flux ) for the problem under consideration are introduced . we assume that eq . ( [ seq14 ] ) has a stable limit - cycle solution of period @xmath177 , @xmath244 which is denoted by @xmath18 . as in the ode case , we first define a phase @xmath245 of the system state @xmath17 on the limit - cycle solution @xmath18 so that @xmath181 increases with a constant frequency @xmath182 as @xmath17 evolves on @xmath18 under eq . ( [ seq14 ] ) . this is performed by identifying the phase with the time multiplied by the frequency . namely , we choose a certain system state @xmath246 on @xmath18 as the origin of the phase , @xmath116 , and assign a phase value @xmath184 to a state @xmath17 on @xmath18 ( @xmath247 ) evolving under eq . ( [ seq14 ] ) from the phase origin @xmath246 . we will denote the system state on @xmath18 with the phase value @xmath15 as @xmath20 henceforth . + next , we need to extend the definition of the phase to the basin of @xmath18 . as in the ode case , we assign the same phase value @xmath181 to the set of system states @xmath248 that eventually converge to the system state @xmath249 on @xmath18 under eq . ( [ seq14 ] ) , namely , @xmath250 here , @xmath251 denotes the @xmath252 norm of a spatial pattern defined as @xmath253 $ ] , and the inner product between two spatial patterns @xmath254 and @xmath255 is defined as @xmath256 = \int { \bf a}({\bf r } ) \cdot { \bf b}({\bf r } ) d{\bf r}.\end{aligned}\ ] ] the integral is taken over the considered spatial domain with appropriate boundary conditions . this introduces a _ phase functional _ @xmath257 that maps a given system state @xmath24 in the basin of @xmath18 to a scalar phase @xmath15 . then , the phase @xmath30 of the state @xmath1 evolving under eq . ( [ seq14 ] ) will constantly obey @xmath258 not only on the limit - cycle solution @xmath18 but also in the whole basin of @xmath18 . using the chain rule for the functional derivatives , the above equation can be written as @xmath259 \cr & = \left [ \left . \frac{\delta \theta\ { { \bf x}({\bf r } ) \ } } { \delta { \bf x}({\bf r } ) } \right|_{{\bf x}({\bf r})= { \bf x}({\bf r } , t ) } , { \bf f}({\bf x}({\bf r } , t ) , { \bf r } ) + { \rm d } \nabla^{2 } { \bf x}({\bf r } , t ) \right ] = \omega , \label{seq32}\end{aligned}\ ] ] where @xmath260 is the functional derivative of @xmath23 with respect to @xmath261 . thus , the phase functional @xmath23 should satisfy @xmath262 = \omega\end{aligned}\ ] ] in the basin of @xmath18 . we call a set of system states sharing the same phase value the _ isochron _ of the rd system , generalizing the same notion for odes ( appendix a ) . the whole basin of @xmath18 is foliated by such isochrons . see fig . [ figs0rd](a ) for a schematic illustration . now we consider the case that the rd system is weakly perturbed as @xmath263 where the perturbation @xmath264 is generally a functional of the state @xmath1 , location @xmath2 , and time @xmath3 . we assume that the original limit - cycle solution @xmath18 is only slightly deformed by the perturbation @xmath265 . if the phase functional @xmath23 is given , then from eq . ( [ seq32 ] ) , the phase @xmath30 of the perturbed system obeys @xmath266 \cr & = \omega + \left [ \left . \frac{\delta \theta\ { { \bf x}({\bf r } ) \ } } { \delta { \bf x}({\bf r } ) } \right|_{{\bf x}({\bf r})= { \bf x}({\bf r } , t ) } , { \bf p}\ { { \bf x}({\bf r } , t ) , { \bf r } , t \ } \right].\end{aligned}\ ] ] however , this is not a closed equation for @xmath181 because the functional derivative of @xmath267 and the perturbation @xmath264 still depend on @xmath1 . therefore , as in the ode case , we approximate @xmath1 in these terms by @xmath20 on @xmath18 , assuming the perturbation @xmath264 to be weak enough so that the system state @xmath1 deviates from @xmath20 on @xmath18 only slightly . then , at the lowest order approximation , a closed phase equation for @xmath181 can be obtained as @xmath268 , \label{seqx2}\end{aligned}\ ] ] which is correct up to @xmath269 . by denoting @xmath270 and introducing a _ phase sensitivity function _ @xmath271 the reduced phase equation ( [ seqx2 ] ) can be concisely written as @xmath272 \label{seqp2}\end{aligned}\ ] ] at the lowest order in the perturbation . the function @xmath273 is the ( functional ) gradient of the isochron estimated at @xmath28 on the limit - cycle solution @xmath18 and plays the key role in the present theory . though we have formally developed a geometric theory of phase reduction for the rd system , it is impossible to obtain @xmath23 explicitly for general rd systems . however , only the phase sensitivity function @xmath37 is actually necessary to write down the lowest - order phase equation ( [ seqp2 ] ) for the weakly perturbed rd systems eq . ( [ seqp1 ] ) . we thus try to derive the equation for @xmath37 as in the ode case , focusing only on the vicinity of the limit cycle @xmath18 . we first note that the phase function @xmath196 for the ode can be linearly approximated as @xmath274 for the oscillator state @xmath6 near @xmath186 on @xmath18 using the phase sensitivity function @xmath209 defined in eq . ( [ seq8 ] ) ( see appendix a ) . in a similar spirit , we make an ansatz that the phase @xmath23 of a state @xmath24 of the rd system near the state @xmath20 on the limit cycle @xmath18 can be linearly approximated as @xmath275 , \label{seq20}\end{aligned}\ ] ] and examine whether this ansatz is reasonable . when @xmath24 is simply a state on the limit cycle @xmath18 with phase @xmath15 , i.e. , @xmath28 , eq . ( [ seq20 ] ) gives @xmath29 . ( [ seq20 ] ) is furthermore valid for arbitrary states @xmath276 sufficiently close to the unperturbed state @xmath20 on @xmath18 , the function @xmath37 will play the role of the phase sensitivity function for the rd system . this actually holds by choosing the function @xmath37 appropriately . see fig . [ figs0rd](b ) for a schematic illustration . we now derive the equation for @xmath37 by generalizing the argument in refs . @xcite for the phase sensitivity function @xmath209 of limit cycles described by odes ( see also appendix a ) . at @xmath233 , we prepare a initial state @xmath277 on the limit cycle @xmath18 with phase @xmath116 , and a slightly perturbed initial state @xmath278 near @xmath279 , where @xmath280 is a small spatial perturbation given to @xmath279 . we evolve these two states without applying further perturbations . linearized equation for @xmath281 can be obtained from eq . ( [ seq14 ] ) as @xmath282 where @xmath34 is a jacobi matrix of @xmath9 estimated at @xmath35 on @xmath18 . from eq . ( [ seq20 ] ) , the phase of the unperturbed state @xmath249 is @xmath283 , and the phase of the perturbed state @xmath284 is given by @xmath285\end{aligned}\ ] ] under the linear approximation . this @xmath231 should also increase with a constant frequency @xmath13 by the definition of the isochron , i.e. , @xmath286 , because no perturbation is given after @xmath233 . therefore , the following equation should hold : @xmath287 \cr & = \left [ \frac{\partial}{\partial t } { \bf q}({\bf r } ; \theta(t)),\ { \bf y}({\bf r } , t ) \right ] + \left [ { \bf q}({\bf r } ; \theta(t)),\ \frac{\partial}{\partial t } { \bf y}({\bf r } , t ) \right ] . \label{seq25}\end{aligned}\ ] ] using eq . ( [ seq23 ] ) , the last term can be transformed as @xmath288 & = \left [ { \bf q}({\bf r } ; \theta(t)),\ { \rm j } ( \theta(t ) ) { \bf y}({\bf r } , t ) + { \rm d } \nabla^{2 } { \bf y}({\bf r } , t ) \right ] \cr & = \left [ { \rm j } ( \theta(t ) ) ^{\dagger } { \bf q}({\bf r } ; \theta(t ) ) + { \rm d}^{\dagger } \nabla^{2 } { \bf q}({\bf r } ; \theta(t)),\ { \bf y}({\bf r } , t ) \right],\end{aligned}\ ] ] where , as usual , partial integration is performed assuming surface terms to vanish or cancel , and @xmath289 denotes matrix transpose . equation ( [ seq25 ] ) now yields @xmath290 = 0,\end{aligned}\ ] ] which should hold for any @xmath291 evolving from arbitrary @xmath280 . therefore , @xmath292 should satisfy the following adjoint equation : @xmath293 or equivalently , by using the relation @xmath294 ( @xmath239 ) , @xmath295 which we presented as eq . ( 3 ) in section ii . since this adjoint equation is linear , we also need to normalize @xmath37 appropriately . as in the ode case , the normalization condition for @xmath37 can be obtained by differentiating the identity @xmath296 by @xmath3 as @xmath297 \cr & = \left [ { \bf q}({\bf r } ; \theta(t)),\ { \bf f}({\bf x}_{0}({\bf r } ; \theta(t ) ) , { \bf r } ) + { \rm d } \nabla^{2 } { \bf x}_{0}({\bf r } ; \theta(t ) ) \right].\end{aligned}\ ] ] therefore , the following normalization condition should be satisfied : @xmath298 . \label{seq30}\end{aligned}\ ] ] note that this condition can also be expressed , using again the relation @xmath294 , as @xmath299 = 1 , \label{seqconst}\end{aligned}\ ] ] which yields the normalization condition eq . ( [ eq : normalization ] ) given in section ii . thus , if the function @xmath37 is the @xmath55-periodic solution to eq . ( [ seq29 ] ) with the constraint ( [ seqconst ] ) , the approximate phase function eq . ( [ seq20 ] ) will satisfy the desired condition eq . ( [ seq25 ] ) in the vicinity of @xmath18 , and @xmath37 will play the role of the phase sensitivity function of the limit - cycle solution @xmath18 of the rd system . note that eqs . ( [ seq29 ] ) and ( [ seqconst ] ) are straightforward generalizations of the conventional adjoint method for the ode , eq . ( [ seq12 ] ) and eq . ( [ seq13 ] ) . generally , the function @xmath37 should be calculated numerically by solving the adjoint eq . ( [ seq29 ] ) with eq . ( [ seqconst ] ) . in numerical calculations , it is useful to integrate the adjoint equation ( [ seq29 ] ) backward in time to avoid numerical overflow with occasional normalization by eq . ( [ seqconst ] ) , as we explained in section iii a. once we obtain the frequency @xmath13 and the phase sensitivity function @xmath37 , we can write down the approximate phase equation ( [ seqp2 ] ) for a slightly perturbed rd system described by eq . ( [ seqp1 ] ) . note that the infinite - dimensional rd system subjected to weak perturbations is reduced to a single one - dimensional phase equation , which drastically simplifies the analysis of weakly perturbed rhythmic spatiotemporal patterns . as a simple example of this phase reduction theory for rd systems , we analyzed synchronization dynamics of a pair of coupled rd systems exhibiting rhythmic patterns in section iii , i.e. , the circulating pulses , oscillating spots , target waves , and rotating spirals of the fitzhugh - nagumo model .

in this paper we prove several results about the lattice of imprimitivity systems of a permutation group containing a cyclic subgroup with at most two orbits . as an application we generalize the first ritt theorem about functional decompositions of polynomials , and some other related results . besides , we discuss examples of rational functions , related to finite subgroups of for which the first ritt theorem fails to be true . epsf.sty [ section ] [ theorem]lemma [ theorem]corollary [ theorem]proposition p v c

0712.3869

let @xmath1 be a rational function with complex coefficients . the function @xmath1 is called _ indecomposable _ if the equality @xmath2 where @xmath3 denotes the superposition @xmath4 of rational functions @xmath5 @xmath6 , implies that at least one of the functions @xmath7 is of degree 1 . a rational function which is not indecomposable is called _ decomposable_. any representation @xmath8 of a rational function @xmath1 in the form f = f_1f_2 f_r , where @xmath9 are rational functions , is called _ a decomposition _ of @xmath10 if all @xmath9 are indecomposable of degree greater than one , then the decomposition @xmath8 is called _ maximal_. two decompositions of a rational function @xmath1 f = u_1u_2 u_k and f = v_1v_2 v_m , maximal or not , are called _ equivalent _ if they have the same length ( that is @xmath11 ) and there exist rational functions of degree one @xmath12 @xmath13 such that @xmath14 in the paper @xcite ritt described the structure of possible maximal decompositions of polynomials . this description can be summarized in the form of two theorems usually called the first and the second ritt theorems ( see @xcite , @xcite ) . the first ritt theorem states that for any two maximal decompositions @xmath15 of a polynomial @xmath1 there exists a chain of maximal decompositions @xmath16 , @xmath17 of @xmath1 such that @xmath18 @xmath19 and @xmath20 is obtained from @xmath21 @xmath22 by replacing two successive functions in @xmath16 by other functions with the same composition . this implies in particular that any two maximal decompositions of a polynomial have the same length . below we will call two maximal decompositions @xmath15 of a rational function @xmath1 such that there exists a chain as above _ weakly equivalent_. this defines an equivalence relation on the set of maximal decompositions of @xmath1 . the first ritt theorem reduces the description of maximal decompositions of polynomials to the description of indecomposable polynomial solutions of the equation ac = bdsuch that the decompositions @xmath23 and @xmath24 are non - equivalent , and the second ritt theorem states if @xmath25 is such a solution then there exist polynomials @xmath26 @xmath27 where @xmath28 such that @xmath29 and up to a possible replacement of @xmath30 by @xmath31 and @xmath32 by @xmath33 either @xmath34 where @xmath35 is a polynomial , @xmath36 @xmath37 and @xmath38 or @xmath39 where @xmath40 are the corresponding chebyshev polynomials , @xmath41 and @xmath42 furthermore , the second ritt theorem remains true for arbitrary polynomial solutions of if to replace equalities @xmath43 by the equalities @xmath44 ( see @xcite , @xcite ) . notice that the classification of polynomial solutions of appears in a variety of different contexts some of which are quite unexpected . for example , this classification is closely related to the problem of description of diophantine equations of the form @xmath45 @xmath46,$ ] having an infinite number of integer solutions ( see @xcite , @xcite ) , and to the problem of description of polynomials @xmath47 satisfying the equality @xmath48 for some compact sets @xmath49 , recently solved in @xcite . notice also that the problem of description of solutions of such that @xmath50 and @xmath51 are polynomials while @xmath52 are allowed to be arbitrary rational ( or even just continuous ) functions on the sphere can be reduced to the description of polynomial solutions ( see @xcite ) . a more detailed account of different results related to the second ritt theorem can be found in the recent papers @xcite , @xcite . the classification of polynomial solutions of essentially reduces to the description of polynomials @xmath52 such that the algebraic curve a(x)-b(y)=0has an irreducible factor of genus zero with one point at infinity . on the other hand , the proof of the first ritt theorem can be given in purely algebraic terms which do not involve the genus condition in any form . indeed , if @xmath53 is the monodromy group of a rational function @xmath1 then equivalence classes of maximal decompositions of @xmath1 are in a one - to - one correspondence with maximal chains of subgroups g_(f)=t_0 < t_1 < ... < t_r = g(f),where @xmath54 is the stabilizer of an element @xmath55 in the group @xmath56 . therefore , any two maximal decompositions of @xmath1 are weakly equivalent if and only if for any two maximal chains of subgroups as above @xmath57 @xmath58 there exists a collection of maximal chains of subgroups @xmath59 , @xmath17 such that @xmath60 @xmath61 and @xmath62 is obtained from @xmath63 @xmath22 by a replacement of exactly one group . it was shown in the paper @xcite ( theorem r.3 ) that the last condition is satisfied for any permutation group @xmath64 containing an abelian transitive subgroup . since the monodromy group of a polynomial always contains a cyclic subgroup with one orbit ( its generator corresponds to the loop around infinity ) , this implies in particular the truth of the first ritt theorem for polynomials . it was also proved in the paper @xcite ( claim 1 ) that if @xmath25 are indecomposable polynomials satisfying such that the decompositions @xmath23 and @xmath24 are non - equivalent then the groups @xmath65 and @xmath66 as well as the groups @xmath67 and @xmath68 are permutation equivalent . since any two maximal decompositions of a polynomial @xmath69 are weakly equivalent , this implies by induction that for any two maximal decompositions of @xmath69 there exists a permutation @xmath70 such that the monodromy groups of @xmath71 and @xmath72 @xmath73 are permutation equivalent ( @xcite ) . the algebraic counterpart of this fact is the following statement : if @xmath74 is a permutation group containing a cyclic subgroup with one orbit then for any two maximal chains @xmath75 the equality @xmath11 holds and there exists a permutation @xmath70 such that the permutation group induced by the action of @xmath76 on cosets of @xmath77 is permutation equivalent to the permutation group induced by the action of @xmath78 on cosets of @xmath79 @xmath80 if a permutation group @xmath64 satisfies this condition , we will say that @xmath64 satisfies the _ jordan - h " older theorem for imprimitivity systems_. in this paper , we extend the above results about the permutation groups @xmath64 containing a cyclic group with one orbit to the permutation groups containing a cyclic subgroup @xmath81 with at most _ two _ orbits and apply these results to rational functions ( or more generally to meromorphic functions on compact riemann surfaces ) the monodromy group of which contains @xmath82 first , we prove that for a permutation group @xmath64 containing @xmath81 the lattice @xmath83 , consisting of subgroups of @xmath64 containing @xmath84 , is lower semi - modular and even a stronger condition of the modularity of @xmath83 holds whenever @xmath83 does not contain a sublattice isomorphic to the subgroup lattice of a dihedral group . it follows easily from the lower semimodularity of @xmath83 that one can pass from any chain of subgroups to any other such a chain by a sequence of replacements as above and therefore the first ritt theorem extends to rational functions the monodromy group of which contains @xmath81 . notice that this implies in particular that the first ritt theorem holds for rational functions with at most two poles . although for such functions the result was know previously ( see @xcite , @xcite , @xcite ) the algebraic proof turns out to be more simple and illuminating . notice also that our description of the lattice @xmath83 for groups @xmath64 containing @xmath81 has an interesting connection with the problem of description of algebraic curves having a factor of genus zero with at most two points at infinity , studied in @xcite , @xcite . further , we prove that if a permutation group @xmath64 contains a cyclic subgroup with two orbits of _ different length _ then the lattice @xmath83 is always modular and @xmath64 satisfies the jordan - h " older theorem for imprimitivity systems . this implies in particular that if @xmath1 is a rational function which has only two poles and the orders of these poles are different between themselves then any two maximal decompositions of @xmath1 have the same length and there exists a permutation @xmath85 such that the monodromy groups of @xmath71 and @xmath72 @xmath86 are permutation equivalent . we also show that the jordan - h " older theorem for imprimitivity systems holds for any permutation group containing a transitive hamiltonian subgroup that generalizes the corresponding results of @xcite , @xcite . for arbitrary rational functions the first ritt theorem fails to be true . the simplest counterexamples are provided by the functions which are regular coverings of the sphere ( that is for which @xmath87 ) with the monodromy group @xmath88 @xmath89 , or @xmath90 these functions were described for the first time by f. klein in @xcite and nowadays can be interpreted as belyi functions of platonic solids ( see @xcite , @xcite ) . for such a function its maximal decompositions simply correspond to maximal chains of subgroups in its monodromy group . therefore , since any of the groups @xmath91 , @xmath92 @xmath93 has maximal chains of subgroups of different length , for the corresponding klein functions the first ritt theorem is not true . although the fact that the klein functions provide counterexamples to the first ritt theorem is a well known part of the mathematical `` folklore '' , the systematic description of compositional properties of these functions seems to be absent . in particular , to our best knowledge maximal decompositions which do not satisfy the first ritt theorem were found explicitly only for the klein function corresponding to the group @xmath91 ( see @xcite , @xcite ) . in the appendix to this paper we provide a detailed analysis of maximal decompositions of the klein functions and give related explicit examples of non weakly equivalent maximal decompositions . in particular , we give an example of a rational function with _ three _ poles having maximal decompositions of different length . this example shows that with no additional assumptions the first ritt theorem can not be extended to rational functions the monodromy of which contains a cyclic subgroup with more than two orbits . 0.2 cm * acknowledgments*. the authors would like to thank i. kovacs , p. mller , u. zannier , and a. zvonkin for discussions of different questions related to the subject of this paper . recall that _ a lattice _ is a partially ordered set in which every pair of elements @xmath94 has a unique supremum @xmath95 and an infimum @xmath96 ( see e.g. @xcite ) . our basic example of a lattice is a lattice @xmath97 of all subgroups of a group @xmath98 where by definition @xmath99 if @xmath100 is a subgroup of @xmath101 ( clearly , @xmath102 is an infimum of @xmath103 and @xmath104 is a supremum ) . a simplest example of the lattice @xmath97 is obtained if @xmath64 is a cyclic group of order @xmath105 . in this case @xmath97 is isomorphic to the lattice @xmath106 consisting of all divisors of @xmath105 , where by definition @xmath107 if @xmath108 a _ sublattice _ of a lattice @xmath109 is a non - empty subset @xmath110 closed with respect to @xmath111 and @xmath112 . for example , for any subgroup @xmath81 of a group @xmath64 the set @xmath113 is a sublattice of @xmath97 . another example of a sublattice of @xmath97 is the lattice @xmath114 ( notice that in our notation @xmath115 means that @xmath116 is a subgroup of @xmath64 while @xmath117 means that @xmath116 is a subset of the set @xmath118 which in general is not supposed to be a group ) . recall that by the dedekind identity ( see e.g. @xcite , p. 8) for arbitrary subgroups @xmath119 of a group @xmath64 such that @xmath120 the equality @xmath121 holds . it follows from the dedekind identity that the mapping @xmath122 is a monomorphism from the lattice @xmath123 into the lattice @xmath124 with the image consisting of all subgroups of @xmath125 which are permutable with @xmath126 . we will call @xmath127 the dedekind monomorphism . for elements @xmath128 of a lattice @xmath109 the symbol @xmath129 denotes that @xmath130 and there exists no element @xmath131 of @xmath109 such that @xmath132 a lattice @xmath109 is called _ semimodular _ if for any @xmath133 the condition ab < a , ab < b , imply the condition b < ab , a < ab.if vice versa condition implies condition , the lattice @xmath109 is called _ lower semimodular_. a lattice @xmath109 is called _ modular _ if @xmath109 is semimodular and lower semimodular . _ a maximal chain _ @xmath134 between elements @xmath128 of @xmath109 is a collection @xmath135 of elements of @xmath109 such that @xmath136 the number @xmath137 is called the length of the chain @xmath134 ( we always assume that in the lattices considered the length of a chain between @xmath138 and @xmath139 is uniformly bounded by a number depending on @xmath138 and @xmath139 only ) . it is well known ( see e.g. @xcite ) that for a semimodular or lower semimodular lattice all maximal chains between two elements have the same length . below , using essentially the same proof , we give a modification of this statement in the spirit of the first ritt theorem . say that two maximal chains between elements @xmath138 and @xmath139 of a lattice @xmath109 are @xmath140-equivalent if there exists a sequence of maximal chains @xmath141 between @xmath128 such that @xmath142 @xmath143 and @xmath144 is obtained from @xmath145 @xmath22 by a replacement of exactly one element . clearly , all @xmath140-equivalent chains have an equal length . let @xmath109 be a semimodular or lower semimodular lattice . then any two maximal chains between any elements @xmath138 and @xmath139 of @xmath109 are @xmath140-equivalent . since after the inversion of the ordering of a lattice the condition of semi - modularity transforms to the condition of lower semi - modularity and vice versa , it is enough to prove the theorem for lower semi - modular lattices . fix @xmath146 . for arbitrary @xmath147 denote by @xmath148 a maximum of lengths of maximal chains between @xmath138 and @xmath139 . we will prove the theorem by induction on @xmath149 for @xmath139 satisfying @xmath150 the theorem is obviously true . suppose that the theorem is proved for @xmath139 satisfying @xmath151 and let @xmath152 be two maximal chains between @xmath138 and an element @xmath147 such that @xmath153 if @xmath154 , then we are done by induction . so , we may assume that @xmath155 . then by the maximality of @xmath156 and @xmath157 in @xmath139 we conclude @xmath158 . hence @xmath159 and therefore by the lower semi - modularity of @xmath109 we have : a_k_1 - 1 b_k_2 - 1 < a_k_1 - 1 , a_k_1 - 1 b_k_2 - 1 < b_k_2 - 1.let @xmath160 be any maximal chain between @xmath138 and @xmath161 and a = c_0 < c_2 < < c_l < a_k_1 - 1be its extension to a maximal chain between @xmath138 and @xmath156 . since @xmath162 is obviously less than @xmath148 , it follows from the induction assumption that the chain @xmath163 obtained from @xmath164 by deleting @xmath165 is @xmath140-equivalent to the chain . therefore , the chain @xmath164 and the chain a = c_0 < c_2 < < c_l < a_k_1 - 1 < b also are @xmath140-equivalent . similarly , the chain @xmath166 is @xmath140-equivalent to the chain a = c_0 < c_2 < < c_l < b_k_2 - 1 < b.since chains and are @xmath140-equivalent , we conclude that the chain @xmath164 is @xmath140-equivalent to the chain @xmath166 . notice that there exist lattices which are not semimodular or lower semimodular such that any two maximal chains between any elements are @xmath140-equivalent . an example of such a lattice is shown on fig . 0.2 cm = 1.8truecm let @xmath167 be a finite set and @xmath168 be a transitive permutation group . recall that a partition @xmath169 of @xmath167 is called an _ imprimitivity system _ of @xmath64 if @xmath169 is @xmath64-invariant . elements of @xmath169 are called _ blocks_. for a point @xmath170 we will denote by @xmath171 a unique block of @xmath169 which contains @xmath172 since the group @xmath64 permutes the elements of @xmath169 transitively , all blocks of @xmath169 have the same cardinality denoted by @xmath173 . denote by @xmath174 the set of all imprimitivity systems of @xmath64 . it is a partially ordered set , where by definition @xmath175 if @xmath176 is a refinement of @xmath177 . notice that if @xmath175 then @xmath178 is an integer denoted by @xmath179 $ ] . it is easy to see that @xmath174 is a lattice where the lattice operations are defined as follows @xmath180 @xmath181 it is well known that the lattice @xmath174 is isomorphic to the subgroup lattice @xmath182 where @xmath170 is an arbitrary fixed point . the correspondence between two sets is given by the formula @xmath183 , where @xmath184 vice versa , an imprimitivity system corresponding to a subgroup @xmath185 is defined as follows @xmath186 . notice that for any @xmath187 we have : @xmath188 moreover , if @xmath175 then @xmath179=[g_{{\f f}(\omega)}:g_{{\f e}(\omega)}].$ ] if a group @xmath64 is the monodromy group of a rational function @xmath1 , then imprimitivity systems of @xmath64 are in a one - to - one correspondence with equivalence classes of decompositions @xmath189 of @xmath1 . namely , suppose that @xmath64 is realized as a permutation group acting on the set @xmath190 of preimages of a non critical value @xmath191 of @xmath192 under the map @xmath193 , and let @xmath194 be the set of preimages of @xmath191 under the map @xmath195 . then blocks of the imprimitivity system of @xmath64 corresponding to the equivalence class of decompositions of @xmath1 containing @xmath189 , are just preimages of the points @xmath194 under the map @xmath196 . more generally , equivalence classes of decompositions of a rational function @xmath1 are in a one - to - one correspondence with chains of subgroups @xmath197 where @xmath64 is the monodromy group of @xmath1 . following @xcite we say that two maximal decompositions @xmath198 of a rational function @xmath1 are weakly equivalent if there exists a chain of maximal decompositions @xmath16 , @xmath17 of @xmath1 such that @xmath199 @xmath200 , and @xmath20 is obtained from @xmath21 @xmath22 by replacing two successive functions in @xmath16 by other functions with the same composition . the remarks above imply that two maximal decompositions of @xmath1 are weakly equivalent if and only if corresponding maximal chains in @xmath83 are @xmath140-equivalent . in particular , the conclusion of the first ritt theorem is true for a rational function @xmath1 if and only if all maximal chains between @xmath84 and @xmath64 in @xmath83 are @xmath140-equivalent . therefore , theorem [ ritt ] implies the following corollary ( cf . @xcite , th . let @xmath1 be a rational function such that the lattice @xmath201 where @xmath64 is the monodromy group of @xmath1 , is semi - modular or lower semi - modular . then all maximal decompositions of @xmath1 are weakly equivalent . the corollary [ df ] shows that the groups @xmath64 for which @xmath83 is semi - modular or lower semi - modular are of special interest for factorization theory of rational functions . the simplest examples of such groups are groups containing a transitive cyclic subgroup . let @xmath202 be a permutation group containing a transitive cyclic subgroup @xmath203 . then the lattice @xmath204 is a modular lattice isomorphic to a sublattice of the lattice @xmath205 since any sublattice of a modular lattice is modular ( see e.g. @xcite ) and it is easy to see that @xmath106 is modular , it is enough to prove that @xmath204 is isomorphic to a sublattice of @xmath205 the transitivity of @xmath203 implies that @xmath206 . therefore , the dedekind monomorphism @xmath207 maps @xmath204 into a sublattice of @xmath208 . on the other hand , @xmath209 0.2 cm note that theorem [ cyc ] implies the following proposition ( cf . @xcite , @xcite ) . let @xmath25 be polynomials such that @xmath210 then there exist polynomials @xmath211 where @xmath212 such that @xmath213 and @xmath214 in particular , if @xmath215 then the decompositions @xmath23 and @xmath24 are necessarily equivalent . let as above @xmath64 be a transitive permutation group . it is easy to see that if @xmath216 is a normal subgroup of @xmath64 then its orbits form an imprimitivity system of @xmath64 . such an imprimitivity system is called _ normal _ and is denoted by @xmath217 . for an imprimitivity system @xmath218 set @xmath219 notice that each block of @xmath176 is a union of @xmath220-orbits and @xmath221 in particular , @xmath220 is a normal subgroup of @xmath64 . let us call a subgroup @xmath222 _ core - complementary _ if @xmath223 . an imprimitivity system @xmath218 is normal if and only if the group @xmath224 is core - complementary . indeed , if g_e()=g_core_g(g_e())= g_g_e then @xmath225 and hence @xmath220 acts transitively on @xmath226 . since @xmath227 , this implies that @xmath220 acts transitively on every block of @xmath176 . thus blocks of @xmath176 are orbits of the normal subgroup @xmath220 . vice versa , if @xmath176 is normal then @xmath228 for some @xmath229 . this implies that @xmath230 and @xmath231 . it follows now from @xmath232 that equality holds . recall that two subgroups @xmath126 and @xmath125 are called _ permutable _ if @xmath233 , or , equivalently , @xmath234 . recall also that if @xmath126 and @xmath125 are subgroups of finite index of @xmath64 then the inequality @xmath235\geq\left[a : a\cap b\right]\ ] ] holds and the equality in attains if and only if @xmath52 are permutable ( see e.g. @xcite , p. 79 ) . denote by @xmath236 the subset of @xmath237 consisting of all core - complementary subgroups . notice that in general @xmath236 is _ not _ a sublattice of @xmath237 the following conditions hold : 1 . if @xmath238 , then @xmath233 for each @xmath239 ; 2 . if @xmath240 , then @xmath241 . \(a ) in order to lighten the notation set @xmath242 . in view of proposition [ normality ] we have : @xmath243 ( b ) set @xmath244 . since @xmath245 and @xmath246 , we have : @xmath247 it follows now from proposition [ normality ] that @xmath248 therefore , @xmath249 and hence @xmath250 by proposition [ normality ] . let @xmath251 be permutable subgroups . if @xmath252 is maximal in @xmath126 and @xmath125 , then @xmath126 and @xmath125 are maximal in @xmath234 . let @xmath253 be a subgroup of @xmath64 satisfying @xmath254 . it follows from @xmath255 that either @xmath256 or @xmath257 . it follows now from the dedekind identity @xmath258 that in the first case @xmath259 while in the second one @xmath260 . if any two subgroups of @xmath237 are permutable , then the lattice @xmath237 is modular . indeed , if @xmath252 is maximal in @xmath126 and @xmath125 then @xmath126 and @xmath125 are maximal in @xmath234 by proposition [ lp ] . suppose now that @xmath126 and @xmath125 are maximal in @xmath118 and let @xmath253 be a subgroup of @xmath64 satisfying @xmath261 . then @xmath262 implies that either @xmath263 or @xmath264 . if @xmath263 , then @xmath265 and therefore @xmath266 on the other hand , if @xmath264 then it follows from @xmath267 that for any @xmath268 there exist @xmath269 and @xmath270 such that @xmath271 . since the last equality yields that @xmath272 , this implies that @xmath273 and hence @xmath274 let @xmath275 be an arbitrary subgroup and @xmath276 . denote by @xmath277 a permutation group arising from the natural action of @xmath64 on @xmath278 . thus @xmath277 is always considered as a subgroup of @xmath279 . notice that if @xmath229 is contained in @xmath81 , then the groups @xmath277 and @xmath280 are permutation equivalent . below we will denote permutation equivalence by @xmath281 . say that a transitive permutation group @xmath282 satisfies the jordan - h " older theorem for imprimitivity systems if any two maximal chains @xmath283 of the lattice @xmath237 have the same length ( that is @xmath11 ) and there exists a permutation @xmath70 such that the permutation groups @xmath284 and @xmath285 @xmath73 are permutation equivalent . notice that if @xmath64 is the monodromy group of a rational function @xmath1 then it follows from the correspondence between imprimitivity systems of @xmath64 and equivalence classes of decompositions of @xmath1 that @xmath64 satisfies the jordan - h " older theorem for imprimitivity systems if and only if any two maximal decompositions of @xmath1 @xmath286 have the same length and the there exists a permutation @xmath70 such that the monodromy groups of @xmath71 and @xmath72 @xmath73 are permutation equivalent . let @xmath64 be a permutation group such that @xmath287 then the lattice @xmath237 is modular and @xmath64 satisfies the jordan - h " older theorem for imprimitivity systems . first of all observe that since by proposition [ properties ] any two subgroups of @xmath237 are permutable it follows from proposition [ modular ] that @xmath237 is a modular lattice . let now @xmath288 be two maximal chains of @xmath237 . since @xmath182 is a modular lattice , it follows from theorem [ ritt ] that @xmath11 and @xmath289 and @xmath290 are @xmath140-equivalent . therefore by induction it is sufficient to prove the theorem for the case when @xmath291 and @xmath292 differs at exactly one place , say @xmath293 ( @xmath294 ) . clearly , in this case we have : @xmath295 in order to lighten the notation set @xmath296 it follows from the equality @xmath297 that @xmath298 . therefore , @xmath299 and hence @xmath300 and @xmath301 since @xmath302 and @xmath303 this implies that @xmath304 by the second isomorphism theorem the group @xmath305 is isomorphic to the group @xmath306 and the image of @xmath307 under this isomorphism is @xmath308 furthermore , it follows from @xmath309 that @xmath310 . therefore , @xmath311 and hence @xmath312 finally , since @xmath313 , @xmath314 and hence @xmath315 replacing @xmath126 and @xmath125 in the above argument we obtain similarly that @xmath316 . recall that a group is called _ hamiltonian _ if all its subgroups are normal . let @xmath64 be a permutation group containing a transitive hamiltonian subgroup @xmath317 . then @xmath237 is a modular lattice isomorphic to a sublattice of @xmath318 and @xmath64 satisfies the jordan - h " older theorem for imprimitivity systems . it follows from the transitivity of @xmath317 that @xmath319 . therefore , by the dedekind monomorphism @xmath237 is isomorphic to a sublattice of @xmath320 clearly , @xmath321 furthermore , since @xmath317 is hamiltonian , the subgroup @xmath322 is normal in @xmath317 and therefore for any @xmath323 the equality @xmath324 holds . this implies that @xmath325 and hence @xmath182 is isomorphic to a sublattice of @xmath318 . since @xmath318 is modular by proposition [ modular ] , @xmath182 is modular as well . by theorem [ jh ] in order to prove that @xmath64 satisfies the jordan - h " older theorem it is enough to show that @xmath326 . observe first that it follows from @xmath327 that for arbitrary @xmath222 the equality core_g(a)=_gk gag^-1holds . on the other hand , since @xmath317 is hamiltonian , @xmath328 . therefore , for each @xmath329 we have : @xmath330 implying ak gag^-1.it follows now from and that @xmath331 and hence @xmath332 since by dedekind s identity @xmath333 , we conclude that @xmath334 for any @xmath222 and hence @xmath326 by proposition [ normality ] . let @xmath1 be a rational function such that its monodromy group contains a transitive hamiltonian subgroup . then any two maximal decompositions of @xmath1 are weakly equivalent . furthermore , for any two decompositions of @xmath1 : @xmath335 there exists a permutation @xmath70 such that the monodromy groups of @xmath71 and @xmath72 @xmath73 are permutation equivalent . notice that the condition of corollary [ ccoo ] is satisfied in particular if @xmath317 is cyclic or abelian . therefore , corollary [ ccoo ] generalizes theorem r.3 and claim 1 of @xcite , and theorem 1.3 of @xcite . 0.2 cm let @xmath167 be a finite set , @xmath336 be a permutation which is a product of exactly two disjointed cycles , and @xmath337 . for the rest of this subsection it is assumed that @xmath282 is a transitive permutation group containing @xmath81 . without loss of generality we may assume that @xmath202 and @xmath338 where @xmath339 @xmath340 say that an imprimitivity system @xmath218 is @xmath81-transitive ( resp . @xmath81-intransitive ) if the action of @xmath81 on blocks of @xmath176 is transitive ( resp . intransitive ) . say that a group @xmath341 is @xmath81-transitive ( resp . @xmath81-intransitive ) if the corresponding @xmath342 is @xmath81-transitive ( resp . @xmath81-intransitive ) . since @xmath81 permutes blocks of @xmath169 , it is easy to see that if @xmath176 is @xmath81-transitive then there exist numbers @xmath343 and @xmath344 @xmath345 such that any block of @xmath169 is equal to @xmath346 where the symbol @xmath347 ( resp . @xmath348 ) denotes a union of numbers from the segment @xmath349 $ ] ( resp . from the segment @xmath350 $ ] ) equal to @xmath351 by modulo @xmath352 on the other hand , if @xmath218 is @xmath81-intransitive then there exist numbers @xmath353 and @xmath354 @xmath355 @xmath356 such that n_1/d_1=n_2/d_2=n_e and any block of @xmath169 is equal either to @xmath357 or to @xmath358 . any @xmath81-intransitive imprimitivity system @xmath218 is normal . in the notation above set @xmath359 and @xmath360 . clearly , we have @xmath361 and therefore any orbit of @xmath220 is a union of orbits of @xmath362 the length of any orbit of @xmath317 on @xmath349 $ ] is equal to @xmath363 on the other hand , the length of any orbit of @xmath317 on @xmath350 $ ] is equal to @xmath364 therefore , the length of any orbit of @xmath220 on @xmath167 is divisible by @xmath365 this implies that orbits of @xmath220 coincide with blocks of @xmath176 and hence @xmath169 is normal . if @xmath81-transitive imprimitivity system @xmath218 is not normal , then @xmath366 and there exists a normal imprimitivity system @xmath367 such that @xmath368=2 $ ] . furthermore , @xmath369 is @xmath81-intransitive , its blocks coincide with the orbits of @xmath220 , and for any @xmath81-intransitive imprimitivity system @xmath370 such that @xmath371 we have @xmath372 . in the notation above set @xmath373 clearly , any block @xmath374 of @xmath176 is a union of exactly two orbits of @xmath317 and @xmath375 since @xmath176 is not normal , this implies that orbits of @xmath220 coincide with orbits of @xmath317 . in particular , since orbits of @xmath220 have the same length the same is true for orbits of @xmath317 and hence @xmath376 the rest statements of the proposition are now obvious . if a transitive permutation group @xmath64 contains a cyclic subgroup with two orbits of different length , then @xmath237 is modular and @xmath64 satisfies the jordan - h " older theorem for imprimitivity systems . it follows from propositions [ normal ] and [ normal1 ] that @xmath377 . now the theorem follows from theorem [ jh ] . let @xmath1 be a rational function such that @xmath1 has only two poles and the orders of these poles are different between themselves . then any two maximal decompositions of @xmath1 are weakly equivalent . furthermore , for any two decompositions of @xmath1 : @xmath335 there exists a permutation @xmath70 such that the monodromy groups of @xmath71 and @xmath72 @xmath73 are permutation equivalent . let @xmath64 be a transitive permutation group . suppose that @xmath237 contains subgroups @xmath378 such that @xmath379 = [ f : e\cap f]=2.$ ] then @xmath380 is normal in @xmath381 and @xmath382 where @xmath383 $ ] . furthermore , @xmath384 since @xmath379 = [ f : e\cap f]=2 $ ] , the subgroup @xmath380 is normal in @xmath385 and @xmath1 simultaneously and therefore @xmath386 since @xmath387 and @xmath388 , @xmath389 , the group @xmath390 is isomorphic to @xmath391 for some @xmath392 ( see e.g. @xcite ) . furthermore , since @xmath393=|{\langle{e , f}\rangle}/(e\cap f)|\ ] ] we have @xmath394=2m$ ] . finally , it is clear that @xmath395 and therefore @xmath384 in the rest of this subsection it is assumed that @xmath282 is a transitive permutation group containing @xmath81 . the lattice @xmath182 is lower semimodular . assume the contrary and let @xmath396 be a subgroup of @xmath64 such that ef < e_1<e , where @xmath397 @xmath398 are maximal in @xmath381 . notice that then @xmath399 if @xmath400 is permutable with @xmath1 , then by @xmath401= \left[e_1:e_1\cap f\right]=\left[e_1:e\cap f\right ] < \left[e : e\cap f\right]\leq \left[{\langle{e , f}\rangle}:f\right].\ ] ] therefore , @xmath402 . since @xmath403 and @xmath1 is maximal in @xmath381 , this implies that @xmath404 hence , @xmath405 and therefore @xmath406 in contradiction with the assumption that @xmath407 suppose now that @xmath1 and @xmath400 are not permutable . then proposition [ properties ] implies that both @xmath400 and @xmath1 are not core - complementary . it follows now from propositions [ normality ] and [ normal1 ] that there exist @xmath408 such that @xmath409=[f : f']=2 $ ] . notice that each of the groups @xmath410 and @xmath411 is permutable with any @xmath412 by proposition [ properties ] . in particular , @xmath413 and @xmath414 . it follows from @xmath415 that either @xmath416 or @xmath417 . if @xmath416 , then the inclusions @xmath418 imply that @xmath419 in contradiction with the assumption that @xmath400 and @xmath1 are not permutable . so , assume that @xmath417 . in this case @xmath420 and hence @xmath421 together with @xmath422 and @xmath423=2 $ ] this implies that e_1=ef = e_1f.in view of proposition [ normal1 ] the last equality yields that @xmath380 is @xmath81-intransitive and @xmath424 . consequently , ef = ef. it follows from @xmath425 that either @xmath426 or @xmath427 . if the equality @xmath426 holds , then and imply the inequality @xmath428\leq [ { \langle{e , f}\rangle}:e]=[ef':e ] = [ f ' : e\cap f ' ] = [ f':e\cap f ] = \frac{1}{2 } [ f : e\cap f]\ ] ] which is impossible . so , assume that @xmath427 . in this case @xmath429 and therefore @xmath430 . together with @xmath431=2 $ ] this implies that either @xmath432 or @xmath433 . furthermore , since in view of the maximality of @xmath1 and @xmath385 in @xmath434 the equality @xmath435 is impossible , we may assume that @xmath436 . in this case @xmath437=2 $ ] . together with and @xmath423 = 2 $ ] this implies that [ f : e_1f]=[e_1:e_1f]=2 . it follows now from proposition [ jk ] that the lattice @xmath438 is isomorphic to the subgroup lattice of a dihedral group @xmath391 , where @xmath439 $ ] . furthermore , it follows from @xmath440 that either @xmath441 or @xmath442 . the first case is impossible since @xmath443 . therefore @xmath442 and hence l(ef , ) = l(e_1f,)l(d_2 m ) . since maximal subgroups of @xmath391 have prime index , it follows from that the number @xmath444 $ ] is prime and hence @xmath445= [ { \langle{e , f}\rangle}:f][f : e\cap f]=2p.\ ] ] on the other hand , by @xmath445=[{\langle{e , f}\rangle}:e][e : e_1][e_1:e\cap f ] = 2[{\langle{e , f}\rangle}:e][e : e_1].\ ] ] therefore , @xmath446[e : e_1]=p.$ ] since this equality implies that at least one of the numbers @xmath446,$ ] @xmath447 $ ] is equal to one , we conclude that there exists no @xmath396 satisfying and therefore the lattice @xmath182 is lower semimodular . let @xmath448 @xmath449 . suppose that @xmath380 is maximal in @xmath385 , @xmath1 . then either @xmath385 and @xmath1 are permutable and @xmath385 , @xmath1 are maximal in @xmath434 , or @xmath450 and @xmath451 for some @xmath392 . furthermore , @xmath384 if @xmath385 and @xmath1 are permutable , then @xmath385 and @xmath1 are maximal in @xmath452 by proposition [ lp ] . so , suppose that @xmath385 and @xmath1 are not permutable and consider the core - complementary subgroups @xmath453 @xmath454 from proposition [ normal1 ] . it follows from @xmath455 that either @xmath456 or @xmath457 . in the first case we obtain @xmath458 that contradicts to the assumption that @xmath385 and @xmath1 are not permutable . therefore @xmath459 , or , equivalently , @xmath460 . since @xmath461=2 $ ] , this implies that @xmath462 . analogously , @xmath436 . thus @xmath463 = [ f : e\cap f ] = 2.\ ] ] now proposition [ jk ] yields the result . let @xmath464 be maximal in @xmath381 . then @xmath380 is maximal in @xmath385 and @xmath1 and either @xmath465 , or @xmath450 and @xmath451 for a prime @xmath466 . by proposition [ lp- ] the group @xmath380 is maximal in @xmath1 and @xmath385 . if @xmath385 and @xmath1 are not permutable , then proposition [ fin ] implies that @xmath450 and @xmath451 for some @xmath392 . furthermore , since @xmath1 is maximal in @xmath381 the group @xmath467 is maximal in the group @xmath468 and therefore @xmath466 is prime . we can summarize propositions [ lp- ] and [ fin ] as follows . let @xmath64 be a transitive permutation group containing a cyclic subgroup with two orbits . then the lattice @xmath237 is lower semi - modular . furthermore , @xmath237 is modular unless there exists an interval of @xmath237 which is isomorphic to the subgroup lattice of a dihedral group . by proposition [ lp- ] the lattice @xmath237 is lower semi - modular . if it is not modular , then existence of an interval isomorphic to @xmath469 follows from proposition [ fin ] . let @xmath1 be a rational function such that its monodromy group contains a cyclic subgroup with at most two orbits . then any two maximal decompositions of @xmath1 are weakly equivalent . furthermore , if @xmath470 are two decompositions of @xmath1 then the set of degrees of the functions @xmath471 @xmath73 coincides with the set of degrees of the functions @xmath472 @xmath80 the first part of corollary follows from theorem [ bi ] and corollary [ df ] . furthermore , it follows from the first part that in order to prove the second part it is enough to establish that if @xmath52 are subgroups of @xmath64 such that @xmath252 is maximal in @xmath52 , and @xmath52 are maximal in @xmath473 then the sets @xmath474,[b : a\cap b]\}$ ] and @xmath475,[a : a\cap b]\}$ ] coincide . if @xmath126 and @xmath125 are permutable , then this is a corollary of formula . on the other hand , if @xmath126 and @xmath125 are not permutable then the property needed easily follows from corollary [ gru ] . 0.2 cm * remark . * the proof of theorem [ bi ] given above is a simplified version of the proof given in the earlier preprint of the authors @xcite . notice that corollary [ bi2 ] and a weaker version of corollary [ gru ] were also independently proved in the preprint @xcite appeared shortly after @xcite . the following result is the algebraic counterpart of proposition 2 in @xcite ( see also theorem 8.1 of @xcite and theorem 3.5 of @xcite ) . let @xmath64 be a group and @xmath52 be non - permutable subgroups of @xmath64 . then there exist non - permutable subgroups @xmath477 of @xmath64 such that @xmath478 @xmath479 and @xmath480 for @xmath481 denote by @xmath482 a maximal number such that there exists a maximal chain of subgroups @xmath483 we use the induction on the number @xmath484 . in order to lighten notation set @xmath485 @xmath486 first of all notice that the subgroups @xmath487 and @xmath488 are not permutable since @xmath489 in particular , @xmath490 and @xmath491 so , if @xmath492 ( that is if both @xmath126 and @xmath125 are maximal in @xmath64 ) , then @xmath493 , @xmath494 , and hence @xmath495 and @xmath496 . since @xmath497 and @xmath229 , this imples that @xmath498 and @xmath499 , and hence @xmath500 therefore , if @xmath492 we can set @xmath501 @xmath502 . assume now that @xmath503 . if @xmath504 or @xmath505 , then the proposition follows from the induction assumption . on the other hand , if @xmath506 and @xmath507 , then as above @xmath493 , @xmath494 , and @xmath500 therefore , we can set @xmath508 . proposition [ fri ] together with previous results allows us to describe non - permutable subgroups of @xmath237 . let @xmath64 be a transitive permutation group containing a cyclic subgroup with two orbits and @xmath464 be non - permutable subgroups of @xmath64 such that @xmath509 . then there exists @xmath229 such that @xmath510 and @xmath511 for some @xmath392 . by proposition [ fri ] there exist non - permutable subgroups @xmath512 of @xmath64 such that @xmath513 @xmath514 and @xmath515 furthermore , proposition [ properties ] implies that both @xmath516 and @xmath517 are not core - complementary . therefore , by propositions [ normal ] and [ normal1 ] [ e:]= 2 , [ f:]= 2,where @xmath518 and @xmath519 . since @xmath520 , we obtain @xmath521 . on the other hand , the inequality @xmath522 implies that @xmath523 is a proper subgroup of both @xmath524 and @xmath525 . it follows now from that @xmath526 and @xmath527 = [ \hat{f}:\hat{f}\cap\hat{f } ] = 2 $ ] . therefore , the theorem follows from proposition [ jk ] taking into account that @xmath528 . theorem [ gru0 ] has an interesting connection with the problem of description of algebraic curves a(x)-b(y)=0having a factor of genus zero with at most two points at infinity . this problem is closely related to the number theory and in this context was studied in the papers @xcite , @xcite . in particular , in @xcite a complete classification of such curves ( defined over any field @xmath137 of characteristic zero ) was obtained . another proof of this classification ( over @xmath529 ) was given in the paper @xcite in the context of description of double decompositions @xmath530 of rational functions @xmath109 , with at most two poles , into compositions of rational functions . the last problem turns out to be more general than the previous one since if curve has an irreducible factor of genus zero with two points at infinity then this factor may be parametrized by some laurent polynomials and therefore there exist laurent polynomials @xmath531 such that the equality l = al_1=bl_2holds . the both proofs of the classification of curves having a factor of genus @xmath532 with at most two points at infinity split into two parts : the first one is the analysis of the condition that , under the assumption that is irreducible , the genus of is zero , and the second one is the reduction of the general case to the case when is irreducible . the first part essentially consists of a straightforward although highly laborious analysis of the formula which calculates the genus of via the branching data of @xmath126 and @xmath125 , while the second part requires some more sophisticated considerations . denote by @xmath64 the monodromy group of @xmath109 and let @xmath533 , @xmath534 be subgroups of @xmath237 corresponding to decompositions . then the condition that is reducible is equivalent to the condition that @xmath535 therefore , theorem [ gru0 ] can be viewed as an algebraic counterpart of the portion of the discussed classification related to the reducible case , and implies easily the corresponding result ( cf . @xcite , theorem 9.3 and @xcite , theorem 7.3 ) . suppose that curve is reducible and has a factor of genus zero with at most two points at infinity . then there exist polynomials @xmath536 @xmath537 @xmath538 @xmath539 where @xmath540 such that a = r a , b = rband either the curve @xmath541 is irreducible , or a=- t_lr , b = t_ls , where @xmath542 are the corresponding chebyshev polynomials with @xmath543 , @xmath544 and @xmath545 without loss of generality we may assume that there exists no polynomial @xmath536 @xmath546 such that holds for some polynomials @xmath547 , or equivalently that @xmath548 if curve is irreducible , then there is nothing to prove so assume that is reducible . in this case @xmath549 @xmath550 are not polynomials since otherwise corollary [ eng ] and the assumption about solutions of imply the equality @xmath551 which in its turn implies easily the irreducibility of curve . therefore , the cyclic subgroup @xmath81 of @xmath64 generated by the permutation corresponding to a loop around infinity has two orbits . it follows now from theorem [ gru0 ] that there exists @xmath229 such that @xmath552 and @xmath511 for some @xmath392 . furthermore , since @xmath229 the action of @xmath64 on cosets of @xmath216 is regular . therefore , @xmath553 and hence there exists a decomposition @xmath554 of @xmath109 such that the monodromy group of @xmath555 is a regular covering of the sphere with the dihedral monodromy group . by the well known classification of regular coverings of the sphere which goes back to klein ( see @xcite and the appendix below ) this implies that @xmath556 where @xmath557 are automorphisms of the sphere . clearly , without loss of generality we may assume that @xmath558 . furthermore , since @xmath109 has poles only at the points @xmath532 and @xmath559 it follows from @xmath554 that @xmath560 for some @xmath561 and @xmath562 . therefore , l=(z^mn+)(cz)and @xmath563 now the proposition follows easily from the description of possible double decompositions of function . in this appendix we describe the structure of maximal decompositions of rational functions which are regular coverings of the sphere that is of the functions for which @xmath564 these functions , appearing in a variety of different contexts from differential equations to galois theory , were first described by klein in @xcite . for such a function @xmath127 its monodromy group @xmath64 is isomorphic to its automorphism group and therefore is isomorphic to a finite subgroup of @xmath565 any such a subgroup is isomorphic to one of the groups @xmath566 @xmath567 @xmath88 @xmath92 @xmath93 and the corresponding function @xmath127 is defined by its group up to a composition @xmath568 where @xmath569 @xmath570 . the klein functions provide the simplest examples of rational functions for which the first ritt theorem fails to be true . indeed , if @xmath127 is a klein function then its maximal decompositions correspond to maximal chains of subgroups of its monodromy group @xmath571 therefore , in order to find counterexamples to the first ritt theorem it is enough to find non @xmath140-equivalent maximal chains of subgroups of @xmath64 . for the groups @xmath203 and @xmath572 such chains do not exist while for the groups @xmath88 @xmath92 @xmath93 they do . for example , it is easy to see that e < c_2 < v_4 < a_4 , e < c_3 < a_4,where @xmath573 ( resp . @xmath574 ) is a cyclic group of order 2 ( resp . 3 ) and @xmath575 is the klein four group , are maximal chains of different length in @xmath91 and therefore for the corresponding klein function the first ritt theorem fails to be true . the fact that the first ritt theorem is not true for arbitrary rational functions was observed already by ritt itself in @xcite . although ritt did not give any indications about the nature of such examples ( see the discussion in @xcite , @xcite , @xcite ) , the fact that the klein functions corresponding to @xmath88 @xmath92 @xmath93 were mentioned by him in a close context in the paper @xcite suggests that he meant exactly these functions . below we give a detailed analysis of decompositions of the klein functions . we show that for a function @xmath127 corresponding to @xmath91 or @xmath89 the number of weak equivalence classes of its maximal decompositions equals two and that two non - equivalent maximal decompositions of @xmath127 are weakly equivalent if and only if they have the same length . on the other hand , we show that the function corresponding to @xmath93 has six weak equivalence classes of maximal decompositions five of which have the same length . besides , we give several related explicit examples of non weakly equivalent maximal decompositions . in particular , we give an example of a rational function with three poles for which the first ritt theorem fails to be true . for the cyclic and dihedral groups the representatives of the corresponding classes of klein functions are @xmath578 and by corollary [ bi2 ] all maximal decompositions of these functions are weakly equivalent . observe that any decomposition of @xmath576 into a composition of two functions is equivalent to the decomposition @xmath579 where @xmath580 while any decomposition of @xmath577 is equivalent either to the decomposition @xmath581 where @xmath580 or to the decomposition @xmath582 where @xmath343 and @xmath583 the subgroup lattice of the group @xmath91 can be described as follows . @xmath91 has tree subgroups @xmath585 of order 2 which are conjugated between themselves and are contained in a unique subgroup of order 4 which is the klein four - group @xmath586 . besides , @xmath91 has four conjugated subgroups @xmath587 of order 3 which are maximal in @xmath91 . this implies that @xmath584 has 7 non - equivalent decompositions corresponding to the chains e < c_2 ^ 1 < v_4 < a_4 , e < c_2 ^ 2 < v_4 < a_4 , e < c_2 ^ 3 < v_4 < a_4,and e < c_3 ^ 1 < a_4 , e < c_3 ^ 2 < a_4 , e < c_3 ^ 3 < a_4 , e < c_3 ^ 4 < a_4 . clearly , all decompositions from the first group are @xmath140-equivalent . the same is true for decompositions from the second group . on the other hand , compositions from the first and the second groups obviously are non - equivalent since they have different lengths . similarly to the case of the group @xmath91 two maximal chains in @xmath89 are @xmath140-equivalent if and only if they have the same length . however , since @xmath89 has already 28 proper subgroups , in order to prove this statement we will use an argument distinct from the examination of all maximal chains . first of all , notice that any maximal subgroup of @xmath89 either is @xmath91 , or is conjugate to @xmath589 or is conjugate to @xmath590 . besides , it is easy to see that any maximal chain of subgroups of @xmath91 has length @xmath591 or @xmath592 . we show now that any two maximal chains @xmath593 of length @xmath591 are @xmath140-equivalent . if @xmath594 , then the statement is clear so we may assume that @xmath595 this implies in particular that @xmath596 is a proper subgroup of the groups @xmath597 and @xmath6 . observe that @xmath596 is non - trivial since otherwise we would have @xmath598 . in order to prove that the chains @xmath8 and @xmath169 are @xmath140-equivalent it is enough to show that the chains @xmath599 are maximal since then @xmath600 first , notice that @xmath601 , since maximal chains in @xmath602 have length 3 . therefore , at least one of the groups @xmath597 , @xmath6 , say @xmath6 , is isomorphic @xmath590 and hence the chain @xmath603 is maximal since @xmath604 . if @xmath605 , then the chain @xmath606 is maximal as well . on the other hand , if @xmath607 , then @xmath608 implying that the chain @xmath609 is one of the chains and , therefore , is maximal . similarly , any two chains @xmath610 of length @xmath592 are @xmath140-equivalent . indeed , if @xmath611 then either @xmath612 or @xmath613 and the statement is true since maximal chains of equal length in the groups @xmath602 and @xmath91 are @xmath140-equivalent . therefore , we may assume that @xmath614 , @xmath615 setting now @xmath616 and observing that @xmath617 , we see that the chains @xmath618 are maximal . since any two chains of equal length inside @xmath602 and @xmath91 are equivalent , this implies that @xmath600 0.2 cm it is easy to see that any maximal subgroup of @xmath93 is conjugated either to @xmath91 , or to @xmath620 , or to @xmath590 and that any maximal chain of subgroups in @xmath619 has length 3 or 4 . in contrast to the groups @xmath91 , @xmath89 in the group @xmath93 we face a new phenomenon : although any two maximal chains of length @xmath591 in @xmath93 are @xmath140-equivalent there exist non @xmath140-equivalent decompositions of length @xmath592 . assume first that @xmath622 and @xmath623 . since @xmath93 is not a product of @xmath620 and @xmath590 , the intersection @xmath596 is non - trivial . therefore the chains @xmath624 are maximal , implying @xmath625 by transitivity of @xmath626 this yields that any two maximal chains of length @xmath591 such that @xmath627 @xmath623 or @xmath628 @xmath629 also are @xmath140-equivalent . let now @xmath630 be a maximal chain such that @xmath631 . then implies that @xmath632 . one can check that the normalizer @xmath50 of any group of order 3 in @xmath93 is isomorphic to @xmath590 . therefore , @xmath290 is equivalent to a maximal chain @xmath633 with @xmath634 . it follows now from the transitivity of @xmath626 that all the chains of length @xmath591 are @xmath140-equivalent . let us show now that two maximal chains of length @xmath592 @xmath635 in @xmath93 are equivalent if and only if their maximal subgroups coincide . clearly , we have @xmath636 . if @xmath637 , then @xmath638 since any two chains of length 4 in @xmath91 are @xmath140-equivalent . assume now that @xmath639 . if the chains @xmath291 and @xmath640 are equivalent , then in the sequence of maximal chains which connects them there should be two chains of the form @xmath641 where @xmath642 . the maximality condition implies that @xmath643 . furthermore , @xmath644 by . on the other hand , @xmath91 contains a unique sylow @xmath645-subgroup of order @xmath592 which is normal in @xmath91 . therefore , @xmath646 , @xmath647 and hence @xmath648 since this contradicts to the simplicity of @xmath93 , we conclude that @xmath291 and @xmath640 are not @xmath140-equivalent . although all the information about maximal decompositions of klein functions can be obtained from the analysis given above , the actual finding of the corresponding decompositions requires some non trivial calculations . in particular , the corresponding maximal decompositions which do not satisfy the first ritt theorem were found explicitly only for the simplest chains ( see @xcite , @xcite ) . it turns out that a convenient tool for such calculations is the grothendieck theory of dessins denfants " which provides an identification of @xmath649 @xmath650 and @xmath619 with the belyi functions of the tetrahedron , cube , and octahedron respectively . below we give several explicit examples of non equivalent maximal decompositions obtained by this method , referring the reader interested in details of calculations to the paper @xcite . first , a calculation shows that the belyi functions for the tetrahedron can be written in the form f_a_4=-and any maximal decomposition of @xmath584 is weakly equivalent either to @xmath651 or to the decomposition f_a_4=-z^3 . on the other hand , one can show that for example the maximal decompositions of @xmath588 ( written in a bit different normalization ) corresponding to the chains 1 < c_2 < c_4 < d_8 < s_4 , 1 < c_2 < s_3 < s_4are : @xmath656 and @xmath657 finally , identifying the chains of subgroups c_2<s_3<s_4 , c_2<v_4<d_8<s_4with maximal decompositions of the function -which is a left compositional factor of @xmath588 , one can show that to correspond the maximal decompositions : @xmath658 and @xmath659 notice that since function has three poles this example shows that with no additional assumptions the first ritt theorem can not be extended to rational functions the monodromy of which contains a cyclic subgroup with more than two orbits . 0.2 cm * remark . * it is interesting to understand how wide is the class of rational function for which the first ritt theorem holds . the following observation can be useful for obtaining some experimental results in this direction . suppose that @xmath1 is a rational function providing a counterexample to the first ritt theorem and let @xmath660 be the galois closure of @xmath1 . then @xmath660 is a regular covering of the sphere and @xmath1 is a left compositional factor of @xmath660 . in particular , the functions entering into decompositions of @xmath1 are left factors of @xmath661 therefore , all possible counterexamples to the first ritt theorem can be obtained by the analysis of regular coverings only . this observation suggests to analyse possible counterexamples to the first ritt theorem relatively to the genus of their galois closure . for example , the rational functions for which this genus equals zero are exactly the klein functions and their compositional left factors considered above . the next case which would be interesting to investigate is the one corresponding to rational factors of regular covering of genus 1 . notice also that the analysis of decompositions of @xmath619 suggests that an other source of possible counterexamples to the first ritt theorem is the functions which admit non - equivalent decompositions of the form @xmath662 . examples of such functions can be found in the papers @xcite , @xcite , @xcite , @xcite .

we analyze the quench dynamics of a one - dimensional bosonic mott insulator and focus on the time evolution of density correlations . for these we identify a pronounced propagation front , the velocity of which , once correctly extrapolated at large distances , can serve as a quantitative characteristic of the many - body hamiltonian . in particular , the velocity allows the weakly interacting regime , which is qualitatively well described by free bosons , to be distinguished from the strongly interacting one , in which pairs of distinct quasiparticles dominate the dynamics . in order to describe the latter case analytically , we introduce a general approximation to solve the bose hubbard hamiltonian based on the jordan wigner fermionization of auxiliary particles . this approach can also be used to determine the ground - state properties . as a complement to the fermionization approach , we derive explicitly the time - dependent many - body state in the non - interacting limit and compare our results to numerical simulations in the whole range of interactions of the bose hubbard model .

1202.5558

in the past two decades , progress in atomic physics , quantum optics , and the nanosciences has propelled quantum many - body theory to meet new challenges . it is indeed now possible to engineer systems that are concrete realizations of some paradigmatic models , which were once introduced to grasp fundamental properties of more complex materials . new frontiers thus have to be explored , among which the dynamics of these isolated quantum models far from equilibrium is probably the least well understood and one of the most exciting . one of the fundamental questions that has to be addressed is how correlations propagate in these systems . the schrdinger equation allows in principle for correlations between distant points to build up in arbitrary short times @xcite . this contrasts with relativistic quantum field theories , where physical effects can not propagate faster than the speed of light and causality relations between two points in space - time can exist only if one lies within the light - cone of the other . in a seminal work @xcite , lieb and robinson however showed that non - relativistic quantum many - body systems can still exhibit some sort of locality : in generic one - dimensional spin models with finite - range interactions , the propagation of correlations appears to be bounded by an effective light cone , outside which correlations are exponentially suppressed . here the role of the speed of light is played by a velocity which is an intrinsic property of the many - body hamiltonian . the existence of so - called lieb robinson bounds has many far - reaching implications . for example , they make it possible to simulate on classical computers the ground state properties as well as the dynamical evolution of such quantum systems @xcite . they also provide a general link between the presence of a finite spectral gap and the existence of a finite correlation length in the ground state of certain lattice systems @xcite . however , the extent to which lieb robinson bounds can be generalized beyond spin systems remains an open question . proofs or evidence for the existence of such bounds have indeed been reported in various systems , ranging from harmonic chains to the bose hubbard model @xcite . but it is also possible to construct models in which the propagation velocity of correlations is explicitly unbounded @xcite . dynamical properties of correlations in a closed system can be probed by studying the time evolution following a sudden change of parameter in the hamiltonian , a situation referred to as a _ quantum quench_. the quench has a particular appeal in the context of ultracold gases in optical lattices as the relevant parameters in the hamiltonian governing these systems can be easily varied in time @xcite . in addition to the existence of an effective light cone , it was discovered in recent theoretical studies that the time evolution of correlations in quenched systems is characterized by a pronounced propagation front @xcite . as for a lieb robinson bound , the velocity at which this front propagates can involve a broad range of the spectrum of the hamiltonian since the system is far from equilibrium . this makes the understanding of this feature particularly challenging : covariant low - energy effective theories would provide a natural description @xcite , but , due to the presence of high - energy excitations , realistic lattice models @xcite show a much richer behavior than their corresponding field theories . gaining more insight into the non - equilibrium properties of quantum systems thus urges the development of new effective models . in a recent work @xcite , the propagation of correlations in a quantum many - body system was studied both theoretically and experimentally in a one - dimensional bosonic gas in an optical lattice and a propagation front could be clearly identified . the observed behavior was interpreted using an exactly solvable effective model derived from the bose hubbard hamiltonian and describing non - interacting fermionic quasiparticles . the key idea behind this model is to use a jordan wigner transformation to cure some of the problems inherent to the slave - boson methods proposed previously @xcite . in the present article , we describe this approach in more detail and use it to derive the ground state as well as the quench dynamics in the mott - insulating phase . we show that its predictions are quantitatively correct in a regime of strong and intermediate interactions . our model , being exactly solvable , allows us to explore the time evolution of the system at long times , and we can show that the velocity of the propagation front exhibits a generic scaling behavior . using numerical simulations , we find that this behavior holds in all interaction regimes , down to the non - interacting limit of free bosons , where explicit solutions are available . the asymptotic value of the velocity of the propagation front , which strongly differs between the strongly and the weakly interacting limits , can be used to characterize the crossover between these two regimes . this article is organized as follows : in sec . [ sec : model ] we present the model that we will study ; in sec . [ sec : fermion_approach ] we carry out the fermionization procedure and derive general relations that enable the calculation of equilibrium ( sec . [ sec : equ_prop ] ) and non - equilibrium ( sec . [ sec : timeevolution ] ) properties . the velocity of the propagation front at weak and strong interactions is analyzed in sec . [ sec : prop - quas ] . in sec . [ sec : conclusions ] we present our conclusions . in this work we consider a one - dimensional system of bosonic atoms in an optical lattice . if the lattice is deep enough , this system can be described by the one - dimensional single - band bose hubbard hamiltonian : @xmath0 where @xmath1 and @xmath2 represent the annihilation and creation operators of a bosonic atom at site @xmath3 and @xmath4 counts the number of atoms at that site . we use a lattice constant @xmath5 and the system is considered to be infinitely large and homogeneous . the kinetic part of the hamiltonian is characterized by the hopping amplitude @xmath6 ; the on - site interaction strength @xmath7 is related to the @xmath8-wave scattering length . we work at fixed commensurate filling @xmath9 , where the model exhibits a quantum phase transition between a superfluid phase at low interaction strengths @xmath10 and a mott - insulating phase at large interaction strengths @xmath11 . at the specific filling @xmath12 , the critical value is given by @xmath13 @xcite . hubbard model is non - integrable @xcite and exhibits complex many - body properties ; in particular , its non - equilibrium properties are far from being fully understood . in order to benchmark the analytical approaches , we will perform exact numerical simulations of model by means of the density matrix renormalization group ( dmrg ) @xcite , an algorithm based on matrix product states @xcite . while the dmrg algorithm gives highly accurate results for the ground state , time evolution @xcite can be calculated only for relatively short periods of time . in the following , we will describe in detail how the bose hubbard model can be mapped onto an effective model of non - interacting auxiliary fermions . the procedure consists of four main steps : ( i ) the local hilbert space is reduced to only three states and ( ii ) auxiliary bosonic operators are introduced that allow switching between these states ( sec . [ sec : auxiliarybosons ] ) ; ( iii ) the auxiliary boson operators are fermionized by a jordan - wigner transformation ( sec . [ sec : fermionization ] ) ; ( iv ) a constraint on the fermionic operators is relaxed so that the effective hamiltonian becomes quadratic and can be diagonalized ( sec . [ sec : unconstrained ] ) . in the mott - insulating phase and away from the critical point , the local density fluctuations around the average filling @xmath14 are limited . it is thus possible to truncate the local basis on a single site @xmath3 to three states only : @xmath15 , with @xmath16 . within this reduced basis , one can represent the bare atomic operators @xmath17 in terms of constrained auxiliary boson operators @xmath18 with two flavors @xmath19 : @xmath20 the ` @xmath21'-bosons correspond to excess particles : @xmath22 , and ` @xmath23'-bosons to holes : @xmath24 . the local fock state @xmath25 represents the vacuum state of the auxiliary particles @xmath26 . bosonic commutation relations are obeyed : @xmath27 } & = \delta_{j , j'}\delta_{\sigma,\sigma'}\,,\notag\\ { [ { b^\dagger_{j,\sigma}},{b^\dagger_{j',\sigma'}}]}&= { [ { b_{j,\sigma}},{b_{j',\sigma ' } } ] } = 0\,,\ ] ] which allow for the unphysical situation of single sites being occupied by more than one auxiliary boson . therefore , the auxiliary operators have to fulfill the hardcore constraint @xmath28 and double occupancies of different species need to be eliminated by imposing @xmath29 this representation in terms of doubly - flavored constrained bosons is slightly different from the one used in slave - particle techniques @xcite , in which one introduces one auxiliary operator for each of the three local states @xmath15 and the number of auxiliary bosons per site is constrained to be exactly one . it is difficult to ensure the operator constraints and in general , and one often resorts to mean - field @xcite and perturbative @xcite approximations . in the special one - dimensional case , however , it is possible to use jordan wigner transformations @xcite which allow on the one hand for the exact treatment of the hard - core constraint ( [ eq : constraint1 ] ) and on the other for suppression of local pairing of auxiliary particles . here , we follow the standard procedure of jordan and wigner @xcite and introduce auxiliary fermion operators @xmath30 with number operators @xmath31 and anti - commutation relations @xmath32 using non - local string operators , @xmath33 we relate the auxiliary bosonic operators to the fermionic ones : @xmath34 the string operator @xmath35 counts the parity of the number of fermions accumulated over all sites @xmath36 ( including the @xmath21-fermion on site @xmath3 if @xmath37 ) and obeys the relations @xmath38 as a consequence , the number operators within both fermionic and bosonic representations coincide : @xmath39 and the original atom number operator can be written as @xmath40 due to the fermionic statistics , the hard - core conditions ( [ eq : constraint1 ] ) are satisfied automatically . the remaining constraint ( [ eq : constraint2 ] ) can be formally accounted for by the global projector @xmath41 , with @xmath42 eliminating states with both species on the same site . it is now possible to show that , within the truncated hilbert space , the original hamiltonian ( [ eq : bhhamiltonian ] ) can be exactly represented by the following fermionic model : @xmath43 we note that the effective hopping amplitudes for the two different flavors differ by the bosonic enhancement factor of eq . ( [ eq : sbmapping ] ) . in practice , it is difficult to take care analytically of the projector @xmath44 . we will thus carry out the calculations in the approximation of unconstrained fermions ( uf ) , @xmath45 , leading to a quadratic hamiltonian that can be diagonalized exactly . we will see that the uf approximation is justified because the main source of creation of double occupancies would be a local pairing mechanism , which , in the fermionic representation , is suppressed by the statistics of the auxiliary particles . the hamiltonian ( [ eq : hfermionic ] ) with @xmath46 can be rewritten in momentum space as @xmath47 with the bare dispersions [ eq : baredisp ] @xmath48 and an antisymmetric pairing parameter @xmath49 which obeys @xmath50 . in analogy to the gutzwiller approximation @xcite , the accuracy of the uf approximation can be estimated _ via _ the translation - invariant expectation value of the local projector @xmath51 this quantity is a measure for the population of unphysical states and gives the order of magnitude of the error in local observables ( due to translational invariance site indices of observables can be dropped ) . additionally , we will study the quality of the relaxation of the constraint ( [ eq : constraint2 ] ) by comparison to the numerically exact dmrg method . the quadratic hamiltonian @xmath52 can be diagonalized via a bogolyubov transformation by introducing the quasiparticle operators [ eq : btransf ] @xmath53 the functions @xmath54 and @xmath55 fulfill the relations @xmath56 and are determined by the following expressions : [ eq : btrans ] @xmath57 we infer from the above equations that the @xmath21modes are excess particles each dressed with absent holes and the @xmath23modes are holes dressed with absent excess particles . this is particularly evident from the perturbative expressions ( [ eq : upert ] ) and ( [ eq : vpert ] ) . we also note that the quasiparticle operators ( [ eq : btransf ] ) can be interpreted as dirac spinors @xcite . using the quasiparticle operators , the hamiltonian can be written in the diagonal form @xmath58 the dispersion relation of the individual quasiparticles is @xmath59 here @xmath60 is the energy of a pair of two distinct types of quasiparticles with opposite momenta , which is given by @xmath61 ^ 2 + 4|\delta(\k)|^2}\label{eq : dispomega } \\ & = u-2j(2\nnot+1)\cos(\k)+\mathcal{o}\left(\frac{j^2}{u}\right)\,.\label{eq : disppert } \end{aligned}\ ] ] the exact dispersion relations for different interaction strengths are displayed in fig . [ fig : disp ] , together with the first order expansion in @xmath62 ( strong coupling expansion ) . one can see that the profiles rapidly differ from their limiting cosine shape as the interactions are lowered . eqs . and show that the width of the energy bands depends only on the hopping amplitude @xmath6 and on the average filling @xmath63 ( _ via _ the bose enhancement factor ) , but does not depend on the interaction strength . at large interaction strengths , the energy gap is proportional to the interaction strength and the gap of the ` @xmath21'-quasiparticles is strictly positive when the interaction is above a certain threshold : @xmath64 below this threshold our uf approximation breaks down . for @xmath65 , the range of validity of our model is thus limited to @xmath66 , which is above the superfluid to mott - insulator transition @xmath67 , but significantly lower than the mean - field transition @xmath68 . a description of the phase transition might be achieved by introducing auxiliary operators on the basis of a coherent - state representation ( see e.g. @xcite ) , but this goes beyond the scope of this work . the slope of the dispersion relations @xmath69 describes the group velocity of the quasiparticles . of particular interest will be the relative velocity of pairs of quasiparticles of distinct types and opposite momenta : @xmath70 whose maximal value @xmath71 plays an important role in the characterization of the non - equilibrium properties . this maximal velocity corresponds to the point where the curvature of @xmath72 changes sign . it is located at @xmath73 at large interaction strengths and is shifted toward lower momenta at smaller interaction strengths , as can be seen in fig [ fig : disp ] . in the relevant interaction regime ( [ eq : regime ] ) , the maximum velocity is well approximated by : @xmath74 in particular , one sees that @xmath75 is a decreasing function of @xmath76 . for the strictly positive quasiparticle energies ( [ eq : regime ] ) , the ground state at a value of @xmath7 and @xmath6 is the quasiparticle vacuum [ eq : gswavefunction ] @xmath77 the ground state at infinitely strong interactions , i.e. the fock state with @xmath14 particles per site , @xmath78 , represents the vacuum of the bare excess particles and holes . we summarize in this sec . some general properties of the correlation functions in the quasiparticle formalism that will be used later to derive ground state and non - equilibrium properties of the system . for the ground state ( [ eq : gswavefunction ] ) , but also for the time - dependent wave functions ( [ eq : timeevolution ] ) which will be introduced in section [ sec : timeevolution ] , the only non - vanishing single - particle correlation functions are [ eq : corrft ] @xmath79 where @xmath80 and the thermodynamic limit has been taken . possible time dependence ( equal time ) is implicit and expectation values are site - independent for the homogeneous systems under consideration . we note that the correlations of the different types are equivalent : @xmath81 therefore , also the quasiparticle densities do not depend on the flavor and we can define a single density of excitations : @xmath82 since the hamiltonian is quadratic , correlations of the occupancies can be related to the single - particle correlations using wick s theorem , which gives us @xmath83 in the special case @xmath84 , the fermionic statistics , together with the symmetry with respect to exchange of fermionic flavor , imply that the on - site correlator @xmath85 vanishes and the local double occupancy factorizes : @xmath86 the density of excitations thus fully determines all local properties , including the atom number fluctuations : @xmath87 atom correlations can be related to correlations of auxiliary particles . making use of eq . ( [ eq : fermionstoatoms ] ) , we can for example express atomic density correlations in the following way : @xmath88 ultracold atom experiments with single - site resolved imaging @xcite can access the parity @xmath89 rather than the density itself . the expression for parity correlations turns out to be similar to that of density correlations : @xmath90 both density and parity correlations are particularly simple to evaluate within the present approach . correlations including the non - local string operator ( [ eq : string ] ) , such as the single - particle correlations @xmath91 , can also be computed , but they require the evaluation of the toeplitz determinant @xcite . interestingly , the fermionic string ( [ eq : string ] ) is equivalent to the string operator recently measured by endres @xcite . in this section we discuss the equilibrium properties of the mott - insulating phase derived within the unconstrained fermion approximation . as argued in the preceding section , the observables are related to single - particle correlations ( [ eq : corrft ] ) , which for the ground state ( [ eq : gswavefunction ] ) can be evaluated straightforwardly : @xmath92 with the coefficients @xmath54 and @xmath55 given in eq . ( [ eq : btrans ] ) . the local density of excitations ( [ eq : nex ] ) can thus be calculated from @xmath93 in the case of strong interactions , one can also derive an explicit expression from the expansion ( [ eq : upert ] ) and ( [ eq : vpert ] ) of the coefficients @xmath94 and @xmath95 : @xmath96 combining eqs . and gives an estimate of the occupation of unphysical states ( [ eq : guwiapprox ] ) , @xmath97 where the right - hand side stems from the evaluation of ( [ eq : neq ] ) at the lowest interaction considered ( [ eq : regime ] ) . for @xmath98 , @xmath99 is less than @xmath100 and we expect the error on local expectation values to be of similar magnitude . with this at hand , we can now consider the behavior of the density correlations in the ground state . in fig . [ fig : gscorr](a ) , the atom number fluctuation @xmath101 is evaluated numerically using ( [ eq : gsnex ] ) and compared to the strong coupling expansion ( see also @xcite ) @xmath102 as well as to the results obtained from dmrg simulations with a truncation of the site occupancy to @xmath103 or @xmath104 ( system size is 256 sites , 400 dmrg - states are retained ) . the predictions of the uf approximation , both from the numerical integration of ( [ eq : gsnex ] ) and from the strong coupling expansion , are in excellent agreement with the dmrg simulations for all interaction strengths satisfying ( [ eq : regime ] ) . the accuracy of the truncation of the local hilbert space to three states only is also confirmed by the dmrg simulations . higher occupancies start to be important only for interaction strengths below the point where the uf approximation breaks down . we further compare our results with the ones derived within a holstein primakov approximation of the slave - boson representation used , e.g. , by huber @xcite . this approach is equivalent to the auxiliary boson representation ( [ eq : sbmapping ] ) when fully relaxing the constraints ( [ eq : constraint1],[eq : constraint2 ] ) . we find that the local observables can not be well described at intermediate interaction strengths , even though the density of excitations is small ( fig . [ fig : gscorr ] ) . a similar instability has been observed with slave bosons in ref . we will analyze the slave - boson approach in more detail later in the context of the non - equilibrium dynamics ( sec . [ sec : holstein - primakov ] ) . we can also evaluate analytically non - local density correlations to second order in @xmath62 : @xmath105 as shown in fig . [ fig : gscorr](b ) , the above expression only slightly overestimates the amplitude of the correlations compared to the full numerical evaluation of the integrals ( [ eq : corrft ] ) and ( [ eq : nncorr ] ) with ( [ eq : diffcorr ] ) . we therefore conclude that local observables and nearest - neighbor correlations in the mott - insulator regime are well described by a perturbation expansion to order @xmath106 . this is no longer the case for longer - range correlations , which are simply vanishing according to the expansion to second order , whereas the exact dmrg predicts that they should be finite and exponentially decaying . as can be seen in fig . [ fig : gscorr](c ) , the numerical evaluation of the uf equations provides a much better agreement with the dmrg results . the correlations at @xmath107 can be almost perfectly reproduced and a similar decay length is found . the amplitude of the correlations for @xmath108 is overestimated , however , and the discrepancy becomes worse as @xmath109 increases . we analyze the quench dynamics of a system prepared initially in a deep mott - insulating state . we first derive the general results for the time evolution of the wave function and the correlation functions and then give explicit expressions for the case where the initial state is a fock state ( infinite interactions ) . these results form the basis for the detailed discussion of the physical properties of the quench dynamics in the subsequent sec . [ sec : prop - quas ] . the initial state considered is the ground state at some values of @xmath6 and @xmath7 satisfying the condition ( [ eq : regime ] ) and takes the form @xmath110 the time - evolution of this state under the hamiltonian @xmath52 reads @xmath111 with @xmath112 for the wave function ( [ eq : timeevolution ] ) , the non - vanishing equal - time single - particle correlations evaluate to [ eq : corrk ] @xmath113 \\ + \delta_{k , k'}v(k)\bar{v}(k ) \left[e^{2i\omega(k)t } v(k ) \bar{u}(k ) - \bar{v}(k){u}(k ) \right ] \end{split } \label{eq : corrk1 } \intertext{and } \begin{multlined}[b ] \av{c^\dagger_{k,\sigma } ( t)c_{\k',\sigma}(t ) } = \delta_{k , k ' } \left[2\cos(2\omega ( k)t)u(k)\bar{u}(k)v(k)\bar{v}(k ) \right.\\ \left . u^2(k)\bar{v}^2(k)-v^2(k)\bar{u}^2(k ) \right]\ , . \end{multlined } \label{eq : corrk2 } \end{gathered}\ ] ] based on these expressions , the expectation values of any observables be either calculated analytically , when the strong coupling expansion holds , or computed numerically for lower interactions ( see [ sec : observables ] ) . for concreteness , we focus now on a quantum quench starting from the fock state with filling @xmath9 by setting @xmath114 , @xmath115 and thus @xmath116 , @xmath117 . in this case , the expansion in @xmath62 leads to [ eq : corrkpert1 ] @xmath118\\ + \mathcal{o}\left(\frac{j^2}{u^2}\right)\ , , \end{split}\label{eq : corrkpert1}\\ \begin{split } \av{c^\dagger_{k,\sigma } ( t)c_{\k,\sigma}(t ) } = \frac{8j^2\nnot(\nnot+1)}{u^2}\sin^2(\k ) \left[\cos(2\omega(k ) t ) - 1\right ] \\ + \mathcal{o}\left(\frac{j^4}{u^4}\right)\ , , \end{split}\label{eq : corrkpert2 } \end{gathered}\ ] ] where @xmath72 stands for the dispersion in the strong coupling expansion ( [ eq : disppert ] ) . within this expansion , the dynamics of the local density of excitations can be expressed in terms of the bessel functions of the first kind , @xmath119 . one gets @xmath120\label{eq : approxnex}\,,\ ] ] with @xmath121 . in the relevant interaction regime , the population of unphysical states @xmath122 thus remains as small as in the ground state ( [ eq : guwiapprox2 ] ) and we expect the uf approximation to be well behaved in general . it is , however , important to note , that the expansion is not rigorous since the approximate dispersion ( [ eq : disppert ] ) is multiplied by time , which is unbounded . the single - particle correlators required to derive non - local density correlations ( [ eq : nncorr ] ) read @xmath123 \notag\\ & = \begin{multlined}[t][0.825\columnwidth ] ( -i)^{d+1}\frac{\sqrt{\nnot(\nnot+1)}j}{u } \left[e^{iut/\hbar } \big({\mathcal{j}}_{d+1}({\tilde{j}}t ) \right.\\ \left.+{\mathcal{j}}_{d-1}({\tilde{j}}t)\big)+\delta_{d,1}\right ] \;. \end{multlined } \label{eq : sgpartcorr } \end{aligned}\ ] ] making use of the identity @xmath124 we obtain the following expressions for the non - local density correlations : [ eq : corrapprox ] @xmath125 we note that the interaction strength @xmath7 is involved only in the magnitude of the correlations for @xmath126 , via the dimensionless parameter @xmath62 . in the case of nearest - neighbor correlations , we find an additional oscillation of the amplitude with frequency @xmath127 . in this section , we analyze the accuracy of the successive approximations that lead from the bose hubbard model to the uf approximation and its strong coupling expansion . for this purpose , we introduce the root - mean - square differences @xmath128 where @xmath129 and @xmath130 are the density correlations predicted using two different level of approximations . by observing the dependency of @xmath131 with the distance @xmath109 , we can verify , in particular , whether a given approximation breaks down at large times . for non - averaged results we refer to sec . [ sec : prop - quas ] . we use @xmath132 , the maximal time accessible by our dmrg simulations . we use a dmrg algorithm in the thermodynamic limit @xcite , with a second - order suzuki trotter decomposition of time step @xmath133 , and we retain 2400 states . the numerical error is always smaller than the symbol size and line width . in fig . [ fig : diff](a ) , we first compare the strong coupling expansion ( [ eq : corrapprox ] ) and the numerical evaluation of the uf approximation . we find that the expansion is relatively accurate ( @xmath134 ) down to interactions @xmath135 , except for the @xmath107 correlation , which only slowly converges to the exact results when @xmath76 is increased . we recall here that a similar accuracy is reached for the ground state correlations ( fig . [ fig : gscorr ] ) . in fig . [ fig : diff](b ) , we then compare the numerical evaluation of the uf approximation with the exact dmrg simulation of the hamiltonian ( [ eq : hfermionic ] ) , i.e. the bose hubbard model when the site occupancy is truncated to @xmath103 . the uf approximation appears to be accurate within @xmath134 for @xmath136 . as we will show in sec . [ sec : corrpropagation ] , the uf approximation still qualitatively describes the dynamics between @xmath137 and @xmath138 , which marks the break down of the quasiparticle picture . finally , we compare in fig . [ fig : diff](c ) the predictions of the dmrg simulation when the local hilbert space is truncated to a maximum site occupancy @xmath103 , corresponding to the model ( [ eq : hfermionic ] ) , or @xmath104 . we observe that the error due to the truncation starts to be significant ( @xmath139 ) only for @xmath140 , i.e. when the interaction energy becomes larger than the width of the quasiparticle band . in order to generalize the description to higher dimensions , it may appear tempting to fully relax the hardcore constraint and work with bosons instead of fermions . the resulting hamiltonian is then equivalent to the one derived by huber @xcite using holstein primakov bosons ( hp ) . the resulting equations for the quasiparticles and their dispersions are very similar to those derived in the fermionic model , except that the coefficient @xmath95 becomes symmetric instead of antisymmetric . as a consequence , local pairing of different species is no longer suppressed and the occupation of the unphysical states becomes much larger than in the fermionic approach . in order to quantify this effect , one first has to calculate the single - particle correlations . to lowest order in @xmath62 , the quasiparticle coefficient takes the form @xmath141 and one finds @xmath142\,,\ ] ] which can be compared to the fermionic version ( [ eq : sgpartcorr ] ) . in particular , one finds that the integral has a finite value at @xmath84 and the overcompleteness @xmath143 becomes of the same order as the density of excitations and the density correlations . this means that physical and unphysical states play an equally important role in the hp approximation and this approach fails to describe the quench dynamics even in the limit @xmath144 , where the density of excitations is low . this can be observed for example in the density correlations , which now read @xmath145^{2}.\ ] ] the change of sign between the two bessel functions compared to the uf expressions has a dramatic effect , since the function in square brackets is now proportional to the derivative of a bessel function , instead of a bessel function itself . this leads in particular to the smearing out of one of the main features of the quench dynamics , namely the propagating correlation peak that we will describe in sec . [ sec : corrpropagation ] . in this section , we complement the preceding analysis of quenches on the strongly interacting side by the extreme situation of a quench from infinite to zero interactions @xcite . at @xmath146 , the time evolution is readily described in the heisenberg picture , @xmath147 in which individual bosons propagate with free dispersion . in the thermodynamic limit , this yields the propagator @xmath148\nonumber \\ & = ( i)^{d}{\mathcal{j}}_{d}\left(\frac{2jt}{\hbar}\right)\ , .\label{eq : vbessel}\end{aligned}\ ] ] using the following relation for the initial fock state : @xmath149 we can derive explicit equations for the density correlations : @xmath150 \nnot^2\big ( \sum_j { \mathcal{j}}_j^2(2jt/\hbar ) \big)^2 \\ + \nnot(\nnot+1 ) \big ( \sum_j { \mathcal{j}}_{j+d}(2jt/\hbar ) { \mathcal{j}}_{j}(2jt/\hbar ) \big)^2 \\ - \nnot(\nnot+1 ) \sum_j { \mathcal{j}}_{j+d}^2(2jt/\hbar ) { \mathcal{j}}_{j}^2(2jt/\hbar ) -\nnot^2 \end{multlined } \\ & = - \nnot(\nnot+1 ) \sum_j { \mathcal{j}}_{j+d}^2(2jt/\hbar ) { \mathcal{j}}_{j}^2(2jt/\hbar ) \ , . \label{eq : corrnonint}\end{aligned}\ ] ] here we have used the properties @xmath151 and @xmath152 for @xmath153 . we now analyze in detail how correlations spread in the quench dynamics starting from the fock state with @xmath63 atoms per site within the bose hubbard hamiltonian ( [ eq : bhhamiltonian ] ) . the description in terms of fermionic quasiparticles for intermediate and strong interactions ( [ eq : timeevolution ] ) , as well as the non - interacting solution ( [ eq : corrnonint ] ) , provide a firm basis for the interpretation of the outcome of the dmrg simulations and of recent experimental results @xcite and allow for their extrapolation at long times , where no analytical solution is available so far . for concreteness , we restrict our discussion in the following to the filling @xmath98 , where the @xmath21-quasiparticles correspond to doublons and @xmath23-quasiparticles to holons . the relevant processes involved in the quench dynamics can be best understood in the expansion of the wave function to lowest order in the auxiliary fermion operators : [ eq : psipert ] @xmath154 in this representation , the state decomposes into two parts : a time - independent part ( [ eq : psipert1 ] ) consisting of the fock state and the symmetric superposition of bound nearest - neighbor doublon - holon pairs , and a time - dependent part ( [ eq : psipert2 ] ) describing the superposition of propagating doublon - holon pairs . the dynamics following the quench is driven by the propagating pairs , whereas the steady state is solely determined by the bound pairs , as the contribution of the propagating pairs phases out at long times . at lowest order in @xmath62 , the steady state simply corresponds to the ground state at the final interaction strength . higher order terms in the strong coupling expansion would describe the population in the excited states . at @xmath155 , the bound and propagating pairs interfere destructively and one recovers the initial fock state @xmath156 . finally , we note that the wave function ( [ eq : psipert ] ) is equivalent to the one obtained within a time - dependent perturbation theory in appendix [ sec : perturbation ] and can be used to derive the perturbative results presented in sec . [ sec : pertexp ] . the doublon and the holon forming a propagating pair are produced initially on neighboring sites by a single hopping event and then move in opposite directions . the two quasiparticles are entangled , since the pair is described by a superposition state : @xmath157 this ensures a constant atomic density and leads to strong bipartite entanglement as the pairs are stretched across the system @xcite . the momentum distribution of the quasiparticle pairs is sine shaped , from which follows that the quasiparticles propagate as a wave packet . the maximal weight of the momentum distribution is located at the wave vector @xmath158 , where the dispersion relation is close to being linear , and is characterized by the maximal group velocity @xmath159 . the wave - packet structure of the propagation can also be made explicit by turning the sum over the momenta in into a sum over the lattice sites : @xmath160 in the above expression , one immediately recognizes the propagation velocity @xmath161 in the argument of the bessel functions . however , we expect a large dispersion of the wave packet due to the width of the momentum distribution . a detailed description of the propagation of the quasiparticle pairs is left for sec . [ sec : corrpropagation ] . the situation is somewhat different for weakly interacting bosons . in the non - interacting solution , the correlation functions result from the interference between free bosons propagating with a relative velocity @xmath162 . unlike the auxiliary particles in the strongly interacting case , the number of free bosons per site is not limited , which leads to some qualitative differences that we will discuss later . the equal - time density correlations @xmath163 in the strongly interacting limit exhibit a very peculiar feature for @xmath164 , namely , the presence of a negative signal , a dip , propagating to larger distances at longer times . this can be seen , for example , in fig . [ fig : quenchcorr1 ] , where we display the time evolution of these correlations for @xmath165 , as predicted by the uf approximation and the dmrg simulation ( which are in remarkable agreement ) . this characteristic signal is already present in the perturbative result and can be attributed to the propagating quasiparticle pairs , illustrating the interest of this picture . the structure of the nearest - neighbor correlation is more complicated . in the long - time limit and within the strong coupling expansion , the nearest - neighbor correlation reaches the value corresponding to the ground state at the final interaction strength . at short time , it exhibits oscillations driven by the interaction strength @xmath7 and corresponding to the interaction of a holon ( doublon ) of the bound pair with a doublon ( holon ) of a propagating pair . in the first order of the strong coupling expansion ( [ eq : psipert ] ) , the bound pairs extend only over a distance @xmath166 but in the full numerical integration they can spread over larger distances [ cf . [ fig : gscorr](c ) ] , leading to additional oscillations for @xmath107 . these oscillations are clearly visible in numerical evaluations of the correlations within the uf approximation , as well as in the dmrg simulations ( fig . [ fig : quenchcorr1 ] ) . while the uf approximation is almost exact at short distances , it overestimates the weak oscillations with period @xmath167 at larger distances . we found that these stem from terms of order @xmath168 dominating the doublon - doublon and holon - holon correlations . these oscillations are also present in the dmrg simulations , but with a much lower amplitude . interestingly , these terms cancel in the parity correlations studied in @xcite , where the uf approximation is in even better agreement with the exact simulations . for quenches to intermediate values of the interaction strength , the dynamics of density correlations exhibits essentially the same features as described above . this can be seen in fig . [ fig : quenchcorr2 ] , where the dynamics following a quench to @xmath169 is depicted . in particular , the characteristic dip corresponding to the propagating quasiparticles is still present . we note that the propagation of this correlation signal is still remarkably well described by the uf approximation , even though this model is close to breaking down at this interaction strength . however , one sees that the strong coupling expansion significantly overestimates the amplitude of the correlations at @xmath166 and that the amplitude of the unphysical oscillations in the numerical evaluation of the uf approximation increases . in the weakly interacting regime , one could expect a different behavior since the relevant quasiparticles are of different nature . however , the main features that characterize the dynamics of density correlations at strong and intermediate interactions are remarkably preserved , as can be seen in fig . [ fig : quenchcorr3 ] . in particular , a propagating dip can be identified in all cases . the main difference between the non - interacting and the strongly interacting cases is the lower velocity and very slow decay of correlations at long times when @xmath170 . at @xmath171 , this long tail is already strongly suppressed at short distances , but is still visible at longer distances . for @xmath172 ( fig . [ fig : quenchcorr3 ] ) , one sees that the overall profile of the propagating correlation signal is already very similar to that in the more strongly interacting case ( figs . [ fig : quenchcorr1 ] and [ fig : quenchcorr2 ] ) . at low interaction strengths , one has to be careful when using the dmrg simulations since the truncation of the local hilbert space to a finite number of states can introduce significant errors . by comparing dmrg simulations to the exact formula obtained in the `` worst '' case @xmath146 , we found that a maximal site occupancy @xmath104 represents a fairly safe approximation , whose accuracy improves with the distance for the times considered ( see fig . [ fig : quenchcorr3 ] ) . to summarize the analysis conducted in this section , we observe that the dynamics of the density correlations is dominated by the propagation of a negative signal ( dip ) . for strong interactions , this dip results from the propagation of the quasiparticle pairs described in sec . [ sec : prop - quas ] . in the next section we will investigate in more detail the shape of that signal and characterize its propagation velocity quantitatively . dynamics of density correlations at distance @xmath173 after a quench from the fock state @xmath156 at infinite interactions to a final interaction @xmath165 . the results for different distances @xmath109 are shifted for clarity by @xmath174 . we display the results obtained from the numerical evaluation of the uf equations , from their strong coupling expansion and from exact dmrg simulations . the shaded blue profiles figure a gaussian fit of the correlation signal from the dmrg simulation , after the high - frequency oscillations have been filtered out . the filled blue circles mark the center of the fitted profile , i.e. the position of the signal . the filled red circles mark the position of the correlation signal obtained in the same way from the numerical evaluation of the uf approximation ( fit not shown ) . the airy function that appears in the analytical formulas derived from the uf approximation is plotted in the inset.,scaledwidth=49.0% ] dynamics of density correlations after a quench from the fock state @xmath156 at infinite interactions to a final interaction @xmath169 . see fig . [ fig : quenchcorr1 ] for more information.,scaledwidth=49.0% ] dynamics of density correlations after a quench from a fock state @xmath156 at infinite interactions to weak final interactions . the results for different distances @xmath109 are shifted for clarity by @xmath175 . unlike in figs . [ fig : quenchcorr1 ] and [ fig : quenchcorr2 ] , the position of the correlation signal of the dmrg results is identified with the position of the absolute minimum and it is denoted by the circles in the corresponding colors.,scaledwidth=49.0% ] we can get a lot of insight into the propagation of the signal from the following approximation of the density correlations at large distances ( we recall that the lattice constant @xmath176 is set to one ) : @xmath177 in the above expression , derived from ( [ eq : corrapprox2 ] ) , we made use of the relation existing between the airy function @xmath178 and the high - order bessel functions @xcite : @xmath179 the airy function is plotted in the inset of fig . [ fig : quenchcorr1 ] . it exhibits a peak located at @xmath180 and surrounded by an exponential tail on the side @xmath181 and by an algebraically - decaying oscillating tail on the side @xmath182 . disregarding the monotonically and slowly varying prefactor in eq . , the profile of the airy function alone allows us to understand several features of the propagation of the correlation signal . for example , it reveals the existence of a well defined propagation front , since the correlations are exponentially suppressed for times @xmath183 , with @xmath184 \,.\ ] ] the signal in the density correlations corresponds to the peak of the airy function . once this peak has passed , that is for @xmath185 , the correlations show an algebraic decay with oscillations . from the definition of @xmath186 , one sees that two terms contribute to the propagation of the correlation signal : the first is simply proportional to the distance , corresponding to a well defined velocity , whereas the second is proportional to @xmath187 . the linear contribution dominates at large distances , leading to a light - cone - like spreading of the correlations . at small distances , however , the dynamics deviates significantly from the asymptotic light cone . this behavior can be accounted for by defining an instantaneous propagation velocity : @xmath188^{-1 } \notag\\ & = \text{v}_\infty\left ( 1-\frac{z_0}{2^{1/3 } 3 } d^{-2/3 } \right ) + \mathcal{o}\big(d^{-5/3 } \big)\,.\label{eq : vscaling } \end{aligned}\ ] ] one sees immediately in the above equation that the asymptotic light cone is characterized by the velocity @xmath189 and is reached algebraically at large distances , whereas the propagation velocity can go down to approximately @xmath162 at short distances . a similar analysis can be carried out for the case @xmath146 . it turns out that the correlation dip is almost completely described by a single term in the infinite sum ( [ eq : corrnonint ] ) . for even @xmath109 , for example , we obtain : @xmath190 the same expression for the instantaneous velocity ( [ eq : vscaling ] ) therefore holds in the non - interacting case as well , but with @xmath191 , which is the velocity of freely propagating bosons . the behavior at @xmath146 mostly differs from the strongly interacting case once the correlation dip has passed ( @xmath185 ) : further terms ( [ eq : corrnonint ] ) beyond eq . then become important , which causes correlations to decay very slowly ( see fig . [ fig : quenchcorr3 ] ) . the width and the height of the correlation dip can also be derived from the expressions ( [ eq : corrairy],[eq : corrairyni ] ) . for both the interacting and the non - interacting case , the width increases proportionally to @xmath187 while the height decreases with @xmath192 in the strongly interacting case and with @xmath193 for @xmath146 . we note that similar power laws have been found for the quantum ising model @xcite . in the following , we show that the approximate scaling of the velocity found in the strongly and non - interacting limits holds for any interaction strength . we first concentrate on large interaction strengths . within the uf approximation , we can evaluate the correlations up to arbitrarily long times and make a rigorous scaling analysis of the instantaneous propagation velocity . we determine the position of the dip by means of a gaussian fit after having filtered out oscillations with a period shorter than @xmath167 using a low - pass filter . [ fig : v](a ) illustrates for a few interaction strengths that the analytical scaling behavior @xmath194 is accurately reproduced at sufficiently large distances @xmath195 . extracting the asymptotic velocities @xmath196 and the exponents @xmath197 with a fit over distances @xmath198 , we obtain values in very good agreement with the approximated analytical predictions . for example , the exponent is found to be the same for all interactions : @xmath199 . the small difference from the value @xmath200 expected from the airy functions is most probably due to the prefactor in ( [ eq : corrairy ] ) , which we neglected when deriving . the asymptotic velocities match the ones that we expect from the quasiparticle dispersion relation eq . , as shown in fig . [ fig : v](c ) . close to the breakdown of the uf approximation , the oscillation frequencies due to the interaction and the finite bandwidth become similar and one can not easily filter out the first one anymore . the instantaneous velocity @xmath201 therefore shows an oscillatory behavior even at very large distances @xmath202 . nevertheless , the scaling behavior remains perfectly obeyed on average and in the long - distance limit . in the non - interacting case , shown in fig . [ fig : v](a ) , we extract accurately the position of the correlation signal by simply locating the first minimum . we again find the scaling exponent @xmath199 and the extracted asymptotic velocity is close to the expected value @xmath191 [ cf . [ fig : v](c ) ] . using dmrg simulations , we can calculate the dynamics exactly for all interaction strengths , but we are restricted to short time and length scales . we therefore fix the scaling exponent to @xmath203 in order to extract the asymptotic velocities . in fig . [ fig : v](b ) we show that the scaling @xmath204 becomes accurate as the distance increases for both strong ( @xmath205 , extracted with low - pass filter and gaussian fit ) and weak interactions ( @xmath206 extracted directly from the peak without low - pass filter ) . despite the limited number of data points available in the scaling region , we can determine the asymptotic velocities with a reasonably small uncertainty . the values that we obtain , gathered in fig . [ fig : v](c ) , are in good agreement with those predicted by the uf approximation . the lack of data in the range @xmath207 results from the mixing of the time scales related to kinetic and interaction processes and which prevents us from locating accurately the position of the correlation signal . the asymptotic velocities in fig . [ fig : v](c ) can be seen as a characterization of a crossover between a regime of quasi - free bosons ( @xmath208 ) , with a renormalized velocity , and the strongly interacting regime described by two flavors of fermions . this crossover is not directly related to the ground - state phase diagram of the bose - hubbard model since the propagation velocity reflects the dispersion in the center of the brillouin zone ( at wave vectors @xmath209 ) , rather than low - wavelength modes . as a consequence , @xmath196 is considerable higher than the sound velocity in the superfluid regime @xcite and a linear propagation with @xmath210 is found at strong interactions , even though at equilibrium the system would be in the mott - insulating phase . as a final remark , we note that the dependency of the propagation velocity on @xmath76 in that system has been studied before by luchli and kollath @xcite , who considered the case of a quench from a small interaction strength to a larger one . surprisingly , the instantaneous spreading velocity has been found to exhibit a maximum at intermediate interaction strength . a possible explanation for this effect could be that bosonic atom number fluctuations present in the initial superfluid state may lead to enhanced velocites as compared to the quench from the fock state . a quantitative comparison between our predictions and this previous work would require an extrapolation of the velocity to large distances which is difficult in the absence of an analytical model . in order to describe accurately the quench dynamics of the one - dimensional bose hubbard model in the mott - insulating regime , we have developed a new analytical approach relying on the fermionization of auxiliary bosons . its predictions regarding both the ground state and the dynamical properties are found in quantitative agreement with exact numerical simulations for large and intermediate interaction strengths @xmath66 . this constitutes a great improvement with respect to the analytical models introduced previously . using this model , we are able to investigate the time evolution of density correlations in the quenched system over exceedingly long times . we observe a characteristic light c / one dynamics , meaning that there exists a distance , linearly growing in time , beyond which correlations between distant sites are exponentially suppressed . more precisely , correlations spread as a wave packet along this light cone , forming a propagation front whose position can be unambiguously identified . a careful analysis of the velocity with which this front propagates reveals a generic scaling behavior characterized by a universal exponent and an asymptotic velocity dependent on the interaction strength . the same behavior is found in the non - interacting limit of freely propagating bosons , where an explicit solution is available , as well as in the intermediate regime @xmath211 , where we rely on exact numerical simulations . the asymptotic velocity , which varies significantly between the weakly and the strongly interacting regime , is a useful quantity to characterize a broad spectral range of the hamiltonian as it does not depend only on its low - lying modes . building upon this first success , we envisage that the representation of the bose hubbard model in terms of fermionic quasiparticles could shed new light on the mechanism for thermalization or serve as a tool to interpret the outcome of spectroscopic measurement on laboratory systems , such as modulation spectroscopy or bragg spectroscopy for ultracold gases in optical lattices . we thank d. baeriswyl , t. giamarchi , v. gritsev , s. huber and a. tokuno for discussions . financial support by anr ( famous ) , snsf under division ii and manep , and eu ( marie curie fellowship to m.c . ) is acknowledged . thanks the university of fribourg for hospitality . in this appendix we develop a complementary perturbative approach to recover the behavior in the strong coupling limit to first non - vanishing order in @xmath62 . the situation we consider is the quench from the fock state @xmath212 at filling @xmath14 to a large final interaction strength @xmath76 . in all generality , the wave function after a sudden change of parameters can be written in the eigenbasis @xmath213 ( with corresponding eigenenergies @xmath214 ) of the final hamiltonian @xmath215 usually the difficulty lies in determining the eigenstates @xmath213 and their corresponding energies @xmath214 in a many - body problem . in this appendix we determine @xmath213 and @xmath214 by perturbation theory in @xmath62 in a system of length @xmath216 with periodic boundary conditions . note , that this is not a full perturbative expansion , since we will not expand the exponential in the corresponding power . we consider the interaction term of the bose hubbard hamiltonian as the unperturbed hamiltonian and the kinetic part as the perturbation . the eigenenergies of the unperturbed hamiltonian are multiples of the interaction strength @xmath7 and the corresponding states are fock states . more precisely , the ground state is the fock state @xmath78 with vanishing energy . the lowest excited states are the states with a single particle - hole excitation with energy @xmath7 . these we denote by @xmath217 with an occupation @xmath14 for all the sites except for site @xmath218 with @xmath219 atoms and the site @xmath220 with @xmath221 atoms , i.e. @xmath222 . using degenerate perturbation theory ( restricted to the same symmetry sector as the initial state ) at first order in @xmath62 , the ground state energy remains zero and the ground state of the final hamiltonian is given by @xmath223 the lowest band of excited states resulting from the single particle - hole excitations is formed by @xmath224 with corresponding energies @xmath225 here @xmath226 ( @xmath227 ) are the symmetric states that diagonalize the kinetic part of the hamiltonian given by @xmath228 note that the index @xmath109 only starts at @xmath229 to avoid the double counting of the fock state . we employed the notation @xmath230 . as we are interested in the time - evolution of the initial fock state , we abbreviated unimportant terms as @xmath231 which are the states beside the fock state that are directly coupled via the kinetic term to the states @xmath232 . using these eigenenergies and states to first order , we can now write the time evolving state @xmath233 as @xmath234 this formula corresponds to the expression ( [ eq : psipert ] ) which one obtains in the unconstrained fermionic approach up to first order . as discussed in the unconstrained fermionic approach this expression gives a lot of insight into the formation and propagation of singly and doubly occupied sites and can be used to compute all the observables that we are interested in . 65ifxundefined [ 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 link:\doibase 10.1103/physrevlett.97.050401 [ * * , ( ) ] link:\doibase 10.1007/bf01645779 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.82.277 [ * * , ( ) ] link:\doibase 10.1063/1.3149556 [ * * , ( ) ] link:\doibase 10.1016/j.aop.2010.09.012 [ * * , ( ) ] http://arxiv.org/abs/1104.1643 http://stacks.iop.org/1367-2630/14/i=2/a=023008?key=crossref.4b77b3c1f75af0c5b967feca59a66322 [ * * , ( ) ] http://www.springerlink.com/index/rq75738643374j55.pdf [ * * , ( ) ] link:\doibase 10.1007/s00220 - 006 - 0030 - 4 [ * * , ( ) ] in @noop _ _ , , vol . , ( , ) pp . http://arxiv.org/abs/1102.0835 [ ] link:\doibase 10.1103/physrevlett.96.136801 [ * * , ( ) ] link:\doibase 10.1088/1742 - 5468/2007/06/p06008 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.06300 [ * * , ( ) ] link:\doibase 10.1007/s00220 - 008 - 0630 - 2 [ * * , ( ) ] link:\doibase 10.1088/1742 - 5468/2008/05/p05018 [ * * , ( ) ] link:\doibase 10.1103/physrevb.79.155104 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.102.240501 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.80.885 [ * * , ( ) ] http://arxiv.org/abs/1111.0776 http://www.nature.com/doifinder/10.1038/nature10748 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.97.156403 [ * * , ( ) ] link:\doibase 10.1103/physreva.80.063619 [ * * , ( ) ] link:\doibase 10.1103/physrevb.81.134305 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.85.3233 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.180406 [ * * , ( ) ] link:\doibase 10.1103/physreva.81.033605 [ * * , ( ) ] @noop link:\doibase 10.1103/physreva.69.022304 [ * * , ( ) ] link:\doibase 10.1103/physreva.74.053616 [ * * , ( ) ] http://stacks.iop.org/1742-5468/2006/p03001 [ ( ) ] link:\doibase 10.1103/physreva.78.012330 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/12/5/055017 [ * * , ( ) ] link:\doibase 10.1103/physrevb.84.165117 [ * * , ( ) ] http://arxiv.org/abs/1201.6674 [ ] http://prl.aps.org/abstract/prl/v89/i25/e250404 [ * * , ( ) ] link:\doibase 10.1103/physrevb.75.085106 [ * * , ( ) ] http://www.ncbi.nlm.nih.gov/pubmed/9982804 [ * * , ( ) ] link:\doibase 10.1103/physrevb.58.r14741 [ * * , ( ) ] link:\doibase 10.1209/epl / i2004 - 10265 - 7 [ * * , ( ) ] link:\doibase 10.1088/1742 - 5468/2010/08/p08011 [ ( ) ] link:\doibase 10.1103/physrevlett.69.2863 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.77.259 [ * * , ( ) ] http://link.aps.org/doi/10.1103/physrevlett.91.147902 [ * * , ( ) ] http://iopscience.iop.org/1742-5468/2004/04/p04005 [ ( ) ] link:\doibase 10.1103/physrevlett.93.076401 [ * * , ( ) ] @noop _ _ ( , , ) http://pra.aps.org/abstract/pra/v68/i4/e043623 [ * * , ( ) ] http://pra.aps.org/abstract/pra/v73/i4/e043607 [ * * , ( ) ] http://arxiv.org/abs/1106.1333 [ ] link:\doibase 10.1007/bf01331938 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.86.1082 [ * * , ( ) ] http://prl.aps.org/abstract/prl/v10/i5/p159_1 [ * * , ( ) ] http://ptp.ipap.jp/link?ptp/53/614/ [ * * , ( ) ] http://prb.aps.org/abstract/prb/v83/i16/e165105 [ * * , ( ) ] http://prb.aps.org/abstract/prb/v79/i17/e174504 [ * * , ( ) ] link:\doibase 10.1038/nature08482 [ * * , ( ) ] link:\doibase 10.1038/nature09378 [ * * , ( ) ] link:\doibase 10.1103/physreva.3.786 [ * * , ( ) ] link:\doibase 10.1126/science.1209284 [ * * , ( ) ] http://stacks.iop.org/0295-5075/26/i=7/a=012 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.98.070201 [ * * , ( ) ] http://arxiv.org/abs/0804.2509 [ ] link:\doibase 10.1103/physreva.78.033608 [ * * , ( ) ] ( , ) chap . link:\doibase 10.1103/physrev.130.1616 [ * * , ( ) ] link:\doibase 10.1103/physreva.71.053606 [ * * , ( ) ]

we study weak convergence of empirical processes of dependent data , indexed by classes of functions . our results are especially suitable for data arising from dynamical systems and markov chains , where the central limit theorem for partial sums of observables is commonly derived via the spectral gap technique . we are specifically interested in situations where the index class is different from the class of functions for which we have good properties of the observables . we introduce a new bracketing number to measure the size of the index class which fits this setting . our results apply to the empirical process of data satisfying a multiple mixing condition . this includes dynamical systems and markov chains , if the perron - frobenius operator or the markov operator has a spectral gap , but also extends beyond this class , e.g. to ergodic torus automorphisms .

1201.2256

let @xmath0 be a stationary stochastic process of @xmath4-valued random variables with marginal distribution @xmath5 . we denote the empirical measure of order @xmath6 by @xmath7 . the classical empirical process is defined by @xmath8)-\mu((-\infty , t]))$ ] , @xmath9 . in the case of i.i.d . processes , the limit behavior of the empirical process was first investigated by @xcite , who proved that @xmath10 converges weakly to a brownian bridge process . this result , known as donsker s empirical process central limit theorem , confirmed a conjecture of @xcite who had observed that certain functionals of the empirical process converge in distribution towards the corresponding functionals of a brownian bridge . donsker s empirical process clt has been generalized to dependent data by many authors . one of the earliest results is @xcite , who considered functions of mixing processes , with an application to the empirical distribution of the remainders in a continued fraction expansion . empirical processes play a very important role in large sample statistical inference . many statistical estimators and test statistics can be expressed as functionals of the empirical distribution . as a result , their asymptotic distribution can often be derived from empirical process limit theorems , combined with the continuous mapping theorem or a functional delta method . a well - known example is the kolmogorov - smirnov goodness - of - fit test , which uses the test statistic @xmath11)- \mu_0((-\infty , t])|$ ] in order to test the null hypothesis that @xmath12 is the marginal distribution of @xmath13 . under the null hypothesis , the limit distribution of @xmath14 is given by the supremum of the gaussian limit of the empirical process . another example are von - mises - statistics , also known as v - statistics . these are defined as @xmath15 , where @xmath16 is a symmetric kernel function . specific examples include the sample variance and gini s mean difference , where the kernel functions are given by @xmath17 and @xmath18 , respectively . @xmath19-statistics can be expressed as integrals with respect to the empirical distribution function , namely @xmath20 . the asymptotic distribution of @xmath21 can then be derived via a functional delta method from an empirical process central limit theorem ; see e.g. @xcite for some recent results . empirical process clts for @xmath22-valued i.i.d . data @xmath0 have first been studied by @xcite , @xcite , @xcite and @xcite . these authors consider the classical @xmath23-dimensional empirical process @xmath24 ) -\mu((-\infty , t]))$ ] , where @xmath25=\{x\in { { \mathbb r}}^d : x_1\leq t_1,\ldots , x_d\leq t_d\}$ ] , @xmath26 , denotes the semi - infinite rectangle in @xmath22 . @xcite , @xcite and @xcite studied weak convergence of the multivariate empirical process in the case of mixing data . @xcite initiated the study of empirical processes indexed by classes of sets , or more generally by classes of functions . this approach allows the study of empirical processes for very general data , not necessarily having values in euclidean space . clts for empirical processes indexed by classes of functions require entropy conditions on the size of the index set . for i.i.d . data , @xcite obtained the clt for empirical processes indexed by classes of sets satisfying an entropy condition with inclusion . @xcite used an entropy condition with bracketing to obtain results for empirical processes indexed by classes of functions . for the theory of empirical processes of i.i.d . data , indexed by classes of functions , see the book by @xcite . limit theorems for more general empirical processes indexed by classes of functions have also been studied under entropy conditions for general covering numbers , e.g. by @xcite who investigate empirical @xmath27-processes . in the case of strongly mixing data , @xcite were the first to obtain clts for empirical processes indexed by classes of functions . @xcite and @xcite study empirical processes for absolutely regular data . @xcite investigate the empirical process and the empirical @xmath27-process for data that can be represented as functionals of absolutely regular processes . for further results , see the survey article by @xcite , the book by @xcite , as well as the paper by @xcite . a lot of research has been devoted to the study of statistical properties of data arising from dynamical systems or from markov chains . a very powerful technique to prove clts and other limit theorems is the spectral gap method , using spectral properties of the perron - frobenius operator or the markov operator on an appropriate space of functions ; see @xcite . when the space of functions under consideration contains the class of indicator functions of intervals , standard tools can be used to establish the classical empirical process clt . finite - dimensional convergence of the empirical process follows from the clt for @xmath28}(x_i)$ ] , and tightness can be established using moment bounds for @xmath29}(x_i)$ ] . @xcite used this approach to establish the empirical process clt for expanding maps of the unit interval . the situation differs markedly when the clt and moment bounds are not directly available for the index class of the empirical process , but only for a different class of functions . recently , @xcite developed techniques to cover such situations . they were able to prove classical empirical process clts for @xmath4-valued data when the clt and moment bounds are only available for lipschitz functions . @xcite extended these techniques to @xmath22-valued data satisfying a multiple mixing condition for hlder continuous functions . under this condition , they proved the clt for the empirical process indexed by semi - infinite rectangles @xmath25 $ ] , @xmath30 . the multiple mixing condition is strictly weaker than the spectral gap condition . e.g. , ergodic torus automorphisms satisfy a multiple mixing condition , while generally they do not have a spectral gap . @xcite proved the empirical process clt for ergodic torus automorphisms . @xcite provide very general conditions under which the classical empirical process clt for @xmath22-valued data holds . the above mentioned papers study exclusively classical empirical processes , indexed by semi - infinite intervals or rectangles . it is the goal of the present paper to extend the techniques developed by @xcite to empirical processes indexed by classes of functions . let @xmath31 be a measurable space , let @xmath0 be a stationary process of @xmath32-valued random variables , and let @xmath1 be a uniformly bounded class of real - valued functions on @xmath32 . we consider the @xmath1-indexed empirical process @xmath33 . as in the above mentioned papers , we will assume that there exists some banach space @xmath34 of functions on @xmath32 such that the clt and a moment bound hold for partial sums @xmath35 , for all @xmath36 in some subset of @xmath34 ; see assumptions 1 and 2 . these conditions are satisfied , e.g. when the perron - frobenius operator or the markov operator acting on @xmath34 has a spectral gap . again , if the index class @xmath1 is a subset of @xmath34 , standard techniques for proving empirical process clts can be applied . in many examples , however , @xmath34 is some class of regular functions , while @xmath1 is a class of indicators of sets . it is the goal of the present paper to provide techniques suitable for this situation . empirical process invariance principles require a control on the size of the index class @xmath1 , as measured by covering or bracketing numbers ; see e.g. @xcite . in this paper , we will consider coverings of @xmath1 by @xmath34-brackets , i.e. brackets bounded by functions @xmath37 . because of the specific character of our moment bounds , we have to impose conditions on the @xmath34-norms of @xmath38 and @xmath39 . we will thus introduce a notion of bracketing numbers by counting how many @xmath34-brackets of a given @xmath40-size and with a given control on the @xmath34-norms of the upper and lower functions are needed to cover @xmath1 . the main theorem of the present paper establishes an empirical process clt under an integral condition on this bracketing number . this paper is organized as follows : section [ sec : main ] contains precise definitions as well as the statement of the main theorem . in section [ sec : examples ] , we will specifically consider the case when @xmath34 is the space of hlder continuous functions . we will give examples of classes of functions which satisfy the bracketing number assumption . in section [ sec : appli ] , we will give applications to ergodic torus automorphisms which extend the empirical process clt of @xcite to more general classes of sets . section [ sec : pf - ep - clt ] contains the proof of our main theorem , while proofs of technical aspects of the examples can be found in the appendix . let @xmath31 be a measurable space , and let @xmath41 be an @xmath32-valued stationary stochastic process with marginal distribution @xmath5 . let @xmath1 be a uniformly bounded class of real - valued measurable functions defined on @xmath32 . if @xmath42 is a signed measure on @xmath31 , we use the notation @xmath43 . we define the map @xmath44 , induced by the empirical measure , @xmath45 the @xmath1-indexed empirical process of order @xmath6 is given by @xmath46 we regard the empirical process @xmath47 as random element on @xmath48 ; this holds as @xmath1 is supposed to be uniformly bounded . @xmath49 is equipped with the supremum norm and the borel @xmath50-field generated by the open sets . it is well known that , in general , @xmath47 is not measurable and thus the usual theory of weak convergence of random variables does not apply . we use here the theory which is based on convergence of outer expectations ; see @xcite . given a borel probability measure @xmath51 on @xmath49 , we say that @xmath52 converges in distribution to @xmath51 if @xmath53 for all bounded and continuous functions @xmath54 . here @xmath55 denotes the outer integral . note that @xmath56 , where @xmath57 denotes the measurable cover function of @xmath58 ; see lemma 1.2.1 in @xcite . in what follows , we will frequently make two assumptions concerning the process @xmath59 , where @xmath60 belongs to some banach space @xmath61 of measurable functions on @xmath32 , respectively to some subset @xmath62 . the precise choice of @xmath34 , as well as of @xmath63 , will depend on the specific example . often , we take @xmath34 to be the space of all lipschitz or hlder continuous functions , and @xmath63 the intersection of @xmath34 with an @xmath64-ball . + [ [ assumption-1-clt - for - mathcal - b - observables ] ] * assumption 1 ( clt for @xmath34-observables ) : * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for all @xmath65 , there exists a @xmath66 such that @xmath67 where @xmath68 denotes the normal law with mean zero and variance @xmath69 . + [ [ assumption-2-moment - bounds - for - mathcal - g - observables ] ] * assumption 2 ( moment bounds for @xmath63-observables ) : * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for some subset @xmath62 , @xmath70 , and @xmath71 , for all @xmath72 , there exists a constant @xmath73 such that for all @xmath74 , @xmath75 \leq c_p\sum_{i=1}^p n^i \|f\|_s^i \log^{2p+ai } ( \|f\|_{{\mathcal b}}+1),\ ] ] where @xmath76 denotes the @xmath40-norm of @xmath2 . + both assumption 1 and assumption 2 have been established by many authors for a wide range of stationary processes . concerning the clt , see e.g. the three - volume monograph by @xcite for mixing processes , @xcite for so - called weakly dependent processes in the sense of @xcite , and @xcite for many examples of markov chains and dynamical systems . @xcite proved @xmath77th moment bounds of the type for markov chains or dynamical systems for which the markov operator or the perron - frobenius operator acting on @xmath34 has a spectral gap . it was generalized to @xmath78-th moment bounds by @xcite . more generally , they gave similar moment bounds for processes satisfying a multiple mixing condition , i.e.assuming that there exist a @xmath79 and an integer @xmath80 such that for all integers @xmath72 , there exist an integer @xmath81 and a multivariate polynomial @xmath82 of total degree smaller than @xmath83 such that @xmath84 holds for all @xmath65 with @xmath85 and @xmath86 , all integers @xmath87 and all @xmath88 . see theorem 4 and the examples in @xcite . note that this multiple mixing condition implies the moment bound with for @xmath89 and @xmath90 . further , the spectral gap property leads to the multiple mixing condition with @xmath91 , and thus to the moment bound with @xmath92 , see @xcite section 4 . we will derive a general statement about weak convergence of the empirical process @xmath47 under the two assumptions and . empirical process central limit theorems require bounds on the size of the class of functions @xmath1 , usually measured by the number of @xmath93-balls required to cover @xmath1 . here we will introduce a covering number adapted to the fact that and hold only for @xmath65 or @xmath94 , respectively , and that both the @xmath34-norm as well as the @xmath95-norm enter on the right hand side of the bound . in our approach , we use @xmath34-brackets to cover the class @xmath1 , which leads to the following definition . [ def : bracket ] let @xmath31 be a measurable space , and let @xmath5 be a probability measure on @xmath31 . let @xmath34 be some banach space of measurable functions on @xmath32 , @xmath62 and @xmath96 . + ( i ) given two functions @xmath97 satisfying @xmath98 , for all @xmath99 , we define the bracket @xmath100:=\{f:{{\mathcal x}}\rightarrow { { \mathbb r } } : l(x ) \leq f(x ) \leq u(x ) , \mbox { for all } x\in { { \mathcal x}}\}.\ ] ] given @xmath101 , we call @xmath102 $ ] an @xmath103-bracket , if @xmath104 and @xmath105 where @xmath106 denotes the @xmath95-norm . + ( ii ) for a class of measurable functions @xmath1 , defined on @xmath32 , we define the bracketing number @xmath107 as the smallest number of @xmath108-brackets needed to cover @xmath1 . our definition is close to the definition of bracketing numbers given by @xcite , but different . in @xcite , no assumptions are made on the upper and lower functions of the bracket other than that they are close in @xmath109 . here , the moment bound forces us to require the extra condition that @xmath39 and @xmath38 belong to the space @xmath34 and that their @xmath34-norms are controlled . obviously , our bracketing numbers are always larger than the ones defined in @xcite , and naturally our condition on the size of @xmath1 are stronger . on the other hand , our results apply to dependent data , while @xcite studies i.i.d . we can now state the main theorem of the present paper . the proof will be given in section [ sec : pf - ep - clt ] . [ th : ep - clt ] let @xmath31 be a measurable space , let @xmath110 be an @xmath32-valued stationary stochastic process with marginal distribution @xmath5 , and let @xmath1 be a uniformly bounded class of measurable functions on @xmath32 . suppose that for some banach space @xmath34 of measurable functions on @xmath32 , some subset @xmath62 , @xmath71 , and @xmath96 , assumptions 1 and 2 hold . moreover , assume that there exist constants @xmath111 , @xmath112 and @xmath113 such that @xmath114 then the empirical process @xmath47 converges in distribution in @xmath49 to a tight gaussian process @xmath115 . \(i ) note that the bracketing number @xmath116 might not be a monotone function of @xmath117 . this is the reason why we take the supremum in the integral . + ( ii ) the proof of theorem [ th : ep - clt ] shows that the statement also holds if condition is only satisfied for some integer @xmath118 satisfying @xmath119 ( iii ) if for some @xmath120 , @xmath121 as @xmath122 , condition is satisfied for all @xmath123 . in the next section , we will present examples of classes of functions satisfying condition . among the examples are indicators of multidimensional rectangles , of ellipsoids , and of balls of arbitrary metrics , as well as a class of monotone functions . in section [ sec : appli ] , we give applications to ergodic torus automorphisms , indexed by various classes of indicator functions . in many examples that satisfy assumptions 1 and 2 , the banach space @xmath34 is the space of lipschitz or hlder continuous functions , see examples in @xcite , @xcite , or @xcite . thus , in this section , we will restrict our attention to the case where @xmath34 is a space of hlder functions and give several examples of classes @xmath1 which satisfy the entropy condition . in this section , we consider a metric space @xmath124 . let @xmath125 $ ] be fixed . we denote by @xmath126 the space of bounded @xmath127-hlder continuous functions on @xmath32 with values in @xmath4 . this space is equipped with the norm @xmath128 for this section we chose @xmath129 . as the approximating class we use the subclass @xmath130):= \{f\in{{\mathcal h}}_\alpha({{\mathcal x } } ) : 0\leq f\leq 1 \}$ ] of @xmath34 . except in example [ exa : zero - centered - balls ] , in all examples we will consider the case where @xmath32 is a subset of @xmath22 equipped with the euclidean norm denoted by @xmath131 , where @xmath132 is some fixed integer . in most of the examples , we will use the transition function given in the following definition which uses the notations @xmath133 for any element @xmath134 and sets @xmath135 , @xmath136 , where we define @xmath137 . let @xmath135 , @xmath138 be subsets of @xmath32 such that @xmath139 . we define the transition function @xmath140:{{\mathcal x}}\rightarrow { { \mathbb r}}$ ] by @xmath141(x):= \frac{d_b(x)}{d_b(x)+d_a(x)},\ ] ] if @xmath135 and @xmath138 are non - empty , @xmath140:=0 $ ] if @xmath142 , and @xmath140:=1 $ ] if @xmath143 but @xmath144 . observe , that we have @xmath140({{\mathcal x}})\subset[0,1]$ ] , @xmath140(x)=1 $ ] for all @xmath145 and @xmath140(x)=0 $ ] for all @xmath146 . [ lem : transition ] for any subsets @xmath135 , @xmath138 of @xmath32 such that @xmath139 , the transition function @xmath140 $ ] is a bounded @xmath127-hlder continuous function and we have @xmath147\|_\alpha \leq 1 + \left(\frac{3}{d(a , b)}\right)^{\alpha}.\ ] ] this lemma is proved in the appendix . we also use the following notations : for a non - decreasing function @xmath148 from @xmath4 to @xmath4 , @xmath149 denotes the pseudo - inverse function defined by @xmath150 where @xmath151 . the modulus of continuity of @xmath148 is defined by @xmath152 constants that only depend on fixed parameters @xmath153 will be denoted with these parameters in the subscript , such as @xmath154 . furthermore the notation @xmath155 as @xmath156 or @xmath157 means that there exists a constant @xmath154 such that @xmath158 for all @xmath159 sufficiently small or large , respectively . here , we consider @xmath160 . in its classical form , the empirical process is defined by the class of indicator functions of left infinite rectangles , i.e. the class @xmath161 } : t\in{{\mathbb r}}^d \}$ ] , where @xmath25 $ ] denotes the set of points @xmath159 such that , we use the partial order : @xmath162 if and only if @xmath163 for all @xmath164 . ] @xmath162 . under similar assumptions as in the present paper , this case was treated by @xcite . we will see that theorem [ th : ep - clt ] covers the results of that paper . the following proposition gives an upper bound for the bracketing number of the larger class @xmath165 } : t , u\in[-\infty,+\infty]^d , t \le u\},\ ] ] where @xmath166 $ ] denotes the rectangle which consists of the points @xmath159 such that @xmath167 and @xmath168 . [ prop - rect ] let @xmath70 , @xmath169 , and let @xmath5 be a probability distribution on @xmath22 whose distribution function @xmath148 satisfies @xmath170 then there exists a constant @xmath171 such that @xmath172 where @xmath173)$ ] . let @xmath174 and @xmath175 . for all @xmath176 and @xmath177 , we define the quantiles @xmath178 where @xmath179 is the pseudo - inverse of the marginal distribution function\times{{\mathbb r}}\times\dots\times{{\mathbb r}})$ ] ] @xmath180 . now , if @xmath181 , we write @xmath182 in the following definitions , for convenience , we will also denote by @xmath183 or @xmath184 the points @xmath185 and by @xmath186 the points @xmath187 . we introduce the brackets @xmath188 $ ] , @xmath189 , @xmath190 , @xmath191 , given by the @xmath127-hlder functions @xmath192,{{\mathbb r}}^d\setminus[t_{k},t_{j-1 } ] \right](x),\ ] ] and @xmath193,{{\mathbb r}}^d\setminus[t_{k-2},t_{j+1}]\right](x),\ ] ] where we have used the convention that @xmath194=\emptyset$ ] if @xmath195 and that the addition of an integer to a multi - index is the addition of the integer to every component of the multi - index . for each @xmath191 , we have @xmath196\setminus[t_{k+1},t_{j-2 } ] \right)\\ & \le & \sum_{i=1}^d |f_i(t_{i , k_i+1})-f_i(t_{i , k_i-2})|+|f_i(t_{i , j_i+1})-f_i(t_{i , j_i-2})| \le 2\frac{3d}{m},\end{aligned}\ ] ] and thus @xmath197 . moreover , since for @xmath198 , @xmath199,{{\mathbb r}}^d\setminus[a , a'])=\inf_{i=1,\dots , d } \inf \left\{|a_i - b_i|,|a'_i - b'_i|\right\},\ ] ] using lemma [ lem : transition ] and ( [ eq - modulus ] ) , we have @xmath200^{-\alpha } \\ & \le & 1 + 3^\alpha\left[\inf\left\{x>0 : c_f|log(x)|^{-s\gamma}\ge \frac{1}{m}\right\}\right]^{-\alpha}\\ & \le & 1 + 3^\alpha \exp \left ( \alpha ( c_f m)^{\frac{1}{s\gamma}}\right),\end{aligned}\ ] ] where @xmath201 is given by . the same bound holds for @xmath202 . thus , there exists a new constant @xmath203 such that for all @xmath204 , @xmath188 $ ] is an @xmath205-bracket . it is clear that for each function @xmath206 there exists a bracket of the form @xmath188 $ ] which contains @xmath2 . further , we have at most @xmath207 such brackets , which proves the proposition . notice that under the assumptions of the proposition , condition ( [ bracket ] ) is satisfied and therefore theorem [ th : ep - clt ] may be applied to empirical processes indexed by the class of indicators of rectangles , taking @xmath34 to be the class of bounded hlder functions . let @xmath208 be an @xmath22-valued stationary process . let @xmath1 be the class of indicator functions of rectangles in @xmath22 and let @xmath173)$ ] . assume that , for some @xmath70 , @xmath71 , and @xmath209 , assumptions 1 and 2 hold , and that the distribution function of the @xmath210 satisfies ( [ eq - modulus ] ) . then the empirical process @xmath211 converges in distribution in @xmath49 to a tight gaussian process . by regarding the class of indicator functions of left infinite rectangles as a sub - class of @xmath1 , we obtain theorem 1 of @xcite as a particular case of the preceding corollary . [ exa : balls ] here , we consider the class @xmath1 of indicator functions of balls on @xmath212^d$ ] , i.e. @xmath213^d , r\ge 0\}\ ] ] where @xmath214^d : |x - y|<r\}$ ] . we have the following upper bound . [ prop - balls ] let @xmath5 be a probability distribution on @xmath215^d$ ] with a density bounded by some @xmath216 and let @xmath70 . then there exists a constant @xmath217 such that @xmath218 where @xmath219^d,[0,1])$ ] . note that the second argument in the bracketing number is different from the one appearing in the condition . in this situation we have a stronger type of bracketing number than in . let @xmath220 be fixed and @xmath221 . for all @xmath222 , we denote by @xmath223 the center of the rectangle @xmath224\times\dots\times[\frac{i_d-1}{m},\frac{i_d}{m}]$ ] . then we define , for @xmath225 and @xmath226 , the functions @xmath227^d\setminus \operatorname{b}\left(c_{i},\frac{j-1}{m}\sqrt{d}\right ) \right](x)\ ] ] and @xmath228^d\setminus \operatorname{b}\left(c_{i},\frac{j+3}{m}\sqrt{d}\right)\right](x),\ ] ] where we use the convention that a ball with negative radius is the empty set . by lemma [ lem : transition ] , these functions are @xmath127-hlder and , since @xmath229 , we have @xmath230 the same bound holds for @xmath231 . since @xmath5 has a bounded density with respect to lebesgue measure , we also have @xmath232 where @xmath233 is the constant @xmath234 ( @xmath235 is the gamma function ) . hence , @xmath236 as @xmath237 , where @xmath238 is a constant depending only on @xmath23 and @xmath138 now , if @xmath2 belongs to @xmath1 , then @xmath239 for some @xmath240^d$ ] , and @xmath241 . thus , there exist some @xmath242 and @xmath226 such that @xmath243 we then have @xmath244 . thus , the @xmath245 brackets @xmath246 $ ] , @xmath225 and @xmath226 , cover the class @xmath1 . therefore , @xmath247 as @xmath237 , which implies that there exists a constant @xmath248 , for which @xmath249 as @xmath237 . [ exa : ellipsoids_cube ] set @xmath250 . here , we consider the class of ellipsoids which are aligned with the coordinate axes , have their center in @xmath215^d$ ] , and their parameters bounded by some constant @xmath251 . without loss of generality , we assume that @xmath252 . for @xmath253^d$ ] and all @xmath254^d$ ] , we set @xmath255 we denote by @xmath1 the class of indicator functions of these ellipsoids , i.e. @xmath256^d , r\in [ 0,d]^d\}.\ ] ] we have the following upper bound . [ pro : ellipsoids_unit_square ] let @xmath5 be a probability distribution on @xmath22 with a density bounded by some @xmath216 . then there exists a constant @xmath257 such that @xmath258 where @xmath173)$ ] . let @xmath259 be fixed and @xmath221 . for all @xmath222 , we denote by @xmath260 the rectangle @xmath224\times\dots\times[\frac{i_d-1}{m},\frac{i_d}{m}]$ ] . then , for @xmath225 and @xmath261 , we define the sets @xmath262 and @xmath263 we introduce the bracket @xmath246 $ ] given by @xmath264(x ) \;\mbox { and } \ ; u_{i , j}(x):=t\left[u_{i , j+1 } , { { \mathbb r}}^d\setminus u_{i , j+2 } \right](x),\ ] ] where we use the convention that an ellipsoid with one negative parameter is the empty set . by lemma [ lem : transition ] , these functions are @xmath127-hlder . further , we have the following lemma which is proved in appendix : [ lem : dist_ellip ] for all @xmath265 , @xmath266 , we have @xmath267 as a consequence we infer that the distance between @xmath268 and @xmath269 is at least @xmath270 and the distance between @xmath271 and @xmath272 is at least @xmath270 . thus , by lemma [ lem : transition ] , we have @xmath273 and the same bound holds for @xmath231 . now , to bound @xmath274 we need to estimate the lebesgue measures of @xmath268 and @xmath271 . recall that , if @xmath275 and @xmath266 , the lebesgue measure of the ellipsoid @xmath276 is given by @xmath277 where @xmath233 is the constant @xmath234 . the set @xmath268 can be seen as the set constructed as follows : start from an ellipsoid of parameters @xmath278 centered at the center of @xmath260 , cut it along its hyperplanes of symmetry , and shift each obtained component away from the center by a distance of @xmath279 in every direction ; @xmath268 is then the convex hull of these @xmath280 components ( see figure [ ellip1 ] for the dimension 2 ) . 2 in dimension 2,title="fig : " ] in dimension 2 ] let us denote by @xmath281 the set that has been added to the @xmath280 components to obtain the convex hull . we can bound the volume of @xmath268 by the volume of the ellipsoid plus a bound on the volume of @xmath281 , that is @xmath282 the set @xmath271 can be seen an the intersection of the @xmath280 ellipsoids of parameters @xmath278 centered at each corner of the hypercube @xmath260 ( see figure [ ellip2 ] for the dimension 2 ) . its volume is larger than the volume of an ellipsoid of parameters @xmath278 minus the volume of @xmath281 . we thus have @xmath283 since @xmath5 has a bounded density with respect to lebesgue measure , we have @xmath284 we infer @xmath285 , as @xmath237 , where the constant @xmath238 only depends on @xmath23 and @xmath138 . now , if @xmath2 belongs to @xmath1 , then @xmath286 for some @xmath287 , and @xmath288^d$ ] . thus , there exist some @xmath242 and @xmath289 such that @xmath290 and for each @xmath291 , @xmath292 we then have @xmath244 . thus , the @xmath293 brackets @xmath246 $ ] , @xmath225 and @xmath289 , cover the class @xmath1 . therefore , there exists a @xmath294 , such that @xmath295 , as @xmath237 . in , we only considered indicators of ellipsoids centered in a compact subset of @xmath22 , namely the unit square . the following lemma will allow us to extend such results to indicators of sets in the whole @xmath22 , at the cost of a moderate additional assumption and a marginal increase of the bracketing numbers . [ lem : extension - lemma ] let @xmath5 be a measure with continuous distribution function @xmath148 , and @xmath70 . furthermore let @xmath296 , where @xmath297 is a class of measurable sets of diameter not larger than @xmath298 , and @xmath173)$ ] . assume that there are constants @xmath299 , @xmath113 , and a function @xmath300 , such that for any @xmath301 we have @xmath302 for sufficiently small @xmath303 , where @xmath304^d \}$ ] . if there are some constants @xmath305 such that @xmath306 for all sufficiently large @xmath307 , then @xmath308 where @xmath309 is the modulus of continuity of @xmath148 . the proof is postponed in appendix . [ cor : ellips_arb]let @xmath310 denote the class of indicators of ellipsoids of diameter uniformly bounded by @xmath251 , which are aligned with coordinate axes ( and arbitrary centers in the whole space @xmath22 ) . if @xmath5 is a measure on @xmath22 with a density bounded by @xmath216 and if furthermore holds for some @xmath311 and @xmath312 , then there exists a constant @xmath257 such that @xmath313 where @xmath173)$ ] . in the situation of change the set of the centers of the ellipsoids @xmath215^d$ ] to @xmath314^d$ ] and apply lemma [ lem : extension - lemma ] . following the proof of proposition [ pro : ellipsoids_unit_square ] we can easily see that condition holds for @xmath315 , @xmath316 and @xmath317 . note that since we have a bounded density , we have @xmath318 and therefore @xmath319 for sufficiently small @xmath303 . in the situation of proposition [ cor : ellips_arb ] for the class @xmath320 of indicators of balls in @xmath22 with uniformly bounded diameter , we can obtain the slightly sharper bound @xmath321 for some @xmath322 by applying lemma [ lem : extension - lemma ] directly to the situation in and using the same arguments as in the previous example . [ exa : zero - centered - balls ] let @xmath124 be a metric space and fix @xmath323 . a @xmath324-centered ball is given by @xmath325 we have the following bound on the bracketing numbers of the class @xmath326 . [ pro : zero - centered ] let @xmath70 and @xmath169 . if for the probability measure @xmath5 on @xmath32 the modulus of continuity @xmath327 of the function @xmath328 satisfies @xmath329 then there is a constant @xmath330 such that @xmath331 where @xmath332)$ ] . note that in the case that @xmath333 , @xmath334 , the metric @xmath23 is given by the euclidean norm , and @xmath335 , an equivalent condition to is @xmath336 fix @xmath220 and choose @xmath337 . let @xmath338 denote the pseudo - inverse of @xmath339 and set for @xmath340 @xmath341 for convenience set @xmath342 and @xmath343 . define @xmath344(x)\;\mbox { and } \ ; u_i(x):=t\left[\operatorname{b}_{i } , { { \mathcal x}}\setminus \operatorname{b}_{i+1 } \right](x)\ ] ] the system @xmath345 : i\in\{1,\ldots , m\}\}$ ] is a covering for @xmath346 . obviously @xmath347 by lemma [ lem : transition ] , we have @xmath348 since by condition @xmath349 for some constant @xmath350 , there is a constant @xmath351 such that @xmath352 analogously , we can show that @xmath353 . this implies that all @xmath354 $ ] are @xmath355-brackets and thus the proposition is proved . [ exa : class - funct ] in this example , we choose @xmath356 . we consider the case of a one - parameter class of functions @xmath357\}$ ] , where @xmath358 are functions from @xmath4 to @xmath4 with the properties : a. [ 6i ] for all @xmath359 $ ] and @xmath360 , @xmath361 ; b. [ 6ii ] for all @xmath362 , @xmath363 ; c. [ 6iii ] for all @xmath364 $ ] , @xmath358 is non - decreasing on @xmath4 . note that all the sequel remains true if in , non - decreasing is replaced by non - increasing . further , for a probability measure @xmath5 on @xmath4 , we define @xmath365 and we say that @xmath366 is lipschitz with lipschitz constant @xmath367 if @xmath368 , for all @xmath369 $ ] . empirical processes indexed by a 1-parameter class of functions arise , e.g. in the study of empirical u - processes ; see @xcite . the empirical u - distribution function with kernel function @xmath370 is defined as @xmath371 then , the first order term in the hoeffding decomposition is given by @xmath372 where @xmath373 . for this class of functions , conditions and are automatically satisfied . condition holds , if @xmath370 is monotone in @xmath159 . this is e.g. the case for the kernel @xmath374 , which arises in the study of the empirical correlation integral ; see @xcite . let @xmath70 and @xmath169 . let @xmath5 be a probability measure on @xmath4 such that its distribution function @xmath148 satisfies @xmath375 and such that @xmath366 is lipschitz with lipschitz constant @xmath367 . then there exists a @xmath171 , such that @xmath376 where @xmath377)$ ] . let @xmath220 and @xmath378 . for @xmath379 , we set @xmath380 we always have @xmath381 , but @xmath324 could be finite or @xmath382 . in order to simplify the notation , in the first case , we change to @xmath383 . we define , for @xmath384 , the functions @xmath385 and @xmath386 as follows . if @xmath387 , we set @xmath388 and @xmath389 , where we have to understand @xmath390 as @xmath391 . if @xmath392 and @xmath393 , we define @xmath394 and @xmath395 by the linear interpolations , @xmath396 @xmath397 with the exceptions that @xmath398 if @xmath399 and @xmath400 if @xmath401 . then it is clear that for all @xmath402 , we have @xmath403 , i.e. @xmath358 belongs to the bracket @xmath404 $ ] . further , being piecewise affine functions , @xmath385 and @xmath386 are @xmath127-hlder continuous functions with hlder norm @xmath405 here we have used the condition and the same computation as for the class of indicators of rectangles . analogously , the same bound holds for @xmath406 . now , @xmath407 first , since @xmath366 is lipschitz , we have @xmath408 for @xmath409 $ ] , since @xmath358 is nondecreasing , we have @xmath410 and @xmath411 , thus @xmath412)\\ & \le \frac1 m \sum_{k=1}^{m-1 } ( |f_{t_j}(x_{k+1})-f_{t_j}(x_{k})|+|f_{t_j}(x_{k})-f_{t_j}(x_{k-1})|)\\ & \le \frac2 m \sum_{k=0}^{m-1 } |f_{t_j}(x_{k+1})-f_{t_j}(x_{k})|\\ & \le \frac2m\end{aligned}\ ] ] since , by monotonicity , @xmath413 . in the same way we get @xmath414 and we infer @xmath415 thus , the number of @xmath416-brackets needed to cover the class @xmath1 is bounded by @xmath417 , which proves the proposition . @xmath418 we can apply theorem [ th : ep - clt ] to the empirical process of ergodic torus automorphisms . let @xmath419 be the torus of dimension @xmath420 , which is identified with @xmath215^d$ ] . if @xmath135 is a square matrix of dimension @xmath23 with integer coefficients and determinant @xmath421 , then the transformation @xmath422 defined by @xmath423 is an automorphism of @xmath424 that preserves the lebesgue measure @xmath425 . thus @xmath426 is a measure preserving dynamical system . it is ergodic if and only if the matrix @xmath135 has no eigenvalue which is a root of unity . a result of kronecker shows that in this case , @xmath135 always has at least one eigenvalue which has modulus different than @xmath427 . the hyperbolic automorphisms ( i.e. no eigenvalue of modulus @xmath427 ) are particular cases of anosov diffeomorphisms . their properties are better understood than in the general case . however , the general case of ergodic automorphisms is an example of a partially hyperbolic system for which strong results can be proved . the central limit theorem for regular observables has been proved by @xcite , see also @xcite for refinements . other limit theorems can be found in @xcite . the one - dimensional empirical process , for @xmath4-valued regular observables , has been studied by @xcite . @xcite proved weak convergence of the classical empirical process ( indexed by indicators of left infinite rectangles ) . we can now generalize this result to empirical processes indexed by further classes of functions . we can get the following proposition , as a corollary of theorem [ th : ep - clt ] and the results of the preceding section . let @xmath428 be an ergodic d - torus automorphism and let @xmath1 be one of the following classes : * the class of indicators of rectangles of @xmath424 ; * the class of indicators of euclidean balls of @xmath424 ; * the class of indicators of ellipsoids of bounded diameter of @xmath424 ; then the empirical process @xmath429 converges in distribution in @xmath49 to a tight gaussian process @xmath430 . [ th : ep - torus ] let @xmath1 be one of the classes of functions and @xmath34 be the class of @xmath127-hlder functions for some @xmath431 . we set @xmath63 the subclass of @xmath34 given by the functions bounded by @xmath427 . we consider the @xmath424-valued stationary process @xmath432 . since the distribution of @xmath433 is the lebesgue measure on @xmath424 , propositions [ prop - rect ] , [ prop - balls ] and [ pro : ellipsoids_unit_square ] show that the condition ( [ bracket ] ) holds for every possible choice of class @xmath1 . for all @xmath434 , the central limit theorem ( [ clt ] ) holds ; see @xcite and @xcite . @xcite , proposition 3 , show that the ergodic automorphisms of the torus satisfy the multiple mixing property for functions of the class @xmath63 , and with the constants @xmath435 and @xmath83 the size of the biggest jordan s block of @xmath428 restricted to its neutral subspace . thus the @xmath78-th moment bound ( [ 2pbound ] ) holds , and theorem [ th : ep - clt ] can be applied to conclude . in the proof of theorem [ th : ep - clt ] , we need a generalization of theorem 4.2 of @xcite . billingsley considers random variables @xmath436 , @xmath437 , @xmath438 , @xmath58 , @xmath439 , with values in a separable metric space @xmath440 satisfying ( a ) @xmath441 as @xmath442 , for all @xmath443 , ( b ) @xmath444 as @xmath445 and ( c ) @xmath446 , @xmath447 as @xmath445 . theorem 4.2 of @xcite states that then @xmath448 . @xcite proved that this result holds without condition ( b ) , provided that @xmath449 is a complete separable metric space . more precisely , they could show that in this situation ( a ) and ( c ) together imply the existence of a random variable @xmath58 satisfying ( b ) , and thus by billingsley s theorem @xmath448 . here we will generalize this theorem to possibly non - measurable random elements with values in non - separable spaces . regarding convergence in distribution of non - measurable random elements , we use the notation of @xcite . in accordance with the terminology of @xcite , we will call a not necessarily measurable function with values in a measurable space a random element . [ sat : bil_erw * ] let @xmath450 , @xmath439 , be random elements with values in a complete metric space @xmath440 , and suppose that @xmath438 is measurable and separable , i.e. there is a separable set @xmath451 such that @xmath452 . if the conditions @xmath453 are satisfied , then there exists an @xmath449-valued , separable random variable @xmath58 such that @xmath454 as @xmath445 , and @xmath455 the proof is postponed to the appendix . for all @xmath456 , there exist two sets of @xmath457 functions @xmath458 and @xmath459 , such that @xmath460 , @xmath461 , @xmath462 and for all @xmath206 , there exists an @xmath463 such that @xmath464 . further , by ( [ bracket ] ) , @xmath465 for all @xmath456 , we can build a partition @xmath466 of the class @xmath1 into @xmath467 subsets such that for all @xmath468 , @xmath464 . to see this define @xmath469 $ ] and @xmath470 \backslash ( \cup_{j=1}^{i-1 } { { \mathcal f}}_j$ ] ) . in the sequel , we will use the notation @xmath471 and @xmath472 if @xmath468 . for each @xmath456 , we introduce the process @xmath473 which is constant on each @xmath474 . further , if @xmath468 , we have @xmath475 we introduce @xmath476 [ prop:1 ] for all @xmath456 , the sequence @xmath477 converges in distribution in @xmath49 to a piecewise constant gaussian process @xmath478 as @xmath479 . since @xmath480 and @xmath63 is a subset of @xmath34 , by assumption ( [ clt ] ) , the clt holds and @xmath481 converges to a gaussian law for all @xmath206 . we can apply the cramr - wold device to get the finite dimensional convergence : for all @xmath482 , for all @xmath483 , @xmath484 converges in distribution to a gaussian vector @xmath485 in @xmath486 . since @xmath487 is constant on each element @xmath474 of the partition , the finite dimensional convergence implies the weak convergence of the process . indeed , consider the function @xmath488 that maps a vector @xmath489 to the function @xmath490 , @xmath491 such that @xmath468 . for @xmath492 we have @xmath493 and thus the continuous mapping theorem guarantees that @xmath487 converges weakly to the random variable @xmath494 which is constant on each @xmath474 . @xmath418 [ prop:2 ] for all @xmath220 , @xmath495 there exists a @xmath496 such that for all @xmath497 @xmath498 for a random variable @xmath499 let @xmath500 denote its centering @xmath501 . if for arbitrary random variables @xmath502 we have @xmath503 then @xmath504 using @xmath505 and @xmath506 for all @xmath206 , we obtain @xmath507 in order to assure @xmath508 , for fixed @xmath6 and @xmath509 , choose @xmath510 , where @xmath511 for each @xmath512 , we obtain @xmath513 by taking @xmath514 , @xmath515 and we get for each @xmath512 , @xmath516 notice that the suprema in the r.h.s . are in fact maxima over finite numbers of functions , since the functionals @xmath517 and @xmath518 ( and thus @xmath487 ) are constant on the @xmath519 . therefore we can work with standard probability theory from this point : the outer probabilities can be replaced by usual probabilities on the right hand side . for each @xmath520 choose a set @xmath521 composed by at most @xmath522 functions of @xmath1 in such a way that @xmath523 contains one function in each non empty @xmath524 , @xmath525 , @xmath526 . then , for each @xmath512 , we have @xmath527 now using markov s inequality at the order @xmath78 ( @xmath118 will be chosen later ) and assumption ( [ 2pbound ] ) , we infer @xmath528 at this point , without loss of generality , we can assume that @xmath529 ( if not , take a larger @xmath530 ) and thus the assumption on @xmath531 reduces to @xmath532 . note that by construction , for each @xmath482 , @xmath533 thus , @xmath534 and if @xmath509 is large enough , @xmath535 where @xmath536 is a new constant which depends on @xmath118 , @xmath537 , and @xmath538 . since @xmath539 is a partition of @xmath1 , we have @xmath540 thus we have @xmath541 because @xmath529 and thus @xmath542 , and where @xmath543 and @xmath544 are positive constants also depending on @xmath118 , @xmath537 , and @xmath538 . as @xmath545 when @xmath118 tends to infinity , there exists some @xmath546 such that @xmath547[cond_on_p ] and thus by , @xmath548 therefore the first summand of goes to zero as @xmath6 goes to infinity and the second summand of goes to zero as @xmath509 goes to infinity . propositions [ prop:1 ] and [ prop:2 ] establish for the random elements @xmath549 , @xmath487 , @xmath550 with value in the complete metric space @xmath49 conditions and of theorem [ sat : bil_erw * ] , respectively . thus theorem [ sat : bil_erw * ] completes the proof of theorem [ th : ep - clt ] . by the triangle inequality , we have for all @xmath551 that @xmath552 therefore , @xmath553(x ) - t[a , b](y)| \\ = \ & \left| \frac{\bigl(d_b(x)-d_b(y)\bigr)\bigl(d_b(y)+d_a(y)\bigr)\ + \ d_b(y)\bigl(d_b(y)+d_a(y)\bigr)-d_b(y)\bigl(d_b(x)+d_a(x)\bigr)}{\bigl(d_b(x)+d_a(x)\bigr)\bigl(d_b(y)+d_a(y)\bigr ) } \right| \\ = \ & \left| \frac{d_b(x)-d_b(y)}{d_b(x)+d_a(x ) } \right| \ + \ \frac{d_b(y)}{d_b(y)+d_a(y ) } \left| \frac{\bigl ( d_b(y)-d_b(x)\bigr ) + \bigl(d_a(y ) - d_a(x)\bigr)}{d_b(x)+d_a(x ) } \right| \\ \leq\ & 3\ \frac{d(x , y)}{d(a , b)}\end{aligned}\ ] ] and thus @xmath554\|_\alpha & : = \|t[a , b]\|_\infty + \sup_{x\ne y } \frac{\bigl|t[a , b](x)-t[a , b](y)\bigr|}{d(x , y)^\alpha}\\ & \leq 1 + \sup_{x\ne y } \left ( \frac{\bigl|t[a , b](x)-t[a , b](y)\bigr|}{d(x , y)}\right)^\alpha \bigl| t[a , b](x)-t[a , b](y ) \bigr|^{1-\alpha } \\ & \leq 1 + \left ( \frac{3}{d(a , b)}\right)^\alpha . \end{aligned}\ ] ] without loss of generality , assume that @xmath555 . for @xmath556 , let @xmath557 denote the diagonal @xmath558-matrix with diagonal entries @xmath559 . we define the operator norm of the @xmath558-matrix @xmath135 by @xmath560 . observe that @xmath561 . we can characterize @xmath562 and @xmath563 by @xmath564 respectively . thus , for any @xmath565 and @xmath566 , @xmath567 since @xmath568 . for any @xmath220 , set @xmath569^d ) \leq 1 - \varepsilon \}$ ] . we will denote the function @xmath570 , @xmath571 by @xmath572 . now , introduce the bracket @xmath573 $ ] , given by @xmath574^d , [ -k_{\varepsilon^s},k_{\varepsilon^s}]^d\right].\end{aligned}\ ] ] obviously , we have @xmath575 . to get a bound for the hlder - norm of @xmath576 , consider the distribution function @xmath577 on @xmath4 , where @xmath578 . observe that the pseudo - inverse @xmath338 of @xmath339 is linked to @xmath572 by the equality @xmath579 . with geometrical arguments we infer @xmath580 where @xmath581 . therefore @xmath582 now by lemma [ lem : transition ] we obtain @xmath583 where we used that @xmath309 is continuous here to replace the infimum by the supremum . then @xmath573 $ ] is an @xmath584-bracket for sufficiently small @xmath303 . since @xmath573 $ ] contains any @xmath585 , by we obtain for all those @xmath303 the bound @xmath586 let us finally consider the growth rate of @xmath587 as @xmath588 . by assumption and since @xmath589 , we have @xmath590 for sufficiently large @xmath307 . therefore , @xmath591 by the definition of @xmath572 , we therefore obtain that @xmath592 which proves the lemma . \(i ) we will first show that @xmath438 converges in distribution to some random variable @xmath58 . we denote by @xmath593 the distribution of @xmath438 ; this is defined since @xmath438 is measurable . moreover , @xmath593 is a separable borel probability measure on @xmath449 . by theorem 1.12.4 of @xcite , weak convergence of separable borel measures on a metric space @xmath449 can be metrized by the bounded lipschitz metric , defined by @xmath594 for any borel measures @xmath595 on @xmath449 . here , @xmath596 , where @xmath597 in addition , the theorem states that the space of all separable borel measures on a complete space is complete with respect to the bounded lipschitz metric . thus it suffices to show that @xmath593 is a @xmath598-cauchy sequence . we obtain @xmath599 for all @xmath600 . for a borel measurable separable random element @xmath438 weak convergence @xmath601 as @xmath602 is equivalent to @xmath603 ; see @xcite . hence by we obtain @xmath604 using lemma 1.2.2 ( iii ) in @xcite , we obtain @xmath605 and therefore @xmath606 where we used the last statement of lemma 1.2.2 in @xcite . now , let @xmath607 be given . by , there exists an @xmath608 such that for every @xmath609 there is some @xmath610 such that for every @xmath611 we have @xmath612 . therefore @xmath613 applying this inequality to , we obtain @xmath614 for all @xmath609 . hence for @xmath615 we have @xmath616 ; i.e. @xmath617 is a @xmath598-cauchy sequence in a complete metric space . \(ii ) the remaining part of the proof follows closely the proof of theorem 4.2 in @xcite , replacing the probability measure @xmath618 by the outer measure @xmath619 where necessary and making use of the portmanteau theorem ; see @xcite , theorem 1.3.4 ( iii ) , and the sub - additivity of outer measures . from part ( i ) , we already know that there is some measurable @xmath58 such that @xmath620 . let @xmath621 be closed . given @xmath607 , we define the @xmath93-neighborhood @xmath622 , and observe that @xmath623 is also closed . since @xmath624 , we obtain @xmath625 for all @xmath626 . by we may choose @xmath627 so large that for all @xmath609 @xmath628 as @xmath620 , by the portmanteau theorem we may choose @xmath629 so large that for all @xmath630 @xmath631we now fix @xmath632 . by we have @xmath633 as @xmath442 . thus an application of the portmanteau theorem yields @xmath634 @xmath635 since this holds for any @xmath607 and @xmath636 , we get @xmath637 for all closed sets @xmath621 . by a final application of the portmanteau theorem we infer @xmath638 . the authors thank the referees and the associate editor for their very careful reading and for their most constructive criticism of an earlier version of this paper that greatly helped to improve the presentation . s. borovkova , r. burton , and h. dehling . limit theorems for functionals of mixing processes with applications to @xmath27-statistics and dimension estimation . _ trans . _ , 3530 ( 11):0 42614318 ( electronic ) , 2001 . h. hennion and l. herv . _ limit theorems for markov chains and stochastic properties of dynamical systems by quasi - compactness _ , volume 1766 of _ lecture notes in mathematics_. springer - verlag , berlin , 2001 . w. philipp . invariance principles for sums of mixing random elements and the multivariate empirical process . in _ limit theorems in probability and statistics , vol . i , ii ( veszprm , 1982 ) _ , volume 36 of _ colloq . jnos bolyai _ , pages 843873 . north - holland , amsterdam , 1984 . m. l. straf . weak convergence of stochastic processes with several parameters . in _ proceedings of the sixth berkeley symposium on mathematical statistics and probability ( univ . california , berkeley , calif . , 1970/1971 ) , vol . ii : probability theory _ , pages 187221 , berkeley , calif . , california press .

triumf experiment 497 has measured the parity violating longitudinal analyzing power , , in elastic scattering at 221.3 mev incident proton energy . this comprehensive paper includes details of the corrections , some of magnitude comparable to itself , required to arrive at the final result . the largest correction was for the effects of first moments of transverse polarization . the addition of the result , , to the parity violation experimental data base greatly improves the experimental constraints on the weak meson - nucleon coupling constants and , and also has implications for the interpretation of electron parity violation experiments .

nucl-ex0211020

this experiment determines the parity - violating longitudinal analyzing power , @xmath5 , in @xmath1 elastic scattering , where @xmath6 and @xmath7 are the scattering cross sections for positive and negative helicity . the measurements were performed in transmission geometry , with beam energy and detector geometries selected to ensure that parity mixing in the lowest order @xmath8 partial wave amplitude did not contribute to the measured @xmath0 , hence leaving a parity violating asymmetry arising almost entirely from @xmath9 mixing @xcite . this amplitude has not previously been studied experimentally , and the possibility is unique to the energy regime accessible with the triumf cyclotron . the energy at which the contribution to @xmath0 from the lowest order @xmath8 mixing vanishes is determined by the well known strong nuclear phase shifts . the scale factors multiplying this and other partial wave contributions are set by the weak interaction . @xmath1 parity violation experiments determine these scale factors experimentally . in the context of the weak meson exchange model @xcite , the triumf measurement of @xmath0 determines primarily the weak @xmath10-meson - nucleon coupling constant @xmath3 = ( @xmath11 ) , where the superscripts refer to isospin change @xcite . precision results already obtained by the sin group at 45 mev @xcite and the bonn group at 13.6 mev @xcite determined essentially the sum @xmath12 , where @xmath13 . with the addition of the triumf result at 221.3 mev , @xmath3 and @xmath4 are now determined separately for the first time . it is important to have experimentally determined values of the weak meson - nucleon couplings , as theoretical calculations of their values are quite uncertain and the correct values are helpful in the interpretation of the results of other parity violation experiments . they are needed , for example , in calculations of the proton anapole moment @xcite , one of the radiative corrections to the electron - nucleon isovector axial form factor in experiments such as sample @xcite and g@xmath14 @xcite . triumf e497 @xcite , was first funded in 1988 , and the required new beamline was completed in 1994 . a major effort to understand and minimize systematic error contributions was then undertaken . following many years of effort that resulted in the reduction of both the helicity correlated beam modulations @xmath15 and the sensitivities @xmath16 , to these modulations , the first significant data set for e497 was acquired in february and march of 1997 , with a statistical error on @xmath0 of @xmath17 , and most systematic errors at or below the 10@xmath18 level . that result represented a major milestone for the experiment @xcite . data taking continued in 1998 and 1999 , the final re - analysis of all the data was completed in early 2001 @xcite , and the result for @xmath0 was published @xcite . the present paper presents detailed descriptions of the experiment , the data analysis , and systematic error corrections , that were not included in the letter @xcite . the principle of the experiment is straightforward . a longitudinally polarized proton beam is passed through a liquid hydrogen target and the change in transmission when the spin of the incident protons is reversed is the parity violating signal . to measure this , the beam current before and after the liquid hydrogen target was measured and the signals were subtracted ( fig . [ principle ] ) . the spin state was then changed in a special pattern at 40 spin states per second and synchronous detection was used to extract that component of the difference signal that was synchronized with the spin flip . unfortunately , beam parameters other than spin changed when the spin was flipped , and great pains had to be taken to measure , and correct for , the false signals resulting from these unwanted helicity correlated changes . technical details of the systems required to do this will be described in detail in what follows , but an idea of the complexity can be obtained from fig . [ general ] , identifying the major pieces of equipment , and fig . [ timing ] , showing the sequencing of the control measurements within a 200 ms , eight spin - state cycle . because of the importance of understanding systematic errors in such an experiment , essentially all the pieces of equipment , from the ion source to the beam dump , had to be considered integral parts of the parity experiment . figure [ general ] shows the main subsystems of the experiment the triumf optically pumped polarized ion source ( oppis ) , the cyclotron , the beam transport , and the specialized parity instrumentation @xcite . a 5 @xmath19a transversely polarized beam was transported to the cyclotron through an approximately 50 m long injection beamline . a 200 na beam at 75 - 80% vertical polarization was extracted at 221.3 mev . spin precession through pairs of solenoid and dipole magnets resulted in delivery of a longitudinally polarized beam to the parity apparatus . there were two complementary states of the spin transport , the `` positive helicity '' and `` negative helicity '' beamline tunes , which transported spin - up in the cyclotron into either @xmath20 or @xmath21 helicity at the parity apparatus . in the last section of the beam line , the longitudinally polarized beam first passed through a series of diagnostic devices two beam intensity profile monitors ( ipms ) and two transverse polarization profile monitors ( ppms ) before reaching the @xmath22 target , which was preceded and followed by transverse electric field , parallel plate ionization chambers ( trics ) to measure the target transmission . a third ipm was located immediately in front of the @xmath22 target , inside the cryostat vacuum . in the ion source , 9 w of 795 nm laser light was used to optically pump a rubidium vapor whose polarization was ultimately transferred to the protons of the @xmath23 beam . the polarization was reversed using small tilting etalons to make rapid frequency adjustments to the two pumping lasers to match either the @xmath6 or @xmath7 component of the rubidium d@xmath24 transition ; the two are separated by 93.5 ghz . no macroscopic electric or magnetic fields were altered . this minimized helicity correlated changes in accelerated beam parameters other than polarization . this was very important to the success of the experiment . the ion source is the ultimate origin of all helicity correlated modulations , and the more that unwanted modulations could be reduced at the source the less corrections were required later . the extracted beam current required by the experiment was not large ( 200 na ) so , to reduce unwanted helicity correlated modulations , it was possible to sacrifice most of the oppis intensity in return for beam quality . for example , the rf bunchers in the injection beamline , which enhance the cyclotron transmission by a factor of five , also amplify the coherent energy modulations by two orders of magnitude , and could not be used during the parity experiment . high - current oppis development @xcite for high energy accelerators proceeded at the same time as development of the triumf source and contributed greatly to the parity violation experiment . the polarization of the rubidium vapor was monitored and controlled on - line by observing the faraday rotation of light from an additional 100 mw , tis probe laser that emitted 780.8 nm light close to the d@xmath25 transition of rubidium . the polarization of the linearly polarized probe laser light rotated through an angle proportional to the rubidium vapor polarization . the faraday measurements also provided confirmation of the helicity state of oppis . the faraday rotation signal was encoded as a frequency to prevent helicity correlated signals from being present in the electronics racks . details of the oppis faraday rotation system are described elsewhere @xcite . every effort was made to tune oppis for minimum intensity change on spin flip . the main technique used to do this was to keep the rubidium polarization , as measured by the faraday rotation , at close to 100% and with the two spin states matched to better than @xmath260.5% . it was not possible to eliminate helicity correlated current modulation completely , but under normal data taking conditions , helicity - correlated current modulations @xmath27 of a few parts in @xmath28 were routinely achieved . to enable measurement of the sensitivity of the experiment to coherent intensity modulation , provision was made at the ion source to produce an intentional , controlled , intensity modulation when desired . this was done using an auxiliary 18 w green argon laser ( spectra physics inc . , operating on all lines with most power at 514 nm and 488 nm ) beam that co - propagated with the @xmath23 beam along the 30 m long horizontal section of the injection beam line , neutralizing through photodetachment a small fraction of the beam . the photodetachment laser could be interrupted synchronously with the parity spin sequence , so that the beam current in every other `` spin off '' ( i.e. with the optical pumping lasers blocked with a shutter ) data taking cycle was modulated at the 0.1% level . after extraction from oppis , the @xmath23 beam passed through a wien filter that was tuned to produce vertical polarization at the entrance to the cyclotron . the injection line from the ion source to the cyclotron used electrostatic elements and was magnetically shielded from the fringe field of the cyclotron . instabilities in the polarized ion source , beamline power supplies , and mechanical vibrations of the whole injector building structure , cause the beam position at the injection point to fluctuate . these fluctuations are converted to energy and current modulations of the accelerated beam . these were significantly reduced by a position stabilization feedback system installed in the vertical section of the injection beamline @xcite . the system was based on two split - plate beam position monitors , with correction voltages applied to electrostatic steering plates . about 50% of the beam was lost on the split - plates , but the result was a significant improvement in the beam stability , and a reduction in noise . the sampling rate of the integrated current feedback amplifier was 1 khz , so spin - flip correlated position modulations produced in the source at 40 hz were also reduced by the position stabilization system . after injection into the cyclotron , the beam was accelerated to 221.3 mev in the cyclotron , and was extracted by a thin stripping foil . various stripping foil designs were tried , but most of the data were taken with a 2.5 mm wide x 26 mm high x 5 mg/@xmath29 thick , pyrolytic graphite foil . this foil was mounted in a special `` bow - saw '' shaped holder that supported the foil from both ends to prevent curling . following extraction from the cyclotron , the proton polarization vector was precessed through @xmath30 in the first solenoid , 88.61@xmath31 in the first ( 40@xmath31 ) dipole , @xmath32 in the second solenoid , and 26.58@xmath31 in the final ( 12@xmath31 ) dipole , resulting in a longitudinally polarized beam that was transported to the parity apparatus . the sign of the solenoid rotation was chosen depending on the desired helicity of the tune and the exact solenoid strengths were fine tuned empirically to produce pure longitudinal polarization in the presence of the cyclotron fringe field . in contrast with systems using only one solenoid and one dipole , which can produce longitudinal polarization only at one energy , the triumf system is capable of producing longitudinal polarization at any proton energy up to 500 mev . @xcite . ideally , the beam transport should produce an achromatic waist downstream of the @xmath22 target , to minimize the effect of first moments of transverse polarization , which are the dominant source of systematic error in this experiment , as they couple to the large ( @xmath33 ) parity allowed transverse analyzing power @xmath34 . the rotation of phase space introduced by the superconducting solenoid magnets was approximately compensated for by rotating the quadrupole doublet before the final 12 degree bend by @xmath35 and the quadrupole triplet following the bend by @xmath36 . to reverse the spin direction at the parity target relative to the spin in the cyclotron , the directions of the solenoidal fields and the signs of the quadrupole rotations were reversed . in principle , that should have been be all that was required . in practice , small adjustments to some quadrupole and steering magnets were necessary , but empirical tunes were developed that kept these changes very small ( 1 - 2% ) . between the second solenoid magnet and the final @xmath37 bend , where the polarization of the beam has both a longitudinal and a sideways component , a four branch polarimeter @xcite was used to measure the transverse polarization components at regular intervals . since the angle of the polarization vector is known at this location , the absolute beam polarization could be determined . once the beam passed the final dipole magnet , the polarization was longitudinal and the ppms determined the small transverse components . the upstream polarimeter was also used , with purely vertically polarized beam , to check the absolute calibration of the ppms . just after the upstream polarimeter , a beam energy monitor ( bem ) @xcite was available to measure the energy of the proton beam . the bem achieves a statistical precision of @xmath26 20 kev in approximately one hour . the absolute calibration is known to @xmath38 kev @xcite . the ( bem ) target could not be inserted during normal data taking , so periodic bem measures were made to check the beam energy . table [ energies ] summarizes the beam energy measurements . lr running period & beam energy + 1997 & @xmath39 mev + 1998 & @xmath40 mev + 1999 & @xmath41 mev + [ energies ] the beam energy of 221.3 mev at the target entrance was chosen so that the @xmath42 contribution would be zero when the acceptance of the detectors and the energy loss in the target was taken into account . this was done using a monte carlo program that simulated the experiment using detailed target and detector geometries , and including the energy dependence of the strong @xmath43 phase shifts . it was found that the beam energy was not overly critical ; a departure of one mev from the energy for which the @xmath44 contribution integrated to zero caused this mixing to contribute only @xmath45 to @xmath0 . theoretical calculations of @xmath0 based on total cross section , and which assume no energy loss in the target , show the @xmath44 zero at 225 mev , where the @xmath46 and @xmath47 strong phase shifts are equal and opposite @xcite . according to the monte carlo simulation , this theoretical @xmath0 at 225 mev should be compared to the triumf @xmath0 multiplied by @xmath48 . very strict constraints had to be imposed on the incident longitudinally polarized beam to limit both random and helicity correlated modulations of the beam intensity , energy , horizontal ( @xmath49 ) and vertical ( @xmath50 ) position and direction , beam width ( @xmath51 and @xmath52 ) , transverse polarization ( @xmath53 and @xmath54 ) , and first moments of transverse polarization ( @xmath55 and @xmath56 ) . the approach followed was to design the experimental apparatus for minimum sensitivity to beam property variations , to monitor helicity - correlated beam properties during data taking , and to correct the data for all significant helicity correlated effects . a program of auxiliary calibration measurements was interwoven in the regular data taking cycle in order to establish the sensitivity of the apparatus to all measurable systematic effects under data taking conditions . the custom - built parity instrumentation occupied approximately the last 10 m of the beamline , between the last quadrupole magnet and the west wall in the triumf proton hall ( beamline 4a/2 ) . transverse field parallel plate ion chambers tric1 and tric2 measured the beam current incident on and transmitted through the target . the parity violation signal was derived from the helicity - correlated analog signal difference between the beam currents measured by the two trics . upstream of the target were two polarization profile monitors ( ppms ) to measure the distributions of transverse polarization @xmath57 and @xmath58 across the beam . the three intensity profile monitors ( ipms ) , measured the intensity distribution of beam current in @xmath49 and @xmath50 . two of the ipms were coupled to a pair of fast ferrite cored steering magnets ( fcsms ) that locked the beam path on the desired axis through the equipment . the ipm signals were based on secondary emission from thin nickel foil strips arranged in harps placed between 8 @xmath19 m thick aluminum foils ( a very similar , earlier version is described in @xcite and @xcite ) . for ipm1 and ipm2 the nickel strips were 3 @xmath19 m thick , 1.5 mm wide , and spaced 2.0 mm center to center . in ipm3 , the nickel strips were 10 @xmath19 m thick , 2.5 mm wide , and spaced 3.0 mm center to center . each ipm contained a vertical and a horizontal harp with 31 strips per harp . ipm1 also contained an aluminum normalization foil that provided a beam current signal to the liquid hydrogen target controller . the 31 signals from the strips of each of the six planes were individually amplified and digitized to provide the beam intensity profiles in each spin state . hardware beam centroid evaluators delivered signals proportional to the beam centroids . corresponding correction signals were used to drive feedback loops to the pair of horizontal and vertical fast fcsms ( fig [ general ] ) . this allowed the beam intensity profile centroids to be kept within @xmath59 m with a dc offset less than 50 @xmath19 m from the desired axis . sensitivities to helicity - correlated position and size modulations were determined with the beam unpolarized and with enhanced modulations introduced using the fast , fcsms synchronized to the spin sequence . blades through the beam . the blades measure 5 mm x 1.6 mm in cross section and extend 85 mm beyond the holders . each ppm has two horizontally scanning and two vertically scanning blades . this figure shows a blade at the middle of a vertical scan.,width=151 ] the ppms are described in detail in @xcite . each ppm has four high density polyethylene ( ch@xmath25 ) blades , 5 mm x 1.6 mm in cross section . the blades were mounted on wheels rotating at 5 revolutions per second , and were scanned through the beam as shown in fig . [ blades ] , two blades scanning vertically through the beam and two blades scanning horizontally . the figure shows one blade in the middle of a vertical scan . protons elastically scattered from hydrogen in the blades were detected by sets of forward and recoil scintillator telescopes , the geometry of which was set to select only @xmath1 elastic events at 17.5@xmath60 laboratory angle , near the maximum in the parity allowed analyzing power , @xmath34 . during vertical scans the up - down asymmetry was measured , and during horizontal scans the left - right asymmetry was measured . information from each blade was stored in 80 time bins of 80 @xmath19s each , for a total time window of 6.4 ms per blade . as the radius of rotation is 215 mm , this corresponds to @xmath61 mm @xmath62 mm centered on the nominal beam axis . from the up - down or left - right asymmetry in each bin , a distribution of transverse polarization across the beam profile was generated . the two ppms were operated with a 180@xmath31 angular mismatch , producing 8 equally spaced profiles in 1/5 second , or 40 profiles per second . ( this is also the spin flip rate [ timing ] ) the master clock for sequencing the whole experiment , including helicity changes , was derived from 2500-line optical encoders mounted on the ppm drive shafts . each shaft encoder produced two 2500 cycle per revolution square waves in phase quadrature , permitting the direction of rotation to be determined from whether one signal was high or low at the positive going transition of the other . by recording both the rising and falling edges of both quadrature signals , the ppm control electronics obtained an effective resolution of 4 @xmath63 2500 = 10,000 lines . the rotation of the two ppms was synchronized by electronic gearing using position information provided by the shaft encoders . during steady state rotation at 5 revolutions per second , the servo system was able to control the angle between the two ppms to @xmath64 milliradian , corresponding to one - third of the 1.6 mm target blade thickness at the location of the proton beam . the @xmath22 target had a flask 0.10 m in diameter and 0.40 m long . the target scattered @xmath654% of the beam . it was important that the windows on the hydrogen volume be flat and parallel to prevent motion of the proton beam from creating intensity noise . the hydrogen was contained by 25 @xmath19 m thick stainless steel windows . upstream and downstream of these inside the vacuum vessel were two helium filled chambers , 21 mm thick on the upstream side and 10 mm thick on the downstream side , ensuring that the inner windows were flat . the outer windows of the helium cells were 50 @xmath19 m thick stainless steel . the cryostat vacuum vessel was sealed with 25 @xmath19 m thick copper windows . although a thick target is desirable from the standpoint of increasing the signal , it was also very important that the tails of the beam profile be well contained in the 150 mm @xmath63 150 mm aperture of the detectors ; beam blow - up due to multiple coulomb scattering limited the target flask length , as well as the thickness of various entrance and exit windows and the thickness of the upstream ipm foils . with the target full , the beam size increased from approximately 4 mm ( @xmath66 of projected profile ) at the center of the the upstream detector to 15 mm at the center of the downstream detector , and 22 mm by its exit . rapid circulation ( 5 @xmath67 ) of the @xmath22 reduced density gradients and prevented boiling . a feedback loop using fast and slow heaters controlled the target temperature . the fast heater responded to a beam current signal from ipm1 ( sec . [ subsubsec : ipm ] ) , and kept the heat load essentially constant when the beam current changed . the slow heater made fine adjustments to hold the lh@xmath25 temperature at 19.3 k to within @xmath68 k over a several week data taking period . each tric ( a smaller earlier version is described in @xcite ) consisted of a cylindrical enclosure filled with 750 liters of ultra - high purity hydrogen gas at a pressure of about 150 torr , and contained an upper cathode plate operated at -8 kv , and a grounded lower signal plate . the trics also contained field shaping electrodes plus guard rings to ensure a uniform sense region , 150 mm wide by 150 mm high by 600 mm long between the parallel electrodes . high - z , low - energy spallation products from the chamber windows can cause large fluctuations in the signal , so the entrance and exit windows were located approximately 900 mm from the center to prevent spallation products from entering the active region . other considerations in the design of the trics were noise due to delta ray ( @xmath69-ray ) production and ion pair recombination . recombination is reduced by lower gas pressure and higher voltage , both of which reduce the space charge density in the active volume . in practice , compromises must be made because too high a voltage causes increased noise from corona discharge and low pressure reduces the desired signal . @xmath69-rays are electrons produced by collisions of the protons with the detector gas . the @xmath69-ray signal is noisy , and the major contribution to this signal is from @xmath69-rays at large angles . the transverse dimension of the ion chamber was a trade - off between a small transverse size which would minimize @xmath69-ray noise and a large transverse dimension which accepts all the beam . the 150 mm transverse dimension was the smallest that gave an acceptably low sensitivity to beam size modulation based on simulations . the main signal from the proton beam increases with the length of the active region . this was limited by the available space in the beamline ; 600 mm was the longest practical length . the noise in the ion chamber signals has two incoherent contributions shot noise from the statistical nature of the proton beam , and noise contributed by the chamber itself . the noise figure , @xmath70 , expresses the chamber noise as a fraction of the shot noise . when the difference between the upstream and downstream chambers is taken , most of the shot noise contribution disappears because , except for the 4% scattered by the target , each upstream proton also passes through the downstream chamber . the chamber noise , on the other hand , does not cancel , and the running time is dominated by chamber noise . in this experiment the counting time was approximately 15 times that which would have been expected from counting statistics alone . the fact that the run time was dominated by chamber noise resulted in the seemingly paradoxical result that better precision was obtained by lowering the beam current , because this reduced the detector noise figure . in a series of test runs at progressively reduced beam currents , the @xmath0 distribution became narrower until approximately 100 na . the 200 na selected for running was a compromise between lower current for better precision on @xmath0 and higher current for better precision from the ppms , which _ were _ limited by counting statistics . the data were taken in @xmath71 second ( 200 ms ) cycles , each cycle consisting of eight @xmath72 second ( 25 ms ) spin states arranged in the pattern ( @xmath73 ) or ( @xmath74 ) . this pattern makes the result insensitive to linear or quadratic drifts . the cycles were further arranged in an eight cycle ( 64 spin state ) `` supercycle '' , with the starting state of each cycle following the same ( @xmath73 ) or ( @xmath74 ) pattern . the initial spin state of each supercycle was chosen at random . in the frequency domain , this switching pattern contains odd multiples of 5 hz , with the largest harmonic content at 15 hz . the data acquisition produced on - line values for the amplitude and phase of the dominant 15 hz component . this was derived from a 16 spin state cycle . the 64 spin state supercycle gave the option of using more advanced digital filtering schemes on the difference signal , but these were only used during early development runs . 20% of the data were taken with all the spin flipping equipment running , but with the pumping lasers blocked with a shutter to guarantee zero polarization . in addition , as mentioned earlier , half of these `` spin off '' data were taken with an artificially enhanced intensity modulation synchronized with the spin flip . as shown in fig . [ spinstate ] , each 25 ms spin state was divided into polarization measuring and asymmetry measuring intervals . following a short dead time to allow the mechanical etalons which change the ion source spin state time to stabilize , was a 6.4 ms window during which one of the 8 blades of the scanning polarimeters passed through the beam . the tric and ipm signals were then integrated over exactly 1/60 second to eliminate sensitivity to 60 hz or harmonics of 60 hz . the dead interval at the end of each state was intended as a buffer to absorb any timing jitter due to imperfect rotation speed of the ppms . as it turned out , the timing jitter was less than 0.1 ms and this buffer zone was more than adequate . the minimum data set for which a full set of helicity correlated beam properties could be extracted was a 0.4 s `` event pair '' corresponding to two full 360@xmath31 rotations of the ppms , as this gave both spin states for each ppm blade . as noted in the ion source section , the spin state was transmitted as a frequency to prevent coupling the spin flip signal into the data acquisition . the effectiveness of this isolation was checked by running the complete data acquisition system including the ion source , but with detector signals supplied by a battery . false @xmath0 from electronic crosstalk was found to be less than @xmath75 , an upper limit determined by the statistics of the test ( crosstalk was probably less ) . the rotation speed of the ppms is locked to a signal derived from the zero crossing of the 60 hz ac line . since the 0.2 s taken for one ppm rotation is exactly 12 cycles of the 60 hz line , one would expect a given ppm blade to always pass through the beam at the same phase of the ac line . to prevent this , a small controlled phase slip was introduced . the rate of slip was programmable . for data taking it was set for one complete cycle of the 60 hz line in 18 minutes . although the phase of one ppm relative to the other is `` quantized '' by the finite resolution of the encoders , the angular frequency of rotation is not , and in principle can be locked to a square wave of any frequency . in practice , roundoff error in the motor control computer limited the choices of the reference frequency . the 18 minutes for 360@xmath31 of slip corresponded to one of the acceptable values . in addition to the regular data taking runs , many dedicated control measurements were made to measure the sensitivity of the apparatus to helicity correlated beam properties , specifically : position , size , intensity , transverse polarization , and energy modulation at the ion source . the parity data used in the final analysis were acquired during three major data taking periods february - march 1997 , july - august 1998 , and may - june 1999 . the data were recorded as a series of `` runs '' of approximately one hour each . statistics for the combined set are * 3.8 million event pairs * 375 runs in positive beamline helicity * 368 runs in negative beamline helicity * 80 position modulation runs * 81 size modulation runs * 40 energy modulation runs * 109 neutral axis scans the runs were grouped into 23 sets , a new set being started following each re - tune of the beamline ( usually involving a reversal of the helicity at the parity apparatus relative to the ion source ) . during data taking every effort was made to maintain the optimum beam conditions ( quiet beam , good longitudinal polarization , low transverse polarization , low first moments of transverse polarization , beam on axis , stable position and size , low intensity modulation ) , and to stop data taking to correct the beam and cyclotron tune when these conditions were not met . nevertheless , some data were recorded for which conditions were not ideal . the raw data sample was therefore subjected to data quality cuts prior to inclusion in the final analysis . the data quality cuts were conservatively chosen to ensure that corrections to the reduced data set based on critical helicity - correlated beam parameters would be consistent with sensitivities measured in ancillary calibration experiments . the data quality cuts first eliminated 46 data runs for which the tric difference signal was anomalously noisy , indicating unstable beam conditions . they also eliminated any data for which diagnostic monitor outputs could be considered spurious . the cuts used are summarized in table [ cuts ] . there were both `` hard cuts '' at a fixed value of beam parameter and `` soft cuts '' at @xmath76 from the mean value . the entire data reduction process reduced the size of the total set by 30% , but significantly improved the quality of the data sample . the @xmath77 for the uncorrected data of the 23 sets went from 11.3 per degree of freedom before the cuts were made to 5.5 after the cuts . it is important to emphasize that the approach used to determine the sensitivity to intrinsic polarization moments , which was the dominant systematic error correction , required consistent and stable beam conditions in particular , the ratio of beam sizes at ipm1/ipm2 and the ratio of intrinsic polarization moments at ppm1/ppm2 must be constant over the data sample used to determine the sensitivities , or the method is invalid . it is important to emphasize that no data cuts were applied to the parity violating asymmetry @xmath0 itself . the ultimate test of the consistency and validity of the data analysis and corrections procedure is the quality of the corrected data set shown in figure [ data ] . the top panel in fig . [ data ] shows the results for the reduced data set , before corrections were made for systematic errors . the bottom panel shows the corrected results from which the final @xmath0 was obtained . this corrected data set was subjected to a regression analysis in which residual sensitivities to helicity correlated beam parameters were explored . no statistically significant residual sensitivities were found , lending confidence to the interpretation of the results . details of the corrections procedure are given in the following sections . lr & + beam parameter & acceptance window + & + neutral axis @xmath78 & @xmath79 mm + position & @xmath80 ( soft ) + width & @xmath81 mm + skew & @xmath68 mm + intensity & 196 - 204 na + intrinsic moments & @xmath82 mm + [ cuts ] sources of error can be divided into those that are related to the beam helicity and those that are not . beam property changes that are synchronized with spin flip ( `` coherent '' or `` helicity correlated '' changes ) can shift the centroid of the @xmath0 distribution and cause a false signal of parity violation . changes that are not helicity correlated , such as detector noise and random variation in beam properties , do not bias the result , but they increase the run time required to reach a given precision . the ultimate limit to the statistical precision of the experiment was set by the counting statistics of the scattered protons . for a target scattering 4% of the 200 na beam , the rate of scattered protons is 50 ghz , which would make it possible to measure @xmath0 to @xmath83 in 20 hours if individual scattered protons could be counted . however , 50 ghz was too high for direct counting , and so current mode detection was used . as noted already in the ion chamber section , the counting time was dominated by detector noise . other random variations in beam properties such as intensity and position also contributed noise and further increased the required run time . it proved to be a net advantage to devote significant beam development time to producing quiet , stable , beam . the approach to minimizing and correcting for the false @xmath0 signal due to helicity correlated beam property changes was as follows : * careful design and operation of the triumf optically pumped polarized ion source and cyclotron made it possible to change the spin direction with very little effect on the other beam properties . * the design of the parity equipment and the operating conditions of the experiment were carefully chosen to minimize the sensitivity to helicity correlated changes . * calibration runs determined the sensitivity to helicity correlated modulations . * the beam properties were continuously monitored during data taking so the actual helicity correlated changes were known , and appropriate corrections were applied . the effects of modulations in beam intensity , position , size , transverse polarization , and energy were considered . measurements made early in the experimental program revealed significant correlations between beam parameters . for example , horizontal and vertical beam motion are often highly correlated . to be sure of extracting the correct sensitivities , separate control measurements were made in which each beam property in turn was artificially modulated , and the effect on apparent @xmath0 was recorded . the sensitivities obtained by this method were consistent with sensitivities extracted by multilinear regression using the natural variation of beam parameters . most importantly , the regression analysis showed no significant sensitivity to _ products _ of beam modulations , so the false @xmath0 was the sum of the contributions of the individual coherent modulations . for the sensitivity calibrations made by varying one parameter at a time , measurements were made using modulations of different size and sign . the false asymmetry was linear in the modulation , passing through zero at zero modulation . to determine the sensitivities used for data reduction , large values of modulation were used to give accurate results in a short time . to determine the sensitivity to first moments of transverse polarization , which gave the largest false asymmetry , much larger transverse polarizations were used for calibration than were present in the parity beam . this was justifiable because false asymmetries must be linear in polarization , and the use of large polarizations allowed the displacements needed for the calibration to be reduced . the measured false asymmetry was linear in polarization moment . as the energy of the beam was 27 mev lower at tric2 than at tric1 , the tric2 gas pressure was lowered relative to tric1 to equalize ion chamber currents . a hardware gain on the tric1 signal was then adjusted to minimize common - mode noise in the amplified difference signal , @xmath84 which was digitized , where @xmath85 is the hardware gain ; @xmath85 is very close to unity . the helicity correlated part of the difference , which should be proportional to @xmath0 , nevertheless contains a coherent current modulation component due to imperfect common mode rejection arising from gain mismatch and nonlinearity of the ion chambers . if the current signal from tric2 is @xmath86 then , expressing the difference in current for the two helicity states as @xmath87 , one can define an `` analog asymmetry '' , @xmath88 where @xmath89 is the nuclear scattering probability in the lh@xmath90target ( @xmath91 ) and @xmath92 . the quantity @xmath93 is a function of tric1 and tric2 gas gains , and is nominally equal to @xmath94 ; it was adjusted with tric2 gas pressure . the hardware gain @xmath85 was set by zeroing the false parity signal when the beam current was modulated by the photodetachment laser . because of the nonlinearity term , @xmath85 must be reset if the beam current changes . a beam current stabilization system was implemented which restricted beam current excursions to the range @xmath95 na @xcite . the current stabilization system took a current signal from the upstream ionization chamber ( tric1 ) and fed a correction voltage to an electrostatic quadrupole lens just upstream of a set of slits in the injection line . the injection line was tuned with a slight excess of current so that , at the set point , about 10% of the beam was skimmed by the slits . to exclude coupling of current modulation to position and energy modulations , the sampling rate was slow ( 0.5 hz ) , much slower than the spin - flip rate of 40 hz . the active range of the current stabilization loop was quite low ( @xmath96% ) to avoid correcting drifts caused by large excursions of cyclotron tune . the current loop operation was a sensitive indicator of cyclotron stability , and operators made adjustments to the cyclotron tune when necessary . even under conditions of constant beam current , periodic adjustments of @xmath85 were needed to compensate for small drifts in the ion chamber gains . the setting for best common mode rejection was checked periodically by turning on the photodetachment laser , creating a ( @xmath97% ) coherent intensity modulation , and performing a scan of subtractor box settings . the results of such a calibration are shown in fig . [ cimtune ] . because the difference signal was multiplied by 1000 , the dc level was a sensitive indication of the setting of the hardware fine gain . a fractional change of @xmath98 in the hardware fine gain caused a dc shift of 0.7 v in the subtractor box output . causes a shift of 0.7 v in the dc level . this particular scan gave an ideal dc level of -1.9 v for zero sensitivity to cim.,width=321 ] during data taking , coherent intensity modulation ( cim ) was measured continuously by the ion chamber ( tric1 ) upstream of the target . as noted in section [ sec : oppis ] , this was normally less than a few parts in @xmath28 . because the sensitivity to cim changed between the dedicated cim runs used to re - set the subtractor box , the sensitivity to cim , like the cim itself , was measured during data taking . periods of enhanced ( 0.1% ) coherent intensity modulation were interleaved with the main data . a typical sensitivity was @xmath99 where @xmath100 is the beam current in @xmath101 . table [ cimsens ] shows the average sensitivities to coherent intensity modulation for the different running periods , and fig . [ cimdata ] shows a histogram of the sensitivities for a typical ( @xmath102 hour ) run . . data have been excluded for which @xmath103% . the mean is 0.23 and the standard deviation 2.6 , in units of ( @xmath98/%).,width=321 ] lcr set & sensitivity @xmath104% ) & cim events + 1997 & @xmath105 & 83,636 + 1998 & @xmath106 & 106,520 + 1999 & @xmath107 & 90,955 + [ cimsens ] coherent position modulation was measured by two of the three ipms . during the 1997 running period , ipm1 and ipm3 were used because their separation was greater and , all else being equal , position control should have been better . unfortunately ipm3 picked up noise from the liquid hydrogen target circulation fan , so for the 1998 and 1999 data , ipm1 and ipm2 , located 1.8 m apart along the beamline , were used . using two ipms permits measuring both tilts and parallel shifts of the beam . the measured coherent modulations are shown in tables [ sizpos97 ] and [ sizpos9899 ] . the sensitivity to beam motion was measured in separate control measurements during which the beam position was modulated in a variety of ways using the fcsms . the false analyzing power arising from helicity correlated beam position was parameterized as @xmath108 where @xmath109 and @xmath110 are the horizontal and vertical helicity correlated beam motion , @xmath49 and @xmath50 are the beam position and the 1 and 2 subscripts refer to ipm1 and ipm2 . the parameters @xmath93 to @xmath111 were extracted from a fit to the calibration data . details may be found in ref . figure [ posmod ] shows the results of such a fit . the abscissa shows the false asymmetry predicted using eq . 2 with the fitted parameters , and the ordinate shows the false asymmetry actually measured with the calibration data . coherent beam size modulation was also measured by the two ipms , and the observed values are listed in tables [ sizpos97 ] and [ sizpos9899 ] . the sensitivity to beam size was determined by control measurements in which the beam size was intentionally modulated by driving the fcsms as quadrupoles . the false analyzing power due to beam size modulation was expressed as @xmath112 where @xmath66 is the rms beam size and @xmath113 is the helicity correlated change in beam size . as with position modulation , the parameters were extracted from a fit to the calibration data . because the quadrupole configuration of the fcsms caused the beam to be steered somewhat when the size was modulated , correction for position modulation had to be included in the fit . figure [ sizemod ] shows the results of a fit to a series of size modulation calibration runs . lcccr & & + beam line : & @xmath20 & @xmath21 & @xmath20 & @xmath21 + @xmath114 & @xmath115 & @xmath116 & @xmath117 & @xmath118 + @xmath119 & @xmath120 & @xmath121 & @xmath122 & @xmath123 + @xmath124 & @xmath125 & @xmath126 & @xmath127 & @xmath128 + @xmath129 & @xmath130 & @xmath131 & @xmath132 & @xmath133 + @xmath134 & @xmath135 & @xmath136 & @xmath137 & @xmath138 + @xmath139 & @xmath140 & @xmath141 & @xmath142 & @xmath143 + @xmath144 & @xmath145 & @xmath146 & @xmath147 & @xmath148 + @xmath149 & @xmath150 & @xmath151 & @xmath152 & @xmath153 + [ sizpos97 ] lcr beam line : & @xmath20 & @xmath21 + @xmath114 & @xmath154 & @xmath155 + @xmath156 & @xmath157 & @xmath158 + @xmath124 & @xmath159 & @xmath160 + @xmath161 & @xmath162 & @xmath163 + @xmath134 & @xmath164 & @xmath165 + @xmath166 & @xmath167 & @xmath168 + @xmath144 & @xmath169 & @xmath170 + @xmath171 & @xmath172 & @xmath173 + [ sizpos9899 ] if the proton spin is not perfectly longitudinal , the small transverse component will reverse with helicity . this can couple to the relatively large ( @xmath33 ) parity allowed analyzing power to cause a false parity violating signal . figure [ firstmom ] uses a pencil beam with a vertical transverse polarization to illustrate the mechanism . because the parity allowed analyzing power is positive , slightly more beam will be scattered to the left in the positive helicity state , and to the right in the negative helicity state . if the beam passes through the center ( neutral axis ) of the detector , as shown in the left - hand panel , the response will be the same for both helicity states and no false effect arises . if , on the other hand , as shown in the right - hand panel , the beam does not pass through the center of the detector , more signal will be recorded in one helicity state than in the other . to a very good approximation , the effect is found to be proportional to the size of the transverse component multiplied by the distance the beam is off center at the detector i.e. to the _ first moment _ @xmath174 of transverse polarization at the detector . in a field - free region , a real particle beam of finite extent is made up of a bundle of straight rays like the pencil beam in this example . the first moment for the beam is the linear sum of the first moments of the individual rays . since the first moment for a ray is proportional to the distance of the ray from the zero axis and varies linearly with distance along the beam , changing sign where the ray crosses zero , the first moment for a particle beam will vary linearly with distance along the beam . a real beam can have a substantial first moment of transverse polarization even if the net transverse polarization is zero . in the example of fig . [ firstmom ] , the vertical polarization could be `` up '' on the right side of the beam and `` down '' on the left side . ) was scanned horizontally and the sensitivity to @xmath54 plotted as a function of @xmath49 projected to the `` magic @xmath175 '' ( @xmath175 location where the false @xmath0 does not depend on angle ) . a beam with @xmath176 mm was found to produce no false signal from vertical polarization components.,width=321 ] to determine the first moment sensitivities , test runs were made with pure vertical and pure horizontal polarization . by scanning the vertically polarized beam horizontally it was possible to generate known moments , @xmath177 , and , by scanning the horizontally polarized beam vertically , known @xmath178 moments could be generated . although in principle the first moments must be known at the two detectors , because first moments vary linearly with distance along the beamline , knowledge of the first moments at _ any _ two locations is sufficient . the transverse polarization profiles were measured by the two polarization profile monitors ( ppms in fig . [ general ] ) located 1.8 m apart along the beamline and , for convenience , the first moment sensitivities were expressed in terms of the sensitivities at the ppms . the first moment scans also defined the beam trajectory for which there was zero sensitivity to average transverse polarization . such a trajectory was referred to as a _ polarization neutral axis_. polarization neutral axes were determined in both the horizontal and vertical directions and , during data taking , the beam was held on this neutral axis by the servo system . figure [ pycal ] shows an example of a horizontal neutral axis scan . another consequence of the linear behavior of the first moments with distance , @xmath175 , is the presence of a `` magic @xmath175 '' downstream of tric2 . _ any _ beam with zero first moment at this @xmath175 will cause no false effects . also notice that the measurements shown in fig . [ pycal ] confirm the linear relation between first moment and false @xmath0 . in analyzing the effect of first moments , _ extrinsic _ first moments , @xmath177 and @xmath179 , caused by a beam which has some net transverse polarization and whose centroid is displaced from the neutral axis , were treated separately from _ intrinsic _ first moments , @xmath180 and @xmath56 , which do not depend on the position of the beam , but rather arise from the distribution of components within the beam . by holding the beam on the polarization neutral axis it was possible to virtually eliminate corrections for _ extrinsic _ first moments . _ intrinsic _ first moments , on the other hand , are independent of beam position , arise in the cyclotron and beamline , and were very hard to control . the false asymmetry arising from extrinsic and intrinsic first moments can be written @xmath181 \nonumber \\ & + \langle p_y \rangle [ b_1(\langle x_1 \rangle - x_1^{na } ) & + b_2(\langle x_2 \rangle - x_2^{na } ) ] \nonumber \\ & + a_1 \langle y p_x \rangle _ 1 & + a_2 \langle y p_x \rangle _ 2 \nonumber \\ & + b_1 \langle x p_y \rangle _ 1 & + b_2 \langle x p_y \rangle _ 2\end{aligned}\ ] ] where the first two lines are the contribution from _ extrinsic _ first moments , and the second two lines are the contribution from _ intrinsic _ first moments . note that the first moment sensitivities , @xmath182 , @xmath183 , @xmath184 , and @xmath185 , are identical for intrinsic and extrinsic moments . the results of the neutral axis scans are shown in table [ fmsens ] . the quantities @xmath186 and @xmath187 depend on what axis @xmath49 and @xmath50 are measured from . they are zero if @xmath188 , @xmath189 , @xmath190 , and @xmath191 are measured relative to the neutral axis . one notes that the sensitivities to first moments are very consistent from run period to run period . in units of @xmath192 per @xmath193 , the sensitivity to @xmath178 is 1.8 at ppm1 and -2.5 at ppm2 . the sensitivity to @xmath194 is -1.5 at ppm1 and 2.0 at ppm2 . lccr & & + sensitivity ( 10@xmath18 ) & feb97 & jul98 & may99 + @xmath195 ) & @xmath196 & @xmath197 & @xmath198 + @xmath199 ) & @xmath200 & @xmath200 & @xmath201 + @xmath186 & @xmath202 & @xmath203 & @xmath204 + @xmath205 ) & @xmath206 & @xmath207 & @xmath208 + @xmath209 ) & @xmath210 & @xmath211 & @xmath212 + @xmath213 & @xmath214 & @xmath215 & @xmath216 + @xmath217 & 1.17 & 1.96 & 2.05 + [ fmsens ] there were no magnetic elements after the first ppm , so the first moments of transverse polarization varied linearly with position along the beamline . furthermore , for a fixed setting of the upstream beamline magnets , the first moments at ppm1 and ppm2 tended to scale together so that , over a wide range of first moments , the _ ratio _ of first moments at ppm1 and ppm2 had a constant value . since the neutral axis scans showed that the first moment sensitivity at ppm1 was of opposite sign to , and 75% of the magnitude of , that at ppm2 , if the ppm2 first moment was always 75% of the ppm1 first moment , then the _ effective _ sensitivity to first moments would be zero . this would correspond to a beam whose first moment of transverse polarization goes through zero at the magic @xmath175 . by adjusting the beam convergence , an attempt was made to achieve this magic moment ratio between ppm1 and ppm2 . unfortunately , the ability of the ppms to measure this ratio to sufficient precision in a reasonable amount of time was very limited , so the cancellation was not perfect and the residual , effective , first moment sensitivity had to be extracted from the data . this could be done either by regressing the first moment ratio from the data and using the sensitivities measured in the neutral axis scans , or by regressing the correlation of @xmath0 with average first moment directly from the data . the two methods agreed with each other , but the latter method produced the smallest statistical spread and was the method used to correct the data . since the beam energy in the downstream tric was on average 27 mev lower than in the upstream tric due to energy loss in the target , helicity - correlated energy modulation caused a systematic error due to the nonlinear energy dependence of the proton beam energy loss in the hydrogen gas of the trics . the sensitivity to coherent energy modulations was determined using an rf accelerating cavity placed upstream of ipm1 in the beam line . the measured sensitivity of false @xmath0 , @xmath218 , was in excellent agreement with predictions based on the variation of stopping power with energy . energy modulation of the extracted beam was caused by position modulations of the radial intensity distribution at the stripping foil ; this converted radial position modulation of the injected beam to energy modulation of the extracted beam . the primary coherent energy modulation produced in the source was converted to position modulation in the injection beamline and then back to energy modulation at the extraction foil . direct measurements using a magnetic spectrometer ( 1.2 m dispersion ) in another beamline ( 4b ) at triumf showed the energy modulation of the extracted beam to be approximately 100 to 200 times greater than the energy modulation at oppis . during the parity runs this direct measure could not be made , but frequent measures of energy modulation at oppis and of @xmath219 were made . the @xmath219 sensitivity was measured by applying a square wave voltage of 0.5 v amplitude to the electrically isolated sodium ionizer in oppis . these @xmath219 runs were then corrected for all known systematic errors , ( dominated by @xmath220 ) , and it was assumed that the residual false @xmath0 arose from energy modulation of the extracted beam . a comparison of the measured @xmath219 to the @xmath221 measured in the beamline with the rf cavity ( sec . [ subsec : emod ] ) indicated that the cyclotron amplified @xmath222 by a factor of about 130 , in agreement with the magnetic spectrometer measurements . the primary energy modulation caused by the optical pumping lasers was measured by using an electrostatic beam energy analyzer in the polarized source and an intensity profile monitor with 16 collector strips 2.5 mm wide , and with 3.0 mm spacing to measure beam position modulation downstream of the steering analyzing plates . the monitor was mounted on a remotely controlled swinging arm . two measurements of coherent position modulation for the right and left monitor positions allowed separation of the energy and position modulation components of the oppis beam . an accuracy of 0.2 mev could be achieved in ten minutes of integration time . the modulation magnitudes were quite sensitive to the pumping laser asymmetry between the two polarization states ; after careful laser tuning , the coherent energy modulation was reduced to 1 - 2 mev and the coherent position modulation to the 20 nm level . lcr + @xmath223 & @xmath224 & @xmath225 + @xmath226 & @xmath227 & @xmath228 + @xmath229 & @xmath230 & @xmath231 + @xmath232 & @xmath233 & @xmath234 + @xmath235 & @xmath236 & @xmath237 + @xmath238 & @xmath239 & @xmath240 + @xmath241 & @xmath242 & @xmath243 + @xmath244 & @xmath245 & + [ helicities ] although the frequent measurement of @xmath246 and @xmath219 helped to set limits on the false @xmath0 from energy modulation of the extracted proton beam , it is significant that the energy modulation of the extracted proton beam could not be measured directly at the parity apparatus . to cancel its effects , use was made of the fact that when the beamline helicity is reversed , the effects of true @xmath0 reverse , but the effects of energy modulation do not . this is because energy modulation can only arise in the ion source and cyclotron . the magnets used to rotate the spin from `` up '' to positive helicity or `` up '' to negative helicity are downstream of all sources of energy modulation and do not affect the beam energy . data were generally taken with alternating beamline helicity , and during the 1997 run data were also taken with the wien filter reversed , an independent method of reversing the proton beam helicity relative to the ion source . the wien filter and beamline settings are detailed in table [ helicities ] . that the @xmath247 for the 23 sets is only 1.08 following corrections , shows that the effects of uncorrected systematic errors , including energy modulation , were small . care was taken to balance the amount of data taken in the two beamline helicities so that if some false @xmath0 from energy modulation was present , it would tend to cancel in the final weighted average . information on the energy modulation at the ion source and the sensitivity to this modulation was used to include the effects of uncorrected energy modulation in the error budget , as described in more detail in the next section . lcr & & + property & average value & @xmath248 + & & ( correction ) + & & + @xmath249 & @xmath250 & + @xmath251 & @xmath252 & @xmath253 + @xmath254 & @xmath252 & @xmath255 + @xmath256 & @xmath257 & @xmath258 + @xmath259 & @xmath260 & @xmath261 + @xmath262 & @xmath263 & @xmath264 + @xmath265 & & @xmath266 + @xmath267 & 715 & @xmath268 + electronic crosstalk & & @xmath269 + total & & @xmath270 + @xmath271 & + @xmath272 & + [ corr ] the false @xmath0 arising from a given coherent modulation was found by multiplying the measured modulation by the sensitivity to that modulation . uncertainties from the variance of the corrected @xmath0 distribution are referred to as `` statistical '' . the uncertainty quoted is the standard error in the mean of the @xmath0 distribution . uncertainties in @xmath0 resulting from uncertainties in the sensitivities are referred to as `` systematic '' because an incorrect sensitivity will cause a systematic shift in the mean of the @xmath0 distribution . these errors are , however , statistical in nature , as they arose from statistical uncertainties in the knowledge of the sensitivities to the various coherent modulations . table [ corr ] summarizes the overall corrections to the parity data . to produce the 23 corrected @xmath0 distributions shown in fig . [ data ] , the following procedure was followed . 1 . the data in each set were grouped into bundles of 10000 event pairs per bundle . each bundle was corrected according to the observed coherent modulations ( except position and size ) for that bundle , giving a corrected @xmath0 for each bundle . the variance of corrected @xmath0 values in a set determined an error bar for that set . this is reported as the `` statistical '' uncertainty . no corrections were made for position and size , because the net correction was consistent with zero and when the corrections were applied it was found that they slightly _ increased _ the residual correlations of @xmath0 with position and size , as well as increasing the variance of the corrected @xmath0 distribution . the uncertainties in the nil correction for position and size modulation were added in quadrature to each of the 23 sets . this is included in what is reported in table [ corr ] as `` systematic '' uncertainty . the uncertainties resulting from uncertainties in the various sensitivities were added in quadrature to the corresponding data . the uncertainties in the sensitivities are independent of what is accounted for in step 2 , so it is justified to add them in quadrature to obtain the total error bar on each of the corrected @xmath0 for the 23 sets . this uncertainty is also included in the `` systematic '' uncertainty in table [ corr ] . the @xmath0 reported is the weighted mean of the 23 data sets with a weight @xmath273 , where @xmath274 is the `` total '' uncertainty , not including energy modulation . the error bars shown in the bottom panel of fig . [ data ] are these `` total '' uncertainties . as mentioned earlier , the correction for extracted energy modulation was complicated by the fact that no direct measurement of energy modulation could be made in the parity beamline . the energy modulation sensitivities depended on the beamline tune , so the distribution of measured oppis energy modulation values and @xmath219 sensitivities was examined for a given beamline tune , and a correction and a `` worst case '' uncertainty in the correction were estimated . + net corrections were calculated for energy modulation on a year by year basis . 1998 and 1999 required two corrections each , one for each beamline helicity . 1997 needed four corrections , as two wien filter settings were used . finally all the energy modulation corrections were combined and one correction was applied to the final @xmath0 . + the energy modulation correction shown in table [ corr ] is the net effect of energy modulation over the three runs . the net correction was zero . the @xmath275 uncertainty comes from the quadrature sum of the uncertainty in energy modulation plus the uncertainty in the energy modulation sensitivities . this was not included in the individual error bars for the 23 @xmath0 numbers because the energy modulation was not known well enough . as a result , the reduced chi squared of 1.08 for the 23 sets is larger than it really `` should '' be if the energy modulation uncertainty was determined individually for each of the 23 sets and included in the individual set by set error bars . and the most precise experimental data at 13.6 mev ( bonn ) , 45 mev ( psi ) and 221 mev ( triumf ) . the solid curve shows the results obtained by carlson _ et al . _ by adjusting the weak coupling constants for the best fit to the experimental data . , width=321 ] the overall result for @xmath0 from the 23 sets is summarized in table [ corr ] . after correcting for systematic errors , the longitudinal analyzing power is found to be @xmath276 at 221.3 mev incident proton energy and the target and detector geometry of this experiment . correcting for finite geometry and target thickness ( see sec . [ subsec : energy ] ) and combining the errors in quadrature , gives @xmath277 at 225 mev for comparison with theoretical calculations . parity violation in @xmath1 scattering has already attracted considerable theoretical interest , and many calculations of @xmath0 have been made @xcite . these calculations are shown in fig . [ theory ] together with the triumf result ( corrected to the @xmath278 zero crossing energy ) and the most precise results at 13.6 mev @xcite and 45 mev @xcite . theoretical predictions for @xmath0 from several models are shown . the model of driscoll and miller @xcite is based on the bonn potential to represent the strong n - n interaction , together with the weak meson - nucleon coupling constants as given by desplanques , donoghue , and holstein ( ddh ) @xcite . the prediction of iqbal and niskanen @xcite has a @xmath279 isobar contribution added to the driscoll and miller model on a semi - ad - hoc basis . the theoretical prediction of driscoll and meissner @xcite is based on a self - consistent calculation , with both weak and strong vertex functions obtained with a chiral soliton model . finally , the quark model calculation of grach and shmatikov @xcite takes explicit account of quark degrees of freedom . none of these predictions are in good agreement with the data , although they all have similar shapes due to the energy dependence of the strong interaction . a major source of uncertainty in these calculations is the value of the weak meson - nucleon couplings . starting with the work of desplanques , donoghue , and holstein ( ddh ) @xcite , many theoretical calculations of these quantities were made @xcite but theoretical uncertainties remained large . until the triumf experiment , the coupling constants were very poorly constrained by experiment . figure [ couplings ] shows the limits on the weak meson - nucleon couplings @xmath3 and @xmath4 now imposed by the low energy results @xcite and the present triumf result . the error bands are based on a calculation by carlson _ _ @xcite assuming the argonne @xmath280 ( av-18 ) potential @xcite , the bonn 2000 ( cd - bonn ) @xcite strong interaction coupling constants , and including all partial waves up to j=8 . although the triumf measurement is not sensitive to @xmath0 from @xmath281 mixing , and the contribution from @xmath282 mixing contains no @xmath283 contribution , there is some @xmath283 dependence arising from the higher partial wave mixings . the net result is that the acceptable band defined by the triumf measurement is almost orthogonal to that defined by the low energy measurements , and greatly reduces the acceptable ranges of both @xmath3 and @xmath4 . adjusting these coupling constants for the best fit to the @xmath43 data , including the triumf 221 mev point , carlson _ @xcite estimate @xmath284 and @xmath285 , compared to the ddh `` best guess '' values of @xmath286 and @xmath287 . the solid curve in fig . [ theory ] is calculated using the carlson _ adjusted couplings . the reduction in the experimentally allowed range for weak meson - nucleon coupling constants also has implications for the analysis of electroweak radiative corrections in backward angle parity violating electron scattering . by combining back angle electron scattering data for hydrogen @xcite and deuterium @xcite , the sample collaboration was able to extract values for both the isovector axial @xmath288-@xmath289 form factor , @xmath290 , and the strange magnetic form factor , @xmath291 at a momentum transfer @xmath292 ( gev / c)@xmath293 . their value @xmath294 differs significantly from the value @xmath295 arrived at by zhu _ _ @xcite by applying one quark and many quark ( proton anapole moment ) radiative corrections to the well known nucleon axial charge as measured in neutron @xmath296-decay . @xmath3 enters in the calculation of the proton anapole moment , but its contribution is small and one would require @xmath297 to bring the value up to the @xmath298 of the sample measurement . such a value of @xmath3 is now clearly ruled out by the @xmath43 parity violation data . the parity violating analyzing power , @xmath0 , in @xmath1 elastic scattering has been measured at 221.3 mev incident proton energy . the result constrains theoretical calculations of @xmath0 in an energy region not previously covered experimentally . in the case of meson exchange calculations , it constrains principally the value of the weak rho meson - nucleon coupling @xmath3 , but , when the low energy @xmath1 data are included , strong constraints are placed on the acceptable values of both @xmath3 and @xmath4 . this result has implications for the interpretation of other experiments . for example , it rules out incorrect values of these couplings as an explanation for the disagreement between the sample isovector axial form factor result @xcite and the calculation of zhu _ et al . _ @xcite . this work was supported in part by the natural sciences and engineering research council of canada , the u.s . department of energy , and the russia - north america scientific collaboration program ; triumf receives federal funding via a contribution agreement through the national research council of canada . m. simonius , in _ interaction studies in nuclei _ , edited by h. jochim and b. ziegler , ( north holland , amsterdam , 1975 ) , p. 3 ; in _ high energy physics with polarized beams and targets _ , edited by c. joseph and j. soffer ( birkhauser verlag , basel , 1981 ) , p.355 . c.d.p . et al . _ , in _ proceedings of the international workshop on polarized beams and polarized gas targets _ ( cologne , 1995 ) , edited by h.p . schieck and l. sydow ( world scientific , singapore , 1996 ) , p. 120 ; a.n . zelenski , _ ibid . _ , p. 111 . _ , _ 15th international conference on cyclotrons and their applications ( cyclotrons 98 ) _ , caen , france , 1998 , edited by e. baron and m. lieuvin , ( inst . phys . , bristol , uk , 1999 ) p. 642

weak gravitational lensing by galaxy clusters on faint higher redshift galaxies has been traditionally used to study the cluster mass distribution and as a tool to identify clusters as peaks in the shear maps . however , it becomes soon clear that peaks statistics can also be used as a way to constrain the underlying cosmological model due to its dependence on both the cosmic expansion rate and the growth rate of structures . this feature makes peak statistics particularly interesting from the point of view of discriminating between general relativity and modified gravity . here we consider a general class of theories and compute the observable mass function based on the aperture mass statistics . we complement our theoretical analysis with a fisher matrix forecast of the constraints that an euclid - like survey can impose on the model parameters . we show that peak statistics alone can in principle discriminate between general relativity and models and strongly constrain the parameters that are sensitive to the non - linear growth of structure . however , further analysis is needed in order to include possible selection function in the peaks redshift determination . cosmology : theory gravitational lensing clusters : general

1204.3148

the gravitational field of galaxy clusters distorts the images of background faint galaxies possibly leading to the formation of spectacular giant arcs ( as first observed by @xcite ) . as pointed out by @xcite in a pioneering work , the most likely effect is , however , a variation in the ellipticity distribution of the background galaxies which can then be used to reconstruct the mass distribution of the lensing cluster ( see , e.g. , @xcite ) . such a mass reconstruction method has now become quite popular @xcite and is actually the most efficient one for intermediate and high redshift clusters where dynamical methods fail because of the difficulties in measuring galaxy redshits . the search for coherent image alignments can also be used as a way to find dark mass concentrations thus offering the opportunity of assembling a mass selected catalog of haloes . in particular , the aperture mass statistics @xcite has emerged as a valuable way to find clusters @xcite and measuring their mass function @xcite . on the other hand , although not originally conceived as a cosmological tool , the aperture mass statistics allows to severely constrain cosmological models . indeed , the number counts of peaks in the weak lensing maps is determined by both the cosmic expansion rate and the growth rate of structures ( entering through its effect on the theoretical mass function ) . peaks statistics has therefore emerged as a promising tool to discriminate among different dark energy models @xcite and constrains primordial non - gaussianity @xcite . actually , dark energy is not the only way to fit the wide amount of data indicating the present day universe as spatially flat and undergoing accelerated expansion @xcite . indeed , rather than representing the evidence of a missing source in the cosmic energy budget , cosmic speed up may also be read as the first signal of breakdown of our understanding of the laws of gravity . modified gravity theories has therefore recently attracted a lot of interest and have been shown to be able to fit the data with the same accuracy as most dark energy models . in this framework , @xmath0 theories represent one of the most natural generalization of general relativity , the basic idea being to replace the scalar curvature @xmath1 in the gravity lagrangian with a generic function @xmath0 . dating back to eddington , this idea has been first reconsidered in the 80s thanks to the possibility of recovering inflation without any scalar field and has then found a renewed interest after the discovery of the acceleration for its ability to achieve cosmic speed up in a matter only universe ( see , e.g. , @xcite and refs . therein ) . it is actually possible to show that , for any dark energy model , it is possible to work out an @xmath0 counterpart providing the same background expansion , i.e. the same @xmath2 , as the given one @xcite . on the contrary , these equivalent models can be discriminated because of the different growth of perturbations driven by a scale and redshift dependent effective gravitational constant and the non vanishing difference between the two bardeen potentials . as already reminded above , the number counts of weak lensing peaks is indeed sensible to both the expansion rate and the growth of structures so that it comes out as an ideal tool to discriminate among dark energy and @xmath0 models sharing the same @xmath2 expression . here , we will therefore investigate whether this is indeed the case for a popular @xmath0 model which has already been shown to excellently fit a wide astrophysical dataset . such a preliminary study can highlight the power of peaks statistics in discriminating among dark energy and modified gravity , but it is not immediately related to what one could actually infer from observations . we will therefore rely on the fisher matrix analysis to forecast the accuracy a realistic survey can achieve on the @xmath0 model parameters thus filling ( although in a simplified way ) the gap between theory and observations . the plan of the paper is as follows . in sect.2 we review the basic of @xmath0 theories and how one can compute the mass function taking care of the peculiarities of this modified gravity model . sect.3 is then devoted to the mass aperture statistics discussing all the steps needed to compute both the signal and the noise in weak lensing maps . the expected number of counts of peaks in @xmath0 theories and how this depends on the model parameters are presented in sect.4 , while the fisher matrix forecasts are given in sect.5 . sect.6 summarizes the main results and our conclusions . being a straightforward generalization of einstein gr , fourth order gravity theories have been investigated almost as soon as the original einstein theory appeared . it is , therefore , not surprising that the corresponding field equations and the resulting cosmology have been so widely discussed in the literature . we here first briefly review the basics of @xmath0 theories and then explain the method used to evaluate the theoretical mass function . in the framework of the metric approach , the field equations are obtained by varying the gravity action @xmath3 } \label{eq : fogaction}\ ] ] with respect to the metric components . we obtain @xmath4 where @xmath1 is the scalar curvature , @xmath5 is the standard matter lagrangian with @xmath6 the matter stress - energy tensor , and the prime ( the dot ) denotes derivative with respect to @xmath1 ( time @xmath7 ) . for @xmath8 , one obtains the usual einstein equations with a cosmological constant @xmath9 , while , in the general case , a further scalar degree of freedom is introduced . in a spatially flat homogenous and isotropic universe , some convenient algebra allows to rearrange the field equations in such a way that a single equation for the hubble parameter @xmath10 is obtained . assuming dust as gravity source and introducing @xmath11 , it is then only a matter of algebra to get : @xmath12 } { f^{\prime } - m^2 ( 1 + z ) ( d\xi / dz ) } \ , \label{eq : hubevsxi}\ ] ] @xmath13 \frac{m^2 f^{\prime \prime}}{1 + z } \frac{d\xi}{dz}\ ] ] @xmath14 \ , \label{eq : xizeq}\ ] ] where @xmath15 is the redshift , @xmath16 and @xmath17 is a convenient curvature scale which depends on the present day values of the matter density parameter @xmath18 and the hubble constant @xmath19 . by inserting eq.([eq : hubevsxi ] ) into eq.([eq : xizeq ] ) , we get a single second order nonlinear differential equation for @xmath20 that can be solved numerically provided @xmath0 and the initial conditions are given . these latter can be conveniently expressed in terms of the present day values of the deceleration ( @xmath21 ) and jerk ( @xmath22 ) parameters . to this end , we first remember that , in a frw spatially flat universe , the curvature scalar @xmath1 reads : @xmath23 so that we get for the present - day values @xcite : @xmath24 it is then only a matter of algebra to get the initial conditions for @xmath25 as @xmath26 because of the definition of @xmath25 , eq.([eq : xizeq ] ) is a single fourth - order nonlinear differential equation for the scale factor @xmath27 so that we need to know the values of the derivatives up to the third order to determine the evolution of @xmath27 thus explaining why the jerk parameter also enters as a model parameter . a key role in fourth order theories is obviously played by the functional expression adopted for @xmath0 . we choose here the hu & sawicki ( 2007 , hereafter hs ) model setting : @xmath28 where @xmath29 is given by ( [ eq : defm ] ) , and @xmath30 are positive dimensionless constants . note that , since @xmath31 we recover an effective @xmath9 term in high curvature @xmath32 environments . in particular , in the limit @xmath33 , the expansion rate @xmath34 , in the early universe , will be the same as in @xmath9cdm with an effective matter density parameter @xmath35 that guarantees that the nucleosynthesis constraints are satisfied . the hs model is determined by the three parameters @xmath30 , but it is actually convenient to replace @xmath36 with @xmath37 . to this end , we follow the method detailed in @xcite and introduce the further parameter @xmath38}$ ] which quantifies the present day deviation of @xmath0 from gr . note that , since we expect @xmath39 , @xmath40 will typically be negative and large . inserting the hs @xmath0 into eqs.([eq : hubevsxi ] ) and ( [ eq : xizeq ] ) and setting the model parameters , one can solve the fourth order nonlinear differential equation for the scale factor @xmath27 and then work out the dimensionless hubble rate @xmath41 . in order to speed up the computation , however , we use here the following approximated solution @xcite : @xmath42 e_{\lambda}(z ) & z \le z_{\lambda } \\ ~ & ~ \\ e_{\lambda}(z ) & z \ge z_{\lambda } \\ \end{array } \right . \label{eq : hubapprox}\ ] ] where @xmath43 is the hubble rate for the phenomenological cpl @xcite parametrization of the effective dark energy fluid equation of state , namely @xmath44 , while @xmath45 . finally @xmath46 is an interpolating function with @xmath47 and @xmath48 depending on @xmath49 . this approximating function excellently reproduces the numerical solution whatever the model parameters @xmath49 are with a rms error which is far lower than @xmath50 over the redshift range @xmath51 . a subtle remark is in order here concerning the value of @xmath18 . indeed , while for @xmath52 , we use the actual matter density parameter , the effective one must be used for @xmath53 . therefore , a discontinuity in @xmath54 is formally present in our approximation . actually , it is easy to show that , for all reasonable model parameters , @xmath18 and @xmath55 are almost perfectly equal so that the discontinuity can not be detected at all and @xmath41 is , to all extents , a continuous function . the extra scalar degree of freedom introduced by the modified gravity lagrangian mediates an enhanced gravitational force on scales smaller than its compton wavelength . in order to hide this boost from local tests of gravity , viable @xmath0 models resort to the chameleon mechanism @xcite which makes the field compton wavelength depending on the environment gravitational potential . as a consequence , for viable @xmath0 models , the gravitational potential reduces to the newtonian one on solar system scales , while nonlinearities in the field equations appear for galaxy scale systems . from the point of view of the mass function , the enhanced gravitational force has a particularly strong impact on the abundance on intermediate mass haloes . indeed , on one hand , the extra force increases the merging rate of low mass haloes into intermediate ones , while the chameleon shuts down the merging of these latter into highest mass ones . an efficient method to take the chameleon effect into account and work out a mass function in agreement with simulations has been recently proposed by @xcite based on a simple interpolation scheme . following their approach , we model the mass function as : @xmath56 where @xmath57 is the number of halos with the ratio between density contrast and variance of the perturbations equal to @xmath58 , @xmath59 is the critical overdensity for collapse and @xmath60 the variance of the perturbations on the scale @xmath1 corresponding to the mass @xmath61 , given by @xmath62 = \frac{1}{(2 \pi)^3 } \int{p_{\delta}(k ) |w(kr)|^2 d^3k } \ . \label{eq : sigmavardef}\ ] ] here , @xmath63 is the fourier transform of the spherical top hat function , while the density power spectrum @xmath64 may be computed as : @xmath65 with @xmath66 the transfer function , @xmath67 the growth factor normalized to 1 at present day and @xmath68 a normalization constant conveniently expressed as function of @xmath69 , the present day variance on the scale @xmath70 . since @xmath0 models reduce to gr in the prerecombination epoch , we can use for @xmath66 the analytical approximation in @xcite although this latter has been obtained assuming gr validity . on the contrary , the growth factor for @xmath0 models gains a scale dependence which is not present for gr dark energy models . indeed , the evolution of the density perturbations , @xmath71 , in fourth order gravity models is ruled by @xmath72 with @xmath73 the wavenumber and @xmath74}{1 + 3 ( k^2/a^2 ) [ f^{\prime \prime}(r)/f^{\prime}(r ) ] } \label{eq : geffdef}\ ] ] the scale dependent effective gravitational constant @xcite . note that , for @xmath8 , @xmath75 reduces to @xmath76 and the growth factor is no more scale dependent . as shown in fig.[fig : geff ] , setting the model parameters to the fiducial values , @xmath77 only slightly deviates from the newton constant . nevertheless , this deviation is both redsfhift and scale dependent so that it has an impact on the final matter power spectrum . in order to compute the mass function through eq.([eq : mf ] ) , one has to choose an expression for @xmath78 . we again follow @xcite and adopt the @xcite function setting @xmath79 \exp{(-a \nu^2/2 ) } \label{eq : stmf}\ ] ] with @xmath80 . it is , finally , worth stressing that both @xmath59 and @xmath81 ( which enters the conversion from @xmath61 to @xmath82 ) actually depends on the matter density parameter @xmath18 and the physics ruling the collapse and virialization of perturbations . previous attempts to fit the numerically derived mass function searched for a @xmath0 derivation of @xmath83 and then use the mass variance @xmath84 evaluated from the @xmath0 linear power spectrum . on the contrary , here we follow the approach of @xcite holding @xmath83 fixed to the values predicted for the gr model having the same background expansion as the @xmath0 one . we then use for the mass variance an interpolation between the @xmath0 and gr ones , i.e. @xmath84 is estimated as : @xmath85 with @xmath86 and @xmath87 computed from the @xmath0 and gr linear power spectra and @xmath88 two interpolation parameters given by @xmath89 finally , we follow @xcite to get @xmath83 as a function of @xmath18 and the dimensionless hubble rate @xmath41 . + the approximated formula ( [ eq : hubapprox ] ) for @xmath41 and the @xcite formalism allows us to estimate the mass function for the hs @xmath0 model provided we set the parameters @xmath90 . moreover , one also has to set the slope of the power spectrum through @xmath91 and fix its normalization through @xmath69 . since we are mainly interested in how the mass function looks in @xmath0 models and whether it is possible to discriminate among the hs model and its corresponding gr based dark energy model with the same expansion rate , we will not vary all the parameters at play . on the contrary , we will first fix the expansion rate to that of a cpl model with @xmath92 second , we fix the shape and the amplitude of the matter power spectrum setting @xmath93 in agreement with @xcite constraints . as a fiducial hs model , we choose the one obtained by setting @xmath94 since it best mimics the cpl expansion rate . moreover , this model is quite similar to the best fit one found by @xcite fitting a large dataset including sneia , grbs , baos and wmap7 distance priors . the mass function ( mf ) for this parameter set ( which we will refer to in the following as the fiducial model ) is plotted in the left panel of fig.[fig : mffid ] for three different redshift values . although not clearly visible , a kink is present in the mf at the scale @xmath95 as a consequence of the interpolation scheme adopted to reproduce the mf from n - body simulations . actually , such a feature is not an artifact , but rather the evidence of the chameleon effect . indeed , the scale at which the kink is present is the same as the one marking the onset of the chameleon mechanism which increases the abundance of intermediate mass haloes . this can also be seen from the right panel where we compare the hs mf with the cpl one . for @xmath96 , the chamaleon effect fully masks the enhancement of the gravitational force introduced by the modified gravity potential thus leading to a mf which is the same as the cpl one . in the opposite regime , the effective field is no more negligible and the gravitational force is significantly boosted leading to a marked increase of the abundance of intermediate mass haloes . since the strength of the effective field is a function of @xmath97 , the threshold chamaleon mass will change with the redshift so that the ratio between the hs and cpl mfs does depend not only on the mass , but on the redshit too . although discriminating between hs and cpl is interesting in its own , it is also instructive to look at how the hs mf depend on the @xmath0 parameters . this can be inferred from fig.[fig : mfhspar ] where we plot the percentage deviation from the fiducial model varying @xmath40 for the fiducial @xmath98 at different redshifts . as an encouraging result , we find that @xmath99 can take quite large values in the range @xmath100 for all @xmath97 . similarly large values can be achieved for larger masses , but we warn the reader not to overcome the results in this range since they refer to mfs which have almost vanished so that even a small difference in value gives rise to a large @xmath99 . explaining the shape of @xmath99 is not a straightforward task . on one hand , the scaling of the chameleon threshold mass @xmath101 as @xmath102 implies that deviations from the fiducial mf are positive or negative depending on the mass being larger or smaller than the corresponding chamaleon mass . on the other hand , since @xmath75 also depends on @xmath40 , the mass variance @xmath84 , which enters the @xcite profile through @xmath103 , will be a function of @xmath40 thus introducing a second way through which @xmath40 affects the mf . it is actually difficult to disentangle the two effects so that the final non monotonic behaviour in fig.[fig : mfhspar ] is not immediate to quantitatively explain . it is , on the contrary , much easier to understand why we find that @xmath98 has only a very minor role in setting the shape and the amplitude of the mf with @xmath99 being smaller than @xmath104 . indeed , since the background expansion is almost the same for each @xmath98 ( see , e.g. , the discussion in @xcite ) , the only way @xmath98 can impact the mf is through the effective gravitational constant @xmath75 . for the hs models we are considering , this dependence is actually quite mild and is further smoothed out by the procedure leading to the final mf . note that this is partially a consequence of how we set the fiducial parameters . indeed , for @xmath94 , we get @xmath105 so that , in the high curvature regime , we get @xmath106 with @xmath98 playing a marginal role in setting the ratio @xmath107 . should we have chosen , e.g. , @xmath108 as fiducial , we would have obtained @xmath109 and hence the term @xmath110 in the high curvature lagrangian would have played a much significant role likely leading to a stronger dependence of the mf on @xmath98 . however , such a large @xmath40 does not lead to a background expansion in agreement with our fiducial cpl one so that we have not considered this kind of models . as a final remark , we stress that both figs.[fig : mffid ] and [ fig : mfhspar ] refer to the theoretical mf . this is not what is actually observed so that the differences between the hs and cpl mfs or among the different hs model parameter sets can be strongly suppressed when the selection effects introduced by the particular method adopted to observationally determine the mf are taken into account . should the observational mf be defined only for @xmath111 or be measured with too large uncertainties at higher @xmath97 , the possibility to discriminate among the hs and cpl models and/or to constrain the hs parameters @xmath112 would be seriously compromised . being the largest and most massive mass concentrations , galaxy clusters are ideal sites to look for gravitational lensing effects . should the source be aligned with the cluster , the formation of spectacular arcs takes place , but , for most cases , the main effect is a coherent distortion of the shape of background galaxies , which appears stretched tangentially around the cluster . the distortion field can be used to reconstruct the shear map and then infer constraints on the cluster mass distribution . on the other hand , a shear map of the sky presents distinct peaks corresponding to the cluster positions thus offering a technique based on weak lensing to identify galaxy clusters . as a peak finder , we consider here the aperture mass defined by @xcite @xmath113 where @xmath114 and @xmath115 $ ] are the convergence and the tangential shear at position @xmath116 and @xmath117 , @xmath118 are compensated filter functions related to each other by the integral equation @xmath119 in order to detect a cluster as a peak in the aperture mass map , we have to preliminarily estimate the @xmath120 variance and then set a cut on the signal - to - noise ( s / n ) ratio . to this end , we have to specify how we compute both the signal and the noise which we address in the two following subsections . for given lens and source redshift @xmath121 , the aperture mass depends on the mass density profile of the halo acting as a lens . we will model cluster haloes using the spherically symmetric nfw profile @xcite : @xmath122 with @xmath123 , @xmath124 and @xmath125 note that we have reparameterized the model in terms of the virial mass @xmath61 , defined as the mass within the virial radius @xmath82 where the mean density equals @xmath126 with @xmath127 the critical density at the lens redshift , and the concentration @xmath128 . according to n - body simulations , the nfw model can be reduced to a one parameter class since @xmath129 correlates with the virial mass @xmath61 . actually , the slope , the scatter and the redshift evolution of the @xmath129-@xmath61 relation are still matter of controversy with different results available in the literature . as a fiducial case , here we follow @xcite setting @xmath130 referring to this in the following as the mc11 relation . however , in order to investigate the dependence of the results on the adopted mass - concentration law , we will also consider the @xmath129-@xmath61 relation empirically found by buote et al . ( 2007 , hereafter b07 ) from observed x - ray galaxy systems spanning the mass range @xmath131 , namely : @xmath132 with @xmath133 as inferred from the fit to the full sample . it is worth noting that both @xmath129-@xmath61 relations are affected by a significant scatter . in order to take this into account , one should convolve the aperture mass for a given @xmath61 value with a lognormal distribution centred on the predicted @xmath129 value and with variance equal to the scatter itself . however , we follow here the common practice of neglecting this issue and simply insert the analytical expression for the convergence @xmath134 of the nfw profile @xcite into eq.([eq : mapdef ] ) with @xmath129 set according to eq.([eq : cvmvmc ] ) or ( [ eq : cvmvbul ] ) . a caveat is in order here . both the nfw profile and the @xmath129-@xmath61 relations have been inferred from n - body simulations carried on in a @xmath9cdm cosmological framework hence implicitly assuming that , on galaxy scales , the gravitational potential is newtonian . on the contrary , we are here investigating @xmath0 theories so that , strictly speaking , the classical newtonian theory does not hold anymore and the potential is modified . actually , a yukawa - like term is added to the newtonian potential with a scale length which , due to the chameleon effect , depends on the environment and the strength of the effective field related to the modified gravity lagrangian @xcite . however , for the typical values of the @xmath0 model parameters we are going to consider , the effective potential is almost identical to the newtonian one so that we do not expect deviations of the halo profile from the nfw one . this is indeed what @xcite have found in their analysis of the halo profiles from n - body simulations carried on for the same hs model we are considering here . although a @xmath129-@xmath61 relation was not reported , we expect that the ( unknown ) actual relation is within the range set by the mc11 and b07 relations . it is worth noting that , being sensitive to all the matter along the line of sight , the observed @xmath120 is actually the sum of the contribution due to both the cluster and the uncorrelated large scale structure projected along the same line of sight , i.e. @xmath136 . in the usual approach , one considers that , being a density contrast , @xmath137 averages to zero so that no bias is introduced in the @xmath120 statistics . on the contrary , the lss will contribute to the variance representing an additional source of noise @xcite . the filter functional form and its parameters are then chosen based on a comparison with simulated data and the characteristics of the survey ( see , e.g , @xcite ) . actually , such a method is not fully efficient in filtering out the lss contribution and is , moreover , related to the underlying cosmology adopted in the reference simulation . a possible way out of this problem relies on the use of the so called _ optimal filter _ introduced by maturi et al . ( 2005 , hereafter m05 ) to take explicitly into account both the shape of the halo signal and the underlying cosmology . following m05 , we set the fourier transform of the filter as @xmath138^{-1 } \frac{\hat{\tau}(\ell)}{p_n(\ell ) } \ , \label{eq : defpsihat}\ ] ] where @xmath139 is the fourier transform of the signal ( in our case , the tangential shear component ) and @xmath140 the noise power spectrum as a function of the angular wavenumber @xmath141 . this latter is made up by the sum of two terms : @xmath142 with @xmath143 the noise contribution from the finite number of galaxies ( with number density @xmath144 ) and their intrinsic ellipticities ( @xmath145 being the variance ) , and @xmath146 the noise due to the lss . note that this latter is set equal to half the convergence power spectrum since we use only one component of the shear . under the limber flat sky approximation , it is ) is the same as in gr based dark energy models which could be surprising at first sight since we are using a modified gravity theory . actually , it has been shown @xcite that , for @xmath0 theories , the convergence power spectrum is still given by eq.([eq : limber ] ) provided the matter power spectrum @xmath147 is computed taking care of the running effective gravitational constant and the result is multiplied by the factor @xmath148 . since , for the models we are considering , @xmath149 with great accuracy , we have neglected this correction when computing @xmath150 . ] @xmath151 with @xmath152 the comoving distance to redshift @xmath97 ( with @xmath153 the distance to the last scattering surface ) , @xmath154 is the lensing weight fucntion @xmath155 and @xmath156 is the source redshift distribution here parameterized by @xcite @xmath157 } \label{eq : pz}\ ] ] and normalized to unity . following @xcite , we set @xmath158 so that @xmath159 is the median redshift of the sources as expected for the euclid mission . following @xcite and @xcite for the shear profile of the nfw model , one finally gets for the fourier transform of the filter : @xmath160^{-1 } \nonumber \\ & \times & \frac{\tilde{\tau}(\ell \theta_s)}{p_n(\ell ) \tilde{{\cal{d}}}(\theta_s ) } \label{eq : fourierpsi}\end{aligned}\ ] ] where @xmath161 is the critical density for lensing ( depending on the lens and source redshift ) , @xmath162 , @xmath163 is the angular scale corresponding to the characteristics radius @xmath164 , and we have defined the dimensionless quantities @xmath165 @xmath166 with @xmath167 the dimensionless shear profile given by @xcite . taking the * back * fourier transform of eq.([eq : fourierpsi ] ) and setting @xmath168 in the aperture mass definition , it is only a matter of algebra to finally get : @xmath169 with ( @xmath170 and @xmath171 ) @xmath172 \cos{(2 \theta ) } d\theta } \ . \nonumber \label{eq : defmaptilde}\end{aligned}\ ] ] * a conceptual remark is in order here . eqs.([eq : endmap ] ) has been obtained starting from eq.([eq : mapdef ] ) and assuming that the shear profile is the same as the cluster one . actually , the tangential shear in eq.([eq : mapdef ] ) is obtained from the observed galaxies ellipticities so that it is not strictly equal to the model shear which one estimates from the cluster model . however , since we are interested in making analytic predictions , such an identification is unavoidable . it is also worth noting that what is actually observable is the reduced shear @xmath173 which is , however , equal to @xmath174 in the weak lensing limit @xmath175 we are interested in here . * a naive look at eqs.([eq : endmap])-(35 ) could lead to the surprising conclusion that the aperture mass does not depend on the halo properties . this is , of course , not the case . first , the halo virial mass @xmath61 and redshift @xmath176 directly enter the definition of @xmath163 which also depends on the adopted @xmath129-@xmath61 relation . second , the halo model sets the dimensionless shear profile @xmath167 which enters both in the filter derivation and the @xmath177 function . on the contrary , the dependence on the source redshift @xmath178 has been integrated out when computing the filter function . this is a consequence of how the optimal filter has been constructed , i.e. , imposing that the filter is optimized to find halos with a given mass profile , virial mass and redsift notwithstanding the source redshift . this latter quantity , however , plays a role when computing the noise given by instead of @xmath179 in eq.([eq : defsigmaap ] ) for the total variance to take into account the contribution of both the shot noise and the large scale structure . however , we are here considering the signal as the sum of the cluster and lss peaks so that only @xmath179 has to be considered as noise term . * ] @xcite @xmath180 ^ 2 } \ell d\ell } \label{eq : defsigmaap}\end{aligned}\ ] ] so that the @xmath135 ratio reads parameters entering the @xmath135 ratio refer to the cluster halo responsible of the signal one is detecting . this halo can also be different from the template model used to generate the filter . however , the @xmath181 template parameters setting the filter normalization cancels out when computing the @xmath135 ratio so that , in order to simplify the notation , we have not used different symbols to differentiate the template and actual halo parameters . * ] @xmath182 where @xmath183 summarizes the parameters the @xmath135 ratio depend on , namely the halo quantities @xmath184 and the source redshift @xmath178 ( entering because of @xmath185 ) . since we are interested in the total @xmath135 ratio , we integrate over the source redshift distribution thus finally obtaining @xmath186 which , for a given @xmath129-@xmath61 relation and mass profile , depends only on the halo virial mass and redshift . eq.([eq : stonend ] ) enables us to estimate the @xmath135 ratio for a nfw halo with virial mass @xmath61 and concentration @xmath129 acting as a lens at redshift @xmath176 . in order to compute @xmath187 , we need first to assign the survey characteristics ( determining the noise power spectrum @xmath179 ) and set the * filter scale * @xmath188 . we consider an euclid - like survey @xcite and take a survey area of @xmath189 with an ellipticity dispersion @xmath190 and a number of galaxies @xmath191 . note that these are the survey goal , but one can easily scale the @xmath135 ratio noting that @xmath192 , while the total number of peaks scales linearly with the survey area . the choice of the * filter scale * @xmath188 asks for some caution . the optimal filter has been designed taking into account the nfw profile in order to maximize the signal . a natural scale would therefore be @xmath193 since most of the mass contributing to the lensing signal is contained within this aperture . for a cluster with @xmath194 at the median survey redshift @xmath195 , we get @xmath196 , while the virial radius subtends an angle @xmath197 ( in our adopted fiducial cosmology ) . however , not all the clusters have the same mass and the same redshift . setting @xmath198 would be an optimal choice for these median values , but will strongly underestimate the signal for clusters at lower redshifts . on the contrary , a varying @xmath188 would allow to maximize the @xmath135 at every redshift , but would introduce an inhomogeneity in the computation of the number of peaks . as a compromise , we therefore set @xmath199 noting that the filter actually cuts the contribution from regions outside few times @xmath163 and vanishes on scales larger than @xmath200 ( which can be smaller than @xmath201 for low mass and/or high redshift clusters ) . it is instructive to look at the @xmath135 ratio as a function of the cluster redshift and virial mass for a given cosmological model . to this end , we consider the fiducial hs model with @xmath202 and look at how @xmath187 depends on the adopted @xmath129-@xmath61 relation . fig.[fig : snratio ] shows @xmath187 as function of the virial mass at different redshift values for the mc11 and b07 cases ( left and central panels , respectively ) . for both models , the trend with @xmath203 is the same : the higher the redshift , the larger must the virial mass be in order to attain a given @xmath187 value . this can be qualitatively explained by noting that the integration in eq.([eq : stonend ] ) is performed on a smaller interval with increasing @xmath97 thus reducing @xmath187 . in other words , the larger is the cluster redshift , the lower is number of sources available for lensing so that the @xmath135 ratio takes smaller values . in order to compensate for this reduction , one has to increase the virial mass since @xmath187 approximately scales as @xmath204 with the slope @xmath205 depending on the adopted @xmath129-@xmath61 relation . the right panel in fig.[fig : snratio ] compares @xmath187 for the mc11 and b01 cases as a function of @xmath61 and @xmath176 showing that the @xmath135 ratio turns out to depend strongly on the @xmath129-@xmath61 relation . in particular , for less massive systems , the mc11 relation predicts a values of @xmath187 which can be up to @xmath206 larger than the b07 ones . the difference is , however , not monotonic reverting its sign for intermediate mass haloes and becoming quite small for very large mass . such a strong sensibility to the @xmath129-@xmath61 model is not unexpected and is a consequence of the different concentrations for the same virial mass . in particular , the ratio @xmath207 depends on both @xmath176 and @xmath61 thus explaining why the @xmath208 may change sign according to which model predicts the larger @xmath129 value . as a final remark , we stress that the above results are almost independent of the underlying cosmological model . this can be easily explained by looking at how cosmology enters in the evaluation of the s / n ratio . first , we note that @xmath187 depends on the integrated hubble rate through the lensing critical density @xmath185 . this quantity only weakly depends on cosmology since it involves an integral of a ratio of distances over the source redshift . on the other hand , the convergence power spectrum @xmath209 enters the filter definition and hence both @xmath120 and @xmath210 so that its impact is reduced when considering the ratio of these two quantities . moreover , for the hs model parameter space we are interested in , the hubble rate @xmath2 is almost the same as the cpl one so that changing the @xmath211 values has a very minor impact on the s / n ratio . this is indeed what we have checked . varying @xmath212 shifts @xmath187 by less than @xmath213 ( but typically a shift larger than @xmath104 is difficult to get ) , while the @xmath135 ratio may be considered independent of @xmath98 within a very good approximation . having detailed how the s / n ratio can be computed and having determined the theoretical mf , we can now estimate the number density of haloes with mass @xmath61 that produces significant peaks in the aperture mass map . to this end , we first have to take into account that a s / n threshold for the weak lensing signal does not correspond to an equally sharp threshold in halo mass because of the scatter in @xmath120 caused by the shot noise from discrete background galaxy positions and the intrinsic ellipticity distribution . a halo of mass @xmath61 has therefore a certain probability @xmath214 to produce an aperture mass @xmath120 which we can model as a gaussian : @xmath215 ^ 2 \right \ } } \label{eq : mapprob}\ ] ] where @xmath216 is the theoretically expected value and @xmath210 the variance in eq.([eq : defsigmaap ] ) integrated over the source distribution . the probability that the @xmath135 ratio will be larger than a given threshold value will then be given by @xcite : @xmath217 \label{eq : snrprob}\ ] ] so that the number density of haloes giving a detectable weak lensing peak finally reads @xmath218 where , hereafter , we will drop the label @xmath219 from @xmath97 to denote the cluster redshift . the total number of peaks generated by cluster haloes and with @xmath135 larger than a threshold value @xmath220 is then obtained by integrating over @xmath97 and multiplying for the survey area @xmath221 thus finally reading : @xmath222 where we set @xmath223 as redshift limits , while @xmath224 is the comoving distance . the number of observed peaks is , however , larger than @xmath225 because of the contamination from the lss . this latter term may be estimated as @xcite @xmath226 } \label{eq : npklss}\ ] ] with @xmath227 and @xmath228 note that @xmath229 only depends on the noise properties and the threshold @xmath135 ratio , but not on the lens mass and redshift since it is not related to the particular cluster at hand , but to the lss . for this same reason , @xmath229 is determined by the matter power spectrum ( and hence the underlying cosmological scenario ) entering @xmath230 . this will actually offers a further opportunity to discriminate among gr based dark energy models and modified gravity theories . the total number of observed peaks will finally be given by : @xmath231 + a look at fig.[fig : npkfid ] helps us to highlight many important issues . first , in the left panel , we plot @xmath232 as a function of the threshold @xmath135 ratio for the fiducial hs model and the mc11 @xmath129-@xmath61 relation showing separately the contribution of the clusters and lss terms . as expected , @xmath233 is dominated by the lss term for small @xmath135 ratio , that is to say , the smaller is the @xmath135 ratio , the larger is the probability that the detected peak is a fake due to the lss rather than the evidence for a cluster . this is in agreement with common sense expectation and previous analysis in literature using different cosmological models and survey parameters @xcite . the term @xmath229 , however , quickly becomes subdominant so that one can confidently be sure that all peaks detected with a @xmath135 ratio larger than @xmath234 are due to clusters . note that we find a value for @xmath220 comparable but larger than what is suggested in @xcite because of differences in both the cosmological model and the survey characteristics . the central panel in fig.[fig : npkfid ] shows the impact of the adopted @xmath129-@xmath61 relation zooming on the high @xmath187 region which is dominated by the clusters term . it turns out that the number of peaks is systematically larger if one adopts the b07 rather than the mc11 relation . this is actually not a fully unexpcted result . indeed , as the right panel of fig.[fig : snratio ] shows , the @xmath135 ratio is larger for the b07 model in the high mass regime . put in other words , for a given @xmath220 , the number of high mass clusters with @xmath235 is larger for the b07 @xmath129-@xmath61 relation . since these are the main responsible of the total number of high @xmath135 peaks , it is then straightforward to explain why we indeed find that @xmath233 is larger for the b07 relation and why the offset increases with @xmath187 . the most interesting issue to address is whether the number count of detectable peaks in the aperture mass maps allows to discriminate between the hs and cpl models notwithstanding their equal expansion rate . the right panel of fig.[fig : npkfid ] offers a first answer to this question . should we rely on clusters peaks setting @xmath236 for detection @xcite , the difference between the two models amounts to a modest @xmath237 with the hs case giving a larger number of peaks . this is somewhat surprisingly considering that the enhanced growth of structures in @xmath0 theories should lead to a larger number of clusters for a given mass . actually , such a result is not fully unexpected if one looks back at the right panel in fig.[fig : mffid ] . here , we have shown that the hs mf is larger than the cpl one for @xmath238 , while the difference becomes negligible at higher masses because of the onset of the chameleon effect . since the @xmath135 ratio for haloes with masses below @xmath239 is quite small ( unless they are at low @xmath97 ) , their contribute to the total number of peaks is strongly suppressed thus lowering the difference with the offset between the hs and cpl predictions . it is , however , worth noticing that this same offset significantly increases ( up to @xmath240 ) and changes sign if we decrease the threshold @xmath135 ratio . this is related to the onset of the lss dominated regime . in such a case , what is important is the difference in the convergence power spectrum entering the ratio @xmath241 , while @xmath209 plays a minor role in @xmath242 . we can therefore conclude that the number of low @xmath135 peaks is a valuable tool to discriminate between hs and cpl models . it is also worth stressing that the offset between the two models predictions is ( within a good approximation ) independent of the adopted @xmath129-@xmath61 relation as can be seen from the closeness of the red and blu curves in the figure . this is an expected result considering that this ingredient only enters the determination of the @xmath135 , but does not change how the number of peaks depend on the underlying cosmological model . the total number of peaks is evaluated by integrating over the full redshift range thus smoothing out the dependence of the mf on @xmath97 . actually , how the mf changes with @xmath97 depends on the underlying cosmological model so that it is worth investigating what can be learned by considering the number of peaks in redshift bins . this is still given by eq.([eq : npktot ] ) provided one replaces @xmath243 with @xmath244 . we set @xmath245 and show the resulting @xmath246 in fig.([fig : npkvsz ] ) . note that here we have turned off the lss term . since these peaks are fake detections due to the lss , it is obvious that no redshift can be attributed to them . we can therefore assume that they have been manually deleted from the sample and simply set . since @xmath247 is the sum of the two terms due to the haloes and the lss , respectively , and the lss one is independent on @xmath97 , we trivially get @xmath248 . ] the left panel suggests that binning the data has not improved the constraints , but this is only part of the story . indeed , the percentage difference between hs and cpl predictions are quite modest @xmath250 up to @xmath251 , but they become as large as @xmath252 in the highest redshift bins . such a behaviour can be qualitatively explained looking back at the right panel of fig.[fig : mffid ] showing that the larger is @xmath97 , the larger is the difference between the hs and cpl mf . probing the high-@xmath97 end of the @xmath253 curve is therefore an indirect method to investigate the mf in the regime where it is most sensible to the underlying gravity theory . as a further remark , we also note that the percentage difference between hs and cpl predictions at a given @xmath97 is smaller for higher @xmath135 thresholds . this can be qualitatively explained noting that , in order to have a large @xmath187 peak , the cluster must be very massive so that the larger is @xmath220 , the more we are investigating the high mass tail of the mf . as already stated before , the hs and cpl mfs become degenerate for @xmath254 so that a sample probing this range will be unable to discriminate between the two competing models . although discriminating between cpl and hs scenarios is likely the most intriguing issue to consider , it is nevertheless of deep interest to investigate how @xmath255 depends on the hs model parameters . to this end , we hereafter adopt the mc11 relation since this gives a lower number of detections thus providing conservative estimates . fig.[fig : npkpar ] shows that only @xmath40 has an important effect on the peak number with deviations which can be of order @xmath256 depending on the redshift bin considered . comparing fig.[fig : npkpar ] with fig.[fig : mfhspar ] clearly shows the impact of moving from a theoretical to an observable quantity . indeed , the deviations from the fiducial model are almost halved the expected values , because only the haloes with masses large enough to lead to a @xmath135 ratio larger than the threshold can now be detected thus removing the part of fig.[fig : mfhspar ] with large deviations . as a consequence , while the possibility of discriminating among different @xmath40 values is still significant , the dependence on @xmath98 almost disappears with deviations being smaller than @xmath104 and only present in the highest redshift bins . finally , although not @xmath0 parameters , we also plot the deviations from the fiducial model due to changes of the matter density @xmath18 and the variance @xmath69 . the bottom left panel of fig.[fig : npkpar ] shows that deviations as high as @xmath257 can be obtained by changing @xmath18 , but the way @xmath258 depends on @xmath97 is different from what happens when varying @xmath40 thus suggesting that no strong degeneracy between these two parameters is present . similarly , the lower right panel tells us that @xmath69 also plays a major role in determining the peak number counts in the highest redshift bins thus suggesting that strong constraints could be put on this parameter . in order to quantify the power of peak number counts to constrain the hs model parameters , we carry on a fisher matrix analysis considering as observed data the total number of peaks with @xmath235 in equally spaced redshift bins centred on @xmath97 and with width @xmath245 over the range @xmath259 . as a first step , we need to define a likelihood function @xmath260 to quantify how well a theoretical model matches the observed number counts . should the errors be gaussian , one can resort to the usual @xmath261 statistics and define @xmath262 . however , when dealing with number counts , one can assume poisson errors and then rely on the so called _ c statistics _ @xcite to define : @xmath263 where @xmath183 is the set of model parameters , @xmath264 the number of redshift bins and , for notational clarity , we have defined @xmath265 and @xmath266 to denote the theoretical and observed number of peaks in the bin centred on @xmath267 ( and with width @xmath245 ) . the fisher matrix elements will then be given by the second derivatives of the logarithm of the likelihood with respect to the parameters of interest evaluated at the fiducial values . starting from ( [ eq : deflike ] ) , one gets @xcite : @xmath268 where @xmath269 is the expected number of peaks in the @xmath73-th bin for the fiducial model . the inverse of the fisher matrix gives us the covariance matrix with its diagonal elements representing the lowest variance that one can achieve on the parameter @xmath270 . it is worth noting that the fisher matrix approximates confidence regions as a gaussian ellipsoids , while the true confidence ones can have broad tails or significant curvature . however , @xcite have checked that this is not the case for number counts by comparing with monte carlo analysis of simulated datasets . we expect that our estimated iso - likelihood contours provide a reliable approximation of the degeneracies in the parameters space . since here we are mainly interested in how peak number counts constrain @xmath0 theories , we will not consider the full eight dimensional parameter space . on the contrary , we will set @xmath271 , while @xmath272 are held fixed to their fiducial values . while this is not what one will actually do when dealing with real data , it is worth noting that here we are only considering peaks statistics as a constraint . actually , the background quantities @xmath273 will be severely constrained by future sneia data , while @xmath91 is determined by the fit to the cmbr spectrum . in a fully realistic approach , we should therefore both explore the full 8d parameter space and add the likelihood terms related to sneia and cmbr data . in order to better highlight the impact of peak statistics , we prefer to deal only with this kind of data and limit the parameter space . we therefore only consider @xmath274 since they are intimately related to the mass function and the @xmath0 model . the errors on these four quantities turn out to depend on the threshold @xmath135 used and are summarized in table[tab : fmres ] , while fig.[fig : fmplot ] shows the iso - likelihood contours in 2d spaces marginalizing over all the remaining parameters for the cases @xmath275 . .fisher matrix forecasts for the accuracy on the parameters @xmath274 from fitting the @xmath276 data for different threshold @xmath135 ratio @xmath220 . [ cols="^,^,^,^,^",options="header " , ] it is interesting to note how the constraints scale with the threshold @xmath135 . on the one hand , increasing @xmath220 has the effect of reducing the overall number of peaks thus one expects a worsening of the constraints because of the poorer statistics . this is indeed the case for @xmath277 , but not for @xmath278 . this is likely related to a second effect of a larger @xmath220 . as a consequence of the mass and redshift scaling of the @xmath135 ratio shown in fig.[fig : snratio ] , varying @xmath220 implicitly leads to probing a different mf regime . since the dependence of the mf @xmath279 on @xmath278 is different depending on the @xmath280 region of the parameter space one is considering , the constraints will turn out to depend in a complicated way from which kind of peaks one is detecting . our main interest in the peaks statistics was motivated by constraining the @xmath0 model parameters . taken at face values , the constraints on @xmath40 could look not so encouraging . we stress , however , that they are still competitive with those from an analysis based on background evolutionary probes as can be appreciated remembering that @xcite obtained @xmath281 from fitting sneia+grbs+bao+cmb distance priors vs the present forecasted @xmath282 value ( for @xmath283 ) using weak lensing peaks counts only . as it is apparent from both the numbers and the plots , the only parameter which remains fully unconstrained is the slope @xmath98 of the hs model . we could naively anticipate this result by simply looking at the top right panel of fig.[fig : npkpar ] where one can see that @xmath255 changes by less than @xmath50 up to @xmath284 so that this parameter has a negligible impact on the observed peak number . this conclusion can be qualitatively explained by looking at the effective gravitational constant @xmath75 in eq.([eq : geffdef ] ) . for the @xmath40 values typical of viable hs models , the dependence of @xmath75 on @xmath98 almost cancels out from the ratio so that the power spectrum @xmath285 is almost independent of @xmath98 with the residual sensitivity being smoothed out from the integration needed to get the variance @xmath286 . it is therefore not surprising that the fisher matrix analysis finally leads to a large uncertainty on @xmath98 . we , nevertheless , caution the reader that this is partly a consequence of how we have set the fiducial model . indeed , as already hinted at in sect.2.3 , for @xmath94 , the @xmath0 lagrangian is quite similar to the @xmath9cdm one with the term depending on @xmath98 giving an almost negligible contribution . investigating how the fisher matrix forecasts change with the fiducial model is , however , outside our aims here . the stronger dependence of @xmath255 on @xmath40 which can be read from the top left panel of fig.[fig : npkpar ] , on the contrary , helps us to understand why peaks statistics is so efficient in constraining this parameter . again , we can look at @xmath75 to qualitatively explain why the results so strongly depend on @xmath40 . first , we note that , for the values of interest , we get @xmath287 within a good approximation . second , the interpolation between @xmath86 and @xmath87 in eq.([eq : sigmalh ] ) is mediated by a mass scale which depends on @xmath40 . small variation in this quantity can therefore significantly change the number of peaks thus leading to stronger constraints . it is worth stressing that the constraints in table[tab : fmres ] are actually quite optimistic since they are based on the assumption that one is able to partition the detected peaks in redshift bins . first , we note that , should the peak be due to the lss , it will not have any actual counterpart in optical images so that the notion of redshift can not be applied . we are therefore forced to consider only the results for the case @xmath283 ( or larger ) . even limiting our attention to these high @xmath187 peaks only , there is still the problem of how to divide them in redshift bins . ideally , one should rely on optical clusters finders @xcite to match the peaks catalog with optical detections and then infer the redshift . * to this end , it is worth noting that most of the optical finders techniques rely on the same kind of data as those available for euclid so that the match with weak lensing detections could be immediate . moreover , since the cluster redshift estimate is based on more galaxies at once supplemented by geometrical constraints , the precision on @xmath97 will be similar ( if not higher ) than that of the photometric redshift of a single galaxy . * in a first approximation , we can therefore assume that the optical finder provides a probability distribution function for @xmath97 which can be modelled as a gaussian with negligible bias and variance @xmath288 . in order a peak with @xmath289 not to be incorrectly assigned to a different redshift bin , we can roughly ask that the @xmath290 uncertainty on @xmath97 is smaller than the bin width . for our assumed value @xmath245 , this translates in @xmath291 . such a precision could be likely achieved if @xmath97 is spectroscopically measured ( @xmath292 according to the euclid red book ) , but could be too demanding if one relies on photometric redshift methods ( @xmath293 for euclid ) . in this second case , one should repeat the above fisher matrix analysis adding a non poissonian uncertainty on @xmath246 and resorting to the pull statistics @xcite . whatever is the method adopted for inferring the peak redshift , it is likely that its precision will also depend on the peak @xmath135 ratio so that the net effect could be included in our analysis by convolving the theoretically computed @xmath255 with an empirically determined selection function . investigating this issue is outside our aims here so that we only warn the reader that the numbers in table[tab : fmres ] should be taken as ( likely optimistic ) upper limits . ( green ) , @xmath294 ( blue ) , @xmath295 ( purple).,width=283 ] as an alternative , one could rely on the total number of peaks as only observational constraint . as shown in fig.[fig : ntotepslog ] , this quantity has a significant dependence on @xmath40 so that it can be a valuable help to narrow down the range for this parameter . however , a single quantity is unable to put constraints in a 4d space so that we have not carried out a fisher matrix analysis for this case . we stress , however , that meaningful constraints could be obtained adding background probes such as , e.g. , sneia and grbs . as an example , we can remember that in @xcite , we found @xmath296 at the @xmath297 cl . should one determine @xmath232 to be consistent with the fiducial @xmath233 value within @xmath213 for @xmath298 , one could safely exclude the three models in fig.[fig : ntotepslog ] thus greatly narrowing the confidence range . a joint analysis is , however , needed to investigate the role of degeneracies with other hs model parameters . the unexpected discovery of the cosmic acceleration promptly raised numerous papers proposing different candidates to drive the accelerated expansion . it soon became clear that one can also go beyond general relativity to modified gravity theories . in particular , @xmath0 theories were shown to be able to provide exactly the same background expansion of dark energy models so that the issue nowadays is , more than checking their validity , discriminating between dark energy and modified gravity . here , we have shown that this is in principle possible by relying on the peak statistics , i.e. the number of peaks in the weak lensing maps constructed using the mass aperture statistics . in order to quantify this possibility , we have carried on a fisher matrix analysis to estimate the accuracy on the model parameters which an euclid - like survey can achieve using peak number counts as the only observational constraint . although the results are quite encouraging , it is worth asking whether they can be improved and refined . first , we note that the theoretical quantity we have actually relied on is the mass function . one can repeat our analysis by using different tracers of the mass function itself such as , e.g. , x - ray catalogues . number counts of x - ray selected clusters have indeed been widely investigated as a possible tool for precision cosmology @xcite so that one can naively believe that a joint analysis of the mass function of x - ray and weak lensing selected clusters can improve the constraints on the @xmath0 model parameters . actually , validating such a prediction is not straightforward . on the one hand , since both kinds of observations probe the same quantity it is indeed possible that degeneracies in the model parameter space are not broken so that no significative improvement is achieved . on the other hand , while lensing probes the full matter distribution notwithstanding its dynamical state , x - ray number counts are subject to the uncertainties related to the cluster mass determination which typically relies on scaling relations which have been determined in a general relativity framework . although great care has to be used to take into account this problem , we nevertheless believe that such a task is worth being carried out in order to complement the peak statistics . as a second issue , we remind the reader that we have only investigated a subset of the full 8d parameter space . allowing for a larger number of parameters to be varied impacts the constraints by introducing further degeneracies and hence widening the confidence ranges . in order to compensate for this degradation of the constraining power of the method , one can complement peak statistics with different dataset . from this point of view , sneia and cmbr are ideal tools . indeed , the sneia hubble diagram is a reliable tracer of the expansion rate over approximately the same redshift range covered by the peak number counts ; this offers the possibility to severely constrain those parameters most related to the distance vs redshift relation such as @xmath273 . similarly , the cmbr anisotropy spectrum strongly depends on @xmath299 so that fitting this dataset allows to set these parameters and hence break the @xmath278 and @xmath300 degeneracies shown in fig.[fig : fmplot ] . as a final comment , we would like to stress that cosmic shear tomography can be efficiently added to peak number counts since it is has been shown to be particularly efficient at constraining the hs parameters @xcite . considering that both the shear power spectrum and the peak number counts rely on the same underlying phenomenon ( the lensing distortion of the images of background galaxies from an intervening mass distribution ) and both will be measured from the euclid mission , we end up with the intriguing possibility to use a single mission and a single probe to give a definitive answer to the up to now unsolved dark energy vs modified gravity controversy . vfc warmly thanks m. maturi for his prompt answers on the optimal filter . we also kindly acknowledge the referee for his / her insightful comments which have helped us to improve the paper . vfc is funded by the italian space agency ( asi ) through contract euclid - ic ( i/031/10/0 ) . vfc and rs acknowledge financial contribution from the agreement asi / inaf i/023/12/0 . sc acknowledges support from fct - portugal under grant ptdc / fis/100170/2008 . sc s work is funded by fct - portugal under post - doctoral grant sfrh / bpd/80274/2011 . ad acknowledges partial support from the infn grant pd51 and the prin - miur-2008 grant ` 2008nr3ebk_003 ` `` matter - antimatter asymmetry , dark matter and dark energy in the lhc era '' .

we have measured the absolute proper motion of the fornax dwarf spheroidal galaxy from a combination of photographic plate material and wfpc2 data that provide a time baseline of up to 50 years . the extragalactic reference frame consists of 8 qso images and 48 galaxies . the absolute proper motion is mas yr and mas yr . the corresponding orbit of fornax is polar , with an eccentricity of 0.27 , and a radial period of 4.5 gyr . fornax s current location is near pericenter . the direction of the motion of fornax supports the notion that fornax belongs to the fornax - leoi - leoii - sculptor - sextans stream as hypothesized by lynden - bell ( 1976 , 1982 ) and majewski ( 1994 ) . according to our orbit determination , fornax crossed the magellanic plane myr ago , a time that coincides with the termination of the star - formation process in fornax . we propose that ram - pressure stripping due to the passage of fornax through a gaseous medium denser than the typical intragalactic medium left behind from the lmc may have caused the end of star formation in fornax . the excess , anomalous clouds within the south galactic pole region of the magellanic stream whose origin has long been debated in the literature as constituents of either the magellanic stream or of the extragalactic sculptor group , are found to lie along the orbit of fornax . we speculate that these clouds are stripped material from fornax as the dwarf crossed the magellanic clouds orbit .

astro-ph0405260

the fornax dwarf spheroidal galaxy ( dsph ) is the second most luminous of the ten dsph satellites of the milky way , just after sagittarius dsph ( sgr ) ( e.g. , grebel , gallagher & harbeck 2003 , hereafter ggh03 , majewski et al . 2003 ) . among the well - known milky - way satellites , sagittarius and fornax are the only dsph that have their own system of globular clusters ( hodge 1961 , da costa & armandroff 1995 ) . sgr has at least five globular clusters associated with it ( pal 12 besides the four clusters in the main body of sgr , dinescu et al . 2000 , martnez - delgado et al . 2002 , cohen 2004 ) , and arguably even more scattered along its orbit ( bellazzini et al . 2003 , majewski et al . 2004 ) as the dwarf is in the process of tidal disruption . fornax is known to have five globular clusters . these clusters show a strong second - parameter effect as recently confirmed by mackey & gilmore ( 2003b ) , while this is not the case for the known sagittarius clusters ( see table 4 in dinescu et al . 2001 ) . also , the fornax clusters are all metal poor ( @xmath5 [ fe / h ] @xmath6 ) and have rather similar masses , while those of sgr span a large range in metallicity ( @xmath5 [ fe / h ] @xmath7 ) and mass ( see strader et al . 2003 , table 7 in gilmore & mackey 2003a , table 4 in dinescu et al . 2001 ) . fornax itself has a wide metallicity range ( @xmath5 [ fe / h ] @xmath8 ) with a mean of -0.9 ( e.g. , pont et al . 2004 ) , a characteristic shared only by sgr among the milky way dsph satellites . fornax s population consists of old , metal - poor stars , a majority of intermediate - age stars ( @xmath9 gyr ) , and young stars formed between 2 gyr and 200 myr ago ( pont et al . 2004 , saviane et al . 2000 , stetson , hesser & smecker - hane 1998 ) . one might expect h i to be detected in fornax , given its recent star formation . however , young ( 1999 ) found no evidence of h i at the galaxy center at a column density detection limit of @xmath10 @xmath11 , nor at a limit of @xmath12 @xmath11 at one core radius ( i.e. , @xmath13 m@xmath14 , ggh03 ) . if star formation in fornax has ceased only recently , then it is important to determine the orbit of the satellite and understand whether its interaction with the galaxy may have been responsible for the abrupt cessation . a much - favored mechanism for removing gas from a dsph is ram - pressure stripping as the satellite moves through the intragalactic medium ( e.g. , ggh03 ) . it has been suggested that the star formation histories of satellite galaxies may be correlated to their orbits and perigalactic passages ( e.g. , ggh03 and references therein , mayer et al . the complex star formation history of fornax makes it an interesting subject from this standpoint . lynden - bell ( 1976 , 1982 ) , kunkel & demers ( 1976 ) and kunkel ( 1979 ) pointed out that the galaxy s dsphs , as well as a number of its globular clusters , seem to collect along distinct planar alignments and proposed that these alignments reflect distinct _ orbital _ planes . one of these orbital planes is the magellanic plane which consists of the magellanic clouds , the h i gas trailing the magellanic clouds ( also known as the magellanic stream , ms hereafter ) , ursa minor , draco , and carina . the second plane includes fornax , leo i , leo ii and sculptor . majewski ( 1994 ) reinvestigated this issue and added sextans and perhaps phoenix to the fornax - leo - sculptor alignment , which is now known as the fl@xmath15s@xmath15 plane . to explain the alignments , a common origin via the disruption of larger satellites orbiting in these planes is usually invoked , with the smaller dsphs forming as tidal condensations or broken - off fragments during the dynamical interaction between our galaxy and a massive satellite ( kroupa 1997 ) , much like the tidal dwarfs seen in extragalactic contexts ( e.g. , knierman et al . 2003 ) . the parent satellite has been proposed to be the large magellanic cloud ( lmc ) for the magellanic plane , and fornax for the fl@xmath15s@xmath15 plane ( lynden - bell 1982 ) . a real physical association of these spatial alignments can be tested by ascertaining whether the member objects share a common angular momentum , and this requires determining absolute proper motions of the dsphs . the issues listed above motivate the present study of the proper motion of the fornax system . a collection of thirteen photographic plates from large aperture telescopes together with numerous @xmath0 frames taken in the field of fornax collectively spanning a time baseline of up to 50 years now make possible a deep , high - precision proper - motion study . during the course of our work with these data a preliminary absolute proper motion of fornax using only @xmath0 data was published by piatek et al . ( 2002 , hereafter p02 ) . their earlier study , with data spanning an epoch difference of up to two years , used qso images as references for the determination of the absolute proper motion . our largely independent measurement allows a comparison with this @xmath0-only proper - motion determination . the paper is organized as follows : we describe the observational data and the astrometric reductions in section 2 , and the absolute proper - motion determination in section 3 . in section 4 we discuss the velocity and orbit of fornax in relation with its star - formation history and the fl@xmath15s@xmath15 plane . we summarize our results in section 5 . the absolute proper motion of the fornax dwarf galaxy was determined from the measurement of thirteen photographic plates and from six @xmath0 wfpc2 data sets taken at different pointings in the field of the dwarf galaxy . the various combinations of plate and @xmath0 wfpc2 data coverage lead to epoch differences varying between 20 and 50 years for different areas of the field . the characteristics of the photographic plates are listed in table 1 , while those of the @xmath0 wfpc2 data sets are listed in table 2 . the oldest plates were taken with the hale 5 m reflector ( scale = @xmath16 mm@xmath2 ) in the early fifties . one of these ( ph805b ) includes fornax cluster # 3 ( ngc 1049 ) and the other two include cluster # 4 . six plates were taken with the ctio 4 m reflector ( scale = @xmath17 mm@xmath2 ) in the mid seventies . of these , we used only three plates in the astrometric reduction , namely 276 , 282 and 287 . the remaining three were taken in the blue passband at large hour angles , and a preliminary proper - motion reduction indicated that these plates have very large color terms that could not be entirely modeled and removed from the proper motions . these plates were however used in the calibration of the photographic photometry . another seven plates were taken with the las campanas du pont 2.5 m reflector ( scale @xmath18 mm@xmath2 ) between 1977 and 2001 . because of the large angular sizes ( [email protected] degrees on 20@xmath1920 inch plates ) and long exposure times for most of the latter plates , special care was taken to observe using two guide probes placed on opposite sides of the plate field . with two probes the effects of differential atmospheric refraction could be observed and partially compensated for by slowly rotating the plate during the exposure . the @xmath0 wfpc2 data were taken between 1995 and 2000 for various purposes : studying the fornax clusters ( e.g. , buonanno et al . 1999 ) , determining the absolute proper motion of the fornax dwarf from @xmath0 measurements only ( p02 ) , and as parallel observations . in figure 1 we show the distribution of the observational material in the field of fornax dsph . in this plot , the center of fornax is at ( 0,0 ) , i.e. , at @xmath20 , @xmath21 , j2000 , ( walcher et al . we have also indicated the five globular clusters in fornax , and the eleven qsos detected by tinney , da costa & zinnecker ( 1997 ) , and tinney ( 1999 ) in the background of the fornax galaxy . the outermost square shows the boundary of the input catalog determined from the scan of du pont plate cd0100 . the large du pont plates have plate centers that vary by a few arcmin ( table 1 ) . by using the same input catalog , objects on the edges of the input catalog may have been missed in some of the large du pont plates . at any rate , these objects were not included in the final proper - motion solution . @xmath0 fields , as well as the ctio and hale areas also also indicated . the photographic plates were scanned with the yale pds microdensitometer . first , we have created input catalogs based on the entire digitization of plates cd0100 , 288 , ph805 and ph863 . the area coverage of measurable images is as follows : 1 ) all du pont plates cover [email protected] degrees , except cd3302 , which is a smaller size plate than the rest at an area of [email protected] degrees , 2 ) all ctio plates cover [email protected] degrees each , 3 ) the hale 5 m plate ph805 covers [email protected] degrees , while hale plates ph863 and ph275 cover [email protected] degrees . although the area coverage for plate cd0100 includes all of the plates listed in table 1 , we made full - area scans of one ctio plate and the hale plates , because these are deeper than cd0100 . from these scans , preliminary positions , object diameters , magnitudes and classification were determined with the focas software ( valdes 1982 , 1993 ) . then , based on these input lists , the plates were scanned in a fine - raster , object - by - object mode . for the du pont 2.5 m and hale 5 m plates we have used a pixel size of 12.7 @xmath22 m , while for the ctio 4 m plates we have used a 10 @xmath22 m pixel , because the plate scale is larger for these latter plates . image positions and instrumental magnitudes for each plate were determined using the yale image centering routines ( two - dimensional gaussian fit ; lee & van altena 1983 ) . objects from various input catalogs were then positionally cross - correlated to create a master list . plate solutions that include transformation of coordinates among same - epoch plates give us an estimate of the centering uncertainty per object and plate . thus , well - measured stars on the hale 5 m plates have a centering uncertainty of 3.0 @xmath22 m ( 33 mas ) ; on the ctio plates , 1.4 @xmath22 m ( 26 mas ) ; and on the du pont 2.5 m plates , 2.0 to 3.0 @xmath22 m ( 22 - 33 mas ) . the 2001 du pont plate is of exceptional quality due to the fine - grained emulsion and seeing conditions . we estimate a centering uncertainty of 1.2 @xmath22 m ( 13 mas ) based on a coordinate transformation into wfpc2 coordinates ( see below ) . the entire archived @xmath0 data set in the field of fornax , as retrieved in 2000 , consists of 156 wfpc2 images taken between 1995 and 2000 . the @xmath0 fields were targeted on all fornax clusters , on qso j0240 - 3434a+b ( tinney et al . 1997 ) , and on two parallel data fields . all of the fornax fields had multiple exposures in one or two filters except for one parallel data field . we chose not to use the @xmath0 data on clusters # 1 and # 5 in this analysis , because these clusters lie at the edge of our field ( fig . 1 ) where geometric systematics of the photographic plate material are severe . the wfpc2 images were processed using iraf routines . first , they were corrected for the 34th row effect described in anderson & king ( 1999 ) , and then cleaned for cosmic rays using the mask technique described in yanny et al . one parallel data field that had only one exposure was left uncleaned . for each @xmath0 pointing , all of the frames were aligned and median stacked to create a high signal - to - noise image . object detection was performed on this stacked image using the daofind routine in iraf . the list of objects detected was then used on each individual frame to determine their centroids and magnitudes . the centroids were determined with the help of the psf algorithm in iraf , for which the analytic part is chosen to be an elliptical moffat function with @xmath23 . the positions thus obtained were corrected for distortion using the coefficients determined by anderson & king ( 2003 ) . two wfpc2 data sets partly overlap in the region of fornax cluster 4 , namely the 1995 set and the 1999 set . a coordinate transformation for the common stars in these two data sets yields a positional uncertainty of 0.06 wfc pixels per star , or 6 mas . this value is larger than the 0.02-pixel nominal wfpc2 positional precision obtained in studies that use an empirically - determined psf in rich stellar fields and a wealth of dithered images ( e. g. , anderson & king 2000 ) . however , the positional uncertainty of the @xmath0 data is significantly better than the centering uncertainty on the photographic plates . thus , the random proper - motion uncertainty is dominated by the centering uncertainty on the photographic plates . we have used only wfc data except for the field containing qso j0240 - 3434a+b which was centered on the pc . in the rest of the data sets , there are either too few stars in the pc that can be matched with stars on the photographic plates , or there are too many stars when the pc is placed on a cluster . in the latter case , the photographic images are hopelessly blended , and no reliable cross - identification can be achieved . from all of the @xmath0 data sets , we were able to cross - identify only a few extragalactic objects with the photographic plates . one is qso j0240 - 3434a+b , a lensed quasar with two distinct images separated by @xmath24 arcsec , and four relatively bright galaxies , all located in the parallel data set taken in 2000 ( see table 1 ) . the reduction procedure is similar to that used in dinescu et al . ( 2000 and 2001 , hereafter d00 and d01 ) . since details are described thoroughly there , here we only briefly mention the necessary steps involved in the reduction . all of the photographic - plate coordinates were first corrected for distortion using the coefficients determined by chiu ( 1976 ) for the hale plates , and by cudworth & rees ( 1991 ) for the ctio and the du pont plates . photographic @xmath25 and @xmath26 magnitudes were obtained by calibrating instrumental magnitudes from the du pont and ctio plates in the appropriate passband with johnson @xmath27 magnitudes from the ccd ground - based study of saviane et al . typical magnitude uncertainties are @xmath28 mags . the @xmath27 colors thus obtained for all of the objects are crucial for modeling color - dependent positional systematics known to affect the photographic astrometry . since the du pont 2.5 m plates have the smallest distortion correction among the three sets of plates , we have selected plate cd2677 to serve as a master plate . this plate has very good image quality and is taken close to the meridian , thus minimizing color systematics in the positions due to differential color refraction . the remaining du pont plates were transformed into this master plate using polynomials up to 6th order in each coordinate and , when needed , color terms . the magnitude - dependent systematics were modeled by using fornax stars selected from the color - magnitude diagram ( cmd ) . fornax stars represent a system with a common motion while random galactic field stars do not have this propertry . thus , using fornax stars , one can separate the guiding - induced magnitude systematic errors in the proper motions from the true , secular proper motion ( for details see d00 and references therein ) . since fornax stars span a magnitude range from @xmath29 down to our plate limit , it is only up to this bright limit that the proper motions are assumed to be free of magnitude systematics . a first iteration that included only the du pont plates , in order to minimize systematics , produced preliminary relative proper motions with a time baseline of 24 years . the stars that define the transformation are well - measured and are predominantly fornax stars . these preliminary proper motions were used to update object positions for each ctio and hale plates which were subsequently introduced into the proper - motion solution and modeled in the same manner as the du pont plates . a second iteration that included all of the photographic plates was done , and then the @xmath0 wfpc2 data were introduced into the solution . each wfpc2 chip was transformed independently using up to 2nd order terms , and between 10 and 50 reference stars . no magnitude or color - dependent systematics were detected in these transformations . for the photographic plates , between 3000 and 20,000 stars were used in the plate solutions , depending on the area coverage and depth of the plate . for each object , the proper motion is calculated from an unweighted linear least - squares fit of positions as a function of time . the error in the proper motion is given by the scatter about this best - fit line . measurements that differ by more than @xmath30 from the best - fit line are excluded . only objects that have at least three measurements and a 16-year epoch difference were kept , and a relative proper motion and proper - motion uncertainty calculated . for the qsos , the images on each plate were visually inspected , and poor images were discarded from the fit . the final proper motion catalog includes 38,059 objects . based on the focas classification of objects on the digitized plate cd0100 , as well as from the plot of image peak as a function of image radius , we selected a preliminary list of galaxies . these were then visually inspected on plate cd0100 to confirm their classification as galaxies rather than blended stars . the final list of galaxies covering the entire area included 315 objects . of these , only 268 galaxies have measurements that allow a proper - motion determination in our catalog . similarly , from the initial list of 12 qso images , only 11 have a proper - motion determination . the qso that failed to have a proper - motion determination is j0242 - 3424 ( tinney 1999 ) , which is the easternmost quasar in fig . 1 . in figure 2 we show the distribution of the proper - motion uncertainty as a function of magnitude for stars , qsos and galaxies , in each coordinate . the median value for the proper - motion uncertainty per star , for well - measured stars ( i.e. , @xmath29 to 20 ) , is between 0.8 and 1 mas yr@xmath2 . galaxies have a median value of the proper - motion uncertainty between 1.3 and 3 mas yr@xmath2 in the range @xmath31 to 20 . the well - measured qsos ( i.e. @xmath32 to 20.5 ) have between 1 and 2 mas yr@xmath2 proper - motion error . in figure 3 we show the relative proper motions for all objects in the catalog as a function of magnitude , color and x , y coordinates . the top panels show the entire magnitude range , while the rest of the panels include objects with magnitudes between @xmath29 and 20.5 . the x coordinate is aligned with the right ascension with x increasing eastward , and the y coordinate is aligned with declination with y increasing toward north , at a scale of @xmath18/mm . therefore , for this small area , @xmath33 is practically identical to @xmath34 , and @xmath35 to @xmath36 . there are no significant trends in the proper motions with either magnitude or color . however , proper motions as a function of position on the plate show that geometric systematics are left in the proper motions determined from the global plate solution . this is because of the limited ability to correct for distortion due to the uncertainty in the position of the distortion center and the uncertainty in the distortion coefficients . to correct for this , we assume that , locally , all proper motions are systematically affected in the same way ( see d00 , d01 ) . thus , for each extragalactic object we define a local reference system comprised of 30 stars . the number of stars to be used in the local solution is determined such that the average radius of the local system is smaller than the size over which systematics become important . in this case the average radius of the local reference systems is 13 mm . we have also performed a solution based on 15 stars in the local reference system , and we found that this solution is indistinguishable from the 30-star solution . the local reference stars are selected from the cmd to belong to the fornax dwarf . they are well - measured stars , i.e. , the proper - motion uncertainty in each coordinate is @xmath37 mas yr@xmath2 , with at least six measurements , and they lie in the magnitude range @xmath29 to 21 . the proper motion of each extragalactic object obtained from the global solution is corrected by subtracting the average proper motion of the local reference system . this average is defined by the median value of the proper motions of the 30 stars . in figure 4 we show the proper motions of galaxies and qsos determined after the local correction has been applied , as a function of @xmath26 magnitude and @xmath38 color . no significant trends are detectable in these plots . the absolute proper motion of the fornax dwarf is determined from a selected sample of the extragalactic objects measured , as follows . from the qsos we keep only those with proper - motion uncertainty @xmath37 mas yr@xmath2 ( see fig . 2 ) . we also discard object j0240 - 3437 from tinney s ( 1999 ) list , since this is a large , bright ( @xmath39 ) galaxy . we are thus left with eight images of seven qsos . from the list of galaxies we keep only objects with @xmath31 to 20 , with proper - motion uncertainty @xmath40 3 mas yr@xmath2 in each coordinate , and that are spatially located within the denser regions of the fornax dwarf , such that the local reference system is appropriately comprised of fornax stars . the spatial extent of the galaxies selected matches closely that of the qsos as seen in figure 5 . the proper - motion error limit for galaxies essentially assures that only well - centered galaxies are used . well - centered galaxies are those galaxies that have image shapes more closely resembling those of stars . within this subsample of 54 galaxies , 6 have proper motions that are larger than 5 mas yr@xmath2 in one of the components . these are considered outliers and are discarded from the final absolute proper - motion estimate ; we are thus left with 48 galaxies . as a check that our local solution represents an improvement over the global solution we calculate the proper - motion dispersion for the 48 selected galaxies as given by the global solution and by the local solution . we obtain for the global solution , @xmath41 mas yr@xmath2 and , @xmath42 mas yr@xmath2 , while from the local solution we obtain @xmath43 mas yr@xmath2 and , @xmath44 mas yr@xmath2 . similarly , for the 8 qso images we obtain from the global solution @xmath45 mas yr@xmath2 and , @xmath46 mas yr@xmath2 , while from the local solution we obtain @xmath47 mas yr@xmath2 and , @xmath48 mas yr@xmath2 . in both cases the proper - motion dispersion decreases from the global to the local solution , an indication that geometric systematics left in the global - solution proper motions were lessened from the local - solution proper motions . we determine a weighted mean average for the absolute proper motion of the fornax dwarf from the selected qsos and galaxies . the weights are given by the square of the proper - motion uncertainty of each extragalactic object . the absolute proper motion of the fornax dwarf galaxy with respect to qsos is @xmath49 mas yr@xmath2 , and @xmath50 mas yr@xmath2 , and with respect to galaxies is @xmath51 mas yr@xmath2 , and @xmath52 mas yr@xmath2 . the two determinations differ by 1.2@xmath53 of their formal uncertainties . in figure 6 we show the proper - motion distribution of the selected galaxies and qsos , and their corresponding averages . for the lensed quasar j0240 - 3434a+b that has two images within @xmath54 of each other , we obtain the following results : 1 ) image a has @xmath55 and @xmath56 mas yr@xmath2 , with @xmath57 and @xmath58 , and 2 ) image b has @xmath59 and @xmath60 mas yr@xmath2 with @xmath61 and @xmath62 . while the proper motions agree within their formal uncertainties , we note that the uncertainty in @xmath36 for the b component is rather large , but not unusual for this magnitude ( see fig . the two qso images have rather different magnitudes and colors , and any residual systematics with these two quantities will be reflected in their proper motions . the final absolute proper motion of fornax is taken to be the error - weighted mean of the qso and galaxy determinations . this is : @xmath63 mas yr@xmath2 and @xmath3 mas yr@xmath2 . the recent absolute proper - motion determination of the fornax dwarf made by p02 , which is based on @xmath0 wfpc2 and stis data alone and uses 4 qso images ( i.e. , the lensed qso in one pc field , and two qsos in two stis fields ) , gives @xmath64 mas yr@xmath2 and @xmath65 mas yr@xmath2 . while the proper motion along right ascension agrees with our value , that along declination is different from ours at the 2-@xmath53 level . left - over systematics are always invoked to explain the discrepancies between different proper - motion determinations . to the best of our abilities we have studied and hopefully eliminated proper - motions systematics with magnitude , colors and positions to a level below our formal uncertainty , as the comparison of our results for galaxies and qsos indicate . the p02 result does agree within uncertainties with our qso - based solution . however , we note that the qsos have colors significantly bluer than the majority of the fornax stars used here and in the p02 study . these stars are basically red giants , and the color range of galaxies better overlaps with that of fornax stars ( see figs . 3 and 4 ) . therefore we regard the result based on galaxies as extremely helpful in gauging residual systematics . the qsos in the p02 study are the brightest ( see their fig . 9 ) and bluest objects on the @xmath0 frames , and while trends with color and magnitude are not necessarily expected , these were not investigated in the p02 study . another possible source of error in the p02 result is the distortion correction . for the pc data , they have used the older distortion coefficients ( baggett et al . 2002 ) rather than the improved , more recent ones determined by anderson & king ( 2003 ) in the densely populated field of @xmath66 centauri . the fact that the orientation of the p02 frames was chosen to be similar at different epochs would alleviate this problem to some extent . there are however indications that the distortion is time - dependent as a result of the change in the scale of the telescope known as the orbital `` breathing '' of the optical telesecope assembly ( see anderson & king 2003 ) . we calculate the velocity components of fornax assuming that the distance to fornax is @xmath67 kpc , and the heliocentric radial velocity is @xmath68 km s@xmath2 ( mateo 1998 ) . we have adopted a standard solar motion of @xmath69 km s@xmath2 ( dehnen & binney 1998 ) with respect to the local standard of rest ( lsr ) . here the @xmath70 component is positive outward from the galactic center , @xmath26 is positive toward galactic rotation , and @xmath71 is positive toward the north galactic pole . the adopted rotation velocity of the lsr is @xmath72 km s@xmath2 , and the solar circle radius is 8.0 kpc . the uncertainties in the derived fornax velocity components are determined based on the uncertainties in the distance , radial velocity and proper motions . we also present the proper motion in the galactic rest frame ( i.e. , with solar and lsr motion subtracted ) , in both equatorial and galactic coordinates . these proper motions and the velocity components are listed in table 3 . the first row for each object lists the proper motion and velocity components for that object , while the second row lists the 1-@xmath53 uncertainty in our measurement of these quantities . the velocity components are in a cylindrical coordinate system , where @xmath73 is positive outward from the galactic center , @xmath74 is positive in the direction of galactic rotation , and @xmath71 is positive toward the north galactic pole . we also list the radial and tangential velocity components , where the radial component is along the direction from the galactic center to the object . along with these , we derive here the same quantities for the fornax proper - motion measurement of p02 , and for the sculptor proper - motion measurement of schweitzer et al . ( 1995 ) . in figure 7 we show density contour plots of fornax as derived from the star counts of the raster scan of plate cd0100 . the fornax globular clusters are also marked . galactic rest frame proper motions are indicated as follows . our proper - motion determination is represented by the shorter arrow ( points ene ) , and its uncertainty is defined by the shaded circle . the longer arrow to the se represents the proper motion as determined by p02 . our proper - motion value indicates a direction of motion closer to the direction of the major axis of the dwarf , while that of p02 is practically perpendicular to the major axis . also , the size of our proper motion is smaller than that of p02 . the position angle of our proper motion is @xmath75 , while that of p02 is @xmath76 . the sculptor dsph is located at a position angle of @xmath77 from fornax . the direction toward sculptor is represented with a gray line in fig . thus , our proper - motion determination indicates that fornax is moving , within errors , along the direction on the sky defined by fornax and sculptor , and away from sculptor ( fig . the position angle of sculptor s proper motion is @xmath78 ( schweitzer et al . 1995 ) , while fornax lies at a position angle of @xmath79 with respect to sculptor . therefore sculptor s direction of motion is some @xmath80 away from the fl@xmath15s@xmath15 plane . we calculate orbital elements by integrating orbits back in time over a period of 20 gyr and using the galactic potential model from johnston , spergel & hernquist ( 1995 ) . the integration time is chosen such that a few orbits are completed , and the orbital elements are averaged over the number of orbits . the uncertainties in the orbital elements are determined from 300 integrations that have initial conditions determined randomly from the uncertainties in the proper motions , radial velocity and distance ( see dinescu , girard & van altena 1999 , dinescu et al . in table 4 we list the orbital parameters for the current fornax absolute proper - motion determination and for that of p02 , and for the sculptor dwarf . the total orbital energy should be regarded as being on a relative scale for comparison between measurements and should not be used as a bound versus unbound criterion for fornax , for example . the orbital inclination is calculated as @xmath81 sin@xmath82 . our proper - motion determination implies a low - eccentricity , polar orbit for the fornax dwarf , with the current location at pericenter . similar integrals of motion for two or more objects indicate membership to the same stream . our measurement for fornax gives @xmath83 , @xmath84 and @xmath85 that are larger than those of sculptor . however , the uncertainties in these quantities do not allow us to confidently rule out the membership of sculptor to a stream that includes fornax , or the proposed fl@xmath15s@xmath15 stream . our estimate of the apocentric radius of fornax is close to the galactocentric radii of leo i ( 250 kpc ) and leo ii ( 205 kpc ) ( mateo 1998 ) , thus reinforcing the hypothesis of the common origin of the fl@xmath15s@xmath15 stream . the sextans dsph with a current galactocentric distance of 86 kpc ( mateo 1998 ) , may qualify as a member of the fl@xmath15s@xmath15 stream . while its location on the sky appears farther away from the orbital plane of fornax ( see figure 9 below ) than those of leo i and leo ii , an overlap of the orbits is within the uncertainties associated with the fornax orbit . the phoenix dsph is located at @xmath86 kpc ( mateo 1998 ) from the galactic center , about twice the apocenter distance of fornax s orbit , and is therefore a rather unlikely member of any daughter stream of the fornax dsph , unless substantial orbital decay occurred for the latter . in contrast to these results , the p02 proper motion measurement for fornax gives a rather eccentric orbit for fornax , with the dsph more likely bound to the local group than to our galaxy . specifically , p02 argue against the membership of fornax to the fl@xmath15s@xmath15 stream . the fornax dsph has a population that consists of predominantly intermediate - age ( @xmath87 gyr ) stars , besides the old , metal - poor stars that are more traditional in dwarf spheroidals ( ggh03 and references therein ) . in addition to these stars , fornax has relatively young stars with ages between 2 gyr and 200 myr ( pont et al . 2004 and references therein ) . stars in fornax have a wide metallicity range , between [ fe / h ] = -2.0 and -0.4 as recently demonstrated by pont et al . more importantly , pont et al . ( 2004 ) determine an age - metallicity relationship which implies that fornax underwent an initial enrichment process up to [ fe / h ] = -1.0 about 3 gyr ago , after which continued star formation that lasted until some 200 myr ago increased the metallicity to [ fe / h ] @xmath88 . the inferred star - formation rate increased substantially over the last 4 gyr ( table 5 in pont et al 2004 ) . given this star - formation history , fornax is expected to have some gas associated with it . however , young ( 1999 ) detected no h i gas within one core radius of the galaxy . specifically , there were no detections at the column density detection limits of @xmath10 @xmath11 at the galaxy center and @xmath12 @xmath11 at one core radius . young ( 1999 ) mentions that there may be undetected h i between the core radius and the tidal radius of fornax . however , if the lack of gas is due to its removal by ram - pressure stripping as fornax moves through a homogeneous intragalactic medium , then this gas is more easily stripped from the least dense regions of the satellite , and these regions are more likely the outskirts rather than the center of the dsph ( ggh03 ) . also , the bright , blue stars in fornax associated with young , main sequence stars ( stetson et al . 1998 ) are located in the central region of the dsph , where there was no h i detection . thus , as pont et al . ( 2004 ) suggest , it seems that fornax is now at a particular moment in its history , just after its `` gas death '' . ggh03 explored whether ram - pressure stripping as fornax moves through a uniform , gaseous halo can explain the lack of gas in fornax . by using the velocity determined by p02 and adopting a gaseous halo density of @xmath89 @xmath90 , ggh03 estimate an upper limit for the fornax gas density of @xmath91 @xmath90 where ram - pressure stripping is efficient . the fornax gas density obtained from this simple , pressure - balance estimation is not that of an environment where stars could have formed recently , as is seen in fornax . therefore the lack of gas in fornax has to be explained by a more powerful gas - removal process than the simple one just described . our lower value of the velocity of fornax compared to that of p02 ( table 3 ) would make even less efficient the ram - pressure stripping caused by its motion through a uniform , low - density gaseous halo . the upper limit for the fornax gas density stated above corresponds to a column density of @xmath92 @xmath11 , if one assumes a uniform density distribution within one core radius . in this estimation , we have used a core radius of @xmath93 ( walcher et al . 2003 ) , which corresponds to 0.6 kpc . from our orbit determination , there may be another plausible explanation for the lack of gas in a system that was capable of producing stars some 200 myr ago . fornax crossed the magellanic plane relatively recently . using the velocity and distance data from van der marel et al . ( 2002 ) for the lmc , we have derived its orbit , and therefore the orientation of the magellanic plane . according to our orbit calculations , fornax crossed the magellanic plane about @xmath94 myr ago , a recent event , considering the 4.5-gyr orbital period of fornax ( table 4 ) . at this crossing , fornax was located @xmath95 kpc from the galactic center , specifically at ( x , y , z ) = ( 5 , -25 , -145 ) kpc ( the sun is at ( 8.0 , 0.0 0.0 ) kpc ) . in figure 8 we show the orbits of the lmc ( dark line ) and of fornax ( gray line ) in the plane of the galactic disk , and the x - z plane perpendicular to the galactic disk . the orbits represent 10-gyr integrations backward in time . the current positions of the lmc and fornax are indicated with filled symbols . we have also marked 200 and 500 myr ago on the orbit of fornax with cross symbols . while in the x - z plot of the orbits , the lmc orbit does not quite reach that of fornax , moderate uncertainties in the orbits due to proper - motion uncertainties easily allow for their actual crossing . it is conceivable that gas from the lmc trails along its orbit beyond what is now known as the ms , a 100@xmath96 tidal filament of neutral hydrogen . the recent n - body / sph simulations of the lmc - galaxy interaction by mastropietro et al . ( 2004 ) show that a large amount of gas ( @xmath97 m@xmath14 ) is lost along the orbit . these simulations include both tidal and ram - pressure stripping . contrary to previous modeling of this interaction , mastropietro et al . ( 2004 ) find that the amount of stripped stars is negligible compared to the amount of stripped gas . the predictions are that the ms forms a great circle on the sky . their figure 2 shows that the lost gas is distributed in the plane of the orbit , and out to distances above and below the galactic plane of @xmath98 kpc . in fact , braun & thilker ( 2004 ) report the detection of a diffuse northern ( @xmath99 ) extension of the ms , that has a column density of @xmath100 @xmath11 . according to the mastropietro et al . ( 2004 ) models , this detection corresponds to the distant , above - the - galactic - plane part of the lmc orbit . thus fornax may have undergone recent , efficient ram - pressure stripping by passing through the denser , inhomogeneous environment of the gas in the magellanic plane , along the orbit of the lmc . whether gas from the lmc in this part of its orbit ( at z = -140 kpc ) has the appropriate density to cause efficient ram - pressure stripping in fornax , remains to be determined from detailed models of the clouds interaction with the galaxy . nevertheless , based on our calculated orbits , the crossing of the magellanic plane took place at a time that coincides with the time when star formation ceased in fornax . to illustrate the geometry , in figure 9 we show the projection on the sky of the entire ( i.e. , one period ) orbit of fornax , and the most recent gyr of the orbit of the lmc . the dark part of fornax s orbit represents the most recent gyr . the current locations of fornax , the lmc , sculptor , leo i , leo ii , sextans and phoenix are indicated with filled circles . the ms is indicated as derived from the h i column density data kindly provided to us by mary putman ( putman et al . 2003 , hereafter p03 ) . in this plot , we highlight the density contour corresponding to n@xmath101 @xmath11 . besides mapping out the large - scale ms structure , p03 also analyze the spatial and velocity distributions of h i clouds in the ms and in the direction of the sculptor group , as cataloged in a previous paper ( putman et al . the sculptor group is an association of galaxies at the south galactic pole ( sgp ) , and at distances ranging from 2 to 4 mpc . roughly half of this group falls on the same line of sight as a part of the ms , namely that section in the vicinity of the sgp . p03 ( see also references therein ) found that there are more clouds in this region than in any other part of the ms . they also found that the distributions of velocity and orientation of the elongated clouds with respect to the ms do not match those of the clouds in the bulk of the ms ( their figures 11 and 14 ) . in addition , they show that the velocity distribution of the clouds in this region of the ms does not match that of the sculptor group galaxies either . unlike distant , extragalactic clouds , these anomalous clouds show diffuse connections between one another , and a likely velocity gradient across the ms ( see their figures 7 and 13 ) . p03 conclude that these `` excess '' clouds ( when compared to the ms cloud distribution ) are more likely associated with some halo material rather than the distant sculptor group . p03 further note that the origin of the clouds may be linked to crossing tidal streams from galactic satellites with polar orbits . this is exactly the type of orbit that we have calculated from our proper - motion measurement of fornax dsph . in figure 10 ( a ) ( left panel ) we show this region of the sky in a gnomonic projection with the sgp at the center of the projection . the orbits of fornax and the lmc are shown . the sculptor , phoenix and fornax dsphs are indicated , and the h i data from p03 are shown , with the n@xmath101 @xmath11 contour highlighted . this panel is to be compared with the density map in p03 ( their figs . 5 and 12 ) , and with the velocity map in p03 ( their fig . 7 ) . a reference feature common to all these figures that can be used to guide the eye , is the bifurcation of the ms at @xmath102 , and @xmath103 . figure 10(b ) is a similar projection in which we show the h i clouds cataloged by putman et al . ( 2002 ) as individual sources . from their catalog we have excluded sources associated with galaxies . each cloud is represented with a line segment that indicates the orientation and size ( major axis ) of the cloud . only elongated clouds ( i.e. , minor / major axis @xmath40 0.7 ) that are small ( i.e. , semi - major axis @xmath104 ) are included in this plot . the putman et al . ( 2002 ) catalog is comprised of clouds with local - standard - of - rest velocities larger than 80 km s@xmath2 in absolute value , a selection that aims to eliminate features associated with the galactic disk h i emission . from fig . 10 ( b ) , one can see that there is an increase in the number of clouds in the sgp region of the ms , specifically near the location on the sky of the sculptor dsph and between @xmath105 to @xmath106 , and @xmath107 . these excess clouds , already mentioned in p03 , are located along the newly determined orbit of fornax . moreover , the orientation of the elongation of these excess clouds is aligned with the orbit of fornax , while the bulk of the clouds along the ms , are aligned with the long axis of the ms . the orientation of the elongation of these two groups of clouds with respect to the long axis of the ms can also be seen in figure 8 of p03 . in addition to the orientation of the clouds , figure 8 of p03 also shows the head - tail structure of the clouds , i.e. , a dense core with diffuse tail structure . p03 note that often the clouds aligned with the ms have the tails point away from the magellanic clouds . from figure 8 of p03 , the clouds in the sgp / sculptor - group region of the ms that have an orientation almost perpendicular to the long axis of the ms have the tails point away from the ms , and , according to our figure 10 ( b ) , away from fornax . therefore the orientation of the elongation and of the head - tail structure of the clouds is similar for these two groups of clouds , provided that for one group the source where they originated is the lmc , and for the other group the source is the fornax dsph . from our orbit calculation , fornax and/or material from fornax has a galactocentric radial velocity ( in the galactic rest frame ) of @xmath108 km s@xmath2 , when it crosses the magellanic plane . this negative , rather low radial velocity indicates that fornax is approaching a turning point in its orbit in this case the pericenter and it is well within the velocity range of the excess clouds discussed in p03 . ( see their fig . 11 , in which the galactic - rest - frame velocities are along the line of sight rather than along the direction to the galactic center , but these directions are very close for objects located at large galactocentric radii . ) therefore we suggest that the excess clouds in the sgp / sculptor - group region of the ms are stripped material from the fornax dsph , as it crossed the magellanic orbit ( fig . according to our orbit , fornax s previous pericenter passage was @xmath109 gyr ago ( tab . 4 ) . since there is a significantly larger intermediate - age ( 4 - 7 gyr , e.g. , ggh03 , tolstoy et al . 2003 ) population than old population in fornax , it is conceivable that the intermediate - age population is the result of the tidal interaction between fornax and our galaxy at the previous pericenter passage of the satellite . between the last pericenter passage and the present time ( which practically coincides with pericenter passage ) , fornax continues to form stars and consumes most of its gas . the recent crossing of the magellanic plane ( and the supposed left - over gas from the lmc ) may have swept away the remaining gas , causing star formation to cease . if fornax is gravitationally bound to the galaxy , then its current velocity is smaller than the escape velocity . by assuming a point - mass galactic gravitational potential , the limit on the escape velocity gives a lower limit for the galactic mass as : @xmath110 where @xmath111 is the galactocentric radius , and @xmath26 is the total velocity of the dsph . by taking @xmath112 kpc , and the total velocity from the values in table 3 we obtain @xmath113 m@xmath14 . the uncertainty in the mass estimate is derived from that in the total velocity . this estimate is in reasonable agreement with recent determinations of the milky way mass from a large sample of halo objects and satellites ( sakamoto , chiba & beers 2003 ) . they find a lower mass limit of @xmath114 m@xmath14 . we have measured the absolute proper motion of fornax dsph from a combination of photographic plates and @xmath0 wfpc2 images with a time baseline of up to 50 years . a total of 8 qso images and 48 galaxies were used in the correction to absolute proper motion . the uncertainty in each proper - motion direction is 0.16 mas yr@xmath2 . this measurement implies a polar orbit , with a low eccentricity , and with the current location of fornax near pericenter . the motion of fornax supports the notion that fornax belongs to the plane defined by leo i , leo ii , sextans , sculptor and fornax . we also note that the orbit of fornax indicates that the dsph crossed the magellanic plane some 200 myr ago , a time that coincides with the termination of the dsph s star - formation process . it is found that the excess , anomalous h i clouds in the sgp / sculptor - group region of the ms ( p03 ) , are located along the orbit of fornax . the orientation of the elongation of these excess clouds seems to better align with the orbit of fornax rather than with that of the lmc . we thus speculate that these clouds were stripped from fornax , as the dsph crossed the magellanic orbit . we acknowledge funding from @xmath0 archive grant 08739 . srm , did and bk were also supported by nsf career award ast-9702521 , nsf grant ast-0307851 , nasa / jpl contract 1228235 , a cottrell scholar award from the research corporation , and a david and lucile packard foundation fellowship to srm . srm thanks eduardo hardy for loan of his early du pont plates of fornax , allen sandage for loan of the 200-inch fornax plates , oscar duhalde and fernando peralta for assistance during srm s du pont photographic observing runs , and david monet for generating early plate scans of the du pont fornax plates . did thanks mary putman for making available the h i density data for the ms . anderson , j. & king , i. r. 1999 , , 111 , 1095 anderson , j. & king , i. r. 2000 , , 112 , 1360 anderson , j. & king , i. r. 2003 , , 115 , 113 baggett , s et al . 2002 , in @xmath0 wfpc2 data handbook , v 4.0 , ed . b. mobasher , baltimore , stsci bellazzini , m. , ibata , r. , ferraro , f. r. , & testa , v. 2003 , , 405 , 577 braun , r. & thilker , d. a. 2004 , , in press buonanno , r. , corsi , c. e. , castellani , m. , marconi , g. , fusi pecci , f. , & zinn , r. 1999 , , 118 , 1671 chiu , l. -t . g. 1976 , , 88 , 803 cohen , j. 2004 , , in press cudworth , k. m. , & rees , r. f. 1991 , , 103 , 470 dehnen , w. & binney , j. j. 1998 , , 298 , 387 da costa , g. s. & armandroff , t. e. 1995 , , 109 , 2533 dinescu , d. i. , girard , t. m. , & van altena , w. f. 1999 , 117 , 1792 dinescu , d. i. , majewski , s. r. , girard , t. m. , & cudworth , k. m. , 2000 , , 120 , 1892 dinescu , d. i. , majewski , s. r. , girard , t. m. , & cudworth , k. m. , 2001 , , 122 , 1916 frinchaboy , p. m. , majewski , s. r. , crane , j. d. , reid , i. n. , rocha - pinto , h. j. , phelps , r. l. , patterson , r. j. , & muoz , r. r. 2004 , , 602 , l21 grebel , e. k. , gallagher , j. s. & harbeck , d. 2003 , , 125 , 1926 kroupa , p. 1997 , new astronomy 2 , 139 kunkel , w. e. 1979 , , 228 , 718 kunkel w. e. & demers , s. 1976 , roy . obs . bull . 182 , 241 hodge , p. 1961 , , 66 , 83 johnston , k. v , spergel , d. n. , & hernquist , l. 1995 , , 451 , 598 knierman , k. a. , gallagher , s. c. , charlton , j. c. , hunsberger , s. d. , whitmore , b. , kundu , a. , hibbard , j. e. , & zaritsky , d. 2003 , 126 , 1227 lee , j. -f . & van altena , w. f. 1983 , , 88 , 1683 lynden - bell , d. 1976 , , 174 , 695 lynden - bell , d. 1982 , observatory , 102 , 202 mackey , a. d. , & gilmore , g. f. 2003a , , 340 , 175 mackey , a. d. , & gilmore , g. f. 2003b , , 345 , 747 majewski , s. r. 1994 , , 431 , l17 majewski , s. r. , skrutskie , m. f. , weinberg , m. d. , & ostheimer , j. c. 2003 , , 599 , 1082 majewski , s. r. , kunkel , w. e. , law , d. r. , patterson , r. j. , polak , a. a. , rocha - pinto , h. j. , crane , j. d. , frinchaboy , p. m. , hummels , c. b. , johnston , k. v. , rhee , j. , skrutskie , m. f. , & weinberg , m. 2004 , , in press martin , n. f. , ibata , r. a. , bellazzini , m. , irwin , m. j. , lewis , g. f. , & dehnen , w. 2004 , , 348 , 12 martnez - delgado , d. , zinn , r. , carrera , r. & gallart , c. 2002 , , 573 , l19 mastropietro , c. , moore , b. , mayer , l. , stadel , j. , & wadsley , j. 2004 , in asp conf . `` satellites and tidal streams '' , la palma 2003 , ed . f. prada , d. martnez - delgado and t. mahoney , in press mateo , m. 1998 , araa , 36 , 435 mayer , l. governato , f. , colpi , m. , moore , b. , quinn , t. , wadsley , j. , stadel , j. , & lake , g. 2001 , , 559 , 754 piatek , s. , pryor , c. , olszewski , e. w. , harris , h. c. , mateo , m. , minniti , d. , monet , d. g. , morrison , h. , & tinney , c. g. 2002 , , 124 , 3198 pont , f. , zinn , r. , gallart , c. , hardy , e. & winnick , r. 2004 , , in press putman , m. e. , de heiji , v. , staveley - smith , l. , braun , r. , freeman , k. c. , gibson , b. k. , burton , w. b. , barnes , d. g. , banks , g. d. , bhathal , r. , de blok , w. j. g. , boyce , p.j . , disney , m. j. , drinkwater , m. j. , ekers , r. d. , henning , p.a . , jerjen , h. , kilborn , v. a. , knezek , p. m. , koribalski , b. , malin , d. f. , marquarding , m. , minchin , r. f. , mould , j. r. , oosterloo , t. , price , r. m. , ryder , s. d. , sadler , e. m. , stewart , i. , stootman , webster , r. l. & wright , a. e. 2002 , , 123 , 873 putman , m. e. , staveley - smith , l. , freeman , k. c. , gibson , b. k. & barnes , d. g. 2003 , , 586 , 170 saviane , i. , held , e. v. , bertelli , g. 2000 , , 355 , 56 schweitzer , a. e. , cudworth , k. m. , majewski , s. r. , & suntzeff , n. b. 1995 , 110 , 2747 stetson p. , b. , hesser , j. e. , & smecker - hane , t. a. 1998 , , 110 , 533 strader , j. , brodie , j. p. , forbes , d. a. , beasley , m. a. , & huchra , j. p. 2003 , , 125 , 1291 tinney , c. g. , da costa , g. s. , & zinnecker , h. 1997 , 285 , 111 tinney , c. g. 1999 , 303 , 565 tolstoy , e. , venn , k. a. , shetrone , m. , primas , f. , hill , v. , kaufer , a. , & szeifert , t. 2003 , , 125 , 707 valdes , f. 1982 , in instrumentation in astronomy iv , s.p.i.e . proceedings , vol . 331 , p. 465 valdes , f. 1993 , focas user s guide , noao document + ( ftp://iraf / noao.edu / iraf / docs / focas / focasguide.ps.z ) van der marel , r. p. , alves , d. r. , hardy , e. & suntzeff , n. b. 2002 , , 124 , 2639 walcher , c. j. , fried , j. w. , burkert , a. , & klessen , r. s. 2003 , , 406 , 847 yanny , b. , guhathakurta , p. , schneider , d. p. , & bahcall , j. n. 1994 , 435 , l59 young , l. m. 1999 , , 117 , 1758 lcrclcc + + + & & & & & + & & & & & & + + + + ph275b & 09.10.50 & 0.15 & 60 & 103a - e gg11+no . 25 & @xmath117 & @xmath118 + ph805b & 04.09.53 & 23.58 & 30 & 103a - d gg11 & @xmath119 & @xmath120 + ph863b & 13.09.53 & 23.15 & 60 & 103a - d corning 3484 & @xmath121 & @xmath122 + + + + 276 & 18.12.74 & 23.06 & 45 & iiia - j gg385 & @xmath123 & @xmath124 + @xmath125 & 18.12.74 & 20.34 & 60 & iiia - j gg385 & @xmath123 & @xmath124 + 282 & 19.12.74 & 21.53 & 70 & iia - d gg495 & @xmath123 & @xmath124 + @xmath126 & 19.12.74 & 20.17 & 50 & iiia - j gg385 & @xmath123 & @xmath124 + 287 & 20.12.74 & 21.80 & 70 & iia - d gg495 & @xmath123 & @xmath124 + @xmath127 & 20.12.74 & 20.30 & 60 & iiia - j gg385 & @xmath123 & @xmath124 + + + + cd0100 & 12.10.77 & 22.43 & 90 & 103a - d w16 & @xmath128 & @xmath129 + cd0103 & 13.10.77 & 22.60 & 60 & 103a - d w16 & @xmath130 & @xmath129 + cd2644 & 16.08.85 & 23.75 & 60 & 103a - o gg385 & @xmath131 & @xmath132 + cd2677 & 20.08.85 & 23.45 & 60 & 103a - o w2c & @xmath131 & @xmath132 + cd3107 & 02.11.94 & 23.33 & 180 & iia - d gg495 & @xmath133 & @xmath134 + cd3110 & 03.11.94 & 0.07 & 225 & iia - d gg495 & @xmath133 & @xmath134 + cd3302 & 23.01.01 & 20.37 & 120 & iiia - f gg495 & @xmath135 & @xmath136 + + lcclcc + + + & & & & + & & & & & + + cluster 4 & 10.03.95 & 1100 & f814w & @xmath137 & @xmath138 + cluster 3[ngc 1049 ] & 04.06.96 & 120 - 700 & f555w / f814w & @xmath139 & @xmath140 + cluster 2 & 05.06.96 & 120 - 700 & f555w / f814w & @xmath141 & @xmath142 + fornax field[parallel data ] & 25.09.98 & 300 & f606w & @xmath143 & @xmath144 + qso[a+b ] & cluster 4 & 10.03.99 & 160 & f606w & @xmath145 & @xmath146 + fornax field[parallel data ] & 08.03.00 & 400 & f606w & @xmath147 & @xmath148 + lrrrrrrrrr + + + & & & & & & & & & + & & & & + + fornax & 0.36 & 0.07 & -0.13 & 0.34 & 197 & 62 & 123 & -23 & 239 + & @xmath149 & 0.16 & 0.16 & 0.16 & 101 & 102 & 44 & 60 & 93 + + fornax ( p02 ) & 0.26 & -0.37 & 0.32 & 0.32 & 152 & -230 & 116 & -38 & 296 + & @xmath150 & 0.13 & 0.13 & 0.13 & 81 & 82 & 36 & 48 & 79 + + sculptor & 0.36 & 0.43 & -0.47 & -0.30 & 60 & 193 & -88 & 96 & 199 + & @xmath151 & 0.25 & 0.22 & 0.25 & 90 & 87 & 12 & 16 & 87 + rrrrrrrrrrr + + + & & & & & & & & & & + & & & & & & & & + + + & & & & & & & & & & + & & & & & & & & & & + + & & & & & & & & & & + & & & & & & & & & & + + & & & & & & & & & & + & & & & & & & & & & +

quantum annealing is explored as a resource for quantum information beyond solution of classical combinatorial problems . envisaged as a generator of robust interferometric probes , we examine a hamiltonian of uniformly - coupled spins subject to a transverse magnetic field . the discrete many - body problem is mapped onto dynamics of a single one - dimensional particle in a continuous potential . this reveals all the qualitative features of the ground state beyond typical mean - field or large classical spin models . it illustrates explicitly a graceful warping from an entangled unimodal to bi - modal ground state in the phase transition region . the transitional ` goldilocks ' probe has a component distribution of width and exhibits characteristics for enhanced phase estimation in a decoherent environment . in the presence of realistic local noise and collective dephasing , we find this probe state asymptotically saturates ultimate precision bounds calculated previously . by reducing the transverse field adiabatically , the goldilocks probe is prepared in advance of the minimum gap bottleneck , allowing the annealing schedule to be terminated ` early ' . adiabatic time complexity of probe preparation is shown to be linear in .

1602.08752

in quantum metrology@xcite we often seek to estimate a continuous time - like parameter associated with unitary evolution . even without a direct hermitian observable for time or phase , one can determine bounds on the mean - squared error of estimated values as a function of @xmath2 , the number of qubits , particles , spins or photons involved in the measurement . the lowest bounds are associated with initializing the instrument in a particular entangled quantum configuration of the @xmath2 qubits , known as a ` probe ' state . without entanglement , the performance can not exceed the precision resulting from sending qubits through the instrument one at a time . large spin and mean - field models used to describe many - body systems typically ignore entanglement altogether . curiously , in a noisy setting the most entangled states do not offer the greatest precision @xcite . in the noiseless case , it has been known for some time that the optimal configuration of the @xmath2 qubits is the noon@xcite or ghz @xcite state . this is an equal superposition of the two extremal eigenstates of the phase - encoding hamiltonian . subsequently , however , we have come to understand that this state offers sub - optimal performance in the presence of realistic noise or decoherence , and recent work has unveiled a new family of optimal probe states for noisy metrology @xcite . unfortunately , this result brings with it the new challenge of generating such probes . the asymptotic analysis that uncovered the optimal states indicates also that , for any large-@xmath2 probes , there will be a precision penalty for those with discontinuities in the distribution of components . ( this is the case with the noon / ghz state . ) . for a spin hamiltonian like @xmath3 associated with phase or frequency estimation , optimal probes typically inhabit the fully - symmetric subspace of largest overall spin @xmath4 . for @xmath5 optimality is achieved by a smooth unimodal distribution of amplitudes , a ground state for a one - dimensional particle trapped between two repulsive coulomb sources@xcite . the optimal distribution width @xmath6 is dependent on the noise strength and is typically wider than @xmath7 , i.e. it is anti - squeezed in the @xmath8-direction . had the optimal probe such a ` square - root ' width , it would be easily produced by rotating an @xmath2 -spin coherent state by @xmath9 around the @xmath10-axis via an optical pulse . this state is a simple product state of the component spins creating ` wider ' optimal probes introduces partial , or ` just the right amount ' of entanglement to the ensemble . in this paper , we explore techniques to generate such quantum probes , balancing sensitivity against robustness . * pseudo - potential * @xmath11 from eqn . that results in the large @xmath2 limit from mapping the quadratic spin hamltonian onto a 1-d particle in a potential well . the evolution from a single well in region iii through a critical region ii to a double - well in region i is apparent as the transverse field is decreased , or equivalently , as the annealing parameter @xmath12 is swept from @xmath13 to @xmath14 . in region iii where the transverse field is strongest , the ground state will be all spins aligned with the field along the x - axis . in the variable @xmath15 , this is a gaussian state of width @xmath16 . gaussian states are the ground state of a quadratic potential . at @xmath17 , however , the pseudo - potential above looks like a semi - circle . there is effectively no disparity because the ground state is very narrow for large n , and has very low probability away from @xmath18 ; within this locale @xmath11 also looks quadratic . the dark blue contour in region ii indicates the quartic form of the potential at the critical point , where the annealing parameter @xmath19 or @xmath20 . for @xmath21 , the right - hand - side vertical sequence shows the ground state of the original spin hamiltonian ( in red ) overlaid by the pseudo - potential ( in blue ) , with potential minima indicated by red dots . for an @xmath22 ensemble , eigenstates @xmath23 with @xmath24 pass through the critical region in the upper four panels ; lower energy states exhibit a _ pitchfork bifurcation _ associated with the phase transition . ( positive amplitudes in red , negative in blue.),width=316 ] bearing in mind the ideal characteristics above , one might start to imagine how such broad , smooth , unimodal probe - state distributions could be engineered . to this end , one of the simplest non - trivial quantum systems that can be investigated is one with an equal @xmath25 coupling between all pairs of qubits in the presence of a transverse field . the field strength increases monotonically with an ` annealing ' parameter @xmath12 : @xmath26 where @xmath27 in this scaled form , @xmath28 ; so @xmath29 corresponds to actual energies . this system exhibits a continuous quantum phase transition , as follows . initializing the system in the ground state of a strong transverse field @xmath30 , all spins are aligned with the @xmath10-axis ( this is the coherent spin state discussed in the introduction ) . then , as the field is gradually attenuated , the parameter @xmath12 decreases to a critical value @xmath19 , at which point the ground state warps _ continuously _ into a qualitatively different bimodal noon - like profile ( exactly a noon state when @xmath31 ) . if the annealing proceeds slowly enough , the spins will remain in the instantaneous ground state at all times ; this is adiabatic passage . how realistic are such annealing dynamics ? quadratic terms like @xmath32 appear frequently in models of two - mode bose - einstein condensates@xcite ( bec ) , describing collisional processes . the two modes may correspond to a single condensate in a double - well potential , or a mixture of atoms in two distinct hyperfine states in a single potential . one of the earliest proposals for generating the spin - spin couplings was introduced in the context of ion traps illuminated by two laser fields@xcite . this hamiltonian is also referred to as the isotropic _ lipkin - meshkov - glick _ ( lmg ) model @xcite , an infininte - range ising model with uniform couplings . the lmg model can provide an effective description of quantum gases with long range interactions@xcite . in terms of metrology , the precision offered by some ising models in a decoherence - free setting was given careful examination recently in ref .. similarities exist with the dynamics of of a single - mode oscillator ( optical field in a cavity ) , coupled adiabatically to a collection of spins or atoms via the dicke hamiltonian @xcite , where the coherence length of the field is much larger than the physical extent of the particle ensemble . the single - mode field introduces an effective ferromagnetic spin - spin coupling . the intensity of light emitted into the dicke super - radiant phase can also be utilized for high - precision thermometry @xcite for probes prepared in close proximity to the critical point . it is not our goal to measure the temperature , evolution time , transverse field or any other ` native ' property of this system . rather , the objective is to utilize the annealing dynamics as a resource for engineering robust high - precision probes for interferometry in noisy environments . the hamiltonian of eqn . has been considered previously for interferometry in works that suggest that it is a good source of squeezing and , as such , should lead to better precision . those prior works@xcite , however , did not describe the dynamics through the critical region , where we believe maximum precision is possible . we now know@xcite that the extent of probe squeezing is not a good quantification of precision in a noisy interferometer ( the most squeezed input states may be some of the most fragile ) . to capture fully the behaviour of this discrete spin system at large @xmath2 throughout the phase transition ( and determine if it has appropriate properties for noisy interferometry ) , we map it onto a continuous particle problem . the idea of mapping a quadratic spin hamiltonian in a transverse field onto a one - dimensional particle in a potential is not a new one ; many examples exist in the literature @xcite . a typical approach involves ` bosonization ' of the spin operators into combinations of @xmath33 and @xmath34 , using either holstein - primakoff @xcite or villain @xcite transformations . then , after identifying quadratures @xmath35 and @xmath36 , an operator differential equation in @xmath10 and @xmath37 is produced that , after some approximations , may resemble a schrdinger equation . ( one may choose to linearize the boson operators about the mean - field direction . ) to understand the behaviour near the critical point , the mapping must remain faithful to the original discrete spin dynamics , both qualitatively and quantitatively ( to leading order , when finite size effects are considered ) . as a caveat , it is admitted that certain subtle phenomena may not be captured , e.g. exponentially small ground - state splitting that occurs for a weak transverse field . this requires precision calculation of small probability ` tails ' deep inside the barrier dividing a double potential well @xcite , e.g. region i of fig.[pseudo ] . luckily , interferometric precision is quantified largely by the bulk probabilities of the probe - state concentrated at the bottom of the potential wells ; any evanescent amplitude in the forbidden region makes an exponentially subordinate contribution . _ ( left plot ) _ * energy levels * @xmath38 for @xmath39 $ ] ( blue curves , left vertical axis ) and ground - state energy gap @xmath40 ( dark red curve , right vertical axes ) for an @xmath21 spin ensemble as the annealing parameter @xmath12 is swept from @xmath14 to @xmath13 through the location of the minimum gap @xmath41 and the critical @xmath42 where the phase transition occurs in the thermodynamic limit , @xmath5 . in regions i and iii , energy levels are uniformly distributed @xmath43 , although in region i neighbouring even and odd numbered levels pair up ; only exponentially small gaps separate them . dashed curve indicates the energy gap in the thermodynamic limit of eqn .. _ ( right plot ) _ * region ii landmarks : * numerical results in the scale - free system are shown for the precision penalty factor ( red curve ) from eqn . , and ground state gap ( blue curve ) near the critical point @xmath44 ( or @xmath45 ) . it is observed that @xmath46 , the point of maximum precision is to the right of the minimum gap @xmath47 , and therefore @xmath48 , also . any annealing schedule from strong to weak transverse field will reach the point of highest precision before it encounters the minimum gap.,width=326 ] using notation @xmath49 for eigenstates of @xmath3 labelled by magnetic quantum numbers @xmath50 one may represent the quantum ground state as a vector of amplitudes @xmath51 in the @xmath52 basis , @xmath53 take the overlap @xmath54 in the eigen - equation as @xmath55 remembering @xmath56 and the definition in terms of ladder operators , @xmath57 where @xmath58 , some book - keeping produces : @xmath59 where @xmath60 is the annealing ` ratio ' . now , assuming @xmath61 one can transform into a continuous variable picture , effectively the reverse technique to solving differential equations numerically by discretizing variables . we assume a small parameter @xmath62 for asymptotic expansions , and introduce a continuous variable @xmath63 $ ] , mapping @xmath64 and @xmath65 . also , assuming features change smoothly on a scale @xmath66 one may define derivatives : @xmath67 having transformed from a difference equation to a differential equation , the eigen - equation becomes a schrdinger equation for a one - dimensional particle of variable mass in a pseudo - potential , as follows : @xmath68 \psi(y ) = \frac{e_0}{\gamma } \psi(y ) \ : , \ ] ] given an inverse mass operator , @xmath69 , and a momentum operator @xmath70 . when solved numerically , the eigenstates of this continuous differential equation map faithfully onto the probability amplitudes for the original quadratic spin problem . see fig.[greenred ] in the appendix . variable - mass schrodinger equations have been tackled analytically previously , e.g. in refs .. written as @xmath71 the kinetic energy operator takes the form of a manifestly hermitian operator . the pseudo - potential is @xmath72 this potential is depicted for @xmath73 , i.e. @xmath5 , in fig.[pseudo ] , taking the form of either a single or double well . for large @xmath2 , the distribution will be concentrated at the bottom of these wells at @xmath74 ( red dots in fig.[pseudo ] ) , and one may make the simplification in the kinetic term : @xmath75 where @xmath74 will be a function of the parameter @xmath76 . as such , when @xmath73 the hamiltonian becomes @xmath77 our hope is that near criticality , the ground state may have properties that make it a promising candidate for noisy interferometry . interestingly , the potential terms quadratic in @xmath15 can be made to vanish at a critical transverse field @xmath42 . for states strongly concentrated near @xmath78 , expand @xmath11 as a taylor series . @xmath79 it is seen that the leading order term of @xmath80 near @xmath81 , ( @xmath19 ) , when @xmath73 . one might expect the quartic ground state also to have a distribution of width scaling greater than @xmath7 and , as such , may be a robust probe in noisy conditions . the width can be checked by employing a symanzik scaling argument . ( a similar approach was used in ref . to recover finite size corrections to the critical exponents at exactly @xmath81 . ) due to the reflection symmetry about @xmath78 the inverse mass @xmath82 has a minimum value @xmath83 there and its first derivatives vanish . again , expand @xmath11 to the fourth power and the schrdinger equation in the vicinity of @xmath78 becomes @xmath84 \psi_n = \tilde{e}_n \psi_n \label{reduced - schrod}\ ] ] where @xmath85 , @xmath86 and @xmath87 . now one may rewrite everything in a scale - free way in terms of a single parameter , ` @xmath88 ' : @xmath89 \phi_n = \epsilon_n ( a ) \phi_n \ ; , \label{scale - free}\ ] ] with scale - free coordinates @xmath90 , @xmath91 , @xmath92 and @xmath93 . at the critical point @xmath94 , and the eigenvalue problem is reduced to that of the pure quartic potential . note that the energy spectrum , including @xmath95 , and the half - width of its ground - state , let s call it @xmath96 , are pure numerical values . scaling back from @xmath95 to @xmath97 indicates that the spectrum is compressed near the critical point , @xmath98 remembering that @xmath99 is a pure number , the energy gap in the original problem is compressed by @xmath100 compared with the strong or very weak transverse field regions i and iii , where it is uniform in @xmath29 , as we shall see in section [ gsregioni ] . the compression of eigenvalues can be inspected in fig.[wavy ] for an @xmath21 ensemble . establishing the true length - scale for @xmath15 involves dividing @xmath101 or @xmath100 , without having to recover any features of the wavefunction explicitly . recall that @xmath102 the width scales as @xmath103 in @xmath104 , or indeed @xmath1 . as we hoped , this partially entangled ` goldilocks ' state at the critical point has greater width than the gaussian separable distribution of width @xmath7 associated with a spin - coherent state ( such as the ground state at @xmath17 ) . from the beginning , our desire has been to prepare a goldilocks probe via quantum annealing we reduce the transverse field adiabatically , keeping the system in the instantaneous ground state at all times . the annealing must proceed especially slowly when the gap between ground and excited states is smallest , avoiding diabatic passage into another eigenstate . it is necessary , therefore , to establish the size and location of the minimum gap during the schedule , as this will be the dominant bottleneck affecting efficient probe preparation . close to the critical point in the thermodynamic limit , @xmath105 . in the scale - free setting , one could potentially treat the @xmath106 term of eqn . perturbatively , @xmath107.it turns out , however , from an exact numerical analysis shown in fig.[wavy ] ( right side ) and fig.[locmin ] ( upper plot ) , that the minimum gap @xmath108 does not correspond to a convergent perturbative regime , as @xmath109 . we focus on the @xmath110 transitions because there is no matrix element between the ground and first excited states ; they have opposite parity . the annealing parameter @xmath41 at the minimum gap may be identified as : @xmath111 again recalling the relation @xmath112 . using the numerical result @xmath113 , the prefactor to the dominant @xmath114 scaling of @xmath115 is approximately @xmath116 . prefactors and scaling for the leading terms are confirmed by comparison with the minimum gap of the original spin problem for different ensemble sizes @xmath117 in fig.[locmin ] ( bottom ) . throughout this paper our goal is to map out this spin system s features in the goldilocks critical region at finite ensemble size @xmath2 . the apparent discontinuities occurring in the thermodynamic limit @xmath118 provides no guidance here , as the goldilocks zone vanishes in this limit . from the shape of the wavefunction @xmath119 at the top of fig.[locmin ] it is apparent that modelling the ensemble approximately as a large coherent spin state , as is done in classical mean - field models , is not valid near the minimum gap ; the distribution is clearly not unimodal here . it is also clearly not accurate to model the state here as a ghz - like bimodal distribution as in region i. a more faithful description , as we have seen , is a ground state of a quartic potential . _ ( upper ) _ * rescaled spectrum : * eigenstates @xmath120 and energy eigenvalues @xmath121 found numerically for the scale - free quartic potential and single parameter @xmath88 . the minimum gap @xmath47 and maximum precision point @xmath46 are indicated . _ ( lower ) _ * minimum gap : * blue points show the difference @xmath122 between the value of the annealing parameter at the minimum gap @xmath41 and the critical point @xmath19 in the original discrete spin hamiltonian . the red curve has function @xmath123 derived in section [ mingapdisc].,width=230 ] ref . proposed annealing all the way from a spin - coherent state in region iii to a ` schrdinger cat ' ( ghz / noon - like ) state in region i. a similar technique was advocated in ref . where atom - atom couplings were generated via interaction with a strong classical driving field in a thermal cavity to produce multi - atom states such as the ghz state . at @xmath124 the ground state is exactly a ghz state and , as the field is ramped back up , its two delta components broaden into symmetrized ( @xmath125 ) or anti - symmetrized ( @xmath126 ) pairs of gaussian lobes , fig.[pseudo ] . close to the well bottoms @xmath127 the potential is predominantly quadratic . one must remember that this is a position - dependent mass problem and that the mass function is well - approximated by its value at @xmath74 . @xmath128 for @xmath129 , where @xmath130 . the mass increases monotonically with decreasing transverse field . to second order @xmath131 close to turning points @xmath132 , where @xmath133 and @xmath134 . overall we have a superposition of twin harmonic oscillators with frequency @xmath135 , minimum gap @xmath136 and ground state energy @xmath137 , indicating the almost degeneracy between the even and odd parity eigenstates , @xmath138 . this approximation at quadratic turning points has been shown robust , even outside the wells extending into much of the forbidden central barrier region @xcite . the width @xmath139 of each gaussian lobe in the @xmath15 variable increases monotonically with the applied field and one may write @xmath140 where coefficient @xmath141 is independent of @xmath2 . the small tunneling probability through the barrier slightly lifts the energy of the anti - symmetric state @xmath126 , but this gap remains exponentially small in @xmath2 . the energy gap to the second and third excited states , also nearly degenerate , is approximately @xmath142 . ( in fact , the sturm - liouville theorem guarantees that there can be _ no _ degenerecies in a one - dimensional system , and that pairs of almost degenerate states are grouped with odd states above the even states see for instance ref .. ) in the strong transverse field of region iii , the pseudo - potential @xmath11 of eqn . is dominated by its quadratic term , @xmath143 and the eigenstates will be approximately those of a harmonic oscillator centered on @xmath78 ; therefore the effective mass remains @xmath144 , and is no longer a function of the applied field for @xmath145 . the schrodinger equation for the strong transverse field is : @xmath146 \psi_n = \left ( \frac{e_n}{\gamma}+\frac{1}{m_2 } \right ) \psi_n\end{aligned}\ ] ] the unnormalized eigenstate is @xmath147 . collecting these results , in the thermodynamic limit ( @xmath5 ) the energy gap of @xmath29 is [ omegat ] @xmath148 we shall see in subsequent sections how these three qualitatively very different ground states : the bimodal distribution in region i ; the broad centrally - weighted goldilocks state in region ii ; and the gaussian state in region iii , compare as interferometric probes . in the appendix , we examine the entanglement present during the annealing . it would seem that the goldilocks state in region ii has some of the right qualitative features for metrology . to quantify the supra - classical precision in e.g. estimation of an interferometric phase @xmath149 , the mean - squared error @xmath150 is lower - bounded by the cramer - rao inequality , @xmath151 where @xmath152 is the number of repetitions of the experiment and @xmath153 is the quantum fisher information ( qfi ) . our objective in quantum metrology is usually to maximize this objective function @xmath153 , which depends on both probe state @xmath154 and the dynamics . the formalism developed in refs . presents the qfi as an exact asymptotic series in powers of @xmath155 or @xmath156 . writing @xmath157 as @xmath158 the qfi for estimation of a phase @xmath149 associated with unitary evolution under @xmath3 in the presence of noise has the form of a generalized ` action ' : @xmath159 excluding cubic and higher powers of @xmath160 , as is valid in the @xmath5 asymptotic limit . the ` noise function ' @xmath161 is responsible for both effective mass and potential in the above action , and is proportional to @xmath2 or @xmath162 it depends on the type and strength of the noise present , see appendix [ decoh - fn ] . from eqn . , it is apparent that , for large ensembles @xmath5 , only those state profiles @xmath163 with smoothly - varying features will be optimal . the term squared in the gradients @xmath158 has a negative sign , penalizing qfi , and therefore precision . consider a combination of classical phase fluctuations of size @xmath164 and local noise @xmath165 . putting @xmath166 ( more details in the appendix ) into eqn . means calculating terms like the second moment @xmath167 , and paying particular attention to the _ penalty _ terms featuring squared gradients : @xmath168 , ( because @xmath169 ) . we might naively propose the ` phase ' state , which has @xmath170 for @xmath171 $ ] as a quantum probe ; it is a completely flat distribution . but it produces a large spurious gradient @xmath172 at the boundary between @xmath173 and @xmath174 remembering the definition of eqn . . also , the role of the probe component variance , or equivalently , the amount of ` squeezing ' is significant ; it is not simply that more is better . optimal probes will have variance dictated by the strength of noise present . for a goldilocks state close to the critical point @xmath175 , let us examine the dominant penalty term , @xmath176 . for eigenstates of @xmath177 it is easy to show @xmath178\rangle = 0 $ ] and calculating this commutator = i \delta$ ] and the inner derivatives : @xmath179 = i \delta \frac{d q(y)}{d y}$ ] and @xmath180 = - i \frac{d r(\hat{p})}{d \hat{p}}$ ] ] provides an expression for @xmath181 . expanding to @xmath182 near @xmath78 and converting to the scale - free variable @xmath8 gives : @xmath183 but we also have , from the schrdinger equation , that @xmath184 $ ] so we can eliminate @xmath185 . identifying @xmath186 using the hellmann - feynman theorem for parameter @xmath88 gives the precision penalty factor ( ppf ) : @xmath187 an exact numerical search reveals @xmath188 . this parameter value @xmath46 minimizes the ppf above at @xmath189 ; this is analogous to how the location @xmath47 of the minimum gap was found earlier , see the right side plot of fig . [ wavy ] . scaled back to the annealing variable @xmath12 , optimal precision occurs at : @xmath190 note that this optimum point on the annealing schedule is not a function of the decoherence strength or type it depends only on the number of qubits ( to leading order ) . the expansion of minimum error is : @xmath191_{\text{ii } } \geq & \ ; \frac{1}{f(\gamma_f ) } \ ; = \kappa^{0 } + \frac{\ : \kappa^{(l)}}{n } \nonumber \\ + & \ ; \frac{2}{3 n^{4/3 } } \left ( \frac{m_2}{2}\right)^{\frac{1}{3}}\bigg ( 2 \epsilon_0(a_f ) - a_f \epsilon'_0(a_f ) \bigg)\end{aligned}\ ] ] ignoring terms @xmath192 and smaller . the leading two terms are independent phase errors from collective phase noise and local noise that has shot - noise scaling @xmath193 . together , they represent the _ ultimate _ upper bound to precision . the next significant term , in @xmath194 , would provide the leading @xmath2 dependence in the absence of local noise . it also dictates how fast the upper bound @xmath195 may be approached ( the more negative the power of @xmath2 in the third term above , the faster the convergence ) . note that both @xmath196 and @xmath165 are absent from this term @xmath197 , its contribution to phase error comes from @xmath181 alone , i.e. the probe shape . the generic @xmath156 scaling of precision for most quantum channels was first derived in ref . ( ) and later , optimal probe shapes were found for lossy interferometry @xcite . * numerics and analytics : * dashed curves show precision @xmath198 for analytic ground states of regions i and iii in the thermodynamic limit ; states discussed in section [ gsregioni ] and precision given in eqn .. here @xmath199 qubit ensembles ( blue , red and green curves ) are subjected to collective dephasing @xmath200 . exact numerical results are unbroken curves of the same colour . first , ground states of the original spin problem are found by direct matrix diagonalization . second , the qfi is calculated by additional diagonalization of the mixed state to which the ground state evolves under decoherence , see eqn . in the appendix . the discrepancy between dashed / unbroken curves in the left figure is due to the absence of goldilocks region ii in the thermodynamic limit . the right plot zooms in on the critical region for larger dephasing @xmath201 and @xmath202 . coloured curves are still numerical results of qfi . the unbroken black curve gives the analytic locus of the predicted qfi maxima @xmath203 for all @xmath2 , using the asymptotic results of eqns . and . the formula for @xmath204 is only valid when the condition @xmath205 is met , as it is in this case.,width=326 ] to contrast , for increasing transverse field @xmath206 the ground - state is an approximately gaussian - distributed wavefunction @xmath207 , becoming eventually a spin - coherent state aligned with the field in the spatial @xmath10-direction , a separable state . for @xmath208 one recovers an exact expression for the lower bound : @xmath209_{\text{iii } } \geq \kappa^{0 } + \left(m_2 \sqrt{3 - 2/\gamma}+\kappa^{(l)}\right)/n \ ; .\ ] ] it would appear that the ultimate upper bound @xmath210 can be approached in the limit @xmath211 ; the associated gaussian wavepacket would , however , have infinite variance . but for @xmath212 and small @xmath213 , there exists a good argument for preparing the probe state as close to @xmath42 as possible . the blow - up in qfi predicted above for region iii at criticality , combined with the more accurate results derived previously for region ii , together promote the vicinity of @xmath42 as optimal for probe preparation . but what about region i ? * annealing : * upper plots show for @xmath21 the overlap @xmath214of the annealed state with the @xmath215 lowest energy states @xmath23 with @xmath216 subject to a linear annealing schedule of total time @xmath217 , so @xmath218 . odd - numbered energy eigenstates are excluded because their overlap with the annealed state is zero at all times . they have odd parity and the annealing hamiltonian respects the even parity of the initial state at @xmath219 . on the left , the annealing schedule is terminated at the critical point @xmath220 , and on the right , it is terminated some way into region i after traversing the minimum gap . when the annealing is halted in the critical region , the instantaneous ground state is still dominant across many orders of annealing time . the lower two plots show the probability @xmath221 to be in the target ground state , again for an annealing cycle halted at @xmath45 and @xmath222 . the horizontal axis is the total annealing time in the linear schedule , in units scaled by the number of qubits : @xmath223 . the blue , green and red curves are for systems of @xmath224 and @xmath225 qubits , respectively . ( in the bottom left plot , the three curves are identical . ) annealing success for a simple linear schedule apparently scales linearly with ensemble size , as discussed in section [ annealtime].,width=307 ] in region i , transitions between the ground state @xmath125 and first excited state @xmath126 are prohibited , due to their opposite parity and because evolution via the time - dependent schrdinger equation is parity - conserving . unfortunately , any noise or decoherence is _ unlikely _ to respect parity , so even at very low temperatures , thermalization occurs to an equal mixture of the two near - degenerate states . such an effectively 2-level ( qubit ) maximally - mixed state will be symmetric under unitary evolution by the interferometric @xmath3 operator , and useless in estimating the associated phase parameter@xcite . let us imagine that the symmetrized superposition of gaussians could be prepared adiabatically . for hybrid noise , the qfi ` action ' integral of eqn . produces : @xmath226_{\text{i } } \geq \kappa^{0 } + \frac{1}{n } \left\{\kappa^{(l ) } ( 1 + y_0 ^ 2 ) + \frac{1}{\sigma^2}\right\ } + o\left ( \frac{1}{n^2}\right ) \nonumber \\ = \kappa^{0 } + \frac{1}{n } \left\{\kappa^{(l ) } \left(2 - \frac{1}{m_\gamma^{2}}\right ) + m_\gamma \sqrt{m_\gamma^2 - 1 } \right\ } + o\left ( \frac{1}{n^2}\right)\end{aligned}\ ] ] approximately valid as long as the wells retain a parabolic shape across the width of the ground - state lobes . notice that , for this noisy scenario , error increases with the width of the central barrier separating the two wells @xmath227 ( shown in fig.[pseudo ] ) and there is a additional precision penalty for narrower gaussian lobes of width @xmath140 ; the opposite behaviour is seen in a noiseless environment , where qfi increases quadratically with the ratio @xmath228 . apparently , wider lobes that are closer together ( less ` cat - like ' ) improve robustness to noise . in the limit @xmath229 , the assumption of smoothly varying ground state amplitudes is no longer valid the state is in reality a noon or ghz state , whose interferometric performance in the presence of noise has been shown elsewhere to scale exponentially badly in ensemble size @xmath2 . for collective dephasing qfi is @xmath230 and in dissipative systems of transmission @xmath231 it is @xmath232 . ( refs . ) . the unbroken curves in fig.[rgb ] show the performance @xmath198 calculated numerically for the original spin system in the presence of interferometric phase noise . these curves asymptote for @xmath5 to give the analytical result in the thermodynamic limit : @xmath233 where @xmath234 is given in eqn . and @xmath235 for @xmath236 and @xmath237 for @xmath238 . the asymptotes ignore the critical region entirely ; recall that it vanishes in @xmath122 near @xmath42 at a rate @xmath239 . comparing @xmath240 with eqn . we see explicitly that the ultimate precision limit @xmath241 is _ only _ asymptotically saturable in region ii ; in the other regions there is an additional contribution to mean squared error of order @xmath156 proportional to the gap @xmath234 . this is a central result of this paper . annealing time can depend on the requirement of adiabaticity whether the system needs to be in the instantaneous ground state at all times . if this can be relaxed , the annealing time can be reduced . roughly speaking , the annealing schedule must progress slowly when the gap between the ground and first excited state is small . the exponential scaling in @xmath2 of the time complexity of certain quantum algorithms can be traced to an exponentially small minimum gap . in the current context , the gaps for @xmath29 in regions i and iii are fixed and independent of @xmath117 or @xmath2 ( derived from the ground state approximation calculated in section [ gsregioni ] ) . that leaves only region ii . choosing the final annealing parameter as @xmath242 and using a prescription from ref . , an optimal annealing time @xmath243 can be calculated as @xmath244 where @xmath245 is the @xmath246-norm of a matrix @xmath247 . for the current hamiltonian @xmath248 for @xmath61 this matrix norm factor is linear in @xmath2 and independent of @xmath12 throughout all regions . as presented in eqns . , irrespective of where the annealing is halted the contributions from @xmath249 in both regions i and iii approaches a constant , leaving only the calculation for region ii . there , @xmath250 , and the gap @xmath251 from eqn . . then we have an annealing time @xmath252 @xmath253 where , in the last step , we have recalled that the goldilocks zone scales @xmath254 . overall time complexity is @xmath255 in all three annealing regions . this estimate could be considered pessimistic as it applies only to adiabatic passage . for a linear annealing schedule , numerical results confirm that , irrespective of where the annealing is halted ( near criticality or all the way to the weak field limit ) , the annealing time for any desired fidelity to the target ground state is linear in @xmath2 , see fig.[anneal - time ] . whether terminating at a ghz - like state or goldilocks probe , the total time differs only by a fixed factor independent of @xmath2 . since the annealing time scales favorably with the size of the ensemble , then decoherence may be less significant during probe preparation . in the scale - free setting of region ii , the pure numerical values of minimum gap and precision penalty factor are predicted to be @xmath256 and @xmath257 , respectively . this latter minimum corresponds to maximum qfi . if instead the maximum qfi is found by brute force numerical diagonalization and eqn . for different ensemble sizes @xmath2 and collective phase noise amplitude @xmath258 , the true penalty factor can be determined including all finite - size corrections . the ground - state gap in the original spin dynamics @xmath259 is also easily converted into that of the scale - free setting . the convergence to the predicted asymptotic values is seen in fig.[ppf ] . from these graphs it is clear that convergence is fairly slow , but by @xmath260 and @xmath261 the numerical values do approach these asymptotic bounds . this observation provides crucial evidence in validating the sequence of approximations that have been made ; mapping from a spin system to a 1-d particle in a potential , restriction to a quartic potential in the critical region , and approximation of the qfi by its two leading terms in the exact asymptotic series . the reason for the slow convergence could be attributed to the first term excluded from the qfi series being only slightly smaller than the last included term , @xmath262 vs @xmath263 . the value of @xmath2 has to become quite large before @xmath263 can dominate . * convergence to asymptotics : * vertical axes plot pure numerical factors expected to converge on @xmath256 ( upper plot ) and @xmath257 ( lower plot ) , a re - arrangement of eqn . and eqn .. the asymptotic predictions are @xmath264 and @xmath265 , respectively for the gap and penalty factor ( blue and red dashed lines ) . since these derive from the scale - free picture , they are independent of both @xmath12 and @xmath2 . , width=268 ] we have examined a network of uniformly - coupled spins in a transverse field as an interferometric probe for use in noisy conditions . mapping the ensemble onto a variable - mass particle in a potential allowed quantitative understanding of the dynamics in the critical region , i.e. we were able to characterize correctly the dominant properties of the continuous phase transition . in terms of annealing parameter @xmath12 , we discovered the ordering and distances between the critical point in the thermodynamic limit @xmath42 , the minimum gap @xmath41 and the point of maximum precision @xmath266 using an exact numerical approach . these latter landmarks are the relevant ones for annealing and metrology , and are sufficiently far from @xmath42 that conventional perturbative techniques would fail . utilizing asymptotic formulae for qfi , we predicted that in a noisy environment , best precision is offered _ only _ by ground states prepared near the critical point . we saw in eqn . that such states asymptotically _ saturate _ the ultimate precision bounds for interferometers subjected to typical noisy environments . we confirmed the accuracy of both the asymptotic qfi expansion and the legitimacy of the continuous particle mapping by brute force matrix diagonalization methods for the original spin system , for @xmath267 $ ] . ( in the asymptotic approach , only the combined parameter @xmath268 is required to be large the product of noise strength and ensemble size . ) we determined that adiabatic probe preparation has a time - complexity scaling linearly with ensemble size . ( in the appendix , the precision of the @xmath2-qubit goldilocks probe is compared numerically with probes prepared by a sudden quench of the transverse field . ) in ref . the calculus of variations dictated that asymptotically the best - performing interferometric state was _ always _ the ground state of a 1d particle in a special pseudo - potential , created between two repulsive coulomb sources and identical to the noise function @xmath269 . in some sense , we have tried to engineer non - linear dynamics that best mimic that optimal potential . although a quartic potential does not much resemble the optimal one , any ground state of width in the variable @xmath15 that narrows with increasing @xmath2 will not ` explore ' the structure of the potential far from @xmath78 ; such a probe can have some of the desired properties in the large @xmath2 limit . decoherence during probe preparation must be strongly suppressed , e.g. fluctuations in the transverse field @xmath76 are amplified in the @xmath88 variable is a gaussian - distributed random variable of width @xmath270 , then @xmath88 is also a random variable , but with a non - gaussian distribution @xmath271 that looks increasingly gaussian for @xmath5 and fluctuations near the mean . the width in @xmath88 is now approximately @xmath272 . remember the interesting range of @xmath273 $ ] so the transverse field noise must be suppressed by a factor @xmath1 if the goldilocks zone is to be located at all . ] by @xmath1 the strength of these fluctuations places an upper limit on the size of the spin ensemble that can be prepared in the critical region . the full effects of decoherence during the annealing schedule we leave to a future publication . the results of this paper promote an alternative perspective on the developing technology that is the quantum annealing machine , e.g. the pioneering work of ref .. typically , the goal of such devices is to prepare a ground state that represents the optimal solution to a combinatorial problem encoded directly in the couplings of an ising hamiltonian . here , it has been proposed that such customizable dynamics might instead be used to prepare some exotic yet useful quantum state of many qubits . perhaps the application is metrology as discussed ; other possibilities include quantum communication , or generation of different types of entanglement@xcite for distributed quantum information . presented in the context of ion traps and optical lattices , the authors of ref . had already recognized the potential of a dicke - ising model of quantum computing for simulation of quantum systems , and as a resource for generating squeezing and entanglement . more recently , the creation of tunable ising systems optically has been proposed in qed cavities@xcite but not yet considered for metrology applications . the inverted challenge in terms of global optimization is to ` reverse engineer ' the ising couplings to prepare a particular known ground state of interest . now , the search objective is the associated hamiltonian couplings and topology . when preparation time is a significant resource , one may have to offset state fidelity against shorter annealing times , if the landscape necessitates annealing through gap regions , or if adiabaticity is not a strict requirement . adding external control fields might avoid proximity to the smallest gaps , and allow adiabatic short cuts@xcite such as transitionless driving@xcite . 10 url # 1`#1 ` urlprefix [ 2]#2 [ 2][]#2 , & . _ _ ( ) . , & . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . . _ _ * * , ( ) . . _ _ * * , ( ) . , & . in _ _ , ( , ) . , & . _ _ ( ) . _ _ * * , ( ) . , , & . _ _ * * , ( ) . & . _ _ * * , ( ) . , & . _ _ * * , ( ) . , , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . & _ _ * * , ( ) . , & . _ _ * * , ( ) . . _ _ * * , ( ) . , , _ _ _ * * , ( ) . , & . _ _ * * , ( ) . & . _ _ * * , ( ) . & . _ _ * * , ( ) . & . _ _ * * , ( ) . _ _ * * , ( ) . . _ _ * * , ( ) . . _ _ * * , ( ) . . _ _ * * , ( ) . , _ _ * * , ( ) . . _ _ * * , ( ) . , & _ _ * * , ( ) . . _ _ * * , ( ) . & . _ _ * * , ( ) . , & . _ _ * * ( ) . & . _ _ * * , ( ) . , & . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . , & in _ _ , ( , ) . & . _ _ * * , ( ) . , , & _ _ * * , ( ) . _ et al . _ . _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . & . _ _ * * , ( ) . , & . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . . _ _ * * , ( ) . , & . _ _ * * , ( ) . & _ _ , vol . , ( , ) . . _ _ * * , ( ) . & . _ _ * * , ( ) . & . _ _ ( ) . & . _ _ * * , ( ) . , & . _ _ * * , ( ) . . _ _ * * , ( ) . solutions to eqn . are shown in fig.[greenred ] for the ground ( red ) and second excited state ( green ) for the variable - mass particle in the one - dimensional potential @xmath11 and different values of the annealing ratio @xmath274 ( remembering the critical point is at @xmath81 ) . compared the discrete ground state amplitudes @xmath51 for the original quadratic spin hamiltonian , good agreement is obtained . fidelity to the original hamiltonian eigenstates improves with larger ensembles and larger values of @xmath76 . for @xmath275 the spin eigenstates contain discrete delta - like components ( ghz state ) and the continuous approximation is no longer valid . also , the continuous variable solution depends on the boundary conditions ; we have chosen @xmath276 at @xmath277 but the discrete amplitude set @xmath278 can be non - zero at @xmath279 for finite @xmath117 . an additional difference between the models is the discrete number of eigenstates for the spin system in contrast , the particle model has no upper bound to the number of excited states . for the three annealing regions , ( i : weak field ) , ( ii : critical field ) , ( iii : strong field ) , discrete ground state amplitudes and second excited state amplitudes for a system of @xmath280 qubits @xmath281 are indicated by black dots . the equivalent continuous wavefunction eigenstates for the variable - mass particle are shown as unbroken green ( ground state ) and red ( second excited state ) lines.,width=172 ] partial entanglement is necessary for probes to offer supra - classical precision in noisy interferometry . it will be a useful exercise , therefore , to quantify the entanglement present during the annealing process , in particular through the phase transition . to characterize entanglement a useful measure is the _ global geometric entanglement _ @xcite ( gge ) which we can define for a pure entangled state @xmath282 as : @xmath283 = \text{min } \left\ { - \log_2 |\langle \chi| \psi \rangle|^2 \right\ } \ ; , \ ; \ ; \ ; \forall \ ; | \chi \rangle \in s\ ] ] i.e. where @xmath284 belongs to the set of separable states @xmath285 and the minimization is performed over all @xmath285 . the function @xmath286 is the negative logarithm of the fidelity of the entangled state to the nearest separable state . the nearest separable state will in fact be a pure ( product ) state since the ground state is pure . the gge is sensitive to bipartite and multi - partite entanglement , although it does not differentiate between different entanglement depths . for the current dynamics the spin ensemble lives in the maximum spin sector ( @xmath287 ) , and therefore the ground state is fully permutation - symmetric . it is simple to argue that the nearest separable state also shares this permutation symmetry . if part of that state lived in a different @xmath117 sector , an orthogonal subspace , it would only _ reduce _ the overlap @xmath288 . the only fully - symmetric pure separable states are in fact the spin - coherent states @xmath289 . finding the gge can be a difficult optimization in general , but for symmetric states it means finding the optimal @xmath290 angle pair the polar and azimuthal angles of the spin vector giving maximum overlap with @xmath154 . the spin - coherent state@xcite has components : @xmath291 the ground state of the quadratic spin system has real coefficients , as does the closest spin coherent state : @xmath292 . also the probability distribution @xmath293 is binomial , with mean @xmath294 and variance @xmath295 . approximating the binomial distribution as gaussian in the @xmath61 limit and converting to @xmath15 produces a mean @xmath296 and standard deviation @xmath297 . obviously @xmath298 is the upper bound on wavefunction ` width ' for a separable state . in region iii the ground state is centered on @xmath78 and the nearest spin - coherent state will also be centered on the origin , being as wide as possible , i.e. @xmath299 . the squared overlap of two gaussian wavefunctions with the same mean but different variances is @xmath300 . the gge for @xmath301 is then : @xmath302 - 1\ ] ] as we have seen , the ground - state passing into region i during an annealing cycle bifurcates into two approximately gaussian lobes . the nearest spin - coherent state will choose one of those lobes and attempt to match both its mean and variance ( to achieve maximum fidelity ) . interestingly , an almost exact matching for both quantities is possible although the spin coherent state is a function of a single parameter @xmath303 . the bi - modal lobes of the ground state have means @xmath304 and standard deviation @xmath305 . fidelity to the spin - coherent state , with @xmath306 and @xmath307 given above , can be close to unity only if : @xmath308 \ ; , \ ; \ ; 0<\gamma < 2\ ] ] remembering that @xmath309 . this gives an asymptotic result for the entanglement in terms of the variable mass : @xmath310\ ] ] valid in the parameter range , @xmath311 . only half the probability ( with some exponentially small correction ) is concentrated in a single lobe , so the overall fidelity to the ground state will quickly converge on @xmath312 in region i. then @xmath313 , which is the known gge for a ghz state , a state with _ only _ @xmath2-partite entanglement . this analysis also indicates how well - conceived is the model of a ` cat ' state with two superposed spin - coherent states , for this spin hamiltonian and @xmath314 , as presented in ref . . in the scale - free setting of fig.[global - ent ] it is seen that for @xmath21 qubits the maximum global entanglement is at @xmath315 . for larger @xmath2 or @xmath117 this maximum will therefore occur in the region @xmath316 because of the properties of the nearest spin - coherent state . in the @xmath15 variable this state has width @xmath317 but in the scale - free variable @xmath318 , the maximum width becomes @xmath319 . the squared overlap of the widest spin - coherent state with the ground state in region ii ( whose variance is just a pure number in the @xmath8-variable near critical annealing ) is going to scale asymptotically as @xmath320 or @xmath321 . the gge in the limit @xmath5 therefore approaches : @xmath322 which confirms the central result of ref . for the ( isotropic ) lipkin - meshkov - glick model . note that the entanglement per copy vanishes in the thermodynamic limit , as @xmath323 . we should not be too shocked by this scaling law as it has been shown that although in general @xmath324 , the maximum entanglement for _ symmetric _ states is @xmath325 . * global entanglement : * upper plot shows the analytic bounds on global geometric entanglement @xmath286 in the thermodynamic limit ( dashed green curve ) as a function of @xmath12 . the unbroken green curves depict the entanglement for finite sized ensembles up to @xmath22 ( black curve ) . the superimposed red and blue distributions show the amplitudes of the nearest spin - coherent state ( red ) to the ground state of the annealed system ( blue ) in the original spin problem . this spin - coherent state is also the nearest ( or highest fidelity ) separable state . finding this state is crucial , due to @xmath286 being the negative logarithm of the square of the overlap between these two states . when the ground state centered on @xmath78 bifurcates into two lobes as it passes from region iii to region i the nearest spin - coherent state ( having an approximately gaussian distribution ) can only track one of the lobes ; its mean @xmath296 starts to move with decreasing @xmath326 . as @xmath12 approaches @xmath14 the entanglement asymptotes to @xmath327 , the entanglement of a ghz state . the lower graph plots the entanglement more closely in region ii for a system of @xmath21 qubits in the scale - free setting . maximum entanglement is seen to occur in the vicinity of the minimum gap , but this maximum will be closer to @xmath328 or @xmath329 for larger @xmath2 . as @xmath2 increases the maximum width of a spin - coherent state scales as @xmath330 in the @xmath8 variable . thus maximum entanglement grows slowly as @xmath331 . the red curve plots the locus of the mean of the nearest spin - coherent state . apparently , as @xmath88 or @xmath12 decreases the ground state has already begun to bifurcate before maximum entanglement is reached.,width=211 ] consider a phase parameter @xmath149 encoded by a spin hamiltonian , e.g. @xmath3 aligned with the spatial @xmath8 direction , acting on a noisy mixed quantum state @xmath332 , of @xmath2 qubits . assume the noise process commutes with the phase rotation , as is the case e.g. dissipation and for collective dephasing . the mixed state is transformed by @xmath333 and for such finite - dimensional systems the calculation of qfi typically involves diagonalization of the density matrix @xcite . for @xmath334 then defining the qfi as @xmath335 , it is : @xmath336 where @xmath337 . the computation becomes increasingly arduous for @xmath5 without introducing any insight into the result . recently , a different formulation was proposed , useful in the large @xmath2 case , where @xmath335 may be expanded as an exact asymptotic series @xcite : @xmath338 \right)\hat{j}_z\right]\right\rangle \ : .\ ] ] here square brackets denote an operator commutator , @xmath339 = a b - b a$ ] , and angular brackets indicate an expectation value taken with the density matrix , @xmath340 . the adjoint endomorphism @xmath341 = \operatorname{ad}_a$ ] acts as a superoperator . thus @xmath341\hat{j}_z=[a,\hat{j}_z]$ ] , and @xmath341 ^ 3 \hat{j}_z=[a[a,[a,\hat{j}_z]]]$ ] are terms in the power series expansion of the hyperbolic tangent ; @xmath342 with @xmath343 $ ] , and acting on the @xmath3 total spin operator . this series expression provides the leading terms in the formulation of the qfi as an action@xcite in eqn .. for pure collective dephasing , ( background phase fluctuations with variance @xmath196 , possibly due to stochastic path length fluctuations inside the interferometer ) , this becomes a constant @xmath344 within the physical box boundary @xmath345 $ ] and infinite outside the boundary . this collective dephasing is the most significant type of noise for a bose - einstein condensate , existing only in the fully symmetric subspace of its constituent atoms , although losses may also occur as atoms leave the condensate . likewise , along with losses , collective dephasing is a dominant noise source in photonic interferometry . the dephasing process can be seen as a convolution of a pure probe state @xmath346 with a gaussian probability distribution @xmath347 with mean @xmath348 and variance @xmath258 : @xmath349 the density matrix is a mixture of these probes , each evolved by a different phase : @xmath350 . the analysis also requires that @xmath351 . for strong phase noise beyond this limit the stochastic phase distribution can no longer be approximately gaussian and localized within a @xmath352 window ; periodic boundary conditions turn the random phase distribution into a wrapped normal distribution . by then , any probe state is so noisy it becomes almost completely insensitive to phase , and precision begins to decay exponentially fast in @xmath258 ( due to the phase uncertainty relation@xcite ) . there are now two bounds on our analysis . including the requirement that the noise parameter @xmath353 is large , so that the asymptotic series can be truncated : @xmath354 collective dephasing is also important in a bayesian estimation scheme , featuring a prior phase distribution that is updated via measurements . dephasing is entirely equivalent to gaussian - distributed _ prior _ phase uncertainty@xcite @xmath164 . in general , there will always be some prior phase uncertainty ( estimation would be otherwise be unnecessary ) and thus collective phase noise is always present adding local noise @xmath355 while the interferometric phase @xmath149 is being acquired is governed by a markovian master equation : @xmath356 \ ; + \ ; \frac{d \zeta}{d \theta } \ ; \sum_{i=1}^n \mathcal{l}_s^{(i ) } \rho \\ \mathcal{l}_s(\rho ) & = \{\hat{s}^{\dagger } \hat{s},\rho \ } - 2 \hat{s } \rho \hat{s}^{\dagger } \end{aligned}\ ] ] ( the latter expression defines the linbladian sueroperator @xmath357 ) . noise processes are local dephasing @xmath358 , excitation @xmath359 and relaxation @xmath360 defined in terms of individual pauli spin operators . note that @xmath361 . in the current analysis all qubits are permutation - invariant and decohere at the same rate ; there are no topological features to this model . the noise function@xcite then becomes : @xmath362 it was shown in ref . that for local noise only the combination of dephasing , relaxation and excitation included in @xmath165 matters asymptotically , not their individual contributions , @xmath363 in this paper , we considered the hybrid noise function @xmath364 in the general form given above . * odd and even ensembles * : the associated precision of the optimal sudden - quench interferometric probe can be better / worse than the goldilocks state in the odd / even @xmath2 cases respectively . for a probe subsequently subjected to collective dephasing , the upper graph indicates , as @xmath258 is increased , the ratio of quantum errors @xmath365 for the optimal probes generated by annealing and a sudden quench , respectively . below ( above ) the dashed horizontal line the annealed ( quenched ) state produces lower error . again , the different performance of odd and even numbered @xmath2 ensembles is apparent . the lower illustrate some of the general characteristic differences between odd and even sudden - quenched probes . circle - insets show the configuration of optimal amplitudes for a sudden quench and large noise parameter . ( for comparison , the optimal annealed profile , the goldilocks probe with asymptotic width @xmath366 , is overlaid as a dashed red curve . ) the widths of optimal sudden - quench states are plotted as open circles and fitted to a model @xmath367 . the best - fit line is shown in red and blue for even and odd cases . , width=288 ] a sidelight on our own proposal is provided by an earlier scheme@xcite , due originally to kitagawa and ulam - orgikh , using entangled quantum spin states for noisy metrology . ( it was analyzed recently and more comprehensively in terms of fisher information in ref .. ) beginning with the same spin - coherent state aligned with the strong transverse field ; subsequently the field is _ abruptly _ and discontinuously stepped to zero . the ensemble then evolves diabatically for some time under the influence of its @xmath368 couplings . after a particular elapsed time @xmath369 , the state is rotated by an angle @xmath149 around @xmath370 to produce an optimal probe for noisy interferometry . the propagator is effectively @xmath371 acting on the spin coherent state . here , for a fair comparison with our scheme , we imagine that state - preparation is decoherence - free . at least numerically , this is a straightforward optimization ; even for large ensembles @xmath372 it involves only two bounded parameters @xmath373 $ ] and @xmath374 $ ] . the optimization landscape is complicated with many local minima , but in the limit of large @xmath353 very few coordinate pairs @xmath375 offer supra - classical precision . fig.[landscape ] depicts this landscape and optima probe shapes for @xmath376 and @xmath377 . it is seen that precision resulting from this method can be better or worse than the goldilocks annealed state , depending on whether the ensemble has an odd or even number of constituent spins , see fig.[oddeven ] . ( this may be due to the non - adiabatic unitary nature of the state preparation . ) one might now ask whether it is more feasible to prepare optimal probes by manipulating two control parameters diabatically , or just a single parameter , under the constraint that it be attenuated adiabatically . for both adiabatic and diabatic schemes the ` sweet - spot ' of supra - classical performance in parameter space ( @xmath12 , or @xmath369 and @xmath149 ) decreases dramatically with increasing @xmath2 : note the very reduced blue zone for the large noise case in the second row of fig.[landscape ] . also , these numerics were carried out for collective phase noise , which we know favors broad probes of width @xmath378 . most other noise types favor narrower probe distributions @xmath379 , according to ref .. calculation of qfi for local noise processes is a sum of terms due to state diffusion into other total spin @xmath380 spaces . necessity of numerical diagonalization in each of these additional subspaces dramatically increases computational overhead with increasing @xmath117 , presenting a challenge for follow - up investigation . * global optimization : * for phase noise @xmath381 this figure shows the quantum fisher information @xmath382 for sudden - quench dynamics of @xmath215 qubits , representing the same data of the top middle pane of fig.[landscape ] . this 3d projection better depicts the difficulty of the optimization task , illustrating the landscape s many local maxima . optimal solutions in the space of @xmath369 and @xmath149 were found by a differential evolution algorithm , which is stochastic and does not use gradient information . by comparison , a sequential least squares quadratic programming approach did not always find the global maximum , even with multiple random starts . a basin - hopping technique or particle swarm approach might have proven effective alternatives.,width=355 ] during the course of this study i had several useful discussions with zhang jiang and sergey knysh at nasa ames research center . assistance with literature review was provided by algorithms developed by the team at lateral.io

the state space of loop quantum gravity admits a decomposition into orthogonal subspaces associated to diffeomorphism equivalence classes of spin - network graphs . in this paper i investigate the possibility of obtaining this state space from the quantization of a topological field theory with _ many _ degrees of freedom . the starting point is a 3-manifold with a network of defect - lines . a locally - flat connection on this manifold can have non - trivial holonomy around non - contractible loops . this is in fact the mathematical origin of the aharonov - bohm effect . i quantize this theory using standard field theoretical methods . the functional integral defining the scalar product is shown to reduce to a finite dimensional integral over moduli space . a non - trivial measure given by the faddeev - popov determinant is derived . i argue that the scalar product obtained coincides with the one used in loop quantum gravity . i provide an explicit derivation in the case of a single defect - line , corresponding to a single loop in loop quantum gravity . moreover , i discuss the relation with spin - networks as used in the context of spin foam models .

0907.4388

background independence and diffeomorphism invariance are key assumptions in various approaches to quantum gravity @xcite . compared to older canonical approaches @xcite , what makes loop quantum gravity @xcite technically possible and mathematically robust is a further assumption consisting in the choice of the loop algebra as the basis for quantization @xcite@xcite . the kinematical hilbert space of the theory turns out to admit a decomposition into orthogonal subspaces @xmath0 associated to diffeomorphism equivalence classes of graphs @xmath1 . spin network states with graph @xmath1 @xcite provide a complete orthonormal basis of @xmath0 . the motivation for this paper comes from the following remark : the mathematical structure of the hilbert space @xmath0 and its physical interpretation in terms of quantum geometries @xcite are the ones proper to the hilbert space of a system with a finite ( but possibly large ) number of degrees of freedom . therefore , while the full theory has an infinite number of degrees of freedom ( as classical general relativity does ) , actually each subspace @xmath0 captures only a finite number of them . the question we want to explore in this paper is if the state space @xmath0 , together with its physical interpretation , can be obtained directly from the quantization of a classical system with a finite number of degrees of freedom . a classical system that is generally considered for this purpose is a discrete gauge theory on an abstract graph @xmath1 @xcite . this model captures some important features of loop quantum gravity . however , as the graph @xmath1 is not embedded into a @xmath2-manifold , in this model the role played by diffeomorphisms is rather obscure . moreover , potential problems with locality have been identified @xcite . on the other hand , here we look for a model such that a manifold is present and diffeomorphisms act on local fields in the standard way , via pullback . the classical system we investigate is a field theory : we consider a theory of locally - flat connections on a manifold . the manifold is assumed to be non simply - connected because of the presence of a network of defects . we quantize this system relying only on standard field theoretical methods @xcite . in particular , the quantization procedure we discuss does not involve the basic assumption about loop observables which characterizes loop quantum gravity . nevertheless , the resulting state space is surprisingly close to the one of ordinary loop quantum gravity . spin networks arise again and provide the appropriate tool for describing gauge- and diffeomorphism - invariant functionals of the connection . the role played by defects and loops in this approach is analogous to the one played by solenoids and wilson loops in the aharonov - bohm effect @xcite . this justifies the title of the paper . the paper is organized as follows : in section [ sec : foundations ] we introduce our model , in section [ sec : line ] we analyze in detail a simple example and in section [ sec : perspectives ] we discuss some perspectives . let @xmath3 be a connected @xmath2-manifold without boundary and consider a network and of curves @xmath4 connecting them . we require that at each of the points @xmath5 at least two curves meet . ] @xmath6 of curves embedded in @xmath3 . in the following we will focus on a particular case : the case where the network @xmath6 arises from the @xmath7-skeleton of a cellular decomposition @xmath8 of the manifold @xmath3 @xcite , @xmath9 we will regard the set @xmath6 as a network of _ defects _ in the manifold @xmath3 . in particular we introduce a new manifold @xmath10 obtained by subtracting the network @xmath6 from the original manifold @xmath3 , @xmath11 the manifold @xmath10 is path - connected but not simply - connected . closed paths around edges of the @xmath7-skeleton @xmath12 are non - contractible , therefore the first homotopy group @xmath13 is non - trivial we refer to appendix [ app : pi1 ] . ] . on the manifold @xmath10 we can consider a locally - flat connection @xmath14 for some gauge group @xmath15 . despite being locally - flat , the connection can have non - trivial holonomy around non - contractible loops in @xmath10 . this is in fact the mathematical origin of the aharonov - bohm effect . we call @xmath16 the space of locally - flat connections @xmath17 and @xmath18 the space of locally - flat connections modulo gauge transformations . the space @xmath18 is finite dimensional and its elements are completely characterized by their holonomy around non - contractible loops . we have in fact that a configuration in @xmath18 can be specified in terms of a homomorphism from @xmath13 into @xmath15 , up to conjugation by @xmath15 . we call @xmath19 this _ moduli _ space , @xmath20 and @xmath21 coordinates ( moduli ) on @xmath19 . more explicitly , a locally - flat connection is labeled by its moduli @xmath22 and a gauge transformation @xmath23 , @xmath24 with the connection @xmath25 satisfying a gauge fixing condition @xmath26 . now we consider the kinematics of general relativity in ashtekar - barbero variables @xcite . the configuration variable is a @xmath27 connection @xmath14 on a @xmath2-manifold @xmath3 . a basic step in attempts to canonically quantize general relativity involves the introduction of a kinematical hilbert space @xmath28 of functionals of the connection . such functionals are required to be invariant under @xmath27-gauge transformations and under diffeomorphisms @xmath29 of the manifold @xmath3 . loop quantum gravity provides a complete implementation of this step . here we follow a similar procedure with two specific differences : * we look for a hilbert space @xmath30 of states @xmath31 $ ] associated to the 3-manifold @xmath10 , instead of @xmath3 ; * besides @xmath27-gauge invariance and @xmath32 invariance , we require _ topological _ invariance of the states @xmath31\,$ ] . the condition can be imposed la dirac as a constraint , @xmath33 = 0\;. \label{eq : top constraint}\ ] ] here @xmath34 is the operator associated to the curvature of the connection . as well known , in a topological gauge theory @xcite@xcite gauge invariance and diffeomorphism invariance are strictly related . therefore we can focus on gauge invariant functionals which satisfy the topological - invariance constraint ( [ eq : top constraint ] ) only . diffeomorphism invariance comes for free . equivalently , we can consider the space of gauge invariant functionals of a locally - flat connection . states belonging to the state space @xmath30 described above depend on the connection @xmath14 only through its moduli @xmath35 . therefore we have that our functionals are labeled by a function @xmath36 so that @xmath37=f(m_1{,.\,.\,,}m_r)\;. \ ] ] there is another characterization of gauge invariant functionals of a flat connection which is slightly redundant , but definitely clearer . let @xmath1 be a graph embedded in @xmath10 . we call @xmath38 its links and introduce the holonomy along a link of @xmath1 as the functional @xmath39 defined in the standard way , @xmath40=\text{p}\exp \int_\gamma a\;. \ ] ] then we introduce a class of gauge invariant functionals of a locally - flat connection : these functionals are labeled by a graph @xmath1 and by a function @xmath41 . the functionals @xmath42 $ ] are given by @xmath43=\eta(h_{\gamma_1}[a]{,.\,.\,,}h_{\gamma_l}[a])\ ; \label{eq : cylgamma}\ ] ] and the function @xmath44 is assumed to be invariant under @xmath27 conjugation at nodes , @xmath45 , in order to guarantee the gauge invariance of the functional @xmath42 $ ] . a remark is in order at this point : as the connection we are considering is locally flat , we have that the functional @xmath42 $ ] depends on the graph @xmath1 only via its ` knotting ' with the skeleton @xmath12 . in particular , such states are invariant under diffeomorphisms @xmath46 connected to the identity , @xmath47=\psi_{\phi^{-1}\circ\gamma,\eta}[a]=\psi_{\gamma,\eta}[a]\;. \ ] ] as a result we have that different couples @xmath48 can actually represent the same state . for instance , for a closed curve @xmath49 which is contractible in @xmath10 , the wilson loop state @xmath50=\text{tr } d^{(j)}(h_\gamma[a])$ ] evaluates to the identity and represents the same state as @xmath51=1 $ ] . in this sense the couples @xmath52 and @xmath53 correspond to the same state . to check if two functionals labeled by different couples @xmath48 represent in fact the same state , we can always go back to the non - redundant description in terms of moduli . in particular we have that there is a unique function @xmath54 on @xmath19 such that @xmath55=f(m_1{,.\,.\,,}m_r ) \ ] ] for the whole class of equivalent states having different couples @xmath48 . in the specific case of a defect - network @xmath6 arising from the @xmath7-skeleton of a cellular decomposition of @xmath3 , there is a class of graphs @xmath1 of particular interest that we describe here : it is the class of graphs @xmath56 _ dual _ to the cellular decomposition @xmath8 , @xmath57 by dual we mean that in each cell of the decomposition @xmath8 we have a node of the graph and that two nodes are connected by a path ( a link of the graph ) if the two corresponding cells are adjacent . we notice that _ loops _ of the graph @xmath1 encircle once the edges of the @xmath7-skeleton of @xmath58 so that the edge - path group @xmath59 of the graph @xmath56 is isomorphic to the first homotopy group of the manifold @xmath10 , @xmath60 as a result we can describe the moduli space @xmath19 in terms of homomorphisms from the edge - path group of the graph to su(2 ) , modulo conjugation , @xmath61 states of the form ( [ eq : cylgamma ] ) , but with graph @xmath56 , now actually depend only on the function @xmath44 . the relation between the function @xmath44 and the function @xmath54 can be derived via a gauge - fixing of the @xmath27 invariance at the nodes of the graph , for instance setting to the identity the group elements @xmath62 associated to links @xmath49 belonging to a maximal tree on @xmath56 . + in order to promote the linear space described above to a hilbert space we need to introduce a scalar product . a topologically invariant functional measure @xmath63 $ ] can be introduced via the requirement that it reduces to an ordinary measure on @xmath19 , @xmath64\ ; \overline{\psi_f[a]}\;\psi_g[a]=\int_{{\mathcal{n } } } d\mu(m_r)\;\overline{f(m_1{,.\,.\,,}m_r)}\;g(m_1{,.\,.\,,}m_r)\;. \label{eq : measure0}\ ] ] the hilbert space of states @xmath30 associated to @xmath10 reduces to a hilbert space @xmath65 of square integrable functions on the moduli space . making sense of the formal functional measure @xmath63 $ ] on the left hand side of ( [ eq : measure0 ] ) amounts to a choice of measure @xmath66 on the moduli space @xmath19 . as a result , the following problem arises : is there a principle that can guide us in the choice of the measure @xmath66 ? our aim here is to use the methods proper to gauge field theory @xcite to make sense of the functional integral over locally - flat connections modulo gauge transformations . such methods , in the case of a finite dimensional moduli space , allow to fully determine a specific measure on the moduli space dimensional gravity with and without point particles @xcite . we point out also a close relationship with works on the field theoretical measure for simplicial geometries @xcite . ] and therefore answer our question . the guiding principle here is _ locality _ , that is : the configurations we are integrating over are local fields . the integral over locally - flat connections @xmath67 $ ] can be thought of as an integral over the moduli @xmath22 _ and _ over gauge transformations @xmath68 . as we are considering only integrals of gauge invariant quantities , the integral over gauge transformations can be factorized and we end up with a finite dimensional integral over moduli , with a non - trivial measure on the moduli . such measure comes from the jacobian of the change of variables on configuration space . this geometrical approach is equivalent @xcite to the more standard and well - known faddeev - popov procedure that we follow here . the measure on locally - flat connections modulo gauge transformations can be formally written as follows @xmath69\;=\int_{{\mathcal{a } } } \big[\prod_x da(x)\big]\,\delta\big(f(a)\big)\,\delta\big(\chi(a)\big)\;\delta_{\text{fp}}(a)\;. \ ] ] here @xmath70 is the space of connections over @xmath10 , @xmath71 imposes that the connection is locally flat , @xmath72 imposes a gauge fixing condition @xmath73 on the connection , and @xmath74 is the standard faddeev - popov determinant @xmath75 i.e. the determinant of the variation of the gauge fixing condition under a small gauge transformation with parameter @xmath76 . as well - know , such expression is ( formally ) independent from the choice of gauge fixing condition @xmath77 . now , the delta function with support on flat connections can be integrated out going to the locally - flat connections @xmath78 of expression ( [ eq : amg ] ) . similarly , the delta function with support on connections satisfying the gauge choice @xmath79 can be integrated out going to the gauge - fixed locally - flat connections @xmath25 which depend only on the moduli @xmath22 . as a result we end up with the following finite dimensional integral with a non - trivial measure @xmath69\;=\int d\bar{a}^{m_r } \;\delta_{\text{fp}}(\bar{a}^{m_r})\ ; = \int_{{\mathcal{n}}}dm_1\cdots dm_r\;j(m_r ) \delta_{\text{fp}}(m_r)\;. \ ] ] the function @xmath80 is the jacobian from gauge - fixed connections @xmath25 to the moduli @xmath22 . a procedure for computing it is described in appendix [ app : j ] . by the function @xmath81 we simply mean the faddeev - popov determinant evaluated on the connection @xmath25 , @xmath82 this is our field theoretical proposal for the measure @xmath83 in expression ( [ eq : measure0 ] ) and thus for the hilbert space @xmath30 : states in @xmath30 are assumed to be normalizable in the following scalar product @xmath64\ ; \overline{\psi_f[a]}\;\psi_g[a]=\int_{{\mathcal{n } } } dm_1\cdots dm_r\;j(m_r)\ , \delta_{\text{fp}}(m_r)\;\overline{f(m_1{,.\,.\,,}m_r)}\;g(m_1{,.\,.\,,}m_r)\;. \label{eq : measurefp}\ ] ] in section [ sec : line ] we investigate this proposal in a rather simple case and provide an explicit expression for the measure @xmath66 . + the field theoretical derivation of the measure @xmath66 described above can be compared to a different construction which is proper to loop quantum gravity . we describe it in the following . using the parametrization of states in terms of the class of graphs @xmath56 dual to the cellular decomposition , we have that a natural choice of measure is the haar measure on links of the graph : @xmath84\ ; \overline{\psi_{\gamma',\eta}[a]}\;\psi_{\gamma',\xi}[a]=\int_{su(2)^l } \prod_{l=1}^l d\mu_h(h_l)\;\overline{\eta(h_1{,.\,.\,,}h_l)}\;\xi(h_1{,.\,.\,,}h_l)\;. \label{eq : measure haar}\ ] ] with this choice of measure , thanks to the peter - weyl theorem , we know that _ spin network _ states with graph @xmath56 provide an orthonormal basis of the hilbert space @xmath30 . for completeness we recall that the spin network basis corresponds to a specific choice of function @xmath85 labeled by a @xmath27 representation ( a _ spin _ ) for each link , and an invariant map on the tensor product of representations ( an _ intertwiner _ ) for each node . more explicitly stands for a contraction of dual spaces as prescribed by the connectivity of the graph . we recall that , if @xmath86 is the vector space where the representation @xmath87 of @xmath27 acts , then the tensor product of representations @xmath88 lives in @xmath89 while the tensor product of intertwiners lives precisely in the dual of this space . ] @xmath90 choosing an orthonormal basis in the intertwiner vector spaces , we have that the orthonormality of spin network states follows , @xmath91 it is important to notice that , for functions @xmath44 and @xmath76 invariant under conjugation at nodes , the haar measure @xmath92 on @xmath93 in ( [ eq : measure haar ] ) reduces to a specific measure @xmath66 on moduli via the weyl integration formula @xcite . therefore it provides a definition of the functional measure in ( [ eq : measure0 ] ) , prescribing a specific measure @xmath66 on moduli space . + while the group theoretical choice of measure described above is certainly a natural one , it of interest to investigate its relation to the field theoretical choice described before . in the following section we investigate these two proposals in a rather simple case and provide an explicit expression for the measure @xmath66 . the result supports the following conjecture : * _ the measure _ ( [ eq : measurefp ] ) _ on the moduli space @xmath19 obtained via the field theoretical construction coincides with the one obtained from the product of haar measures _ ( [ eq : measure haar ] ) _ via weyl integration formula_. if such conjecture turns out to be robust , then we have that spin network states with embedded graph @xmath56 provide an orthonormal basis of the hilbert space @xmath30 built via the standard field theoretical construction . in this section we consider the simplest possible example : we assume that the topology of the original manifold @xmath3 is trivial , @xmath94 , and that the defect - network @xmath6 consists of a single line @xmath95 . the line @xmath96 is assumed to be unknotted in @xmath3 . then , following the construction described in the previous section , we introduce a new manifold @xmath10 obtained from @xmath3 subtracting the line @xmath96 : @xmath97 the manifold @xmath10 has a non - trivial fundamental group @xmath13 generated by the homotopy class @xmath98 $ ] of loops which encircle once the line @xmath96 . a flat connection in @xmath10 is locally pure gauge , @xmath99 . here we are interested in identifying the moduli space of such connections modulo gauge transformations . in order to identify a representative in each gauge orbit , we proceed introducing a gauge fixing condition . such condition has to break the local symmetries of the problem : a convenient choice is to introduce an auxiliary metric @xmath100 on @xmath3 and consider the coulomb - like gauge @xmath101 . in order to simplify the calculation we can assume that the line @xmath96 is a geodesic in this metric . we assume also that the metric is euclidean so that , choosing cartesian coordinates so that in the following we will raise and lower space indices freely . ] @xmath102 , the line is so to say straight and coincides with the @xmath103-axes . a locally - flat connection in @xmath10 , satisfying the gauge fixing condition @xmath104 is given by @xmath105 with @xmath106 given by @xmath107 this is in fact the general solution . ] . we have that its holonomy along a loop @xmath49 is simply given by @xmath40=\exp [ i \big(\int_\gamma \alpha_a dx^a\big ) \frac{\phi^i}{2\pi } \tau_i\,]\;. \ ] ] the integral @xmath108 is @xmath109 times the winding number of the loop @xmath49 around the line @xmath96 . as a result the holonomy @xmath110 $ ] provides a homomorphism from @xmath13 to @xmath27 . now we would like to provide a physical interpretation for the parameters @xmath111 . we introduce the non - abelian magnetic field @xmath112 . as the connection is locally flat in @xmath10 , we have that the magnetic field vanishes everywhere in @xmath10 and has support only on the line @xmath96 . it is given by @xmath113 . the defect - network consists of the single line @xmath96 . as the connection is locally - flat , the magnetic field vanishes everywhere except along the line @xmath96 . the line can be thought as a thin solenoid which confines the magnetic field . the flux of the magnetic field through a surface @xmath114 ( in figure ) determines the holonomy of the connection along a curve @xmath49 bounding the surface.,scaledwidth=40.0% ] we can compute the flux of the magnetic field through a surface @xmath114 which intersects the line @xmath96 as in figure [ pic : solenoid ] . we have that @xmath115=&\;\int_s\;b^a_i(x(\sigma))\,{\varepsilon}_{abc}\frac{\partial x^b}{\partial \sigma^1}\frac{\partial x^c}{\partial \sigma^2}\;d\sigma^1 d\sigma^2\\ = & \;\int_l ds\int_s d\sigma^1 d\sigma^2\;\phi_i\ ; { \varepsilon}_{abc}\dot{x}^a(s)\frac{\partial x^b}{\partial \sigma^1}\frac{\partial x^c}{\partial \sigma^2 } \delta^{(3)}(x(\sigma)-x(s))\ ; = \;\phi_i \ ] ] therefore the parameters @xmath116 describe in fact the flux of the magnetic field through the defect line @xmath96 , which can be thought of as a thin _ solenoid_. despite the fact that the magnetic field vanishes outside the solenoid , there are non - local observable effects in the region @xmath10 . such effects are captured by the wilson loop , i.e. by the holonomy of the connection around non - contractible closed paths . the non - abelian stokes theorem assures in fact that the exponential of the flux through a surface @xmath117 \tau^i$ ] coincides with the holonomy of the connection along the loop given by the boundary of the surface , @xmath118 . the magnetic flux @xmath111 provides a parametrization of the moduli space of locally - flat connections . to find its range we have to identify different magnetic fields which correspond to the same holonomy . thus the parameter space has the topology of the unit three - sphere @xmath119 , i.e. of the group manifold of @xmath27 . therefore , functionals of a locally - flat connection in @xmath10 actually depend only on the flux @xmath111 . moreover , from the requirement that such functionals are gauge invariant , we have that they can depend only on the modulus @xmath120 of the non - abelian flux @xmath111 . in fact a global @xmath27 transformation rotates the flux , @xmath121 . as a result , states in @xmath30 are labeled by a function @xmath122 @xmath123=f(\phi)\ ; , \ ] ] and the moduli space is given by @xmath124\}\;. \ ] ] now we would like to investigate the field theoretical proposal ( [ eq : measurefp ] ) for the measure . in order to determine the measure @xmath125 on the moduli space , we have to compute the jacobian @xmath126 and the faddeev - popov determinant @xmath127 . the jacobian can be easily computed going from cartesian to spherical coordinates , @xmath128 , so that @xmath129 . for a more systematic derivation we refer to appendix [ app : j ] . the faddeev - popov term @xmath130 is given by the determinant of an operator @xmath131 defined as the functional derivative of the gauge fixing condition ( [ eq : chi ] ) with respect to a gauge parameter . we have @xmath132 its eigenfunctions and eigenvalues can be easily found ( see appendix [ app : fp ] ) . in particular we are interested in the eigenvalues @xmath133 . they are given by @xmath134 with @xmath135 and are twice degenerate . as a result , the @xmath111 dependence of the faddeev - popov determinant can be extracted considering the appropriately regularized ratio @xmath136 where in the last equality we have used the product representation of the sine . the constant @xmath137 is undetermined and is fixed in the following in such a way that the measure of @xmath19 is normalized to one . finally we have that the scalar product on @xmath30 defined by this measure is @xmath138\overline{\psi_f[a]}\psi_g[a ] = \int_{{\mathcal{n}}}d\phi\ ; j(\phi)\delta_{\text{fp}}(\phi)\;\overline{f(\phi)}\,g(\phi)\\ = & \frac{1}{\pi}\int_0^{2\pi}d\phi\ ; \sin^2\frac{\phi}{2}\;\;\overline{f(\phi)}\,g(\phi ) \ ] ] this measure can be compared to the one obtained from the group theoretical construction ( [ eq : measure haar ] ) via weyl integration formula . in this setting , states are labeled by the homotopy class @xmath98 $ ] of loops encircling once the line @xmath96 , and by a complex - valued function @xmath44 on @xmath27 , @xmath139=\eta(h_{\gamma}[a])\;. \ ] ] gauge invariance at the base point of the loop @xmath49 requires that @xmath44 is a class function , @xmath140 . the scalar product can then be written in terms of the haar measure on @xmath27 and reduces to an integral over the _ class angle _ @xmath141 . defining @xmath142 , we have @xmath143\;\overline{\psi_{\gamma,\eta}[a]}\psi_{\gamma,\xi}[a ] = \int_{su(2)}d\mu(h_\gamma ) \;\overline{\eta(h_\gamma)}\,\xi(h_\gamma)\\ = & \frac{1}{\pi}\int_0^{2\pi}d\phi\ ; \sin^2\frac{\phi}{2}\;\;\overline{f(\phi)}\,g(\phi ) \ ] ] which coincides with the field theoretical one derived above . this provides support to our conjecture and leads to a physical interpretation of the class angle as the modulus of the flux of the magnetic field through the defect line . the analysis we have presented can be developed in a number of directions . in this section we point out some of them . * _ momenta_. in ashtekar - barbero variables the momentum conjugate to the connection @xmath144 is a non - abelian electric field @xmath145 . in the setting described in this paper , the connection is assumed to be locally flat and thus labeled by a finite number of moduli @xmath22 . consistently , the electric field @xmath145 is labeled by momenta @xmath146 canonically conjugate to the moduli . therefore , the reduced phase space of the system is given by the cotangent bundle to the moduli space , @xmath147 . at the quantum level , the canonical couple @xmath148 can be promoted to self - adjoint operators satisfying canonical commutation relations . in particular , we have that in the schroedinger representation the momenta @xmath146 are represented as differential operators @xmath149 with @xmath150 given by @xmath151 in the case of a single defect line discussed in section [ sec : line ] , we have that the momentum conjugate to the modulus @xmath120 is given by @xmath152 and that the spectrum of the operator @xmath153 is discrete . the eigenvalues are labeled by an half integer @xmath87 and are given by @xmath154 this result is to be related to the spectrum of the area of a surface with boundary on the defect - line , as discussed in the following two points . * _ electric field and surfaces_. a basic ingredient in loop quantum gravity is the flux of the electric field through a surface @xmath155 within the setting described in this paper , surfaces @xmath114 belonging to the @xmath156-skeleton of the cellular decomposition play a distinguished role . the reason is the following . let us consider two adjacent cells @xmath157 and @xmath158 in a cellular decomposition @xmath8 . we call @xmath114 the surface shared by the two regions and @xmath159 the portion of the @xmath7-skeleton given by the boundary of this surface , @xmath160 . then we consider a second surface @xmath161 diffeomorphic to @xmath114 and having the same boundary with the topology of a ball such that its boundary is the surface @xmath162 . ] we have that the fluxes @xmath163 and @xmath164 coincide ( up to a global @xmath27 transformation ) . this is a consequence of the gauss constraint @xmath165 , together with the fact that the connection is locally - flat and the region we are considering is simply - connected . therefore , quantities which are gauge- and diffeomorphism - invariant can be build up from the fluxes and the holonomy . an example is the quantity @xmath166 . the quantization of these observables is straightforward once they are written in terms of the moduli and of their conjugate momenta @xmath146 . * _ the dual picture of quantum geometry_. in the ashtekar - barbero formulation of general relativity , the electric field plays the role of spatial metric : we have that @xmath167 , where @xmath168 is a riemannian metric on a @xmath2-manifold @xmath3 and @xmath169 is its determinant . therefore , geometrical quantities as the volume of a region or the area of a surface can be written in terms of the electric field . in loop quantum gravity these quantities correspond to operators having discrete spectra and eigenstates given by spin network states @xcite . the area and the volume operator provide an interpretation of spin network states in terms of quantum geometries : quanta of volume are associated to regions dual to nodes of the spin network graph , and quanta of area are associated to surfaces dual to links of the graph . within the setting described in this paper , a cellular decomposition is present from the very beginning and a spin network with graph dual to it arises as the appropriate tool to describe gauge- and diffeomorphism - invariant functionals of the connection . it would be interesting to understand if , in this setting , operators build out of the flux of the electric field can attach a geometric meaning to the cells of the decomposition . this perspective was first proposed in @xcite and is currently under study . we point out some key differences between the spin network states discussed in this paper and the ones used in ordinary loop quantum gravity . in loop quantum gravity , embedded spin networks are not diffeomorphism invariant . in fact , states with diffeomorphic graphs which do not coincide are orthogonal . however , once the diffeomorphism constraint is imposed , most of these states are identified : abstract spin networks ( also called s - knots ) actually depend only on the diffeomorphism equivalence class of the graph and on the @xmath27 labels attached to it . the diffeomorphism equivalence class of graphs knows about the connectivity of the graph , about its knotting with the non - trivial topology of the manifold , about the self - knotting of the graph , and about some continuous parameters ( moduli ) related to the tangents to the links at each node of the graph @xcite . these continuous parameters are due to the assumed differentiable structure at the nodes and disappear if a larger class of diffeomorphisms ( the so called @xmath170 ) is considered @xcite . on the other hand , the spin networks considered in our analysis arise as a tool for describing functions on the moduli space of locally - flat connections . as a result , even if the graph of our spin networks is embedded , the state they define is actually diffeomorphism - invariant . moreover , they do not depend on the self - knotting of their graph , nor on the continuous moduli associated to the differentiable structure at the nodes . the only thing they actually depend on is their knotting with the manifold @xmath10 , i.e. with the network of defects and with the topology of the original manifold @xmath3 . from this point of view , they are closer to the use that is made of spin networks as boundary states within the context of spin foams . this fact leads us to the next two points . * _ dynamics and spin foams_. the construction discussed is purely kinematical . a natural tool for implementing its dynamics is provided by spin foams . spin foam models @xcite can in fact be considered as topological field theories of the @xmath171 type @xcite defined on a manifold with defects . this statement is supported by the following two observations : ( i ) at the classical level , general relativity can be formulated as a topological @xmath171 theory with a set of constraints on the @xmath172 field @xcite ; at the quantum level , in spin foams , these constraints are imposed _ only _ on the @xmath156-boundary of the @xmath173-cells of a simplicial complex . thus a @xmath156-skeleton of defects where the constraints are imposed is identified . a second remark is that ( ii ) the currently studied spin foam models @xcite are based on the quantization of a version of regge simplicial gravity @xcite . at the classical level , regge action can be thought as the evaluation on a locally - flat metric of the einstein - hilbert action for a manifold with defects . the boundary of the manifold is given by the defect - network and regge action consists in fact solely of the gibbons - hawking - york boundary term @xcite , i.e. of the integral of the area density times the extrinsic curvature at the defect . the theory has finitely many degrees of freedom which are in fact associated to the defect - network . * _ infinitely many degrees of freedom_. spin foam models provide a sum over histories . an history can be obtained from a cellular decomposition of a @xmath173-manifold @xmath174 . then we can introduce a new manifold @xmath175 obtained subtracting from the original manifold the @xmath156-skeleton of the cellular decomposition . as a result , a foliation of @xmath175 results in a family of @xmath2-manifolds @xmath10 with different networks of defects . a model of this kind allows transitions between hilbert spaces @xmath30 associated to different defect - networks . this fact points to the need of considering a larger ( but still separable ) hilbert space @xmath176 obtained la fock @xcite . the analogous of the @xmath177-particle state space would be the hilbert space @xmath178 associated to defect - networks having moduli space of dimension @xmath177 , so that @xmath176 is the closure in norm of the span of states belonging to @xmath179 . the hilbert space @xmath176 is the one proper to a system with infinitely many degrees of freedom . the idea that purely topological theories might play an important role at the planck scale has already been noticed in the past . in the first pioneering papers on loop quantum gravity @xcite , topological invariants played a key role ( see also @xcite ) . more recently , t hooft @xcite introduced a locally finite model for 4d gravity which is topological and has dynamical string defects . it would be interesting to explore its quantum kinematics within the setting described in this paper . we conclude this section with a remark about an intriguing long - sighted perspective : * _ locality and effective description_. topological field theories have only global degrees of freedom . however , in presence of defects , some of the degrees of freedom are associated to the defect - network . as a result , there is a finite number of degrees of freedom associated to each region of the manifold , thus leading to a notion of locality . the discrete spectrum of quantum - geometry operators sets the scale of the region . this fact points to an appealing scenario for quantum gravity where the theory has no trans - planckian degrees of freedom because it is topological ( and therefore finite ) at small scales , while at larger scales it has finitely many degrees of freedom which can be described effectively in terms of a local quantum field theory . loop quantum gravity may provide a realization of this scenario . in this paper we have studied the state space of a theory of locally - flat connections on a manifold with defects . we briefly summarize our analysis : * the space @xmath18 of locally - flat connections modulo gauge transformations is finite - dimensional . therefore a state , i.e. a gauge- and diffeomorphism - invariant functional of the connection , is actually given by an ordinary function of a finite number of parameters ( moduli ) . a scalar product between states is introduced in terms of a functional integral over @xmath18 . the functional integral reduces to an ordinary integral over the space of moduli with a non - trivial measure . the final result is given in equation ( [ eq : measurefp ] ) . the measure consists in the product of the jacobian from gauge - fixed connections to moduli , times a faddeev - popov determinant expressed in terms of the moduli . * the field theoretical derivation of the scalar product can be compared to an a priori unrelated construction which uses the techniques proper to loop quantum gravity . this construction consists in considering states labeled by an embedded graph and build out of the holonomy of the connection . as the connection is locally - flat , these states actually depend only on the knotting of the graph with the network of defects . we consider a class of graphs @xmath56 having edge - path group @xmath59 isomorphic to the first homotopy group @xmath13 of the manifold @xmath10 . in the particular case of a defect - network arising as the @xmath7-skeleton of a cellular decomposition , the graphs @xmath56 is dual to the decomposition . these states correspond to specific functions on moduli space . moreover they suggest a natural proposal for the choice of measure involved in the definition of the scalar product : this is the haar measures for group elements associated to the links of the graph . spin network states with graph @xmath56 provide an orthonormal basis of the hilbert space . we notice that the group theoretical measure induces a non - trivial measure on moduli space via weyl integration formula . * we argue that the measure on moduli space obtained via the field theoretical construction and the one obtained from the product of haar measures coincide . proving this conjecture would establish spin network states as an orthonormal basis of the hilbert space @xmath30 build via the field theoretical construction . * we provide a test of this conjecture in the case of a single defect - line . the moduli are identified with the flux of the magnetic field through the defect . the faddeev - popov determinant is computed explicitly and the result is shown to coincide with the measure on the class - angle induced by the haar measure . the physical picture proper to the aharonov - bohm effect arises . the interest in this analysis lies in its relation with loop quantum gravity and in the new perspectives it opens . thanks to giorgio immirzi , pietro menotti and carlo rovelli for many insightful discussions and for encouragement . i wish also to thank benjamin bahr , marc knecht , antonino marcian , leonardo modesto , roberto pereira , daniele pranzetti , alejandro perez , michael reisenberger and matteo smerlak for comments and suggestions on a preliminary version of this work . we recall the definition of the first homotopy group and discuss its presentation @xcite . let @xmath10 be a connected @xmath2-manifold . a loop based at @xmath180 in @xmath10 is a closed path starting and ending at the base point @xmath181 . two loops @xmath182 and @xmath183 based at @xmath180 are said to be _ homotopic _ if one can be continuously deformed into the other without moving the base point . homotopy is an equivalence relation : we denote @xmath98 $ ] the class of loops homotopic to @xmath49 . the set of all homotopy classes of loops based at @xmath180 is denoted @xmath184 . this set is indeed a group , the fundamental group or first homotopy group , with the product given by composition of loops , the inverse given by the loop traversed in the opposite direction , and the identity given by the class of contractible loops . the first homotopy group depends both on the topology of the manifold @xmath10 and on the base point @xmath180 . as we have assumed that the manifold @xmath10 is connected , the dependence on the base point can be dropped as different base points correspond to isomorphic groups . in this paper , the manifold @xmath10 arises from a manifold @xmath3 and a network of defects @xmath6 , so that @xmath185 . we point out that the group @xmath186 is well - studied for some specific choices of @xmath6 . when the defect - network @xmath6 consists of a collection of non - intersecting loops , the group @xmath186 is called the _ knot _ group . a convenient presentation ( a set of generators and defining relations ) for the knot group is provided by wirtinger presentation @xcite . in the paper we are mainly interested in the case of a network @xmath6 arising as the @xmath7-skeleton of a cellular decomposition of the manifold @xmath3 @xcite , @xmath187 . in this case the first homotopy group of the manifold @xmath185 is discussed by regge in @xcite ( see also @xcite ) . we have a generator for each defect - line and a number of relations for each point where defect - lines join . a related presentation can be obtained via the following construction . we build a graph @xmath1 dual to the cellular complex , i.e. topologically equivalent to the @xmath7-skeleton of the dual to @xmath8 . then we consider the edge - path group @xmath188 of the graph @xmath1 . its elements are closed paths on the graph @xmath1 , based at a vertex of the graph . the edge - path group of @xmath1 is isomorphic to the first homotopy group of @xmath189 , @xmath190 , as can be shown considering a fattening of the graph @xmath1 . then , the standard presentation of @xmath188 can be used @xcite . in order to compute the jacobian from the gauge - fixed connections to the moduli we can use a procedure similar to the one introduced by polyakov in @xcite . we consider the space of gauge - fixed locally - flat connections . a point in such space is a flat connection @xmath191 satisfying a gauge - fixing condition @xmath192 and labeled by the moduli @xmath22 and by a global @xmath27 transformation @xmath193 which is left unfixed by the condition @xmath192 . the tangent space at a point @xmath191 is spanned by vectors @xmath194 . such vectors can by decomposed in terms of the variation of the moduli @xmath195 and the variation of the global @xmath27 element @xmath196 , @xmath197 we can introduce a scalar product on this tangent space : @xmath198 such scalar product involves the auxiliary metric @xmath199 introduced via the gauge - fixing condition . the scalar product can be used to provide a gaussian normalization of the formal measure on gauge - fixed connections and define the jacobian @xmath80 , @xmath200 the jacobian is independent of the group element @xmath193 as the scalar product is invariant under global @xmath27 transformations . in order to clarify the construction we compute the jacobian @xmath126 in the simple case of a single line defect of section [ sec : line ] . we write the magnetic flux @xmath111 in terms of its modulus @xmath120 and a given direction @xmath201 , @xmath202 , so that @xmath203 the tangent vector at a configuration is simply given by @xmath204 as a result we have that , defining @xmath137 the integral of @xmath205 , the scalar product @xmath206 is given by @xmath207 with @xmath208 the projector in @xmath209 orthogonal to the vector @xmath201 . using relation ( [ eq : def j ] ) we find @xmath210 which matches with the simpler derivation discussed in section [ sec : line ] . here we compute the faddeev - popov determinant @xmath127 used in section [ sec : line ] . the gauge - fixing function we consider is the coulomb - like gauge @xmath211 . the variation of @xmath212 with respect to a gauge transformation defines an operator @xmath213 which depends on the connection . the quantity @xmath130 is given by the functional determinant of this operator . computing it is in general hard unless one resorts to perturbative techniques . in the case studied in this paper , however , the connection involved is locally - flat so that the moduli space is finite dimensional . thus the faddeev - popov determinant is an ordinary function of few parameters the moduli and we can hope to determine it non - perturbatively , at least in simple cases . in particular , in the case of a single line defect discussed in section [ sec : line ] , the moduli space is one - dimensional and the faddeev - popov determinant @xmath127 is in fact an ordinary function on the interval @xmath214 $ ] . it is given by the ( appropriately regularized ) product of the eigenvalues of the differential operator @xmath215 . in the following we study this operator . we work in cylindrical coordinates @xmath216 . in order to simplify the analysis of the boundary conditions to be imposed on the defect line at @xmath217 , we actually focus on a improved gauge - fixing function @xmath218 . this improvement does not change the @xmath120-dependence of the faddeev - popov determinant . we have that the differential operator @xmath219 is given by @xmath220 where @xmath201 is a given direction that we have fixed in the previous section . to derive ( [ eq : k ] ) we have used the fact that @xmath221 and @xmath222 . the operator is defined on functions @xmath223 with @xmath224 and boundary conditions at @xmath217 @xmath225 such boundary conditions on gauge transformations guarantee that the affine part of the connection is unchanged on the defect line . the eigenvalue problem we have to solve is @xmath226 eigenfunctions can be found via separation of variables . we consider an ansatz of the form @xmath227 and introduce two separating constants , @xmath228 and @xmath229 . we have that @xmath230 the first equation is solved by @xmath231 . the second equation is solved by the two periodic functions @xmath232 where the third component for @xmath233 imposes that its component in the direction @xmath201 vanishes . ] corresponds to the direction @xmath201 , @xmath177 in an integer and the separating constant @xmath229 is given by @xmath234 the general solution of equation ( [ eq : bessel ] ) is a linear combination of modified bessel functions @xmath235 the boundary conditions ( [ eq : boundary condit ] ) impose that @xmath236 and @xmath237 . as a result we have that eigenfunctions are labeled by the integer @xmath177 and a sign . the eigenvalues are @xmath238 in the following we will denote them simply by @xmath239 , with @xmath240 . the eigenvalues are twice degenerate , the corresponding eigenfunctions being @xmath241 . the determinant of the operator @xmath215 can be computed as a product of its eigenvalues ( with the appropriate degeneracy ) , defined via zeta - function regularization . the dependence of the faddeev - popov determinant on the modulus @xmath120 can be obtained considering the ratio @xmath242 where @xmath137 is an undetermined constant . c. j. isham , `` canonical quantum gravity and the problem of time '' , http://arxiv.org/abs/gr-qc/9210011[gr-qc/9210011 ] . 19th international colloquium on group theoretical methods in physics , salamanca , spain 1992 . c. rovelli , `` quantum gravity '' . cambridge university press , 2004 . a. ashtekar and j. lewandowski , `` background independent quantum gravity : a status report '' , _ class . _ * 21 * ( 2004 ) r53 , http://arxiv.org/abs/gr-qc/0404018[gr-qc/0404018 ] . t. thiemann , `` modern canonical quantum general relativity '' . cambridge university press , 2007 . c. rovelli , `` loop quantum gravity '' , _ living rev . rel . _ * 11 * ( 2008 ) 5 . c. rovelli and l. smolin , `` knot theory and quantum gravity '' , _ phys . lett . _ * 61 * ( 1988 ) 1155 . c. rovelli and l. smolin , `` loop space representation of quantum general relativity '' , _ nucl . phys . _ * b331 * ( 1990 ) 80 . a. ashtekar and j. lewandowski , `` representation theory of analytic holonomy c * algebras '' , http://arxiv.org/abs/gr-qc/9311010 [ gr - qc/9311010 ] . in quantum gravity and knots , ed . by j. baez , oxford univ . a. ashtekar and j. lewandowski , `` differential geometry on the space of connections via graphs and projective limits '' , _ _ * 17 * ( 1995 ) 191230 , http://arxiv.org/abs/hep-th/9412073[hep-th/9412073 ] . j. lewandowski , a. okolow , h. sahlmann , and t. thiemann , `` uniqueness of diffeomorphism invariant states on holonomy- flux algebras '' , _ commun . math . * 267 * ( 2006 ) 703733 , http://arxiv.org/abs/gr-qc/0504147[gr-qc/0504147 ] . c. fleischhack , `` representations of the weyl algebra in quantum geometry '' , _ commun . _ * 285 * ( 2009 ) 67140 , http://arxiv.org/abs/math-ph/0407006[math-ph/0407006 ] . c. rovelli and l. smolin , `` spin networks and quantum gravity '' , _ phys . rev . _ * d52 * ( 1995 ) 57435759 , http://arxiv.org/abs/gr-qc/9505006[gr-qc/9505006 ] . j. c. baez , `` spin network states in gauge theory '' , _ adv . math . _ * 117 * ( 1996 ) 253272 , http://arxiv.org/abs/gr-qc/9411007[gr-qc/9411007 ] . c. rovelli and l. smolin , `` discreteness of area and volume in quantum gravity '' , _ nucl . phys . _ * b442 * ( 1995 ) 593622 , http://arxiv.org/abs/gr-qc/9411005[gr-qc/9411005 ] . a. ashtekar and j. lewandowski , `` quantum theory of geometry . i : area operators '' , _ class . * 14 * ( 1997 ) a55a82 , http://arxiv.org/abs/gr-qc/9602046[gr-qc/9602046 ] . a. ashtekar and j. lewandowski , `` quantum theory of geometry . ii : volume operators '' , _ adv . * 1 * ( 1998 ) 388429 , http://arxiv.org/abs/gr-qc/9711031[gr-qc/9711031 ] . a. ashtekar , a. corichi , and j. a. zapata , `` quantum theory of geometry . iii : non - commutativity of riemannian structures '' , _ class . * 15 * ( 1998 ) 29552972 , http://arxiv.org/abs/gr-qc/9806041[gr-qc/9806041 ] . s. a. major , `` operators for quantized directions '' , _ class . * 16 * ( 1999 ) 38593877 , http://arxiv.org/abs/gr-qc/9905019[gr-qc/9905019 ] . t. thiemann , `` a length operator for canonical quantum gravity '' , _ j. math . * 39 * ( 1998 ) 33723392 , http://arxiv.org/abs/gr-qc/9606092[gr-qc/9606092 ] . e. bianchi , `` the length operator in loop quantum gravity '' , _ nucl . * b807 * ( 2009 ) 591624 , http://arxiv.org/abs/0806.4710[0806.4710 ] . g. immirzi , `` quantum gravity and regge calculus '' , _ nucl . * 57 * ( 1997 ) 6572 , http://arxiv.org/abs/gr-qc/9701052[gr-qc/9701052 ] . a. barbieri , `` quantum tetrahedra and simplicial spin networks '' , _ nucl . * b518 * ( 1998 ) 714728 , http://arxiv.org/abs/gr-qc/9707010[gr-qc/9707010 ] . f. markopoulou , `` dual formulation of spin network evolution '' , http://arxiv.org/abs/gr-qc/9704013[gr-qc/9704013 ] . j. c. baez , `` an introduction to spin foam models of bf theory and quantum gravity '' , _ lect . notes phys . _ * 543 * ( 2000 ) 2594 , http://arxiv.org/abs/gr-qc/9905087[gr-qc/9905087 ] . f. markopoulou and l. smolin , `` disordered locality in loop quantum gravity states '' , _ class . _ * 24 * ( 2007 ) 38133824 , http://arxiv.org/abs/gr-qc/0702044[gr-qc/0702044 ] . g. t hooft ( ed . ) , `` 50 years of yang - mills theory '' . world scientific , 2005 . y. aharonov and d. bohm , `` significance of electromagnetic potentials in the quantum theory '' , _ phys . _ * 115 * ( 1959 ) 485491 . t. t. wu and c. n. yang , `` concept of nonintegrable phase factors and global formulation of gauge fields '' , _ phys . * d12 * ( 1975 ) 38453857 . m. peshkin and a. tonomura , `` the aharonov - bohm effect '' . springer , 1989 . a. ashtekar , `` new variables for classical and quantum gravity '' , _ phys . * 57 * ( 1986 ) 22442247 . a. ashtekar , `` new hamiltonian formulation of general relativity '' , _ phys . _ * d36 * ( 1987 ) 15871602 . j. f. barbero g. , `` real ashtekar variables for lorentzian signature space times '' , _ phys . _ * d51 * ( 1995 ) 55075510 , http://arxiv.org/abs/gr-qc/9410014[gr-qc/9410014 ] . e. witten , `` topological quantum field theory '' , _ commun . _ * 117 * ( 1988 ) 353 . e. witten , `` quantum field theory and the jones polynomial '' , _ commun . * 121 * ( 1989 ) 351 . g. t. horowitz , `` exactly soluble diffeomorphism invariant theories '' , _ commun . * 125 * ( 1989 ) 417 . d. birmingham , m. blau , m. rakowski , and g. thompson , `` topological field theory '' , _ phys . * 209 * ( 1991 ) 129340 . a. s. cattaneo , p. cotta - ramusino , j. frohlich , and m. martellini , `` topological bf theories in three - dimensions and four- dimensions '' , _ j. math * 36 * ( 1995 ) 61376160 , http://arxiv.org/abs/hep-th/9505027[hep-th/9505027 ] . a. m. polyakov , `` quantum geometry of bosonic strings '' , _ phys . * b103 * ( 1981 ) 207210 . o. alvarez , `` theory of strings with boundaries : fluctuations , topology , and quantum geometry '' , _ nucl . * b216 * ( 1983 ) 125 . s. carlip , `` a phase space path integral for ( 2 + 1)-dimensional gravity '' , _ class . quant . grav . _ * 12 * ( 1995 ) 22012208 , http://arxiv.org/abs/gr-qc/9504033[gr-qc/9504033 ] . l. cantini and p. menotti , `` functional approach to 2 + 1 dimensional gravity coupled to particles '' , _ class . grav . _ * 20 * ( 2003 ) 845858 , http://arxiv.org/abs/hep-th/0209061[hep-th/0209061 ] . e. witten , `` ( 2 + 1)-dimensional gravity as an exactly soluble system '' , _ nucl . phys . _ * b311 * ( 1988 ) 46 . e. witten , `` topology changing amplitudes in ( 2 + 1)-dimensional gravity '' , _ nucl . * b323 * ( 1989 ) 113 . g. t hooft , `` quantization of point particles in 2 + 1 dimensional gravity and space - time discreteness '' , _ class . * 13 * ( 1996 ) 10231040 , http://arxiv.org/abs/gr-qc/9601014[gr-qc/9601014 ] . a. jevicki and m. ninomiya , `` functional formulation of regge gravity '' , _ phys . * d33 * ( 1986 ) 1634 . p. menotti and p. p. peirano , `` faddeev - popov determinant in two - dimensional regge gravity '' , _ phys . * b353 * ( 1995 ) 444449 , http://arxiv.org/abs/hep-th/9503181[hep-th/9503181 ] . p. menotti and p. p. peirano , `` functional integration on two dimensional regge geometries '' , _ nucl . _ * b473 * ( 1996 ) 426454 , http://arxiv.org/abs/hep-th/9602002[hep-th/9602002 ] . p. menotti and p. p. peirano , `` diffeomorphism invariant measure for finite dimensional geometries '' , _ nucl . phys . _ * b488 * ( 1997 ) 719734 , http://arxiv.org/abs/hep-th/9607071[hep-th/9607071 ] . z. bern , e. mottola , and s. k. blau , `` general covariance of the path integral for quantum gravity '' , _ phys . rev . _ * d43 * ( 1991 ) 12121222 . e. mottola , `` functional integration over geometries '' , _ * 36 * ( 1995 ) 24702511 , http://arxiv.org/abs/hep-th/9502109[hep-th/9502109 ] . a. kirillov , `` representation theory and noncommutative harmonic analysis i '' . springer , 1994 . n. grot and c. rovelli , `` moduli - space structure of knots with intersections '' , _ j. math . phys . _ * 37 * ( 1996 ) 30143021 , http://arxiv.org/abs/gr-qc/9604010[gr-qc/9604010 ] . w. fairbairn and c. rovelli , `` separable hilbert space in loop quantum gravity '' , _ j. math . * 45 * ( 2004 ) 28022814 , http://arxiv.org/abs/gr-qc/0403047[gr-qc/0403047 ] . j. c. baez , `` spin foam models '' , _ class . grav . _ * 15 * ( 1998 ) 18271858 , http://arxiv.org/abs/gr-qc/9709052[gr-qc/9709052 ] . d. oriti , `` spacetime geometry from algebra : spin foam models for non - perturbative quantum gravity '' , _ rept . * 64 * ( 2001 ) 14891544 , http://arxiv.org/abs/gr-qc/0106091[gr-qc/0106091 ] . a. perez , `` spin foam models for quantum gravity '' , _ class . grav . _ * 20 * ( 2003 ) r43 , http://arxiv.org/abs/gr-qc/0301113[gr-qc/0301113 ] . j. w. barrett and l. crane , `` relativistic spin networks and quantum gravity '' , _ j. math . _ * 39 * ( 1998 ) 32963302 , http://arxiv.org/abs/gr-qc/9709028[gr-qc/9709028 ] . j. engle , e. livine , r. pereira , and c. rovelli , `` lqg vertex with finite immirzi parameter '' , _ nucl . * b799 * ( 2008 ) 136149 , http://arxiv.org/abs/0711.0146[0711.0146 ] . l. freidel and k. krasnov , `` a new spin foam model for 4d gravity '' , _ class . * 25 * ( 2008 ) 125018 , http://arxiv.org/abs/0708.1595[0708.1595 ] . t. regge , `` general relativity without coordinates '' , _ nuovo cim . _ * 19 * ( 1961 ) 558571 . g. w. gibbons and s. w. hawking , `` action integrals and partition functions in quantum gravity '' , _ phys . d _ * 15 * ( 1977 ) 2752 . j. w. . york , `` role of conformal three geometry in the dynamics of gravitation '' , _ phys . _ * 28 * ( 1972 ) 1082 . r. f. streater and a. s. wightman , `` pct , spin and statistics , and all that '' . addison - wesley , 1989 . c. di bartolo , r. gambini , j. griego , and j. pullin , `` consistent canonical quantization of general relativity in the space of vassiliev knot invariants '' , _ phys . * 84 * ( 2000 ) 23142317 , http://arxiv.org/abs/gr-qc/9909063[gr-qc/9909063 ] . g. t hooft , `` a locally finite model for gravity '' , _ found . * 38 * ( 2008 ) 733757 , http://arxiv.org/abs/0804.0328[0804.0328 ] . j. c. baez and a. perez , `` quantization of strings and branes coupled to bf theory '' , _ adv . theor . math . * 11 * ( 2007 ) 3 , http://arxiv.org/abs/gr-qc/0605087[gr-qc/0605087 ] . w. j. fairbairn and a. perez , `` extended matter coupled to bf theory '' , _ phys . * d78 * ( 2008 ) 024013 , http://arxiv.org/abs/0709.4235[0709.4235 ] . w. j. fairbairn , `` on gravitational defects , particles and strings '' , _ jhep _ * 09 * ( 2008 ) 126 , http://arxiv.org/abs/0807.3188[0807.3188 ] . i. m. singer , j. thorpe , `` lecture notes on elementary topology and geometry '' . springer , 1976 .

for the anisotropic$]-theory with we calculate the imaginary parts of the renormalization - group functions in the form of a series expansion in , around the isotropic case . dimensional regularization is used to evaluate the fluctuation determinants for the isotropic instanton near the space dimension . the vertex functions in the presence of instantons are renormalized with the help of a nonperturbative procedure introduced for the simple-theory by mckane et al .

hep-th9508172

more than twenty years ago , brezin , le guillou and zinn - justin ( bgz ) studied the phase transition of a cubic anisotropic system by means of renormalization group equations @xcite . within a @xmath4-expansion , they found that to lowest nontrivial order in @xmath5 , the only stable fixed point for @xmath6 is the @xmath7-symmetric one , where @xmath8 is the number of field components appearing in the cubic anisotropic model . they interpreted this as an indication that the anisotropy is irrelevant as long as @xmath8 is smaller than four . for @xmath9 , the isotropic fixed point destabilized and the trajectories crossed over to the cubic fixed point . recently , our knowledge of perturbation coefficients of the renormalization group functions of the anisotropic system was extended up to the five - loop level by kleinert and schulte - frohlinde @xcite . since the perturbation expansions are badly divergent , they do not directly yield improved estimates for the crossover value @xmath10 where the isotropic fixed point destabilizes in favor of the cubic one . an estimate using pad approximants @xcite indicates @xmath10 to lie below @xmath11 , thus permitting real crystals to exhibit critical exponents of the cubic universality class . for a simple @xmath12-theory , the pad approximation is known to be inaccurate . in fact , the most accurate renormalization group functions for that theory have been obtained by combining perturbation expansions with large - order estimates and using a resummation procedure based on borel - transformations @xcite@xcite . it is the purpose of this paper to derive the large - order behavior of the renormalization group functions for the anisotropic @xmath13-theory . in a forthcoming paper we will combine these results with the five - loop perturbation expansion of kleinert and schulte - frohlinde to derive the precise value for the crossover value @xmath10 . for the simple @xmath13-theory , the large - order behavior of perturbation coefficients has been derived by lipatov @xcite , bgz @xcite and others @xcite@xcite in a number of papers . the generalization to the @xmath7-symmetric case was given in @xcite . an equivalent method for calculating the large - order behavior is based on the observation that for a negative coupling constant green functions possess an exponentially small imaginary part due to the fact that the ground state is unstable @xcite . the imaginary part is associated with the tunneling decay rate of the ground state . it determines directly the large - order behavior of the perturbation coefficients via a dispersion relation in the complex coupling constant plane . in the semiclassical limit , the imaginary part of all green functions can be calculated with the help of classical solutions called _ instantons_. for a massless @xmath13-theory in @xmath14 space dimensions , these instantons can be found analytically . the imaginary part is a consequence of a negative frequency mode in the spectrum of the fluctuation operator , whose determinant enters the one - loop correction to the instanton contribution . mckane , wallace and de alcantara bonfim @xcite found a way to continue the results of the @xmath13-theory in @xmath14 to a field theory in @xmath15 dimensions . they proposed an extended dimensional regularization scheme for nonperturbative renormalizing the imaginary parts of vertex functions . in the present work we have to extend this scheme to the case of a @xmath16-theory with cubic anisotropy , where the energy functional has the following form : @xmath17 \,\ , .\ ] ] for @xmath18 , the corresponding model in quantum mechanics was first studied by banks , bender and wu @xcite who used multidimensional wkb - techniques to derive the large - order behavior of the perturbation expansion for the ground state energy . in @xmath19 janke @xcite presented a more efficient calculation using a path integral approach . in the present work , this approach will be generalized to quantum field theory and extended by a careful discussion of the region near the isotropic limit @xmath20 . this is important , since the infrared - stable cubic fixed point is expected to appear very close to the @xmath7-symmetric one . in fact , it will be sufficient to give the quantum - field theoretical generalization of @xcite in terms of an expansion about the isotropic case in powers of @xmath1 . the paper is organized as follows . the method is developed by treating first the case @xmath21 . in we derive the feynman rules for the power series expansion of all green functions around the isotropic limit . in we calculate the small - oscillation determinants for the transversal and longitudinal fluctuations . in we use the extended renormalization scheme of @xcite to find the full ( real and imaginary ) vertex functions , and derive the renormalization constants to one loop . in we calculate the imaginary parts of the renormalization - group functions and thus the large - order behavior of the perturbation coefficients . in , finally , we extend the results to the physically relevant case @xmath22 . for positive coupling constants @xmath23 and @xmath1 , the system defined by the energy functional ( [ hn ] ) is stable and the green functions are real . on the other hand , if both coupling constants are negative the system is unstable and the green functions acquire an imaginary part . the corresponding functional integrals can be calculated by an analytical continuation from positive to negative coupling constants , keeping the factor @xmath24}$ ] real . we perform this analytical continuation by means of a joint rotation in the two complex planes , substituting @xmath25 and @xmath26 , and rotating the azimuthal angle @xmath27 from @xmath28 to @xmath29 . at the same time , we rotate the contour of integration in the field space . the convergence of the functional integrals is maintained by the field rotation @xmath30 . a natural parameter for the anisotropy of the system is the ratio @xmath31 . the isotropic limit corresponds to @xmath32 . during the joint rotation of @xmath23 and @xmath1 , the parameter @xmath33 remains constant . thus , @xmath33 is a good parameter for the anisotropy at both positive and negative couplings @xmath23 and @xmath1 . we shall use the coupling constants @xmath34 and @xmath33 for a calculation of the green functions . these are given by the functional integrals @xmath35\ } } { \int d \phi \exp\ { -h[\phi]\ } } \ , , \nonumber \\\end{aligned}\ ] ] where the subscripts @xmath36 run through the @xmath8 components of the field @xmath37 . as explained in the introduction , we shall first study the case @xmath21 with the free energy functional @xmath38 = } \nonumber \\ \nonumber \\ & & \!\!\!\!\!\!\!\ ! \int d^dx \left\{\frac{1}{2 } \left[\phi_1 ( -{\nabla}^2 ) \phi_1 + \phi_2 ( -{\nabla}^2 ) \phi_2 \right]+\frac{g}{4 } \left[\phi_1 ^ 4 + 2(1-\delta ) \phi_1 ^ 2 \phi_2 ^ 2 + \phi_2 ^ 4 \right ] \right\}.\end{aligned}\ ] ] when expressed in terms of the coupling constants @xmath39 and @xmath33 , the green functions possess an imaginary part for @xmath40 . for the reasons explained above , we shall derive an expansion of the imaginary parts of the green functions ( [ g2 m ] ) around the isotropic case , i. e. , in powers of @xmath33 : @xmath41\ , .\ ] ] each power @xmath42 has its own @xmath43-dependent imaginary part . given such an expansion , the large - order estimates for the coefficients of the powers @xmath44 follows from a dispersion relation in @xmath39 : @xmath45 for a general discussion of the relationship between imaginary parts and large - order behavior see , for example , ch . @xmath46 of @xcite . if the power - series expansion of @xmath47 is @xmath48 the coefficients @xmath49 have the asymptotic behavior @xmath50.\ ] ] these estimates apply to perturbation coefficients , in which the maximal power of @xmath39 is much larger than the power of @xmath33 . being interested in the region close to the isotropic limit , this restricted large - order estimation will be sufficient . an expansion of the green function ( [ g2 m ] ) around the instanton yields exponentially small imaginary parts in both numerator and denominator . in order to isolate the imaginary part of the numerator , we simplify the denominator in ( [ g2 m ] ) and calculate first the imaginary part of the approximate green functions @xmath51\ } } { \int d \phi \exp\ { -h_0 [ \phi]\ } } \ , , \nonumber \\\end{aligned}\ ] ] where the denominator contains only the free energy functional @xmath52 . with the aim of calculating ( [ img ] ) , we expand the fields around the classical solution of the isotropic limit @xmath32 for the space dimension @xmath14 . thus we write : @xmath53 where @xmath54 is the well - known @xmath13-instanton in four dimensions @xmath55 and @xmath56 is the rotation angle of the isotropic instanton in the @xmath57-plane . the fields @xmath58 and @xmath59 correspond to the degrees of freedom orthogonal to the rotation of that instanton . the parameters @xmath60 and @xmath61 are position and scale size of the instanton , respectively . inserting the expansion ( [ exp ] ) in @xmath62 , we obtain the expression : @xmath63 where @xmath64 collects all terms in @xmath58 and @xmath59 of order @xmath65 . they are given by : + ( 1 ) linear terms : @xmath66 ^ 2 } \xi \nonumber \\ & & + \frac{\delta}{(-g)^{1/2}}\ , \sin^2(2\vartheta)\,8(2)^{1/2 } \int d^dx \frac{\lambda^3}{[1+\lambda^2(x - x_0)^2]^3 } \xi \nonumber \\ & & + \frac{\delta}{(-g)^{1/2}}\ , \sin(4\vartheta ) \,4(2)^{1/2 } \int d^dx \frac{\lambda^3}{[1+\lambda^2(x - x_0)^2]^3 } \eta\end{aligned}\ ] ] ( 2 ) quadratic terms : @xmath67 ^ 2}(\xi^2-\eta^2 ) \nonumber \\ & & + \delta \ , \sin(4\vartheta)\ , 6 \int d^dx \frac{\lambda^2}{[1+\lambda^2(x - x_0)^2]^2 } \xi \eta \ , , \end{aligned}\ ] ] where @xmath68 and @xmath69 are the operators @xmath70 ^ 2}\ , , \,\ , m_t = -{\nabla}^2- \frac { 8(1-\delta ) \lambda^2}{[1+\lambda^2 ( x - x_0)^2]^2}\ ] ] ( 3 ) cubic terms : @xmath71 \nonumber \\ & & -\delta \ , \frac{\sin^2(2\vartheta)}{2 } \ , ( -8g)^{1/2 } \int d^dx \frac{\lambda}{1+\lambda^2(x - x_0)^2 } ( 3 \eta^2 \xi-\xi^3 ) \nonumber \\ & & -\delta \ , \frac{\sin(4\vartheta)}{4 } \ , ( -8g)^{1/2 } \int d^dx \frac{\lambda}{1+\lambda^2(x - x_0)^2 } ( \eta^3 - 3 \eta \xi^2)\end{aligned}\ ] ] ( 4 ) quartic terms : @xmath72 \nonumber \\ & & -\delta \ , g \ , \frac{\sin^2(2\vartheta)}{8 } \int d^dx ( \xi^4+\eta^4 - 6\xi^2 \eta^2 ) \nonumber \\ & & + \delta \ , g \ , \frac{\sin(4\vartheta)}{4 } \int d^dx ( \xi \eta^3-\eta \xi^3)\end{aligned}\ ] ] the linear terms ( [ h1 ] ) contain factors @xmath5 or @xmath33 , since the expansion of the fields around the isotropic instanton is extremal only in the four - dimensional isotropic limit . the classical contribution of the instanton is @xmath73 -\frac{\lambda^{\varepsilon } \delta } { g } \ , \frac{8\pi^2}{3 } \frac{\sin^2(2\vartheta)}{2}+ { \cal o } \left(\frac { \delta } { g}\ , \varepsilon \right ) \ ! . \nonumber\end{aligned}\ ] ] the first term in @xmath74 is evaluated up to the first order in @xmath5 , because the one - loop renormalization will require replacing @xmath75 $ ] , where @xmath76 is the renormalized version of the coupling constant @xmath39 . a contribution of order @xmath5 is not needed in the second , @xmath33-dependent term , where it would produce a further factor @xmath33 , and thus be part of the neglected terms @xmath77 in ( [ img ] ) . due to the anisotropy of the action , the fluctuation expansion ( [ h1])([hclas ] ) contains @xmath56-dependent parts . however , all these terms are of order @xmath33 and can therefore be handled by straightforward perturbation theory . note that the angle @xmath56 disappears when the isotropic instanton is directed along the coordinate axis . then the isotropic instanton coincides with the exact solution in @xmath14 for @xmath78 . this is also seen by inspecting the potential in ( [ hn2 ] ) . for @xmath78 , the term @xmath79 is larger than @xmath80 , so that the `` tunneling - paths '' of extremal action are obviously straight lines along the coordinate axis . for the region @xmath81 , the exact instanton follows from the known fact that the action ( [ hn2 ] ) is invariant under the orthogonal transformation : @xmath82 with the new coupling constants : @xmath83 satisfying @xmath81 for @xmath84 . in contrast to the method in @xcite , the treatment of the fluctuations in a power series in @xmath33 does not allow us to deal with all modes perturbatively . near the @xmath85-symmetric case , the gaussian approximation for the rotation of the instanton becomes invalid , and the gaussian integral must be replaced by an exact angle integration . the separation of the rotation mode must be done with the help of collective coordinates . the jacobian of the relevant transformation can be deduced from the isotropic system . the field @xmath59 of small oscillations must not contain modes of @xmath69 with eigenvalues of the order @xmath33 , since these would vanish for @xmath32 . the discussion of the longitudinal part is given in @xcite . in dimensional regularization , all eigenvalues of @xmath68 which would be zero for @xmath14 are of order @xmath5 in @xmath86 dimensions . to avoid eigenvalues of the order @xmath87 in the propagator for the longitudinal fluctuations , all collective coordinates of the four - dimensional case must be retained in @xmath88 dimensions . therefore the field @xmath58 in ( [ exp ] ) contains no modes of @xmath68 with eigenvalues proportional to @xmath5 . in order to calculate the fluctuation factor and to separate the almost - zero modes from @xmath89 and @xmath90 , it is convenient to do a conformal transformation onto a sphere in @xmath91 dimensions leading to a discrete spectrum for the transformed differential operators @xmath68 and @xmath69 @xcite . their products of eigenvalues can be given in terms of the riemann @xmath92-function which we easily expanded near @xmath93 . simple @xmath87-poles , which are characteristic of dimensional regularization , appear directly from the known singularity of that function . in place of the fields @xmath58 and @xmath59 , we define the fields @xmath94 and @xmath95 on the unit sphere in @xmath91 dimensions : @xmath96 where @xmath97 . the instanton is supposed to be centered at the origin and rescaled by @xmath61 . the spherical operator corresponding to the differential operator @xmath98 is @xmath99 where @xmath100 is the total angular momentum operator on the @xmath101-dimensional sphere . the eigenfunctions of @xmath102 are the spherical harmonics @xmath103 in @xmath91 dimensions @xcite . they satisfy @xmath104 and have the degeneracy @xmath105 after the transformation , the angle - independent quadratic part of @xmath106 in ( [ h2 ] ) becomes @xmath107 \phi_2 \ , , \ ] ] where @xmath108 is the surface element of the unit sphere in @xmath91 dimensions . after rewriting the entire energy functional([hall ] ) in terms of the new fields , we can summarize the feynman rules for the diagrammatic evaluation of the functional integrals as follows : + @xmath109 propagators ( from the @xmath56-independent part of @xmath106 ) : + + longitudinal propagator : @xmath110^{-1}\ ] ] transversal propagator : @xmath111^{'}\right\}^{-1}\ ] ] the prime indicates that the almost - zero modes are omitted when forming the inverse . + @xmath112 tadpoles @xmath113 : @xmath114 the bold dot stands for the @xmath56-dependence of the vertex . + @xmath115 two - point vertices ( @xmath56-dependent part of @xmath106 ) : + @xmath116 @xmath117 cubic vertices @xmath118 : @xmath119 @xmath120 quartic vertices @xmath121 : @xmath122 in order to obtain the leading corrections we must consider all connected diagrams of order @xmath123 , @xmath124 , and @xmath125 . the contributions of these diagrams are added to @xmath126 in ( [ hclas ] ) . we consider first the contributions to @xmath123 , which arise from the connected tree diagrams @xmath127 and their @xmath128 insertions of zeroth order in @xmath39 : @xmath129 since each @xmath128-vertex produces an @xmath5-factor , the smallest order in @xmath5 is two . however , terms of order @xmath130 are negligible for our calculation even after one - loop renormalization , since : @xmath131 $ ] and @xmath132 . the diagrams of @xmath124 can be generated from those of @xmath123 by one of the substitutions : @xmath133 by inspection , we see that the only diagram with an @xmath5-power smaller than two is given by @xmath134 which is of order @xmath135 . after the renormalization , a term of order @xmath136 appears . but this term can be neglected in comparison with the @xmath137-term from @xmath126 in expression ( [ hclas ] ) . hence all the tree diagrams would enter only in a higher - order calculation . contributions to order @xmath138 can appear from one - loop diagrams . the only possible candidates are @xmath139 where the @xmath87-pole from the loop integration and the @xmath5-factor of the @xmath128-vertex cancel . following the method of @xcite , we have found that these diagrams contribute to the coefficient of the imaginary part of @xmath140 in ( [ tilg ] ) a factor @xmath141 the angle - independent quadratic form is decomposed into a longitudinal and a transverse part . for the longitudinal fluctuations we obtain the determinant @xmath142^{-\frac{1}{2 } } \!\!= \prod^{\infty}_{l=0 } \left[\frac{(l+\frac{1}{2 } d-3)(l+\frac{1}{2 } d+2 ) } { ( l+\frac{1}{2 } d-1)(l+\frac{1}{2 } d ) } \right]^{-\frac{1}{2 } \nu_l(d+1)}_{,}\ ] ] which coincides with that appearing in the one - component @xmath3 theory , and is therefore known @xcite . it contains a bound state at @xmath143 , this being responsible for the expected imaginary part of @xmath140 , and @xmath91 almost - zero eigenmodes of order @xmath5 associated with the dilatation and translation degrees of freedom of the instanton in @xmath86 dimensions . extracting these modes from the product ( [ lprod ] ) in the framework of collective coordinates @xcite , we obtain the well - known formal replacement rule for the determinant @xmath144 with @xmath145 \ , , \ ] ] and the jacobian of the collective coordinates transformation : @xmath146 \ , .\ ] ] the expression for @xmath147 contains the euler constant @xmath148 , and the derivative of the riemann - zeta function @xmath149 at the value @xmath150 . note the simple pole in @xmath5 , which is related to the ultraviolet divergence on the one - loop level . the transverse fluctuations contain neither negative nor zero modes for @xmath33 larger than zero . the corresponding fluctuation determinant is given by @xmath151 } { \det(-v_0 ) } \right\}^{-\frac{1}{2 } } \nonumber \\ \nonumber \\ & = & \prod^{\infty}_{l=0 } \left[\frac{(l+ \frac{1}{2 } d-1)(l+\frac{1}{2 } d)-2(1-\delta ) } { ( l+\frac{1}{2 } d-1)(l+\frac{1}{2 } d ) } \right]^{-\frac{1}{2 } \nu_l(d+1)}.\end{aligned}\ ] ] just one zero eigenmode appears for @xmath143 in the isotropic limit @xmath152 , due to the rotational invariance in that case . for @xmath78 , the numerator in ( [ tprod ] ) contributes a factor @xmath153 . to avoid this artificial zero - mode divergence for @xmath152 we separate , as announced earlier , the angular degree of freedom from the integral measure via the collective - coordinates method . similar to the longitudinal case , we make the formal substitution @xmath154 with @xmath155 which coincides with the jacobian of the isotropic model . since the integration interval for the angle @xmath56 is compact , no singularity appears in the limit @xmath152 . it is useful to illustrate the appearance of a divergent factor @xmath153 in a careless use of gaussian integral . assuming @xmath78 , we expand the angle - dependent classical action around @xmath156 up to quadratic order . the ensuing gaussian integral can be evaluated after replacing the finite integration interval @xmath157 $ ] by the noncompact one @xmath158 $ ] . in this way , the integral @xmath159\ ] ] is evaluated to @xmath160 = \frac{1}{\sqrt{2 \delta}}\ ] ] showing the spurious would - be zero - mode singularity , as the consequence of false rotation - mode treatment . excluding the @xmath143 -mode in the numerator on the right - hand side of ( [ tprod ] ) , we obtain for @xmath161 : @xmath162 \nonumber \right . \\ \nonumber \\ & & \left . -\frac{1}{2 } \sum_{l=1}^{\infty } \frac{(2l+d-1 ) \gamma(l+d-1 ) } { \gamma(d ) \gamma(l+1 ) } \ln \left [ 1+\frac{2\ , \delta}{(l+\frac{d}{2}+1 ) ( l+\frac{d}{2}-2 ) } \right ] \right\}. \nonumber \\\end{aligned}\ ] ] the first sum is the well - known contribution from the isotropic limit . the second sum is to be expanded in powers of @xmath33 . for large @xmath163 , the sum diverges as @xmath5 tends to zero . in order to separate off the divergence as a simple @xmath87-pole , we use the zeta - function regularization method described in @xcite . a somewhat tedious but straightforward calculation leads to @xmath164\ , .\ ] ] the series in front of the @xmath5-singularity ends after the quadratic power of @xmath33 , thereby leading to the exact @xmath165-coefficient of the simple @xmath5-pole . collecting all contributions to the one - loop expression of @xmath166 , we obtain the imaginary parts @xmath167 \right\}\end{aligned}\ ] ] with @xmath168\ ] ] and @xmath169 . the expression contains a standard factor @xmath170 , since for symmetry reasons each saddle point contributes only one half of the gaussian integral . it remains to perform the collective coordinates integration over the dilatation @xmath171 , translation @xmath172 , and the rotation @xmath173 degree of freedom . having obtained the imaginary parts of the green functions , we go over to the bare imaginary part of the vertex functions . taking the fourier transform and excluding the @xmath174-factors , an amputation of the external legs of @xmath175 leads to : @xmath176 \right\ } [ 1+{\cal o}(\varepsilon , g)]\end{aligned}\ ] ] with @xmath177 where @xmath178 is the modified bessel - function . the coefficient @xmath179 is given by @xmath180.\end{aligned}\ ] ] an @xmath5-singularity results from the ultraviolet divergence of one - loop diagrams . as discussed in the following section , this simple pole in @xmath5 is removed by a conventional coupling constant renormalization . the contribution from the imaginary part of the denominator in equation ( [ g2 m ] ) follows for @xmath181 in ( [ v2 m ] ) . we observe that it is of higher order in @xmath39 in comparison with the imaginary part of the numerator , and can therefore be neglected . so the imaginary part of ( [ tilg ] ) supplies the desired imaginary part of ( [ g2 m ] ) . in the preceding section we have calculated the contribution of the quadratic fluctuations around the isotropic instanton . the almost - zero eigenvalues of translation , dilatation and rotation have been extracted and treated separately . of course , the resulting expressions are useless , if we are not able to renormalize the theory . this is the most difficult additional problem which arises in the transition from quantum mechanics to higher - dimensional field theories . a systematic scheme to renormalize both the real and imaginary part of vertex functions for a @xmath3 theory in @xmath86 dimensions was introduced by mckane , wallace and de alcantara bonfim . they calculated the full ( real and imaginary ) renormalization - group constants using an extended minimal subtraction scheme to one loop ( the conventional one is given in @xcite ) . we have extended their method to the case of two coupling constants in view of applications to critical phenomena . for an illustration we consider first the four - point vertex function . the result ( [ v2 m ] ) of the one - loop calculation about the instanton reads : @xmath182^{2n } \nonumber \right . \\ \nonumber \\ & & \left . \mbox { } \times \left [ -c_{\rm b } \int\limits_0^{\infty } \frac{d\lambda}{\lambda } \lambda^{\varepsilon } \left(-\frac{\lambda^{\varepsilon}a}{g}\right)^{\frac{d+6 + 2n}{2 } } \!\!\ ! \exp \left(\frac{\lambda^{\varepsilon}a}{g } \right ) \prod^{4}_{\nu=1 } \left(\frac{q_{\nu}}{\lambda}\right)^2 \ ! \tilde{\phi } \left ( \frac{q_{\nu}}{\lambda } \right ) \right ] \right\}\end{aligned}\ ] ] with @xmath183 and @xmath184 . it remains to evaluate the integral over the parameter @xmath61 . this is a relict of the introduction of collective coordinates in the instanton calculation . the integral converges for small @xmath61 due to the exponentially decreasing of modified bessel - function @xmath185 . for large @xmath61 , the product @xmath186 behaves like @xmath187 \nonumber \\ & & + \left(\frac{q_{\nu}}{2\lambda}\right)^2 \bigg [ \left(\frac{q_{\nu}}{2\lambda}\right)^{\varepsilon } \left(\frac{2}{\varepsilon}+\gamma-\ln \pi+{\cal o}(\varepsilon)\right ) \nonumber \\ & & \quad -\left(\frac{2}{\varepsilon}+1-\gamma-\ln \pi + { \cal o}(\varepsilon ) \right ) \bigg ] + { \cal o}\left(\frac{q_{\nu}}{2\lambda}\right)^4 \bigg\ } \ , , \end{aligned}\ ] ] so that the integral diverges logarithmically as @xmath5 goes to zero . this divergence causes a @xmath87-pole in the imaginary part . since after the above approximation of @xmath188 the integral diverges for @xmath189 , a small @xmath61 cutoff @xmath190 has to be introduced . in this way , the @xmath61-integral in ( [ pol ] ) takes the form @xmath191 \ , .\ ] ] the coupling constant @xmath39 is chosen to lie on top of the tip of the left - hand cut in the complex @xmath39-plane . being integrated only near @xmath192 , the integral can be evaluated perturbatively in @xmath39 . using @xmath193= \frac{1}{\varepsilon}\frac{|g|}{a } \mu^{a\varepsilon } \exp\left[-\left(\frac{\mu^{\varepsilon } a}{|g|}\right)\right ] [ 1+{\cal o}(g)]\ , , \ ] ] we get for ( [ scal ] ) the expression @xmath194 after expanding the factor @xmath195 in powers of @xmath5 , we obtain the typical finite contribution @xmath196 . since this term can be omitted by a minimal subtraction , we are left with @xmath197 \right\}\!,\end{aligned}\ ] ] where @xmath198 and @xmath199 the tensor structure arises explicitly upon doing the integrals @xmath200 for all combinations of indices in @xmath201 . the first term in ( [ gim ] ) contains only mixed index combinations , whereas the second term has only a contribution from equal indices . now we can proceed to renormalize the bare results . the real part of @xmath202 is the perturbative one , and it is easily calculated to one loop . in terms of our coupling constants , the real parts @xmath203 take the form @xmath204 g^2 \mu^{-\varepsilon } \right\ } \ { s_{ijkl}-\delta_{ijkl } \ } \nonumber \\ \nonumber \\ & & \!\!\!\!\ ! + \left\{-6g+\left[\frac{3}{4\pi^2\varepsilon } \left(10 - 2 \delta+{\delta}^2\right ) + { \cal o}({\varepsilon}^0 ) \right ] g^2 \mu^{-\varepsilon } \right\ } \ , \delta_{ijkl}\ , , \end{aligned}\ ] ] where @xmath190 is the arbitrary momentum scale introduced above in ( [ scal ] ) . in the absence of an instanton , the wave function renormalization has no one - loop contribution . in the consequence , the expression ( [ greal ] ) is rendered finite by a coupling constant renormalization only . the subscript r of the coupling constants indicates renormalized expressions in the conventional perturbative manner , i. e. , @xmath205 inserting this into ( [ gim ] ) and ( [ greal ] ) we find the perturbatively renormalized vertex functions @xmath206\ ] ] with @xmath207 and @xmath208 the coefficient @xmath209 is given by @xmath210 \nonumber \\ \nonumber \\ & = & \!\!\ ! 3^{1/2 } 2^{-2/3 } \pi^{-2 } \exp \bigg[\frac{1}{3\varepsilon } \left(10 - 2 { \delta}_{\rm r}+{\delta}_{\rm r}^2 \right)- \frac{1}{3\varepsilon } \left(10 - 2 { \delta}_{\rm r}+{\delta}_{\rm r}^2 \right ) \nonumber \\ \nonumber \\ & & \qquad \qquad \qquad \quad + \frac{4\zeta^{'}(2)}{\pi^2}-\frac{7}{3}\gamma-\frac{2}{3 } + { \cal o}({\delta}_{\rm r } ) + { \cal o}({\delta}_{\rm r } \ , g_{\rm r } ) \bigg].\end{aligned}\ ] ] the @xmath87-pole terms cancel . thus , @xmath209 remains finite for @xmath132 . for the leading expansion ( [ img ] ) we can take the perturbative renormalized expression ( [ cr ] ) at the position @xmath211 . the only remaining singularity is the @xmath87-factor in @xmath212 . it requires a further renormalization . in our special choice of coupling constants , the disastrous @xmath213-divergence of perturbation series appears in the expansions coefficients of @xmath44 . since @xmath33 is a dimensionless anisotropy measure for positive as well as negative @xmath39 , we have expanded the imaginary part of vertex functions around the isotropic case in a well defined manner . for the derivation of nonperturbative corrections to the renormalization - group functions and for later applications , however , it is convenient to avoid ratios of coupling constants . therefore we proceed with @xmath214 and @xmath215 instead of @xmath136 and @xmath215 . then @xmath216 receive the rule of @xmath136 . now we apply the extended minimal subtraction and perform a second ( nonperturbative ) renormalization step to eliminate the @xmath87-pole term : @xmath217\ , , \nonumber \\ \nonumber \\ \nonumber \\ v_{\rm r}\!\!&=&\!\!v_{\rm r}-i 2 ^ 7 3 \pi^4 c_{\rm r } \frac{1}{\varepsilon } \displaystyle{\sum_{n=0}^{\infty } } \bigg [ \frac{v_{\rm r}^n}{n ! } \left(\frac{3}{4\pi^2}\right)^n \left(\frac{4n}{2n+3}\right ) b \left(n+\frac{1}{2},n+\frac{5}{2 } \right ) \nonumber \\ & & \qquad \qquad \qquad \quad \times \left(-\frac{a}{g_{\rm r } } \right)^{\frac{d+4 + 4 n}{2 } } \exp \left(\frac{a}{g_{\rm r } } \right ) \bigg]\ , .\end{aligned}\ ] ] the subscript r denotes the fully renormalized coupling constants . the required wave function renormalization in nonperturbative terms give no leading contributions to the imaginary part ( see below ) . therefore the @xmath87-singularity in ( [ gren ] ) is removed by a nonperturbative coupling - constant renormalization only . we now proceed to investigate the two - step renormalization for the wave function . there are two additional complications for the @xmath218-functions . first , @xmath219 is quadratically divergent . second , there is a momentum - dependent divergence in the imaginary part of @xmath218 . for the @xmath3-theory it was shown in @xcite that the undesirable momentum - dependence disappears during the nonperturbative renormalization process . this is comparable to the situation in conventional two - loop perturbation expansion . we continue to show that this non - obvious cancellation still works in the more complex case of two coupling constants and more than one field components . according to ( [ v2 m ] ) the imaginary part of the unrenormalized two - point vertex function reads @xmath220 ^ 2 \nonumber \right.\\ \nonumber \\ & & \left . \qquad \qquad \times \frac{1}{2 } \int\limits_0^{2\pi } d \vartheta u_{l\,i}(\vartheta ) u_{l\,j}(\vartheta ) \exp\left[a \sin^2(2\vartheta)\right ] \right\}.\end{aligned}\ ] ] the integration over the angle @xmath56 can be done : @xmath221 } \nonumber \\ \nonumber \\ & & \!\!\!\!\!\!\ ! = \delta_{ij } \sum_{n=0}^{\infty } \frac{a^n}{n!}\ , 4^n \frac{\gamma\left(n+ \frac{1}{2 } \right ) \gamma\left(n+\frac{3}{2 } \right ) } { \gamma(2n+2 ) } = \delta_{ij } \sum_{n=0}^{\infty } \frac{a^n}{n!}\ , 4^n b \left(n+ \frac{1}{2},n+\frac{3}{2 } \right ) . \nonumber \\\end{aligned}\ ] ] thus , we get the usual form : @xmath222 . the wave function renormalization is contained in @xmath223 . following a similar procedure as in the case of @xmath224 , we obtain the imaginary part @xmath225 \left(-\frac{\mu^{\varepsilon}a}{g } \right)^{\frac{d+2 + 2n}{2 } } \exp\left ( \frac{\mu^{\varepsilon}a}{g}\right ) \right\ } \ , . \nonumber \\\end{aligned}\ ] ] apart from the appearance of the @xmath226 term , a very similar expression is found for @xmath227 . in order to show the cancellation of this term during the second stage of renormalization we need the two - loop expression for the real part of @xmath223 @xmath228 \ , , \ ] ] where p denotes some number which is the contribution of order @xmath229 . at @xmath230 , ( [ wfreal ] ) is commonly defined as the inverse wave function renormalization constant @xmath231 . in terms of the perturbative renormalized coupling constants @xmath232 is given by @xmath233 and together with ( [ wfim ] ) , we find after the first step of renormalization @xmath234 \nonumber \\ \nonumber \\ & + & \!\!i\,2 ^ 4 \pi^2 3 \,c_{\rm r}\ , \frac{1}{\varepsilon } \sum_{n=0}^{\infty } \left\{\frac{v_{\rm r}^n}{n ! } \left(\frac{3}{4\pi^2}\right)^n b \left(n+\frac{1}{2},n+\frac{3}{2 } \right ) \nonumber \right . \\ \nonumber \\ & & \left . \mbox { } \times \left[\gamma-\frac{1}{2}+\ln \left(\frac{q}{2\mu}\right ) \right ] \left(-\frac{a}{g_{\rm r } } \right)^{\frac{d+2 + 4n}{2 } } \exp\left ( \frac{a}{g_{\rm r}}\right ) \right\ } \ , .\end{aligned}\ ] ] we have again introduced the coupling constants @xmath216 and @xmath215 which are more convenient for practical applications . this expression reads in the fully renormalized couplings ( [ ren ] ) @xmath235 \nonumber \\ \nonumber \\ & & \times \left(-\frac{a}{g_{\rm r } } \right)^{\frac{d+2 + 4n}{2 } } \exp\left ( \frac{a}{g_{\rm r}}\right ) \bigg\ } \ , .\end{aligned}\ ] ] by using the definition of the @xmath236-function in terms of gamma - functions one can easily read off the cancellation of @xmath226 singularities for every power @xmath43 of the coupling constant @xmath237 . therefore we are left with @xmath238 \nonumber \\ \nonumber \\ & + & \!\ ! i\,2 ^ 4 \pi^2 3 \,c_{\rm r}\ , \frac{1}{\varepsilon } \left(2 ^ 4 \pi^2 a\,\,{\rm p}+\gamma-\frac{1}{2}-\ln2 \right ) \nonumber \\ \nonumber \\ & \displaystyle{\times}&\!\!\ ! \sum_{n=0}^{\infty } \bigg[\frac{v_{\rm r}^n}{n ! } \left(\frac{3}{4\pi^2}\right)^n b \left(n+\frac{1}{2},n+\frac{3}{2 } \right ) \left(-\frac{a}{g_{\rm r } } \right)^{\frac{d+2 + 4n}{2 } } \exp\left ( \frac{a}{g_{\rm r}}\right ) \bigg ] \ , . \nonumber \\\end{aligned}\ ] ] now we can define the nonperturbative minimally subtracted wave function renormalization to remove the @xmath87-pole : @xmath239 \ , .\end{aligned}\ ] ] the full renormalization constant is defined by @xmath240 and follows from the product of ( [ zp ] ) and ( [ znp ] ) . a further important green function is associated with the composite field @xmath241 : @xmath242 by a straightforward application of the techniques discussed in the preceding sections we obtain for the corresponding vertex function @xmath243 \hat{\phi } \left(\frac{q_3}{\lambda}\right ) \frac{1}{2 } \int\limits_0^{2\pi } d \vartheta u_{l\,i}(\vartheta ) u_{l\,j}(\vartheta ) \exp\left[a \sin^2(2\vartheta)\right ] \bigg\ } \ , , \nonumber \\\end{aligned}\ ] ] where @xmath244 now we can proceed as in the case of wave function renormalization . performing the collective coordinates integration over @xmath61 and @xmath56 we obtain @xmath245 \left(-\frac{\mu^{\varepsilon}a}{g } \right)^{\frac{d+4 + 2n}{2 } } \exp\left ( \frac{\mu^{\varepsilon}a}{g}\right ) \right\ } , \nonumber \\\end{aligned}\ ] ] which agrees with the usual form @xmath246 . at @xmath247 , @xmath248 , the vertex function @xmath249 defines a third renormalization constant , called @xmath250 , via @xmath251 the perturbative @xmath250 follows from the real part of @xmath252 : @xmath253 where @xmath254 is a number of order @xmath229 . in terms of the perturbative renormalized coupling constants , the one - loop expression of @xmath255 is given by @xmath256 including @xmath257 and @xmath258 we obtain after the perturbative step of renormalization @xmath259 \nonumber \\ \nonumber \\ & + & \!\!i\,2 ^ 4 \pi^2 3 ^ 2 \,c_{\rm r}\ , \frac{1}{\varepsilon } \sum_{n=0}^{\infty } \left\{\frac{v_{\rm r}^n}{n ! } \left(\frac{3}{4\pi^2}\right)^n b \left(n+\frac{1}{2},n+\frac{3}{2 } \right ) \nonumber \right . \\ \nonumber \\ & & \left . \mbox { } \times \left[\gamma+\ln \left(\frac{q_3}{2\mu}\right ) \right ] \left(-\frac{a}{g_{\rm r } } \right)^{\frac{d+4 + 4n}{2 } } \exp\left ( \frac{a}{g_{\rm r}}\right ) \right\ } \ , .\end{aligned}\ ] ] in the second step of renormalization the @xmath260 singularities cancel , and we find @xmath261 \nonumber \\ \nonumber \\ & & \quad -i\,2 ^ 4 \pi^2 3 ^ 2 \,c_{\rm r}\ , \frac{1}{\varepsilon } \left(\frac{8\pi^2}{3 } { \rm q}-\gamma+\ln2 \right ) \nonumber \\ \nonumber \\ & & \quad \displaystyle{\times } \sum_{n=0}^{\infty } \bigg[\frac{v_{\rm r}^n}{n ! } \left(\frac{3}{4\pi^2}\right)^n b \left(n+\frac{1}{2},n+\frac{3}{2 } \right ) \left(-\frac{a}{g_{\rm r } } \right)^{\frac{d+4 + 4n}{2 } } \exp\left ( \frac{a}{g_{\rm r}}\right ) \bigg ] \ , . \nonumber \\\end{aligned}\ ] ] hence , the nonperturbative @xmath250 is @xmath262.\end{aligned}\ ] ] with respect to the leading expansion ( [ img ] ) this is equal to @xmath263 . after having calculated the imaginary parts of the renormalized couplings and of the renormalization constants @xmath264 and @xmath250 , we are in the position to derive the nonperturbative renormalization group functions @xmath265 , @xmath266 and @xmath267 in the usual way . first we derive the @xmath268-functions from the expressions ( [ ren ] ) for @xmath269 and @xmath270 , respectively , and obtain : @xmath271\ , , \nonumber \\ \lefteqn{\beta_v(g_{\rm r},v_{\rm r})= } \nonumber \\ & & \mu \frac{\partial}{\partial \mu } v_{\rm r } = \mu \frac{\partial}{\partial \mu } v_{\rm r}- i 2 ^ 7 3 \pi^4 c_{\rm r } \frac{1}{\varepsilon } \displaystyle{\sum_{n=0}^{\infty } } \mu \frac{\partial}{\partial \mu } \bigg [ \frac{v_{\rm r}^n}{n ! } \left(\frac{3}{4\pi^2}\right)^n b \left(n+\frac{1}{2},n+\frac{5}{2 } \right ) \nonumber \\ \nonumber \\ & & \qquad \qquad \qquad \qquad \qquad \times \left(\frac{4n}{2n+3}\right ) \left(-\frac{a}{g_{\rm r } } \right)^{\frac{d+4 + 4 n}{2 } } \exp \left(\frac{a}{g_{\rm r } } \right ) \bigg]\ , .\end{aligned}\ ] ] the derivatives are taken at fixed bare couplings @xmath39 and @xmath1 . these functions will be finite as @xmath132 . furthermore , they will have a left - hand cut in the complex @xmath39-plane . on top of the tip of this cut we find @xmath272 \ , , \nonumber \\ \nonumber \\ \beta_v(g_{\rm r},v_{\rm r})= \beta_v^{\rm p}(g_{\rm r},v_{\rm r } ) \!\!\!\!&+&\!\!\!\!i 2 ^ 7 3 \pi^4 c_{\rm r } \displaystyle{\sum_{n=0}^{\infty } } \bigg [ \frac{v_{\rm r}^n}{n ! } \left(\frac{3}{4\pi^2}\right)^n b \left(n+\frac{1}{2},n+\frac{5}{2 } \right ) \nonumber \\ \nonumber \\ & & \times \left(\frac{4n}{2n+3}\right ) \left(-\frac{a}{g_{\rm r } } \right)^{\frac{d+6 + 4 n}{2 } } \exp \left(\frac{a}{g_{\rm r } } \right ) \bigg ] \ , .\end{aligned}\ ] ] the real parts are the familiar perturbative @xmath268-functions @xmath273 in the leading contributions the pair of coupling constants @xmath274 may be replaced by @xmath275 in ( [ bet1 ] ) . this can be proven using ( [ ren ] ) . the result is @xmath276\ , , \nonumber \\ \beta_v(g_{\rm r},v_{\rm r})= \beta_v^{\rm p}(g_{\rm r},v_{\rm r } ) \!\!\!\!&+&\!\!\!\!i 2 ^ 7 3 \pi^4 c_{\rm r } \displaystyle{\sum_{n=0}^{\infty } } \bigg [ \frac{v_{\rm r}^n}{n ! } \left(\frac{3}{4\pi^2}\right)^n b \left(n+\frac{1}{2},n+\frac{5}{2 } \right ) \nonumber \\ \nonumber \\ & & \times \left(\frac{4n}{2n+3}\right ) \left(-\frac{a}{g_{\rm r } } \right)^{\frac{d+6 + 4 n}{2 } } \exp \left(\frac{a}{g_{\rm r } } \right ) \bigg ] \ , .\end{aligned}\ ] ] note that the @xmath268-functions have a form expected within a minimal subtracted scheme : @xmath277 and @xmath278 , where the @xmath279 denotes the @xmath268-functions in four dimensions . the critical large - distance behavior of the correlation functions is defined by the anomalous dimension of the field @xmath37 , which is defined by @xmath280 with ( [ zp ] ) for @xmath232 and ( [ znp ] ) for @xmath281 we get the @xmath5-independent expression @xmath282 \ , .\end{aligned}\ ] ] the divergence of the length scale at critical temperature is governed by the anomalous dimension of the composite field @xmath283 : @xmath284 using ( [ ppp ] ) and ( [ nppp ] ) , we obtain @xmath285 . \nonumber \\\end{aligned}\ ] ] for the leading contributions to the imaginary part , it was possible to replace @xmath274 by @xmath275 . we remark that the expansions ( [ betar]),([gamr ] ) and ( [ npp2 ] ) have the correct isotropic limit for @xmath286 . generalizing ( [ exp ] ) to @xmath22 we write : @xmath287 where @xmath27 and @xmath56 are the rotation angles of the isotropic instanton . this change of variables yields the following expansion of the energy functional : @xmath288 ^ 2 } \xi \nonumber \\ \nonumber \\ & -&\!\!\ ! ( -8g)^{1/2 } \int d^dx \frac{\lambda}{1+\lambda^2(x - x_0)^2 } \left[\xi^3+(1-\delta)(\xi \eta^2+\xi \chi^2 ) \right ] \nonumber \\ \nonumber \\ & + & \!\!\ ! \delta \,\ , f(g,\vartheta,\theta,\xi,\eta,\chi)\end{aligned}\ ] ] with @xmath68 and @xmath289 of equation ( [ mlt ] ) . for brevity , we have written down explicitly only the terms responsible for the leading contributions in the expansion ( [ img ] ) . the remaining terms are denoted collectively by @xmath290 . the classical contribution to the energy functional is @xmath291 \nonumber \\ \nonumber \\ & & \!\!\!\!\ ! -\frac{\lambda^{\varepsilon } \delta } { 2 g } \ , \frac{8\pi^2}{3 } \left[\sin^4{\theta } \sin^2(2\vartheta)+\sin^2(2\theta ) \right]+ { \cal o}\left(\frac { \delta } { g}\ , \varepsilon \right).\end{aligned}\ ] ] with similar steps as for @xmath21 we obtain the bare imaginary parts of the vertex functions : @xmath292 \bigg\ } \left[1+{\cal o}(\varepsilon , g ) \right ] \ , , \end{aligned}\ ] ] where @xmath293 \ , , \quad ( i_k\ , = \ , 1,2,3)\ , , \end{aligned}\ ] ] with @xmath294 , @xmath295 as in ( [ aa ] ) . after evaluating the integrals over the collective coordinates @xmath61 , @xmath27 , and @xmath56 we obtain for the imaginary part of the four - point vertex function @xmath296 \bigg\}\ , .\end{aligned}\ ] ] similarly the bare imaginary parts of @xmath297 and @xmath298 are found to be @xmath299 } \nonumber \\ \nonumber \\ & & \times \frac{1}{\varepsilon } \sum_{n=0}^{\infty } \bigg[\frac{(-2\ , \delta)^n}{n ! } \ , i(n ) \ , \left(-\frac{\mu^{\varepsilon}a}{g } \right)^{\frac{d+3 + 2n}{2 } } \exp\left ( \frac{\mu^{\varepsilon}a}{g}\right ) \bigg ] \ , , \end{aligned}\ ] ] and @xmath300 \left(-\frac{\mu^{\varepsilon}a}{g } \right)^{\frac{d+5 + 2n}{2 } } \exp\left ( \frac{\mu^{\varepsilon}a}{g}\right ) \right\ } , \nonumber \\\end{aligned}\ ] ] with @xmath301 now we apply the extended renormalization scheme . using the coupling constants @xmath1 and @xmath39 , we obtain for the nonperturbative renormalized couplings @xmath302 \ , , \nonumber \\ \nonumber \\ v_{\rm r}\!\!\!&=&\!\!\!v_{\rm r } -i 2 ^ 7 3 \pi^4 c_{\rm r}^{(3 ) } \frac{1}{\varepsilon } \displaystyle{\sum_{n=0}^{\infty } } \bigg [ \frac{v_{\rm r}^n}{n ! } \left(\frac{3}{4\pi^2}\right)^n j_v(n ) \left(-\frac{a}{g_{\rm r } } \right)^{\frac{d+5 + 4 n}{2 } } \exp \left(\frac{a}{g_{\rm r } } \right ) \bigg]\ , , \nonumber \\\end{aligned}\ ] ] with @xmath303 \ , b \left(3 + 2n - p , p+\frac{1}{2}\right ) \nonumber \\ \nonumber \\ & & \qquad \qquad \qquad \qquad \quad \times b \left(n - p+\frac{1}{2},n - p+\frac{5}{2 } \right ) \bigg\ } \ , , \end{aligned}\ ] ] and @xmath304\ , .\ ] ] for the nonperturbative renormalization constants @xmath264 and @xmath250 we find @xmath305 \,\end{aligned}\ ] ] and @xmath306.\end{aligned}\ ] ] to calculate the nonperturbative renormalization group functions @xmath265 , @xmath266 and @xmath267 , we simply repeat the calculations of the case @xmath21 . using the expressions ( [ c3r ] ) for @xmath307 and @xmath237 yields the @xmath268-functions @xmath308\ , , \nonumber \\ \lefteqn{\beta_v(g_{\rm r},v_{\rm r})= } \nonumber \\ & & \beta_v^{\rm p}(g_{\rm r},v_{\rm r } ) + i 2 ^ 7 3 \pi^4 c_{\rm r}^{(3 ) } \displaystyle{\sum_{n=0}^{\infty } } \bigg [ \frac{v_{\rm r}^n}{n ! } \left(\frac{3}{4\pi^2}\right)^n j_v(n ) \left(-\frac{a}{g_{\rm r } } \right)^{\frac{d+7 + 4 n}{2 } } \exp \left(\frac{a}{g_{\rm r } } \right ) \bigg]\ , . \nonumber \\ \nonumber \\\end{aligned}\ ] ] the expressions for the nonperturbative anomalous dimension of the field @xmath37 and the composite field @xmath283 are given by @xmath309 \ , , \end{aligned}\ ] ] and @xmath310 \ , , \end{aligned}\ ] ] respectively . our main results are the real and imaginary parts ( [ b3])([npp3 ] ) of the renormalization - group functions @xmath268 , @xmath148 and @xmath311 in terms of the nonperturbative renormalized couplings @xmath307 and @xmath237 . via the dispersion relation ( [ dr ] ) , we obtain the large - order behavior ( [ lo2 ] ) . it is useful to check the expansion terms proportional to powers of @xmath312 accompanying the @xmath313 in the imaginary part of each renormalization - group function . if the power of @xmath312 is denoted by @xmath314 , we note that @xmath314 is the same as for the corresponding vertex function before the integration over the dilatation parameter @xmath61 . this integration with the measure @xmath315 reduces @xmath314 by @xmath316 [ see ( [ lambda ] ) ] . however this effect is compensated by the differentiation with respect to the logarithm of the scale parameter @xmath190 in the definition of the renormalization group functions . the power @xmath314 in the vertex functions can simply be explained : for a @xmath317-point vertex function with @xmath318 @xmath283-insertions there is first a power @xmath319 from the fields . this is multiplied by a factor @xmath320 from the @xmath88 translational , @xmath321 rotational , and one dilatational would - be zero modes . the classical contribution to the energy functional yields a factor @xmath322 for each power of the anisotropic constant @xmath1 . hence , for the renormalization group function which follows at the one - loop level from the vertex function @xmath323 , the value of @xmath314 is given by @xmath324 inserting the imaginary parts into a dispersion relation ( [ dr ] ) , we can estimate the large - order coefficients of the corresponding series expansions @xmath325 where @xmath326 . the function @xmath327 stands for @xmath328 , @xmath329 , @xmath148 and @xmath311 , respectively . the rules how to go from the imaginary parts of @xmath327 to @xmath330 are known from ( [ lo2 ] ) . for instance , the large - order coefficients of the @xmath268-functions for @xmath22 are given by @xmath331 \ , , \nonumber \\ \nonumber \\ \beta_{v , \,kn}\!\!&=&\!\!-c_{\rm r}^{(3)}2 ^ 7 3 \pi^3 \left(\frac{3}{4\pi^2}\right)^n \frac { j_v(n ) } { n ! } \nonumber \\ \nonumber \\ & & \qquad \times ( -1)^k \left(\frac{3}{8\pi^2}\right)^k k ! \ , k^{2n+(d+5)/2 } \left[1+{\cal o}(1/k)\right ] \ , , \end{aligned}\ ] ] with @xmath332 , @xmath333 from ( [ j ] ) and ( [ cc33 ] ) , respectively . the leading large-@xmath318 behavior can be checked by means of a simple combinatorial analysis . we start with a series in terms of the standard couplings @xmath23 and @xmath1 [ see ( [ hn ] ) ] . for illustration , we consider the contribution of the pure @xmath23-powers . it is known from a theory with only one coupling constant , that the coefficients have the large - order behavior @xmath334 changing over to our couplings ( @xmath335 ) , the right - hand side of ( [ lao ] ) contributes @xmath336 with @xmath337 replacing @xmath163 by @xmath338 , the contribution of @xmath339 has the form @xmath340 now , for large @xmath318 , the @xmath341-functions can be approximated by @xmath342 thus in the region @xmath326 we obtain @xmath343 in agreement with ( [ concl ] ) . as stated in the introduction , the stable cubic fixed point is expected to lie in the vicinity of the isotropic fixed point . since @xmath1 becomes very small in this region , reasonable results should be obtained by resumming the @xmath39-series accompanying each power @xmath313 . this will be done in a forthcoming publication .

using a renormalization group approach , we determine the phase diagram of an extended quasi - one - dimensional electron gas model that includes interchain hopping , nesting deviations and both intrachain and interchain repulsive interactions . -wave superconductivity , which dominates over the spin - density - wave ( sdw ) phase at large nesting deviations , becomes unstable to the benefit of a triplet-wave phase for a weak repulsive interchain backscattering term , despite the persistence of dominant sdw correlations in the normal state . antiferromagnetism becomes unstable against the formation of a charge - density - wave state when exceeds some critical value . while these features persist when both umklapp processes and interchain forward scattering ( ) are taken into account , the effect of alone is found to frustrate nearest - neighbor interchain- and-wave pairing and instead favor next - nearest - neighbor interchain singlet or triplet pairing . we argue that the close proximity of sdw and charge - density - wave phases , singlet-wave and triplet-wave superconducting phases in the theoretical phase diagram provides a possible explanation for recent puzzling experimental findings in the bechgaard salts , including the coexistence of sdw and charge - density - wave phases and the possibility of a triplet pairing in the superconducting phase .

cond-mat0510744

the theory of low - dimensional metals has exerted a strong influence on our understanding of ordered phases in quasi - one - dimensional ( quasi-1d ) organic conductors . the description of the extensively studied bechgaard salts series ( ( tmtsf)@xmath5x ) and their sulfur analogs , the fabre ( ( tmttf)@xmath5x ) salts , has served over more than two decades to illustrate this view.@xcite for these materials , a direct correspondence can be traced between the various modulated spin and charge ordered states of their phase diagram and the possible states of the quasi-1d electron gas model when the couplings are repulsive and the filling of the band is commensurate with the underlying lattice.@xcite in its standard form , this generic model is defined by the phenomenological expression of the direct interaction between electrons in terms of weak _ _ intra__chain backward ( @xmath6 ) and forward ( @xmath7 ) electron - electron scattering processes , to which umklapp ( @xmath8 ) scattering amplitudes are added at commensurate band filling.@xcite the quasi-1d character of the model is defined by adding an interchain single electron hopping integral @xmath9 , which is at least an order of magnitude smaller than its longitudinal counterpart . in virtually all compounds of the above series , superconductivity is observed next to a spin - density - wave state for some critical value on the pressure scale,@xcite whereas antiferromagnetic spin correlations are found to dominate the metallic state precursor to superconductivity over a wide range of temperatures.@xcite on the theoretical side , however , for spin - independent repulsive couplings and for a fermi surface with good nesting properties , the coexistence of spin - density - wave ( sdw ) and superconducting ( sc ) correlations is essentially excluded from the model phase diagram.@xcite it is only when deviations from perfect electron - hole symmetry ( nesting ) are introduced and the long - range component of the sdw order is suppressed that superconductivity can be actually realized in place of magnetism.@xcite in the anisotropic metallic phase where this suppression takes place , interchain cooper pairing is enhanced and superconductivity emerges from the coupling between the weakened electron - hole and the still singular electron - electron scattering channels . recent calculations @xcite using the renormalization group ( rg ) method did confirm the existence of such an electronic pairing mechanism beyond the level of single - channel rpa - like approaches.@xcite a smooth crossover from the sdw state to superconductivity has then been found whenever the amplitude of nesting deviations reaches some threshold a result in accordance with the sequence of transitions observed as a function of pressure.@xcite the rg approach shows nevertheless that whenever antiferromagnetism stands out as the dominant correlation in the normal state , the most stable interchain pairing is invariably a spin singlet state corresponding to a ` @xmath0-wave ' symmetry gap with nodes on the fermi surface . although some experimental findings in the bechgaard salts and their sulfur analogs do agree with this type of pairing , @xcite other series of observations have rather been interpreted in support of a triplet order parameter,@xcite thus challenging the singlet scenario for superconductivity . though there is too little so far to favor one scenario over another , these observations bring us to the question of whether triplet superconductivity can be possible or not when short - range antiferromagnetic correlations are dominant in the metallic state . as is well known for the electron gas model , there is a region of the phase diagram where triplet ` @xmath10-wave ' superconductivity does exist as the most stable state . it has been suggested , on a phenomenological basis , that such a state is realized in the bechgaard salts.@xcite within a microscopic approach , the region where @xmath10-wave superconductivity is stable is defined by irrelevant umklapp scattering and by a backward scattering coupling that is much larger than forward scattering . however , this description pattern for superconductivity can be considered unsatisfactory given the unrealistic constraint it puts on the coupling constants , and for the suppression of both the antiferromagnetism at short distance and the mott pseudo gap in the charge sector.@xcite these flaws can hardly be reconciled with the related phenomenology of these molecular compounds observed around the critical pressure for superconductivity , when either the temperature or the magnetic field is varied.@xcite another way to look at this problem is to consider more closely the effect of charge fluctuations on superconductivity . staggered charge fluctuations are known to favor triplet pairing at odd but large angular momentum , a mechanism whose roots go back to the early work of kohn and luttinger about cooper pairing in the presence of charge friedel oscillations in isotropic fermi systems.@xcite for the quasi-1d electron gas model and its version for lattice electrons , recent calculations have shown that triplet @xmath1-wave pairing is indeed enhanced when intrachain couplings are chosen to boost charge - density - wave ( cdw ) fluctuations close to the level found in the sdw channel.@xcite however , for realistic repulsive couplings , @xmath0-wave pairing still remains tied to the highest critical temperature and hence to the most stable state for superconductivity . all this goes to establish the robustness of @xmath0-wave pairing for the model with repulsive intrachain couplings and nesting deviations . it turns out , however , that the model is incomplete when charge fluctuations are found along the chains since then _ _ inter__chain coulomb interaction is also present in practice . the inclusion of direct interchain electron - electron scattering processes , which will be denoted by @xmath11 in the following , defines the quasi-1d electron gas model in its extended form.@xcite at large momentum transfer , the interchain interaction is well known to favor a cdw ordered state.@xcite this mechanism is mostly responsible for cdw long - range order observed in several organic and inorganic low - dimensional solids.@xcite the physical relevance of interchain interactions in the bechgaard salts , besides the intrachain @xmath12 and @xmath9 , is supported by x - ray studies , which revealed that the sdw phase of these compounds is actually accompanied by cdw order.@xcite on the theoretical grounds , very little is known about the impact of adding direct interchain interactions on the structure of the phase diagram , especially in the repulsive sector when both a finite @xmath9 and nesting deviations are present . in this work we wish to determine the possible density - wave and superconducting states of the extended quasi-1d electron - gas model . in order to tackle this problem we shall apply the renormalization group method , which at the one - loop level has proved to be suited to reach a controlled description of interfering density - wave and superconducting channels of correlations . among the results reported below , we have the unexpected finding that a small repulsive interchain backscattering term @xmath2 is sufficient to make @xmath0-wave superconductivity unstable to the benefit of a triplet @xmath1-wave phase . this occurs despite dominant sdw correlations in the metallic state and stable itinerant antiferromagnetism at lower nesting deviations . under the latter conditions , sdw order becomes in turn unstable to the formation of a cdw state when the amplitude of @xmath3 exceeds some critical value . while these features persist when commensurability effects are taken into account and small half - filling umklapp scattering is included , the effect of interchain forward scattering is found to frustrate nearest - neighbor interchain @xmath0- and @xmath1-wave pairing and to favor instead superconductivity with next - nearest - neighbor interchain pairing . part of these results have been reported in refs . . in section [ technique ] , we introduce the model and the rg scheme employed for the four point vertices and the response functions . by way of illustration , the rg results at the one - loop level are given for purely intrachain interactions . in section [ innter ] , we present the results for non zero interchain backward and forward interactions by which the different possibilities of ordered states in the phase diagram are obtained in the incommensurate case . the influence of umklapp processes in the half - filled case is examined in section [ umklapp ] . a discussion of the results is given in section [ nnandexp ] , where a possible connection between theory and experiments is made . we consider a lattice of @xmath13 coupled metallic chains described by the partition function @xmath14 , which is expressed as a functional integration over anticommuting ( grassmann ) fields @xmath15 . here @xmath16 and @xmath17 are the noninteracting and interacting parts of the action , respectively . the former part is given by @xmath18 where @xmath19 , @xmath20 , @xmath21 is the fermionic matsubara frequency , @xmath22 , the wave vector , and @xmath23 , the spin of the fermion field . the kinetic energy is given by @xmath24 where @xmath25 is the chemical potential taken as temperature independent . here we have linearized the spectrum in the chain direction , using @xmath26 as the longitudinal fermi velocity , @xmath27 being the parallel fermi wave vector if @xmath28 . throughout this work both the chain lattice constant @xmath29 and the interchain distance @xmath30 are put equal to unity . the interchain single electron hopping @xmath9 is considered small with respect to the longitudinal bandwidth @xmath31 . in the following , we will take @xmath32 , which is a typical figure for the anisotropy ratio in quasi-1d conductors like the bechgaard and fabre salts.@xcite the next nearest - neighbor hopping @xmath33 in the transverse direction gives the amplitude of nesting deviations . in eq . ( [ perpkin ] ) , we have neglected the possibility of hopping in the third direction , which does not have any sizeable effect on our calculations . its existence is of course crucial for the stabilization of true long - range order at finite temperature . in the framework of the quasi-1d electron gas model,@xcite the electron - electron interaction is parameterized by means of the g - ology approach . one first distinguishes between right and left moving fermions , depending on their velocity along the chains , so that the grassmann variables @xmath34 become @xmath35 using these definitions , the interaction part of the action takes the form @xmath36 where @xmath37 is the number of lattice sites and @xmath38 the reciprocal lattice vector along the chains , which is involved in the umklapp scattering at half - filling . here we have omitted the so - called @xmath39 contribution for the interaction of electrons of the same branch since this coupling does not contribute to the singular channels of correlation in the renormalization group flow at the one - loop level . if we restrict ourselves to intrachain and nearest - neighbor chain interactions , the amplitudes of the bare interactions are given by @xmath40 where the only transverse momentum dependence comes from the interchain interaction . in this work we analyze the properties of the model for repulsive @xmath41 , @xmath42 and finite umklapp scattering , which is the physically relevant sector for real materials like the bechgaard salts and their sulfur analogs . the renormalization group can be applied to the problem of quasi-1d systems of interacting electrons in different ways.@xcite in the following , we shall adopt the so - called one - particle irreducible ( 1pi ) momentum shell rg scheme , as developed by honerkamp _ _ et al.__@xcite ( cf . also binz _ _ et al.__@xcite ) . in this scheme , the 1pi vertex functions @xmath43 of a physical system with infrared cutoff @xmath44 are calculated . only degrees of freedom with energies equal to or greater than the cutoff are integrated out , which allows one to establish a direct link with the low energy effective action @xmath45 with an _ ultraviolet _ cutoff , as obtained in a kadanoff - wilson renormalization group scheme . according to morris,@xcite if @xmath46 , the effective interactions @xmath47 that appear in @xmath45 are simply @xmath48 . the free particle propagator is suppressed for energies below an infrared cutoff @xmath44 , @xmath49 where @xmath50 is the step function . setting all vertices involving more than two particles to zero , the renormalization group equations to the one - loop level are obtained . we also neglect self - energy corrections . it is practically impossible to take into account the complete functional dependence of the vertices , which will in general vary with the wave vectors @xmath51 as well as with the frequencies @xmath21 of the incoming and outgoing particles . by means of scaling arguments , one can show that the frequency dependence is irrelevant in the renormalization group sense , cf . for example ref . . the same is true for the dependence on the _ distance _ of the wave vector @xmath51 from the fermi surface . irrelevant variables are not necessarily negligible , in the sense that they may in principle have an influence on the rg flow before they vanish . however , the @xmath52 dependence , with @xmath52 the projection of the wave vector on the fermi surface , remains by far the most important one in the low energy limit . we will hence only consider the dependence of the functions @xmath53 on the positions of the wave vectors along the fermi surface , which we will parameterize by @xmath54 . due to momentum conservation , it is in general impossible to have all four arguments of a given vertex on the fermi surface . usually , the flow of the vertices with three arguments on the fermi surface is calculated . in the case of imperfect nesting , such a procedure will underestimate the sdw correlations . in our case , the strong anisotropy of the dispersion relation allows us to make a slightly different choice . we fix the values of @xmath55 and @xmath21 such that the most divergent contributions to the renormalization group flow are taken into account . if we define the cooper , landau , and peierls momentum variables @xmath56 we can write the one - loop rg equations in the form @xmath57 in this equation , @xmath58 , @xmath59 and @xmath60 are functions of the internal loop momentum - frequency @xmath61 . the most divergent contributions to the rg flow are obtained if @xmath62 for the frequencies , as well as @xmath63 for the longitudinal components . note that , for either choice of the @xmath55 dependence , one does not obtain a closed set of equations . the renormalization group equations for vertices for which relations ( [ leadingqpara ] ) are valid contain vertex functions with _ different _ values of @xmath64 , @xmath65 . we will therefore need to replace these functions by those calculated by imposing conditions ( [ leadingqpara ] ) . for the dimensionless vertices @xmath66 , we thus obtain the equations@xcite @xmath67 \biggr|_{{\genfrac{}{}{0pt}{2 } { k{_{\bot}}'=-k{_{\bot}}+{q_{{\cal c}\bot } } } { { q_{{\cal c}\bot}}=k_{\bot 1}+k_{\bot 2 } } } } \\ \nonumber & & + \frac{1}{n{_{\bot } } } \sum_{k{_{\bot } } } \biggl [ b_{{{\cal p}}}(k{_{\bot}},{q_{{\cal p}\bot } } ) \\ \nonumber & & \times \bigl\ { \bigl ( \tilde{{\gamma}}_2(k_{\bot 1}'k{_{\bot}}k_{\bot 1 } k{_{\bot } } ' ) - \tilde{{\gamma}}_1(k_{\bot 1}'k{_{\bot}}k{_{\bot}}'k_{\bot 1 } ) \bigr ) \tilde{{\gamma}}_1(k{_{\bot}}'k_{\bot 2}'k_{\bot 2 } k{_{\bot } } ) \\ \nonumber & & + \tilde{{\gamma}}_1(k_{\bot 1}'k{_{\bot}}k{_{\bot}}'k_{\bot 1 } ) \bigl ( \tilde{{\gamma}}_2(k{_{\bot}}'k_{\bot 2}'k{_{\bot}}k_{\bot 2 } ) - \tilde{{\gamma}}_1(k{_{\bot}}'k_{\bot 2}'k_{\bot 2 } k{_{\bot } } ) \bigr ) \\ \nonumber & & + \bigl ( \tilde{{\gamma}}_3(k_{\bot 1}'k{_{\bot}}k_{\bot 1 } k{_{\bot } } ' ) - \tilde{{\gamma}}_3(k_{\bot 1}'k{_{\bot}}k{_{\bot}}'k_{\bot 1 } ) \bigr ) \tilde{{\gamma}}_3(k{_{\bot}}'k_{\bot 2}'k_{\bot 2 } k{_{\bot } } ) \\ \nonumber & & + \tilde{{\gamma}}_3(k_{\bot 1}'k{_{\bot}}k{_{\bot}}'k_{\bot 1 } ) \bigl ( \tilde{{\gamma}}_3(k{_{\bot}}'k_{\bot 2}'k{_{\bot}}k_{\bot 2 } ) - \tilde{{\gamma}}_3(k{_{\bot}}'k_{\bot 2}'k_{\bot 2 } k{_{\bot } } ) \bigr ) \bigr\ } \biggr ] \biggr|_{{\genfrac{}{}{0pt}{2 } { k{_{\bot}}'=k{_{\bot}}+{q_{{\cal p}\bot } } } { { q_{{\cal p}\bot}}=k_{\bot 1}'-k_{\bot 1 } } } } \ , \end{aligned}\ ] ] @xmath68 \biggr|_{{\genfrac{}{}{0pt}{2 } { k{_{\bot}}'=-k{_{\bot}}+{q_{{\cal c}\bot } } } { { q_{{\cal c}\bot}}=k_{\bot 1}+k_{\bot 2 } } } } \\ \nonumber & & + \frac{1}{n{_{\bot } } } \sum_{k{_{\bot } } } \biggl [ b_{{{\cal p}}}(k{_{\bot}},{q_{{\cal p}\bot } } ) \\ \nonumber & & \times \bigl\ { \tilde{{\gamma}}_2(k_{\bot 1}'k{_{\bot}}k_{\bot 2 } k{_{\bot } } ' ) \tilde{{\gamma}}_2(k{_{\bot}}'k_{\bot 2}'k{_{\bot}}k_{\bot 1 } ) \\ \nonumber & & + \tilde{{\gamma}}_3(k_{\bot 1}'k{_{\bot}}k_{\bot 2 } k{_{\bot } } ' ) \tilde{{\gamma}}_3(k{_{\bot}}'k_{\bot 2}'k{_{\bot}}k_{\bot 1 } ) \bigr\ } \biggr ] \biggr|_{{\genfrac{}{}{0pt}{2 } { k{_{\bot}}'=k{_{\bot}}+{q_{{\cal p}\bot } } } { { q_{{\cal p}\bot}}=k_{\bot 1}'-k_{\bot 2 } } } } \ , \end{aligned}\ ] ] @xmath69 \biggr| _ { { \genfrac{}{}{0pt}{2 } { k{_{\bot}}'=k{_{\bot}}+{q_{{\cal p}\bot } } ' } { { q_{{\cal p}\bot } } ' = k_{\bot 1}'-k_{\bot 1 } } } } \\ \nonumber & & + \frac{1}{n{_{\bot } } } \sum_{k{_{\bot } } } \biggl [ b_{{{\cal p}}}(k{_{\bot}},{q_{{\cal p}\bot } } ) \\ \nonumber & & \times \bigl\ { \tilde{{\gamma}}_2(k_{\bot 1}'k{_{\bot}}k_{\bot 2 } k{_{\bot } } ' ) \tilde{{\gamma}}_3(k{_{\bot}}'k_{\bot 2}'k{_{\bot}}k_{\bot 1 } ) \\ \nonumber & & + \tilde{{\gamma}}_3(k_{\bot 1}'k{_{\bot}}k_{\bot 2 } k{_{\bot } } ' ) \tilde{{\gamma}}_2(k{_{\bot } } ' k_{\bot 2 } ' k{_{\bot}}k_{\bot 1 } ) \bigr\ } \biggr ] \biggr|_{{\genfrac{}{}{0pt}{2 } { k{_{\bot}}'=k{_{\bot}}+{q_{{\cal p}\bot } } } { { q_{{\cal p}\bot}}=k_{\bot 1}'-k_{\bot 2 } } } } \ , \end{aligned}\ ] ] where the dot denotes derivation with respect to @xmath70 . the particle - particle ( @xmath71 ) and particle - hole ( @xmath72 ) loops , after summation over matsubara frequencies and longitudinal wave vectors , are given by @xmath73 @xmath74 for continuity reasons , we always take @xmath75 . the starting values are given by the bare interactions , cf . section [ model ] . it is useful to consider the following limiting cases for the rg equations . as long as @xmath76 , we may take their 1d limit ( @xmath77 ) . in this regime , the cooper and peierls renormalization channels are entirely coupled , in the sense that all vertices are strongly renormalized in both channels ( except for umklapp processes , which only appear in the peierls channel ) . when @xmath78 , the coupling between the cooper and peierls channels is weak in the sense that , depending on the arguments @xmath79 , most vertices are strongly renormalized in only one ( or no ) channel at a time . nevertheless , the remaining interplay between the channels is at the origin of spin- and charge - fluctuation - induced superconductivity in the weak coupling regime . for purely repulsive interactions , the peierls channel is the most important one as long as deviations from perfect nesting may be neglected , _ _ i . e. _ _ for @xmath80 . in the presence of attractive effective interactions , the cooper channel plays an important role , and will be dominant when @xmath81 . the rg equations are written in these limits in appendix [ glims ] . a similar dimensional crossover may be observed with respect to the temperature instead of @xmath44 . as long as @xmath82 , the functions we calculate do not vary very strongly with the transverse wave vector , contrary to the low temperature case . an example is given in fig . [ chits](a ) . in order to evaluate susceptibilities , we add a term of the form @xmath83 to the action . the first term describes pairing and the second term density - wave correlations . the external fields @xmath84 are taken to be infinitesimal . they couple to pairs of fermionic variables which we will define in the following . let us first introduce the particle - particle operators @xmath85 for singlet ( @xmath86 ) and triplet ( @xmath87 ) pairs . the spin dependence is given by the coefficients @xmath88 ( @xmath89 ) . the pair operator appearing in @xmath90 is defined as @xmath91 the function @xmath92 describes the orbital symmetry of the superconducting order parameter . we classify these order parameters by their behavior on the fermi surface . they are parameterized by @xmath54 and @xmath93 . a list of the superconducting order parameters we shall examine is given in table [ opetab ] . with these , we have @xmath94 @xmath95 similarly , we introduce particle - hole operators @xmath96 for charge ( @xmath97 ) and spin ( @xmath98 ) excitations . here , @xmath99 is the @xmath100 identity matrix , and @xmath101 are the pauli matrices . at half filling , one has to distinguish between bond and site density waves . in direct space , these are given by @xmath102 where @xmath103 , and @xmath104 is the lattice periodicity along the chains . since we have @xmath105 , the associated fourier transforms are , for @xmath106 , @xmath107\ ] ] with @xmath108 for site and @xmath109 for bond density waves , and @xmath110 . ( for the case @xmath111 , we have neglected a constant imaginary factor . ) note that @xmath112 , and that @xmath113 and @xmath114 do not vanish , but contribute to the associated susceptibilities . away from half filling , we only consider site density waves . in this case , only the first term in eq . ( [ odsitelien ] ) contributes . in the presence of @xmath90 , the one - particle vertex contains a nondiagonal part with a contribution linear in @xmath115 of the form @xmath116 ( if @xmath117 ) . also due to @xmath90 , vertices with two outgoing , but no incoming , particles are now non - zero . their linear part in @xmath118 takes the form @xmath119 ( if @xmath117 ) . the vertex parts @xmath120 determine the renormalization of the susceptibilities @xmath121 as follows . in the beginning of the flow , we have @xmath122 and @xmath123 . the rg equations for density waves ( @xmath124 ) are : @xmath125 \ , \\ \dot{\tilde{\chi}}_{\alpha}^m ( q{_{\bot } } ) & = & - \frac{2}{n{_{\bot } } } \sum_{k{_{\bot } } } b_{{{\cal p}}}(k{_{\bot}},q{_{\bot } } ) { \left| { \zeta}_{\alpha}^m ( k{_{\bot}},k{_{\bot}}+q{_{\bot } } ) \right|}^2 \ .\end{aligned}\ ] ] for cooper pairs ( @xmath126 ) , we have : @xmath127 @xmath128 is defined as @xmath129 . we have introduced the linear combinations of vertex functions @xmath130 interactions @xmath131 ( @xmath132 ) and @xmath133 are defined analogously . using these functions as well as the particle - particle and particle - hole pair operators defined above , and neglecting umklapp processes , we can rewrite the interaction part of the action in the following way : @xmath134 @xmath135 ( resp . @xmath136 ) thus describes the interaction between particles forming a singlet or triplet pair ( resp . particle - hole pair ) . the relation between these couplings is as follows : @xmath137 we will make use of these relations in the following , when we discuss density fluctuation induced cooper pairing . note that umklapp processes do not couple to the cooper channel , and therefore affect cooper pair formation only indirectly _ via _ their effect on @xmath138 , @xmath139 . the rg equations are solved numerically , using a fourth order runge - kutta algorithm @xcite with fixed step sizes ( taking large steps in the beginning of the flow and short steps close to the divergence ) . the graphs showing transition temperatures are obtained with the help of an adaptive stepsize algorithm.@xcite as far as the @xmath54 dependence of the vertex functions is concerned , we discretize the fermi surface using 32 patches for each sheet . taking advantage of all symmetries of the problem , we thus have to calculate 9010 different function values for each @xmath140 . the use of twice the number of patches does not significantly modify our results . the integral of the functions @xmath141 must be calculated more precisely ; we use a fourth order runge - kutta algorithm with adaptive step size . to illustrate our method , we shall first consider the by now well known case of a quasi-1d electron gas model with purely intrachain interactions and no umklapp scattering . this case has been studied in detail by duprat and bourbonnais @xcite using a kadanoff - wilson rg scheme and an approximation where only two independent momentum variables for the vertices were taken into account . we here use the 1pi scheme and retain the full three variable dependence for the vertices . our results confirm those of reference . rg phase diagram for the quasi-1d electron gas model at incommensurate filling ( @xmath142 ) with intrachain interactions ( @xmath143 ) and nesting deviations , parameterized by @xmath144 . the circles indicate the transition temperature for sdw and the triangles that for sc@xmath0 . all figures in this article are obtained using the bare intrachain interactions , and the anisotropy ratio @xmath145.,width=264 ] the rg calculations that follow are performed for the values @xmath146 and @xmath147 , which are representative of the couplings likely to be found in practice in low dimensional conductors . thus for not too large nesting deviations , the renormalization group flow scales to strong coupling , which leads to a singular behavior in the susceptibility of a particular channel of correlations . this signals an instability of the normal state towards an ordered phase . the phase diagram obtained as a function of @xmath148 is shown in fig . [ localphas ] . for good nesting , there is an sdw phase with a modulation wave vector that corresponds to the best nesting vector of the spectrum ( [ spectrum ] ) . the transition temperature obtained for perfect nesting ( @xmath149 ) is @xmath150 . if we take , we have , which falls in the range of the experimental @xmath151 for systems like the bechgaard salts at low pressure . by increasing @xmath144 , the transition temperature decreases until the threshold value is reached , where the sdw is suppressed and replaced by @xmath0-wave superconductivity ( sc@xmath0 ) . the maximum temperature for the sc@xmath0 state is found to be . these estimates for @xmath152 and @xmath151 are comparable to the experimental results.@xcite temperature dependence of the susceptibilities in the normal phase ( @xmath142 , @xmath143 ) : ( a ) above the sdw phase ( @xmath153 ) . the continuous ( dashed ) line corresponds to the sdw response at @xmath154 ; the dashed - dotted line to sc@xmath0 correlations . it is interesting to observe that the temperature scale @xmath155 below which the @xmath156 curve separates and levels off corresponds to the so - called single particle dimensionality crossover.@xcite our rg scheme thus captures this effect correctly . ( b ) above the sc phase ( @xmath157 ) . the continuous line corresponds to the sdw response with transverse modulation @xmath158 , the dashed - dotted line to sc@xmath159 ( @xmath160 ) , the dotted line to sc@xmath161 ( @xmath162 ) and the dashed line to sc@xmath163 ( @xmath164 ) correlations.,width=264 ] the order parameter of the low temperature phase can be identified in two different ways , which give equivalent results . the first one follows from the identification of the most singular behavior in the temperature dependence of the various susceptibilities , as shown in fig . [ chits ] for values of @xmath148 below and above the threshold for superconductivity . @xmath165 in the sdw regime ( @xmath166 ) close to the divergence ( a ) , in the sc@xmath0 regime ( @xmath167 ) at an intermediate stage of the flow ( b ) and close to the divergence ( c ) . the vertical scale is arbitrary.,width=264 ] an alternative way to determine the nature of the ordered phase is to look at the wave vector dependence of the renormalized vertex functions close to the divergence . in fig . [ gks ] , we have plotted the sdw vertex function @xmath168 in the @xmath169 plane . in the sdw regime , only processes involving particle - hole pairs at wave vector are found to be singular . on the other hand , near the superconducting transition temperature , we obtain the separable form @xmath170 a plot of the singlet interaction amplitude @xmath171 gives essentially the same picture . from eq . ( [ zstrenorm ] ) , it is clear that the most divergent vertex part @xmath172 in such a situation is the one proportional to @xmath160 . recall that the vertex functions @xmath53 we calculate in our rg scheme are equal to the effective interactions @xmath173 of a wilsonian low energy effective theory . quite early during the flow , the vertex associated to the spin density , @xmath174 , and hence @xmath175 , develops a peak structure similar to the one close to the sdw transition , but less pronounced , cf . [ gks](b ) . there are thus important spin fluctuations at temperature or energy scales above the transition temperature for superconductivity , as is confirmed by the behavior of the associated susceptibility , fig . [ chits](b ) . the same peak structure appears in the effective interaction between electrons forming a singlet ( or triplet ) pair , @xmath176 . we can decompose the peak of @xmath177 at intermediate cutoff @xmath44 in terms of the variables @xmath178 and @xmath179 . neglecting the weak @xmath179 dependence , the result is , schematically , @xmath180 @xmath181 corresponds to the distance between the chains where the two interacting electrons are located . in the sdw regime , [ gks](a ) , @xmath182 has the form of a @xmath183 peak , so that all @xmath184 will be equal . in the superconducting regime , fig . [ gks](b ) , the peak is slightly enlarged , so that @xmath184 will be some decreasing function of @xmath181 . we thus see that , due to the spin fluctuations , the effective interaction between particles forming a singlet pair contains attractive contributions at all chain distances . ( note that , according to our definitions eqs . ( [ csstvertexdef ] ) and ( [ sigalf ] ) , an `` attractive '' interaction corresponds to _ positive _ @xmath135 . ) the most important one is the nearest - neighbor chain one , @xmath160 ( sc@xmath0 ) , followed by @xmath185 ( sc@xmath163 ) , @xmath186 etc . , as it can also be seen from the calculation of the related pairing susceptibilities , cf . [ chits](b ) . note that , according to the relations ( [ gstvongcs ] ) , @xmath139 also gives an attractive contribution to the triplet channel . however , all the three components of a spin - one boson sdw fluctuation contribute to the superconducting coupling in the singlet channel , whereas only one contributes to the triplet channel . antiferromagnetic fluctuations thus favor singlet pairing as compared to triplet pairing . on the other hand , according to eq . ( [ gstvongcs ] ) , strong charge fluctuations should in principle be able to change this tendency . in this section , we examine the role of interchain interactions in the phase diagram of the extended quasi-1d electron gas model . we will first consider the influence of interchain backward scattering @xmath187 and forward scattering @xmath188 separately , before studying their combined effect . the intrachain interactions are kept fixed to their values used in the previous section . concerning the nesting quality , the overall picture remains the same . for weak deviations from perfect nesting , we find density - wave instabilities , superconductivity for more important deviations , and discarding possible kohn - luttinger effects at low temperatures beyond the numerical accuracy of our rg calculations a metallic phase when the nesting is deteriorated even more . low temperature phases versus @xmath187 , keeping @xmath189 and @xmath190 . circles indicate a sdw phase , squares a cdw phase , triangles sc@xmath0 ( @xmath160 ) and crosses sc@xmath1 ( @xmath191 ) . in the region below the dotted line , spin fluctuations dominate over charge fluctuations in the normal phase.,width=264 ] transition temperatures for different values of @xmath192 , when @xmath193 and @xmath190 : ( a ) @xmath194 ( phase sequence sdw@xmath195sc@xmath0 , continuous line ) , @xmath196 ( sdw@xmath195sc@xmath0 , dotted line ) and @xmath197 ( cdw@xmath195sc@xmath1 , dashed - dotted line ) ; ( b ) @xmath198 ( sdw@xmath195sc@xmath0 , dotted line ) and @xmath199 ( sdw(@xmath195sc@xmath0)@xmath195sc@xmath1 , dashed line ) . note that in the latter case , the sc@xmath0 phase is extremely narrow.,width=264 ] temperature dependence of the susceptibilities in the normal phase above the sc@xmath0 phase ( ( a ) @xmath200 and ( b ) @xmath201 ) and above the sc@xmath1 phase ( ( c ) @xmath202 ) , keeping @xmath203 , @xmath190 and @xmath204 . the continuous line corresponds to sdw , the dotted line to cdw , the dashed line to sc@xmath0 and the dashed - dotted line to sc@xmath1 correlations.,width=264 ] by switching on @xmath187 gradually on the positive scale , one observes that at small values of @xmath187 , the sdw phase remains essentially unaffected , whereas the transition temperature for @xmath0-wave superconductivity is considerably reduced by the presence of a finite @xmath187 ( fig . [ tcbann ] ) , and the region where superconductivity is stable shrinks ( fig . [ g1nnphas ] ) . for higher @xmath187 ( @xmath187 of the order of @xmath205 of the intrachain backscattering @xmath6 ) , @xmath0-wave superconductivity turns out to be no longer stable and an @xmath1-wave triplet superconducting phase appears . in this parameter range , @xmath206 charge fluctuations are strongly enhanced by interchain backward scattering ( fig . [ chitvgl ] ) . however , spin fluctuations remain important , and in a sizeable region of the phase diagram , triplet superconductivity is preceded in temperature by dominant spin fluctuations in the normal state ( figs . [ g1nnphas ] and [ chitvgl ] ) . for the values of intrachain interactions used in this section , and @xmath207 , one finds a maximum @xmath208 k , that is of the same order of magnitude as that of the @xmath0-wave case . @xmath209 increases with the amplitude of @xmath187 , and the superconducting phase widens ( figs . [ g1nnphas ] and [ tcbann ] ) . once triplet superconductivity occurs , @xmath187 starts to affect the density wave phase : the sdw state is suppressed and replaced by a cdw . the values of @xmath210 for which sc@xmath1 and cdw phases first appear depend on the values of intrachain interactions and increase with the value of the ratio @xmath211 . the origin of the @xmath1-wave sc and cdw phases can be understood by considering the contribution of the @xmath212 s to the ( bare ) scattering amplitudes in the singlet and triplet particle - particle channels , as well as in the charge and spin channels : @xmath213 @xmath214 from these equations , it can be easily seen that the interchain repulsion @xmath187 contributes positively to @xmath215 at @xmath216 , and therefore induces cdw correlations with a phase difference of @xmath158 between neighboring chains . in the cooper channel , @xmath3 favors triplet @xmath1-wave and singlet @xmath161-wave pairing , whereas its contribution to singlet @xmath159-wave and triplet @xmath217-wave pairing is negative . as for @xmath4 , it tends to suppress both singlet and triplet pairings on nearest - neighbor chains . in addition to this ` direct ' contribution there is also an indirect effect due to the exchange of density fluctuations . upon renormalization , @xmath206 cdw correlations are enhanced beyond the level expected from a mean - field treatment of @xmath187.@xcite these cdw fluctuations enhance triplet @xmath1-wave pairing but suppress singlet pairing , whereas the sdw fluctuations are well known to favor singlet pairing ( see eqs . ( [ gstvongcs ] ) ) . ( [ spinfluktpeak ] ) shows that the latter reinforce @xmath159- but suppress @xmath161-wave pairing . regardless of the values of the interaction constants @xmath6 , @xmath7 and @xmath4 , the cdw and triplet @xmath1-wave phases always appear almost simultaneously when @xmath3 increases . this suggests that cdw fluctuations ( rather than the direct effect of @xmath3 in the cooper channel [ eqs . ( [ gnnstvongnn12 ] ) ] ) provide the dominant driving force leading to @xmath1-wave superconductivity . low temperature phases for @xmath188 varying , keeping @xmath218 and @xmath190 . circles indicate a sdw phase , triangles sc@xmath0 and diamonds sc@xmath163 ( @xmath164).,width=264 ] transition temperatures for different values of @xmath219 , if @xmath194 and @xmath190 : @xmath193 ( sdw@xmath195sc@xmath0 , continuous line ) , @xmath220 ( sdw@xmath195sc@xmath0 , dashed line ) and @xmath221 ( sdw@xmath195sc@xmath163 , dotted line).,width=264 ] we now consider the effect of interchain _ forward _ scattering @xmath4 alone , setting @xmath218 . from eqs . ( [ gnnstvongnn12 ] ) , it can easily be seen that @xmath188 contributes negatively to both @xmath222 and @xmath223 and then favors the suppression of the nearest - neighbor - chain cooper pairing induced by spin fluctuations . our results ( figs . [ g2nnphas ] and [ tcfonn ] ) show that this is indeed the case . we have seen , however , in section [ local ] , that sdw fluctuations can generate a smaller yet present attractive interaction between electrons on next - nearest - neighbor chains , which is not affected by @xmath188 . it follows that when @xmath0-wave superconductivity is sufficiently weakened by @xmath188 , it is replaced by @xmath163-wave singlet pairing ( @xmath224 ) . this is shown in fig . [ g2nnphas ] . it is worth noting that the instability of the normal state with respect to superconductivity with high angular momentum pairing can be seen as a result of the kohn - luttinger effect originally predicted for isotropic metals.@xcite at variance with more isotropic systems , however , the rg results show that for a quasi-1d metal , the transition temperature of high angular momentum superconducting phases , like sc@xmath163 , remains experimentally accessible . we find @xmath225 . finally , the sdw phase remains nearly unaffected by @xmath188 . this can be understood from eqs . ( [ gnncsvongnn12 ] ) , which indicate that its contribution averages out over the fermi surface in the particle - hole channel of the rg equations . low temperature phases for @xmath226 , @xmath142 . circles indicate a sdw phase , black squares a cdw phase , triangles sc@xmath0 ( @xmath160 ) , white diamonds sc@xmath163 ( @xmath164 ) and crosses sc@xmath1 ( @xmath191 ) . below the dotted line , spin fluctuations dominate over charge fluctuations in the normal phase.,width=264 ] we now consider the combined effect of @xmath187 and @xmath188 . [ diagphas ] shows the results for @xmath226 , where all aforementioned phases appear . in the presence of both @xmath187 and @xmath188 , @xmath0-wave pairing is suppressed even faster than by @xmath187 or @xmath188 alone . the appearance of @xmath1-wave pairing is retarded by @xmath188 which , as previously mentioned , is detrimental to nearest neighbor chain pairing ( eq . ( [ gnnstvongnn12 ] ) ) . however , for our choice of intrachain interactions , no triplet phase with pairing on next - nearest - neighbor chains is found for @xmath226 , since @xmath187 favors nearest - neighbor triplet pairing at the outset . as for cdw s , they are found to occur at slightly higher values of @xmath187 in the presence of a finite @xmath188 . as mentioned before , @xmath188 has no important effect in the particle - hole channel alone . the slight suppression of the cdw due to @xmath188 must therefore come from the 1d regime , where the correlation channels are coupled . this can be checked from the rg equations for @xmath77 given in appendix [ glims ] . the phase diagram ( fig . [ diagphas ] ) depends quantitatively and qualitatively on the bare intrachain interactions . when the ratio @xmath211 increases , higher values of the interchain interactions are necessary for obtaining cdw and sc@xmath1 phases because @xmath227 is more negative at the outset . for @xmath228 , we even find a triplet sc@xmath1 order parameter of the form @xmath229 , corresponding to second - nearest - neighbor chain pairing , instead of @xmath230 . this is coherent with the fact that , in the 1d regime , the renormalization of @xmath231 is proportional to @xmath232 , so that a more negative @xmath227 reduces @xmath231 and hence nearest - neighbor chain triplet pairing ( appendix [ glims ] ) . conductors like the bechgaard and fabre salts are slightly dimerized in the direction of the organic chains . it follows that at low energy or temperature , the hole band can be considered as effectively half filled rather than quarter - filled and this gives rise to umklapp scattering processes with amplitudes @xmath8 and @xmath233 . to leading order , the bare amplitude of @xmath234 is proportional to the dimerization gap @xmath235,@xcite which yields a @xmath8 that is rather weak as compared to @xmath236 . we assume a similar ratio between the interchain couplings @xmath233 and @xmath237.@xcite transition temperatures when umklapp processes are taken into account . ( a ) without interchain umklapp scattering : @xmath190 , @xmath238 ( sdw@xmath195sc@xmath0 , continuous line ) , @xmath239 , @xmath238 ( sdw@xmath195sc@xmath0 , dashed line ) , @xmath239 , @xmath240 , @xmath241 , ( sdw@xmath195sc@xmath1 , dotted line ) . ( b ) effect of interchain umklapp processes : @xmath239 , @xmath240 , @xmath241 ( sdw@xmath195sc@xmath1 , dotted line ) and @xmath242 ( sdw@xmath195sc@xmath0@xmath195sc@xmath1 , continuous line).,width=264 ] low temperature phases in the presence of umklapp processes , with @xmath243 ( a ) and @xmath244 ( b ) , taking @xmath245 . circles indicate a site - sdw phase , squares a bond - cdw phase , triangles sc@xmath0 ( @xmath160 ) and crosses sc@xmath1 ( @xmath191 ) . below the dotted lines , spin fluctuations dominate over charge fluctuations in the normal phase.,width=264 ] temperature dependence of the susceptibilities in the normal phase above the sc@xmath0 phase ( ( a ) @xmath246 , @xmath201 ) and above the sc@xmath1 phase ( ( b ) @xmath246 and ( c ) @xmath247 , both for @xmath248 ) , taking @xmath239 and @xmath249 . the continuous line corresponds to sdw , the dotted line to cdw , the dashed line to sc@xmath0 and the dashed - dotted line to sc@xmath1 correlations.,width=264 ] intrachain umklapp processes enhance the formation of ` site ' centered sdw correlations and ` bond ' centered cdw correlations and weaken bond sdw and site cdw fluctuations , as can be seen from eqs . ( [ zdwrenorm ] ) . thus , contrary to the incommensurate case , site and bond density - wave orders no longer join together and must be considered separately with their distinct strengths at half filling . in addition , intrachain umklapp scattering reinforces the effective interactions @xmath138 and @xmath139 at momentum transfers @xmath206 , thus increasing spin and charge density fluctuations of both site and bond type . these effects are known in the 1d case ( see _ _ e . g. _ _ ref . and also appendix [ glims ] ) , as well as from calculations restricted to the particle - hole channels ( ref . and appendix [ glims ] ) . even though umklapp processes do not enter the particle - particle channel directly , they do enhance cooper pairing , because they amplify the peak at @xmath206 in the effective interactions @xmath250 and thus the attraction between electrons on neighboring chains [ see eq . ( [ spinfluktpeak ] ) ] . since density - wave correlations are enhanced with respect to the incommensurate case , transition temperatures are also higher.@xcite this is true for the formation of density - wave states as well as for superconductivity ( fig . [ umtc](a ) ) . the ratio @xmath251 remains essentially unaffected by @xmath8 . the critical @xmath152 needed to destroy the density - wave phase is also higher in the presence of umklapp scattering . let us now turn to the effect of interchain interactions in the presence of umklapp processes . we will first consider the effect of a finite intrachain @xmath8 , for @xmath252 and @xmath243 . a comparison of figs . [ diagphas ] and [ umklphas](a ) shows that the regions of @xmath0-wave and @xmath1-wave superconductivity are now enlarged , and there is no more sc@xmath163 phase corresponding to next - nearest - neighbor chain singlet pairing . this is a consequence of the fact that density - wave correlations are reinforced by umklapp scattering . a stronger @xmath188 would then be needed to destroy nearest - neighbor chain pairing , but owing to the condition @xmath226 the sc@xmath1 phase is also reinforced and in turn stabilized once the sc@xmath0 phase is suppressed . when we finally add interchain umklapp scattering @xmath253 , the picture does not change significantly ( fig . [ umklphas](b ) ) . the only effect is that the occurrence of sc@xmath1 and cdw phases takes place at slightly higher values of the interchain interactions in the phase diagram , whereas transition temperatures are scarcely lower ( fig . [ umtc](b ) ) . in the 1d regime ( @xmath254 ) , @xmath253 enhances @xmath138 ( appendix [ glims ] ) , but this is only a second order effect . on the other hand , the bare contribution of @xmath253 , eq . ( [ gstarting ] ) , reduces the peak in the effective interaction @xmath255 at @xmath256 , which is connected to the charge density fluctuations ( appendix [ glims ] ) . this influence is linear in @xmath253 and therefore stronger than the 1d effect , thus explaining the overall weakening of cdw and , in turn , of sc@xmath1 correlations . there is no direct effect of @xmath253 on sdw correlations . however , the rg equations ( appendix [ glims ] ) show that @xmath253 weakens the renormalization of @xmath8 in the 1d regime and in turn the increase of transition temperatures in comparison to the incommensurate situation . in fig . [ umchi3in1 ] , we show the typical behavior of the most important susceptibilities as a function of temperature in the normal phase above the superconducting phases . fig . [ umchi3in1](a ) shows that for sufficiently important interchain interactions , triplet correlations are already strongly enhanced , although the ordered phase still corresponds to spin singlet pairing . sdw correlations are always important . they are the dominant fluctuations in the normal phase in the major part of the parameter range we have explored , sometimes even above the triplet sc phase ( cf . figs . [ umchi3in1](c ) and [ umklphas ] ) . in this work , we have determined the possible electronic phases of the extended quasi-1d electron gas model that includes both intrachain and interchain repulsive interactions , interchain electron hopping and the influence of nesting deviations . our results reveal that in correlated quasi-1d metals both for zero and non zero umklapp scattering , interchain interactions can act as a key factor in expanding the range of possibilities of ordered states compared to the case where only repulsive intrachain interactions are present.@xcite at large momentum transfer , the interchain electron - electron coupling acts as a short - range interaction that is responsible for the enhancement of cdw correlations , consistently with what was found long ago in the absence of interchain hopping.@xcite for finite @xmath9 , however , interchain interaction leaves the amplitude of sdw correlations essentially unaffected and a relatively small critical value of @xmath3 coupling is then needed to make the cdw ordered state possible , and this , on equal footing with the sdw phase , which is known to dominate the phase diagram when only repulsive intrachain interactions are present . when nesting deviations are cranked up beyond some threshold value @xmath152 , density - wave order is suppressed and interchain interactions turn out to affect correlations of the cooper channel too . thus an important conclusion that emerges from this work is the gradual suppression of interchain @xmath0-wave pairing when repulsive short - range interchain interaction increases . as a result of the growth of cdw correlations in the normal state , it ultimately yields the stabilization of a triplet sc@xmath1 superconducting phase corresponding to an order parameter @xmath257 having nodes on the warped fermi surface that are at the same locus as for the sc@xmath0 case . the normal phase is still dominated by strong sdw fluctuations over a sizable region of the phase diagram in this sector . it is interesting to consider how far the rg results of this work are applicable to quasi-1d organic conductors . in the case of the bechgaard ( tmtsf)@xmath5x salts with centrosymmetric anions x for example , the observation of a sdw - cdw coexistence below the critical pressure for superconductivity@xcite indicates that interchain electron repulsion is a relevant interaction in these materials besides intrachain interactions and interchain hopping . accordingly , the rg calculations show that a relatively small and realistic amplitude of repulsive interchain interaction is sufficient to bring the stability of cdw order close to sdw , indicating that this part of interaction would indeed play an important role in the emergence of density - wave order in these materials . the suppression of the sdw state ( and presumably of cdw as well ) followed by the emergence of superconductivity is well known to constitute the closing sequence of transitions that characterizes virtually all members of the bechgaard and fabre salts series as one moves along the pressure scale . since pressure introduces alterations of the electron spectrum , it prompts deviations from perfect nesting . in our model , these simulate the main effect of pressure , which together with a reasonable set of parameters , yield a ` pressure ' profile of the critical temperature that agrees quite well with the characteristic variation seen in experiments.@xcite as to the nature of the superconductivity in these materials , our results based on a purely electronic model , indicate that given the observation of the close proximity between sdw and cdw ordered states , not only sc@xmath0 but also triplet sc@xmath1 order parameter become serious candidates for the description of the superconducting phase in these compounds ( figs . [ g1nnphas],[diagphas ] , and [ umklphas ] ) . however , as pressure also affects the normalized amplitudes @xmath258 through the band width and the dimerization gap ( umklapp ) , and @xmath259 through the interchain distance,@xcite the actual trajectory in the phase diagram under pressure can not be determined with great precision . it follows that besides the possibilities sdw@xmath260 sc@xmath0 or sdw @xmath260 sc@xmath1 , sequence of transitions such as sdw @xmath260 sc@xmath0 @xmath260 sc@xmath1 , where one can pass from singlet to triplet sc order under pressure , can not be excluded . it is worth remarking that in this sector of the phase diagram , the addition of a small magnetic field as actually used in many experiments@xcite or accounting for the small but yet finite spin anisotropy would tend to tip the balance in favor of a triplet order parameter.@xcite experimental features of the normal phase also argue in favor of this region of the phase diagram for the bechgaard salts . this is the case of the puzzling growth of cdw correlations seen in optical conductivity in the low temperature part of the metallic phase above the superconducting transition.@xcite cdw correlations are found to be significantly enhanced in a temperature region where nmr experiments reveal the existence of strong sdw correlations.@xcite this feature can not be captured for realistic intrachain interactions alone . it can find , however , a natural explanation in the framework of the extended quasi-1d electron gas model , for which interchain interactions can boost the amplitude of cdw correlations besides those of the sdw channel that are kept essentially unchanged . in the case of non - centrosymmetric anions ( e. g. x = clo@xmath261 ) , our approach should be refined in order to take into account the doubling of the unit cell in the transverse direction due to the anion ordering taking place below 24 k in the normal phase . the concomittant reduction of the brillouin zone yields two electronic bands at the fermi level and in turn multiple nesting vectors.@xcite an accurate description of this fermi surface should be incorporated in the rg approach , and a modification of the node structure for the sc gap is expected.@xcite a previous ( simplified ) rpa - like calculation has shown that the nodes of a @xmath0-wave sc order parameter , which are located at @xmath262 , are precisely found where a gap opens due to clo@xmath261 anion ordering , thus making the sc phase effectively nodeless at low temperature.@xcite the results presented in this work may also be relevant for the phase diagram of other series of quasi-1d organic superconductors . in this matter , the case of the two - chain compounds ttf [ m(dmit)@xmath5]@xmath5 ( m = ni , pd ) is of particular interest . these compounds are characterized by an incommensurate cdw state that takes place on the m(dmit)@xmath5 stacks at low pressure.@xcite at some critical pressure , they become superconducting.@xcite the variation of the critical temperature under pressure has been analyzed in detail in the case of m = pd , which shows close similarity with the one found for the bechgaard and fabre salts.@xcite the temperature scale for the cdw instability at low pressure is relatively large , however , and owing to the pronounced anisotropy of the band parameters in these compounds,@xcite this indicates that the interchain interaction is likely to be a key coupling in the stabilization of a 3d ordered state in these materials . according to our model , these conditions would be favorable to the existence of a triplet sc@xmath1 state in these systems under pressure ( see _ _ e.g . _ _ fig . [ diagphas ] ) . another result that is highlighted by our analysis in the incommensurate case is the occurrence of a singlet sc@xmath163 state when the interchain electron scattering dominates for small momentum transfer . the @xmath4 coupling tends to suppress electron pairing between the first nearest - neighbor chains , which therefore suppresses sc@xmath0 or sc@xmath1 type of superconductivity . however , an instability of the normal state remains possible . it results from the oscillating tail of density - wave correlations in the transverse direction which favors longer range pairing between electrons separated by more than one interchain distance . the fact that an instability of the normal state persists in the cooper channel illustrates how the quasi-1d geometry for electrons , with its inherent interference between peierls and cooper channels , is prone to magnify the kohn - luttinger mechanism for the ( cooper ) instability of a fermi liquid for repulsive interactions.@xcite one can easily infer from our results that the addition of longer range interchain interactions will frustrate short - range interchain pairing and shift it to larger interchain distances , thus unfolding possibilities of superconductivity at even larger angular momentum pairing . j. c. n. is grateful to the gottlieb daimler- und karl benz - stiftung for partial support . c. b. thanks d. jrome , y. fuseya , m. tsuchiizu , y. suzumura , l. g. caron and s. brown for useful discussions and comments , and the natural sciences and engineering research council of canada ( nserc ) , le fonds qubecois de la recherche sur la nature et les technologies du gouvernement du qubec ( fqrnt ) , and the institut canadien de recherches avances ( ciar ) for financial support . n. d. thanks the condensed matter theory group of the universit de sherbrooke for its hospitality . in order to obtain the one - dimensional limit of the rg equations given in section [ renorm ] , we neglect interchain hopping ( @xmath264 ) and take into account intrachain and nearest - neighbor - chain interactions only . we thus obtain@xcite @xmath265 \ , \\ \nonumber \dot{\tilde{{\gamma}}}^{\bot}_1 & = & { { \tilde{\gamma}}^{\bot}}_1 \bigl [ -2 \tilde{{\gamma}}_1^{(1d ) } + \tilde{{\gamma}}_2^{(1d ) } - { { \tilde{\gamma}}^{\bot}}_2 \bigr ] - \tilde{{\gamma}}_3^{(1d ) } { { \tilde{\gamma}}^{\bot}}_3 \ , \\ \nonumber \dot{\tilde{{\gamma}}}_2^{(1d ) } & = & - \frac{1}{2 } \bigl ( \tilde{{\gamma}}_1^{(1d ) } \bigr)^2 + \frac{1}{2 } \bigl ( \tilde{{\gamma}}_3^{(1d ) } \bigr)^2 \ , \\ \nonumber \dot{\tilde{{\gamma}}}^{\bot}_2 & = & - \frac{1}{2 } \bigl ( { { \tilde{\gamma}}^{\bot}}_1 \bigr)^2 + \frac{1}{2 } \bigl ( { { \tilde{\gamma}}^{\bot}}_3 \bigr)^2 \ , \\ \nonumber \dot{\tilde{{\gamma}}}_3^{(1d ) } & = & \tilde{{\gamma}}_3^{(1d ) } \bigl [ - \tilde{{\gamma}}_1^{(1d ) } + 2 \tilde{{\gamma}}_2^{(1d ) } \bigr ] - 4 { { \tilde{\gamma}}^{\bot}}_1 { { \tilde{\gamma}}^{\bot}}_3 \ , \\ \nonumber \dot{\tilde{{\gamma}}}^{\bot}_3 & = & { { \tilde{\gamma}}^{\bot}}_3 \bigl [ -2 \tilde{{\gamma}}_1^{(1d ) } + \tilde{{\gamma}}_2^{(1d ) } + { { \tilde{\gamma}}^{\bot}}_2 \bigr ] - \tilde{{\gamma}}_3^{(1d ) } { { \tilde{\gamma}}^{\bot}}_1 \ . \end{aligned}\ ] ] typical pictures of the interactions @xmath168 ( a ) and @xmath266 ( b ) , for initial values @xmath267 , @xmath239 , if the renormalization is restricted to the peierls channel . the result for @xmath268 is identical to that for @xmath269.,width=264 ] alternatively , we may restrict the renormalization to a given correlation channel in certain situations . for the particle - hole channel , it is convenient to rewrite the rg equations in terms of the vertex functions @xmath270 and @xmath271 , as defined in eqs . ( [ csstvertexdef ] ) . remember however that @xmath272 and @xmath273 are not independent : @xmath274 we obtain @xmath275 @xmath276 @xmath277 these equations are remarquably symmetric with respect to spin and charge density correlations , cf . _ _ _ the results when we take @xmath278 from the beginning ( fig . [ sympeilbild ] ) . for low temperature and energy cutoff @xmath44 , the particle - hole loop integral @xmath279 is strongly peaked for momentum transfers equal to the best nesting vector , e. _ _ @xmath216 . as can be seen from the preceding equations , this generates a peak at @xmath280 in @xmath269 ( responsible for the sdw ) and in @xmath268 ( responsible for the cdw ) . in @xmath281 , peaks are created at @xmath280 as well as @xmath256 . the peak in @xmath281 for @xmath280 enhances the spin density correlations , whereas the one for @xmath256 supports the charge density correlations . let us now consider the influence of interchain interactions [ eq . ( [ gstarting ] ) ] on these structures . interchain umklapp scattering @xmath253 reduces the latter peak and thus weakens the charge fluctuations , whereas it has no direct influence on the spin correlations . @xmath187 and @xmath188 have no direct effect on the spin fluctuations either . interchain backward scattering @xmath187 , on the contrary , reinforces the peak in @xmath138 at @xmath280 and thus supports the formation of charge density waves .

the kpii equation is an integrable nonlinear pde in 2 + 1 dimensions ( two spatial and one temporal ) , which arises in several physical circumstances , including fluid mechanics where it describes waves in shallow water . it provides a multidimensional generalisation of the renowned kdv equation . in this work , we employ a novel approach recently introduced by one of the authors in connection with the davey - stewartson equation , in order to analyse the initial - boundary value problem for the kpii equation formulated on the half - plane . the analysis makes crucial use of the so - called d - bar formalism , as well as of the so - called global relation . a novel feature of boundary as opposed to initial - value problems in 2 + 1 is that the d - bar formalism now involves a function in the complex plane which is discontinuous across the real axis .

1006.0458

the kadomtsev - petviashvili ( kp ) equation @xmath0 with the operator @xmath1 defined by @xmath2 is one of the most notable integrable nonlinear equations in 2 + 1 dimensions ( i.e. evolution equations in two spatial dimensions ) . it is the natural generalisation of the celebrated korteweg - de vries ( kdv ) equation from one to two spatial dimensions and , as such , it appears in various physical problems . in particular , it was first derived in the study of waves of long wavelength in shallow water . in this specific application , if the surface tension dominates over the gravitational force , then @xmath3 and equation is called kpi @xcite , whereas if the gravitational force is dominant , then @xmath4 and equation is called kpii @xcite . the initial - value problem for the kdv equation was solved via the so - called inverse scattering transform ( ist ) in 1967 @xcite ; kpi and kpii equations were formally solved in @xcite and @xcite respectively , using a nonlocal riemann - hilbert formalism and a d - bar formalism respectively ( see also @xcite , @xcite , @xcite @xcite ) . the solution of initial - boundary value ( ibv ) problems is considerably more complicated than the solution of pure initial - value problems . a unified transform method for solving linear and integrable nonlinear evolution pdes in one spatial dimension was introduced in @xcite . in particular , interesting results for the linearised version of the kdv and for the kdv itself formulated on the half - line are presented in @xcite , @xcite , @xcite and @xcite . the generalisation of these results from one to two spatial dimensions for linear and for integrable nonlinear equations is presented in @xcite and @xcite respectively . here , we employ the general methodology of @xcite in order to analyse the kpii on the half - plane . it has been emphasised by one of the authors that the solution of the linearised version of a given nonlinear pde via a lax pair approach provides a useful starting point before analysing the nonlinear pde itself . hence , we will first solve the linearised version of the kpii equation on the half - plane , @xmath5 where @xmath6 before analysing the kpii equation , @xmath7 our approach to both problems involves the following steps : 1 . _ the formulation of the pde in terms of a lax pair . _ for the spectral variable @xmath8 , and for some _ sectionally analytic _ function @xmath9 , the linearised kp equation admits the lax pair : + [ laxintro ] @xmath10 @xmath11 + similarly , the kpii equation possesses the lax pair : + [ laxintro ] @xmath12 @xmath13 + where the operator @xmath14 is defined by @xmath15 2 . _ the direct problem . _ by applying a fourier transform in @xmath16 , it is possible to analyse the two equations defining the lax pair _ simultaneously _ in order to obtain an expression for @xmath17 which is bounded for all @xmath18 . it turns out that @xmath17 has different representations in different parts of the complex @xmath19-plane , namely @xmath20 where @xmath21 denote the four quadrants of the complex @xmath19-plane . + the above expression for @xmath22 depends on @xmath23 , @xmath24 and @xmath25 . the derivation of the global relation . _ this relation is an algebraic equation coupling the so - called spectral functions . for the linearised kp these functions are appropriate transforms of @xmath26 namely , the global relation couples @xmath27 , @xmath28 , @xmath29 and @xmath30 , where + @xmath31 + and @xmath32 for the kpii , the spectral functions can also be expressed in terms of via linear integral equations . 4 . _ the inverse problem . _ by using the fact that the function @xmath17 defined in step 2 is bounded for all @xmath33 , it is possible to obtain an alternative representation for this function using a d - bar formalism ( or more precisely the so - called pompeiu s formula ) . in order to achieve this , it is necessary to : ( a ) compute @xmath34 , ( b ) compute the jumps of @xmath22 across the real and imaginary @xmath19-axes ; actually the jump across the imaginary @xmath19-axis vanishes . for the linearised kp , @xmath34 and @xmath35 can be expressed in terms of @xmath28 , @xmath29 and @xmath30 . then , using pompeiu s formula , it is possible to express @xmath17 in terms of @xmath28 , @xmath29 and @xmath30 . in the nonlinear case , the d - bar derivatives and the jumps can be expressed in terms of the spectral functions . in summary , @xmath17 can be expressed via the spectral functions and hence via @xmath36 , @xmath37 and @xmath38 . after obtaining @xmath17 it is straightforward to obtain a formula for @xmath39 . [ [ notations - and - assumptions ] ] notations and assumptions + + + + + + + + + + + + + + + + + + + + + + + + + * the complex variable @xmath19 is defined as @xmath40 * a bar on top of a complex variable will denote the complex conjugate of this variable ; in particular , @xmath41 . * for the solution of the inverse problem we will make use of the so - called pompeiu s formula : if @xmath42 is a smooth function in some piece - wise smooth domain @xmath43 , then @xmath44 is related to its value on the boundary of @xmath45 and to its d - bar derivative inside @xmath45 via the equation @xmath46 where @xmath47 * we will denote the initial value by + @xmath48 + and the boundary values @xmath24 and @xmath25 by @xmath49 and @xmath50 + we will assume that @xmath51 , where @xmath52 denotes the space of schwartz functions . * we will seek a solution which decays as @xmath53 for all fixed @xmath54 and which also decays as @xmath55 for all fixed @xmath56 . * throughout this paper we will _ assume _ that there exists a solution @xmath23 with sufficient smoothness and decay in @xmath57 , which denotes the closure of the domain @xmath58 . * a hat `` @xmath59''above a function will denote the fourier transform of this function in the variable @xmath16 . * the index `` o '' on a function will denote evaluation at @xmath60 . * whenever we write @xmath39 , we mean that @xmath39 depends on the physical variables @xmath61 . * whenever we write @xmath17 , we mean that @xmath17 depends on the physical variables @xmath61 and on the spectral variables @xmath62 , @xmath63 . we consider the linearised kp equation with the associated lax pair . differentiating equation with respect to @xmath64 and equation with respect to @xmath65 , we find : @xmath66 and @xmath67 the requirement @xmath68 implies equation , thus is indeed a lax pair for equation . [ lkpdpprop ] assume that there exists a solution @xmath69 to an initial - boundary value problem for equation . then , there exists a solution @xmath17 of the lax pair which is bounded @xmath70 . this solution can be represented in the form @xmath71 ( see figure [ eigenfunctions ] ) , where @xmath22 can be expressed in terms of @xmath72 by the following formulae : @xmath73 and @xmath74 where @xmath75,\ ] ] and the functions @xmath37 and @xmath38 are defined by equations and . let @xmath76 denote the fourier transform of @xmath17 with respect to @xmath16 , i.e. @xmath77 with the inverse fourier transform given by @xmath78 then , the fourier transform of equation gives : @xmath79 we solve this equation by integrating either from @xmath80 or @xmath81 : @xmath82 we look for a solution which is bounded @xmath70 . in the first expression in equation , @xmath83 , thus we require that @xmath84\geq 0 $ ] , i.e. @xmath85\cup[-2k_r,\infty),\ k_r\leq0 \\ \\ l\in(-\infty,-2k_r]\cup[0,\infty),\ k_r\geq0 . \end{array } \right.\ ] ] analogous considerations are valid for the second expression in equation , where @xmath86 . let @xmath87 inverting equation we obtain : @xmath88 and @xmath89 where @xmath90 denote the value of @xmath91 at the boundary @xmath80 , i.e. @xmath92 equation evaluated at @xmath80 yields @xmath93 using the definition with @xmath80 , we find @xmath94 thus , @xmath95 does not satisfy any constraints for @xmath96\cup [ -2k_r , \infty)$ ] , whereas it does satisfy the following constraint for @xmath97 $ ] : @xmath98.\ ] ] the situation is similar for @xmath99 , namely @xmath100 does not satisfy any constraint for @xmath101 , whereas it does satisfy the following constraint for @xmath102 $ ] : @xmath103.\ ] ] in order to determine @xmath104 , we use the second of the equations defining the lax pair , evaluated at @xmath80 : @xmath105 where @xmath106.\end{aligned}\ ] ] the fourier transform in @xmath16 of equation yields @xmath107 thus , integrating with respect to @xmath64 either from @xmath60 or from @xmath108 , we find @xmath109 moreover , integration by parts of @xmath110 implies @xmath111 where @xmath112 is the fourier transform of the function @xmath113 defined by equation and @xmath114 denotes the evaluation of @xmath115 at @xmath108 . we look for a solution @xmath116 which is bounded @xmath70 . nothing that @xmath117=12lk_i(l+2k_r)(\tau - t),\ ] ] it follows that the exponential @xmath118 is bounded , provided that @xmath119 in order to analyse this inequality , we consider the following cases : * if @xmath120 , then @xmath121 implies that @xmath122 , whereas @xmath123 implies that @xmath124 . recalling our earlier analysis for @xmath95 , we can choose @xmath125 and then relation implies @xmath126 consequently , equations yield @xmath127 * if @xmath128 , then @xmath129\cup[-2k_r , \infty)$ ] implies that @xmath124 , whereas @xmath130 $ ] implies that @xmath122 . following steps similar to those used above , we find @xmath131\ ] ] and @xmath132\cup[-2k_r , \infty).\ ] ] then , equations imply @xmath133\cup[-2k_r , \infty)\\ \\ \\ \int_{0}^{t}d\tau \ , e^{-4il(l^2 + 3kl+3k^2)(\tau - t ) } \hat q(l,\tau , k_r , k_i)\\ \\ -e^{4il(l^2 + 3kl+3k^2)t}\int_{0}^{\infty}d \eta \ , e^{l(l+2k)\eta}\hat q_0(l,\eta ) , \quad l\in[0,-2k_r].\end{array}\right.\end{aligned}\ ] ] * similarly , we can show that for @xmath134 , @xmath135\cup[0 , \infty)\\ \\ \\ -\int_{t}^{t}d\tau \ , e^{-4il(l^2 + 3kl+3k^2)(\tau - t ) } \hat q(l,\tau , k_r , k_i)\\ \\-e^{-4il(l^2 + 3kl+3k^2)(t - t)}\int_{0}^{\infty}d \eta \ , e^{l(l+2k)\eta}\hat q(l,\eta , t ) , \quad l\in[-2k_r,0].\end{array}\right.\end{aligned}\ ] ] * for @xmath136 , @xmath137\cup[0 , \infty)\\ \\ \\ \int_{0}^{t}d\tau \ , e^{-4il(l^2 + 3kl+3k^2)(\tau - t ) } \hat q(l,\tau , k_r , k_i)\\ \\ -e^{4il(l^2 + 3kl+3k^2)t}\int_{0}^{\infty}d \eta \ , e^{l(l+2k)\eta}\hat q(l,\eta,0 ) , \quad l\in[-2k_r , 0].\end{array}\right.\end{aligned}\ ] ] inserting equations - in equations and , we find equations and respectively . @xmath138 [ grprop ] define @xmath139 , @xmath140 and @xmath141 by [ transfs ] @xmath142 where @xmath143 the functions defined in satisfy the so - called global relation : @xmath144 , \ \ k_1\in \mathbb r , \ \mathrm{im } k_2\leq 0.\ ] ] the linearised kp equation can be written in divergence form as : @xmath145_t+3\bigg[e^{-ik_1x - ik_2y+\omega(k)t}\bigg(-ik_1q_y - ik_2q_x\bigg)\bigg]_y\nonumber\\ & + \bigg[e^{-ik_1x - ik_2y+\omega(k)t}\bigg(-ik_1q_{xxx}+ik_1 ^ 3q_x+k_1 ^ 2q_{xx}+3ik_2q_y\bigg)\bigg]_x=0.\end{aligned}\ ] ] green s theorem in the plane implies @xmath146 integrating equation , we obtain the identity @xmath147 valid for @xmath148 and @xmath149 , which is equation . @xmath138 * remark 2.1 * we also note that if @xmath64 is replaced by @xmath150 , equation becomes @xmath151\ ! , \ k_1\in \mathbb r , \ \mathrm{im } k_2\leq 0,\ ] ] with @xmath29 and @xmath30 defined by equations analogous to and . * remark 2.2 * the global relation can also be obtained for the lax pair . in fact , we have that @xmath152 and @xmath153 . \end{split}\ ] ] these two equations are compatible iff @xmath154\right)_y.\end{aligned}\ ] ] hence , green s theorem over a domain @xmath155 yields @xmath156dy=\\ & = \left\{e^{l(l+2k)y-4il(l^2 + 3kl+3k^2)t}\int_{-\infty}^{\infty}dx\ , e^{-ilx}\ , 3\bigg[g_\xi+2ikg+\partial_\xi^{-1}h\bigg]\right\}dt . \end{split}\ ] ] in particular , for the domain depicted in figure [ la1 ] we find the identity @xmath157,\quad l(l+2k_r)\leq 0.\end{aligned}\ ] ] the constraint on @xmath158 is imposed in order for the exponential @xmath159 to be bounded . we will use pompeiu s formula in order to reconstruct @xmath17 via a d - bar problem . since the functions @xmath160 and @xmath161 depend explicitly both on @xmath19 and on @xmath62 , we need to compute the jumps of @xmath17 across the real and the imaginary @xmath19-axes , as well as the @xmath19-bar derivatives of @xmath17 in each of the four quadrants . [ lkpipprop ] let @xmath17 be defined by equation . this function admits the following alternative representation in terms of appropriate transforms of @xmath162 , @xmath163 , @xmath164 : @xmath165 where @xmath166 and @xmath167 are defined by @xmath168,\label{sttilde}\end{aligned}\ ] ] and the contours @xmath169 and @xmath170 are the boundaries of the second and the third quadrant , with the orientation as shown in figure [ d1d2 ] . the definitions and of @xmath160 and @xmath171 yield the following equations : @xmath172 @xmath173 @xmath174 and @xmath175 equations and imply @xmath176 thus @xmath17 does not have a discontinuity across the imaginary @xmath19-axis . on the other hand , equations and imply @xmath177 and @xmath178 where the function @xmath179 is given by @xmath180 with @xmath113 defined by equation . hence , @xmath17 is discontinuous across the real @xmath19-axis . pompeiu s formula for the contour depicted in figure [ pompeiu ] yields @xmath181 in order to compute the associated d - bar derivatives , we recall the definition @xmath182 thus , by differentiating equation , we find @xmath183 similarly , using equation , we find @xmath184 thus , the d - bar derivatives of @xmath185 and @xmath186 are equal , i.e. , @xmath187 moreover , employing the global relation together with the identity @xmath188 we can express the above d - bar derivatives in terms of the initial and boundary values of @xmath39 and its derivatives : @xmath189 and @xmath190,\\ \frac{\partial \mu_{2}^-}{\partial \bar \nu}&=-\frac{1}{2\pi}\ , e^{-2i\nu_rx+4i\nu_r\nu_iy+8i\nu_r(3\nu_i^2-\nu_r^2)t}\left[\tilde q_0(\nu_r , \nu_i)+\tilde s_t(\nu_r , \nu_i)\right],\end{aligned}\ ] ] where @xmath36 and @xmath191 are defined by equations and respectively . thus , inserting the above expressions in pompeiu s formula , we obtain the following expression for @xmath17 : @xmath192 in the first two integrals on the rhs of equation , we make the change of variables @xmath193 the jacobian of this transformation is given by @xmath194 the domains of integration are transformed as shown in figure [ transform1 ] ; the first two terms on the rhs of equation become @xmath195 we next introduce the transformation @xmath196 then , integrating by parts and recalling definitions and , we find @xmath197\nonumber\\ & = -\tilde s_t(\nu_r,\nu_i).\end{aligned}\ ] ] thus , the first two integrals on the rhs of equation are equal to the following expression : @xmath198 using this expression into equation , we find equation . @xmath138 * remark 2.3 * in equation , @xmath191 can be replaced by @xmath199 , where @xmath200.\ ] ] indeed , we have that @xmath201 , \ \nu_r\geq 0 \\ \\ \arg \nu_i \in \left[\pi,2\pi\right ] , \ \nu_r\leq 0 \end{array } \right.\end{aligned}\ ] ] and @xmath202\leq 0 & \leftrightarrow\left\{\begin{array}{ll } \arg \nu_i\in \left[\dfrac{\pi}{2},\pi\right]\cup \left[\dfrac{3\pi}{2},2\pi\right ] , \ \nu_r\geq 0 \\ \\ \arg \nu_i\in \left[0,\dfrac{\pi}{2}\right]\cup \left[\pi,\dfrac{3\pi}{2}\right ] , \ \nu_r\leq 0 . \end{array } \right.\end{aligned}\ ] ] hence , the exponential involved in the integrals along the contours @xmath169 and @xmath170 is bounded and goes to zero as @xmath203 for @xmath204 , and for @xmath205 . hence , by cauchy s theorem and jordan s lemma in the regions @xmath206 and @xmath207 the term which involves the integral @xmath208 vanishes . the representation for @xmath17 suggests that @xmath209 consequently , equation allows us to write @xmath39 in terms of @xmath17 : @xmath210.\ ] ] hence , by differentiating equations we find : @xmath211 where @xmath166 and @xmath191 are defined by equations and respectively . according to remark 2.3 we can replace @xmath191 by @xmath199 , thus an alternative representation for @xmath39 is given by @xmath212 where @xmath166 and and @xmath199 are defined by equations and respectively . let us introduce the variables @xmath213 by the equations @xmath214 the jacobian of this transformation is given by @xmath215 thus , equation becomes @xmath216 where @xmath28 and @xmath199 are defined by equations and , @xmath217 is given by equation and the contours of integration are shown in figure [ d1d2 ] . the real part of @xmath217 is equal to @xmath218 and , since @xmath122 , the exponential term in equation is bounded in the domains @xmath219 and @xmath220 . hence , the contour @xmath170 can be deformed onto @xmath221 ( see figure [ d1d2 ] ) . then , equation implies @xmath222 integration by parts in the definition of @xmath199 yields that @xmath223,\end{aligned}\ ] ] hence , recalling the definitions and , we find the following expression : @xmath224\nonumber\\ & + \frac{1}{(2\pi)^2}\int_0^{\infty}\!\!\ ! dk_1\int_{\partial d_2^+}dk_2\ , e^{ik_1 x+ik_2 y-\omega(k)t}\,3\left[\frac{k_2}{k_1}\,\tilde g_t(k_1,k_2)-\frac{i}{k_1}\,\tilde h_t(k_1,k_2)\right],\end{aligned}\ ] ] where @xmath225 is defined by equation , @xmath28 , @xmath226 and @xmath227 are defined by equations , and respectively , and the contours @xmath169 and @xmath221 are shown in figure [ d1d2 ] . the solution @xmath39 of an ibv problem for the linearised kp in the domain @xmath58 is given by equation . according to remark 2.3 , an equivalent formula for @xmath39 is given by @xmath228\nonumber\\ & + \frac{1}{(2\pi)^2}\int_0^{\infty}\!\!\ ! dk_1\int_{\partial d_2^+}\!dk_2\ , e^{ik_1 x+ik_2 y-\omega(k)t}\,3\left[\frac{k_2}{k_1}\,\tilde g_t(k_1,k_2)-\frac{i}{k_1}\,\tilde h_t(k_1,k_2)\right]\!\!,\end{aligned}\ ] ] with @xmath29 and @xmath30 defined analogously to @xmath226 and @xmath227 . the formula defined by equation in terms of @xmath229 , @xmath24 , @xmath230 represents the solution to the linearised kp equation in @xmath58 , provided that the functions @xmath28 , @xmath29 , @xmath30 satisfy the global relation . an equivalent representation is given by equation . since equations and are equivalent , we can use either of them in the calculations that follow . by differentiating equation , it is straightforward to show that @xmath23 satisfies the linearised kp equation . : evaluating the expression at @xmath60 and using the fact that @xmath231 we find @xmath232 thus the initial condition is satisfied . : at @xmath80 , equation becomes @xmath233\nonumber\\ & + \frac{1}{(2\pi)^2}\int_0^{\infty}\!\!\ ! dk_1\int_{\partial d_2^+}\!dk_2\ , e^{ik_1 x-\omega(k)t}\,3\left[\frac{k_2}{k_1}\,\tilde g_t(k_1,k_2)-\frac{i}{k_1}\,\tilde h_t(k_1,k_2)\right].\end{aligned}\ ] ] under the summetry transformation @xmath234 of @xmath217 , the global relation assumes the form : @xmath235 while we notice that the functions @xmath29 and @xmath30 , which depend on @xmath236 only through @xmath217 , remain invariant . solving this expression for @xmath237 and substituting into equation , we find @xmath238 recalling the definition of @xmath27 , we note that the last term on the rhs of the above equation vanishes due to jordan s lemma on the complex @xmath236-plane . also , recalling the definition of @xmath36 , equation becomes @xmath239 the definition of @xmath225 implies @xmath240 hence we can deform the contours @xmath169 and @xmath221 onto the real @xmath19-axis in the third term on the rhs of equation , i.e. @xmath241 the first and the third term on the rhs of this equation cancel , while the change of variables @xmath242 on the second term yields @xmath243 using the identity , we find @xmath244 in a similar way , it can be shown that @xmath245 . the first lax equation suggests that there exists a solution @xmath17 with the asymptotic behaviour @xmath246 [ kpiidpprop ] assume that there exists a solution @xmath247 , for an initial - boundary value problem of equation . then , the function @xmath17 , which is bounded @xmath70 and is defined as @xmath248 in the different quadrants of the complex @xmath19-plane ( see figure [ eigenfunctions ] ) , admits the following representations in terms of @xmath23 , @xmath24 and @xmath25 : @xmath249 @xmath250 where @xmath251 the function @xmath252 is given by @xmath253\ ] ] and @xmath254 as defined by equations and . define the fourier transform of @xmath17 with respect to @xmath16 by [ ftpair ] @xmath255 , \quad l \in \mathbb r , \ k\in \mathbb c,\ ] ] with the inverse fourier transform given by @xmath256 applying the fourier transform on the first lax equation yields an ode in @xmath65 : @xmath257 where @xmath258 we solve this equation by integrating with respect to @xmath65 either from @xmath80 or from @xmath259 : @xmath260 where we have assumed that @xmath261 as @xmath53 . we look for a solution which is bounded @xmath70 . in the first expression in equation , @xmath83 , thus we require that @xmath84\geq0 $ ] , i.e. @xmath85\cup[-2k_r,\infty),\ k_r\leq 0 \\ \\ l\in(-\infty,-2k_r]\cup[0,\infty),\ k_r\geq 0 . \end{array } \right.\ ] ] analogous considerations are valid for the second expression in equation : @xmath86 implies that @xmath84\leq0 $ ] , i.e. @xmath262,\ k_r\leq 0 \\ \\ l\in[-2k_r , 0],\ k_r\geq 0 . \end{array } \right.\ ] ] applying the inverse fourier transform on equations and writing @xmath17 according to the definition , we obtain @xmath263 @xmath264 evaluating equation at @xmath80 and recalling the definition we find @xmath265 the definition yields @xmath266 this equation implies that @xmath95 does not satisfy any constraints for @xmath129\cup[-2k_r , \infty)$ ] , whereas it does satisfy the following constraint for @xmath97 $ ] : @xmath267.\ ] ] the second lax equation evaluated at @xmath80 yields @xmath268 where @xmath269 is defined as @xmath270.\end{aligned}\ ] ] by applying the fourier transform on equation , we obtain the following ode in @xmath64 : @xmath271 integrating this expression with respect to @xmath64 either from @xmath60 or from @xmath108 , we find : @xmath272 integration by parts of the integral in @xmath273 implies @xmath274 where the quantity @xmath252 is defined by equation , and @xmath275 in order for @xmath115 to be bounded @xmath70 , we require that the exponential involved in equation is bounded . noting that @xmath276=12k_il\left(l+2k_r\right),\ ] ] we are led to consider the following cases ( see also figure [ eigenfunctions ] ) : * if @xmath277 , then @xmath129\cup[-2k_r , \infty)$ ] implies that @xmath122 , whereas @xmath130 $ ] implies that @xmath124 . thus , recalling our analysis for @xmath95 , we can choose @xmath278\cup [ -2k_r , \infty),\ ] ] equation imposes @xmath279.\ ] ] therefore , using equation we find : @xmath280 . \end{array}\right.\end{aligned}\ ] ] similar calculations for @xmath19 in each of the other three quandrants yield the following results : * for @xmath281 , @xmath282\!\cup\![-2k_r , \infty)\\ \\ \int_{0}^{t}d\tau \ , e^{-4il(l^2 + 3kl+3k^2)(\tau - t ) } \widehat { h\phi_1 ^ -}\vspace{2mm}\\ -e^{4il(l^2 + 3kl+3k^2)t}\int_{0}^{\infty}d \eta \ , e^{l(l+2k)\eta}\widehat{q\mu_1 ^ -}\big|_{t=0},\ l\in[0,-2k_r].\end{array}\right.\end{aligned}\ ] ] * for @xmath283 , @xmath284\cup[0 , \infty)\\ \\ \!\!\!\!-\int_{t}^{t}d\tau \ , e^{-4il(l^2 + 3kl+3k^2)(\tau - t ) } \widehat { h\phi_2^+}\vspace{2mm}\\ \!\!\!\!\!-e^{-4il(l^2 + 3kl+3k^2)(t - t)}\!\int_{0}^{\infty}\!\!d \eta \ , e^{l(l+2k)\eta}\widehat{q\mu_2^+}\big|_{t = t},\ l\in[-2k_r,0].\end{array}\right.\end{aligned}\ ] ] * and for @xmath285 , @xmath286\!\cup\![0 , \infty)\\ \\ \int_{0}^{t}d\tau \ , e^{-4il(l^2 + 3kl+3k^2)(\tau - t ) } \widehat { h\phi_2 ^ -}\vspace{2mm}\\ -e^{4il(l^2 + 3kl+3k^2)t}\int_{0}^{\infty}d \eta \ , e^{l(l+2k)\eta}\widehat{q\mu_2 ^ -}\big|_{t=0},\ l\in[-2k_r , 0].\end{array}\right.\end{aligned}\ ] ] inserting equations - into equations and , we obtain for @xmath287 and @xmath288 the representations and respectively . @xmath138 [ grpropkpii ] define @xmath36 , @xmath37 and @xmath38 by equations , and respectively . also , define the functions @xmath115 and @xmath252 by equations and . then , these functions satisfy the so - called global relation : @xmath289 equation yields @xmath290 also , the second equation of the lax pair implies @xmath291 the above equations imply @xmath292 using green s theorem over the domain depicted in figure [ la1 ] , we obtain the global relation : @xmath293 note that the constraint on @xmath158 is needed in order for the exponential @xmath159 to be bounded . @xmath138 it will be shown later ( see equation ) that @xmath294 assuming that @xmath17 is continuous across the imaginary @xmath19-axis , pompeiu s formula @xmath295 applied in the simple domain @xmath45 ( see figure [ pompeiu ] ) , yields @xmath296 where @xmath297 therefore , we need to derive expressions for the discontinuities of @xmath17 across the real @xmath19-axis and for the d - bar derivatives of @xmath17 in the complex @xmath298plane . ( the d - bar derivatives)[kpiid - barprop ] define the functions @xmath185 and @xmath186 by equations , and . then , @xmath299 _ and _ @xmath300 where @xmath301 and @xmath302 , @xmath303 are defined in the appropriate quadrant of the complex @xmath19-plane by the following formulae : @xmath304 where the functions @xmath305 and @xmath252 are defined by equations and respectively . differentiating equation according to the definition of the d - bar derivative , we find the following expression : @xmath306 this expression can be written as an integral equation for @xmath307 : @xmath308 where @xmath309 is defined by @xmath310 using the identity , we can write equation in the following form : @xmath311 where @xmath312 denotes the heaviside function . the global relation for @xmath313 yields @xmath314 hence , equation implies that @xmath315 where @xmath316 are defined in proposition [ kpiidpprop ] . then , equation becomes @xmath317 let us introduce the notations @xmath318 and @xmath319 then , equation can be written in the form @xmath320$}\right.\nonumber\\ & -\frac{1}{2\pi}\int_{0}^{-2k_r } dl \int_{-\infty}^{\infty}d\xi\int_y^\infty d\eta\ , e^{-i(l+2k_r)(\xi - x)+(l(l+2k)+4ik_rk_i)(\eta - y ) } q \mathcal m_1^\pm.\label{m1}\end{aligned}\ ] ] let @xmath321 in the integrals on the rhs of equation . consequently , @xmath322 dropping the prime , equation becomes @xmath323$}\nonumber\\ & -\frac{1}{2\pi}\int_{2k_r}^{0 } dl \int_{-\infty}^{\infty}d\xi\int_y^\infty d\eta\ , e^{-il(\xi - x)+l(l-2\bar k)(\eta - y ) } q\mathcal m_1^\pm.\end{aligned}\ ] ] employing the transformation @xmath324 in the definition of @xmath252 , we find @xmath325 introducing the notation @xmath326 equation becomes : @xmath327 thus , it follows that this equation is identical with equation hence , assuming that the solution is unique , it follows that @xmath328 or , equivalently , @xmath329 therefore , using the definition , we obtain equation for the d - bar derivative of @xmath185 . in a similar way , we can derive the expression for the d - bar derivative of @xmath186 . @xmath138 in order to show that @xmath17 is continuous across the imaginary @xmath19-axis , we evaluate the expressions and at @xmath330 and then subtract the resulting equations : @xmath331$}\mbox{\large $ \big|$}_{k_r=0}.\end{aligned}\ ] ] this equation evaluated at @xmath80 implies @xmath332$}\mbox{\large $ \big|$}_{k_r=0};\end{aligned}\ ] ] hence , assuming uniqueness we find @xmath333 . then , it follows that @xmath334 satisfies a homogeneous integral equation , hence under the assumption of uniqueness , @xmath335 next , we compute the jump of @xmath17 across the real @xmath19-axis . ( the jumps across the real @xmath19-axis)[kpiidiscprop ] define the functions @xmath185 and @xmath186 , by equation , and . these functions satisfy the following discontinuity relations across the real @xmath19-axis : @xmath336 where @xmath337 are defined by equation , and @xmath338 the functions @xmath339 , @xmath340 , @xmath341 and @xmath342 are the solutions of the following systems of volterra integral equations : @xmath343 and @xmath344 with @xmath345 , @xmath346 and @xmath347 defined by @xmath348 and @xmath38 given by equation . evaluating equations at @xmath349 and subtracting the resulting expressions yields the following equation for @xmath350 : @xmath351 furthermore , using the identity , we can rewrite this expression in the following form : @xmath352 in what follows , by analysing the lax pair , we will construct an equation similar to equation . this equation will be satified by an appropriate combination of the functions @xmath353 and @xmath354 . therefore , assuming uniqueness , the solution @xmath355 of equation equals the above combination . consider the fourier transform pair , but now let @xmath356 . this transformation will allow us , after applying a suitable change of variables , to split the exponential terms into two parts : one which will match the exponentials on the rhs of equation and one which will be _ independent _ of @xmath158 : @xmath357 and @xmath358 we now follow a similar approach to the one presented in section [ kpiidp ] : applying the transform on the lax equation , we find @xmath359 using the change of variables @xmath360 equation becomes : @xmath361 with @xmath362 defined by equation . hence , @xmath363 where @xmath364 we solve this equation by integrating with respect to @xmath65 either from @xmath80 or from @xmath259 : @xmath365 where we have assumed that @xmath366 as @xmath53 . we are interested in obtaining an expression for the difference @xmath350 , thus we assume that @xmath367 . also , since @xmath287 must be bounded for @xmath367 , the exponential involved in equation must be bounded and hence , we need to control the sign of @xmath368 . we consider two cases for @xmath369 , namely @xmath370 or @xmath371 . according to the definition , the distinction between @xmath287 and @xmath288 is made on the basis of the sign of the real part of @xmath19 . however , the transformation implies that @xmath372 , hence @xmath373 in the first case , whereas @xmath374 in the second case . inverting the expression ( by making use of equation ) , recalling the definition and relabeling @xmath375 to @xmath158 , we find the following equations : @xmath376 which is valid for @xmath377 and @xmath378 which is valid for @xmath379 following the approach used in section [ kpiidp ] , we can employ the second lax equation in order to eliminate the dependence in @xmath65 from the second term on the rhs of the above expressions . this yields the following equations : @xmath380 which is valid for @xmath381 and @xmath382 which is valid for @xmath383 , where @xmath384 is defined by equation , @xmath385 is given by @xmath386 and @xmath252 is defined by equation . in order for @xmath387 to satisfy an equation similar to equation , we need to split the integrals with respect to @xmath158 . however , in order to ensure that the resulting integrals remain bounded , we impose two additional conditions : in equation , where @xmath370 , we actually require the stronger condition that @xmath388 , and in equation , where @xmath371 , we demand the stronger condition that @xmath389 . under these new restrictions , the expressions and become @xmath390 which is valid for @xmath391 and @xmath392 which is valid for @xmath393 . the quantities @xmath394 and @xmath395 are given by the following expressions : @xmath396 which is defined for @xmath397 , and @xmath398 which is defined for @xmath393 . we need to verify that the exponentials involved in the expressions , , and are bounded . this is straightforward for the exponentials @xmath399 , @xmath400 and @xmath401 in and . concerning @xmath394 , we have @xmath402 thus , the exponentials @xmath399 and @xmath403 , which are involved in the definition of @xmath394 , are bounded . concerning @xmath395 , we have @xmath404 thus , the exponentials @xmath399 and @xmath403 , which are involved in the definition of @xmath395 , are also bounded . the expressions for @xmath394 and @xmath395 can be rewritten in the following form : @xmath405 and @xmath406 applying the transformation to the identity , and then replacing @xmath158 by @xmath407 , we find the following identity : @xmath408 note that if @xmath158 satisfies the restrictions imposed in the domain of @xmath395 and @xmath394 then the exponential @xmath403 is bounded , hence the global relation is satisfied . thus , employing the global relation in the expressions and , we find @xmath409 and @xmath410 where @xmath411 and @xmath345 and @xmath346 are defined by equations and . in summary , we have obtained two integral equations for @xmath412 and @xmath413 , namely equations and , with the quantities @xmath394 and @xmath395 given by equations and respectively . next , we multiply equation by @xmath414 and integrate with respect to @xmath369 from @xmath415 to 0 ; similarly , we multiply equation by @xmath416 and integrate with respect to @xmath369 from @xmath417 to @xmath418 ( the precise form of the functions @xmath416 and @xmath414 will be determined later ) . adding the two resulting expressions and defining @xmath419 by @xmath420 we find @xmath421 where @xmath385 is defined by equation , and @xmath422 in order to identify equation with equation , the functions @xmath416 and @xmath414 must be defined in such way that equations and have the same forcing , i.e. so that @xmath423 where @xmath424 is defined by equation and @xmath347 is given by equation . this implies that @xmath416 and @xmath414 must satisfy the following condition : @xmath425 where @xmath158 has been relabeled as @xmath369 on the rhs . our aim is to obtain appropriate integral equations for @xmath416 and @xmath414 . in this respect , we note that we can interchange the order of integration with respect to @xmath369 and @xmath158 : @xmath426 @xmath427 @xmath428 and @xmath429 [ int1_2 ] therefore , equation becomes : @xmath430 in addition , it follows from the definitions and of @xmath401 and @xmath424 , that @xmath431 hence the transformation @xmath432 in the terms involving the exponential @xmath424 yields @xmath433 this condition can be identically satisfied provided that we set both curly brackets equal to zero , i.e. provided that @xmath434 and @xmath435 which are equations and respectively . note that the restrictions on @xmath369 originate from equation ; furthermore , these restrictions ensure that @xmath345 and @xmath346 are well defined , see equations and . equations and can be derived in a similar way . @xmath138 consider the initial - boundary value problem for the kpii equation , with the initial and the boundary values defined by equations - . this problem is well posed provided that one of the two boundary values , or , is known . it is then possible to obtain the unknown boundary value in terms of the initial condition and the given boundary condition via the global relation . now , we will define a map from the initial condition @xmath36 and from the boundary values @xmath37 and @xmath38 to the spectral functions of propositions [ kpiid - barprop ] and [ kpiidiscprop ] : @xmath436 where @xmath437 , @xmath367 for @xmath438 and @xmath439 for @xmath440 , @xmath441 for @xmath442 , @xmath443 for @xmath444 and @xmath445 . 1 . @xmath446 , for @xmath447 and @xmath441 , where the functions @xmath448 and @xmath449 are defined in terms of @xmath36 via the following linear integral equations : + [ maprhoa ] @xmath450 @xmath451 2 . @xmath452 , + for @xmath437 and @xmath443 , + where the functions @xmath453 and @xmath454 are defined in terms of @xmath455 via the following linear integral equations : + [ maprhob ] @xmath456 @xmath457 + and + [ mapphi ] @xmath458 @xmath459 + with @xmath252 defined by equation . + the motivation for the above definitions emanates from proposition [ kpiidpprop ] . indeed , evaluating equations and at @xmath60 and letting @xmath460 , we recover equations - . furthermore , applying the inverse fourier transform for @xmath305 on equations and , we find equations and . 3 . @xmath461 , + where the functions @xmath462 and @xmath303 are defined by the equations : + [ mapsp ] @xmath463 4 . @xmath464 , with @xmath465 , + where the functions @xmath466 and @xmath347 are defined by the following equations : + [ maprp ] @xmath467 + with @xmath401 given by equation and @xmath468 5 . @xmath469 , + where the functions @xmath470 and @xmath471 are defined via the following integral equations : + @xmath472 + and + @xmath473 in summary , steps 1 - 5 provide the map from @xmath455 to the spectral functions . as mentioned in the beginning of section [ invprob ] , the function @xmath17 satisfied pompeiu s formula . we notice that , under the definition of the exponential @xmath424 , equations and can be written as @xmath474 and @xmath475 using these expressions for the d - bar derivatives , as well as equations and for the discontinuities across the real @xmath19-axis , pompeiu s formula yield the following expressions : @xmath476(x , y , t,\nu_r,\lambda ) \nonumber\\ & + \frac{1}{2i\pi}\int_{0}^{\infty } \frac{d\nu_r } { \nu_r - k}\,\bigg[\int_{-\infty}^{-2\nu_r}d\lambda\,\psi_2\left(e_{\lambda}\breve \mu_2^+\right)+\int_{0}^{\infty}\!d\lambda\,\psi_1\left(e_{\lambda}\breve \mu_1^+\right)\!\bigg](x , y , t,\nu_r,\lambda)\nonumber\\ & + \frac{1}{\pi}\int_{0}^{\infty}d\nu_r\int_{0}^{\infty}\frac{d\nu_i}{\nu - k}\,\gamma_2^{+}(\nu_r,\nu_i)e(x , y , t,\nu,-2\nu_r ) \mu_1^+(x , y , t,-\nu_r , \nu_i)\nonumber\\ & -\frac{1}{\pi}\int_{-\infty}^{0}d\nu_r\int_{0}^{\infty}\frac{d\nu_i}{\nu - k}\,\gamma_1^{+}(\nu_r,\nu_i ) e(x , y , t,\nu,-2\nu_r ) \mu_2^+(x , y , t,-\nu_r , \nu_i)\nonumber\\ & + \frac{1}{\pi}\int_{0}^{\infty}d\nu_r\int_{-\infty}^{0 } \frac{d\nu_i}{\nu - k}\,\gamma_2^{-}(\nu_r,\nu_i)e(x , y , t,\nu,-2\nu_r ) \mu_1 ^ -(x , y , t,-\nu_r , \nu_i)\nonumber\\ & -\frac{1}{\pi}\int_{-\infty}^{0}d\nu_r\int_{-\infty}^{0 } \frac{d\nu_i}{\nu - k}\,\gamma_1^{-}(\nu_r,\nu_i)e(x , y , t,\nu,-2\nu_r ) \mu_2 ^ -(x , y , t,-\nu_r , \nu_i).\end{aligned}\ ] ] using the transformation @xmath477 the second and the third terms on the rhs of equation become @xmath478 [ j1c1 ] moreover , letting @xmath479 , we find @xmath480 the definitions and imply @xmath481 replacing the tilde by the hat and letting @xmath482 , we can write the second and the third term as follows : @xmath483 the same transformations applied on the fourth and the fifth term on the rhs of equation yield @xmath484 dropping the hats and inserting equations and into equation , we conclude that @xmath485 [ qfinal ] assume that @xmath23 satisfies an ibv problem for the kpii equation with given initial condition @xmath486 . let @xmath487 assume that the following global relation is valid : @xmath488 where @xmath489 are bounded @xmath70 , @xmath490 , @xmath491 are defined by equations , and @xmath252 is given by , i.e. by @xmath492.\ ] ] given @xmath162 , @xmath163 and @xmath164 , define the functions @xmath493 by @xmath494 where the functions @xmath495 are given by @xmath496 assume that the following systems of volterra integral equations have a unique solution for the functions @xmath497 and @xmath498 respectively : @xmath499 and @xmath500 where @xmath501 and @xmath401 is given by equation . given @xmath502 , define @xmath17 as the solution of equation , where @xmath503 and assume that equation has a unique solution . define @xmath23 by @xmath210.\ ] ] then @xmath23 solves the kpii equation . * remark 3.1 * by differentiating equation with respect to @xmath16 we can write an integral representation for the solution @xmath39 : @xmath504 * remark 3.2 * ( the case of zero initial condition ) consider the special case @xmath505 . then , equations - imply that the functions @xmath470 and @xmath471 are given explicitly by @xmath506 with @xmath347 defined by equation . also , equations and yield @xmath507 thus , according to equations - : @xmath508 and @xmath509 where @xmath510 are defined by equations and . hence , equation for @xmath17 becomes significantly simpler , i.e. @xmath511 where @xmath512 , the functions @xmath347 and @xmath510 are given by @xmath513 and @xmath305 are defined by equations . suppose now that @xmath39 is small , i.e. let @xmath514 then , the @xmath515 term of the nonlinear kpii equation yields a linear equation which , as expected , is the linearised kp equation : @xmath516 the following integral representation of the solution of an ibv problem was derived in section [ lkp ] : @xmath517 -5 mm where -5 mm @xmath518 for the nonlinear problem , equations and suggest that , in the linear limit , @xmath519 . moreover , equations , and imply that @xmath520 and then , equations and suggest that @xmath521 and @xmath522 . in fact , we can write @xmath523 for some functions @xmath524 , @xmath525 , @xmath526 and @xmath527 , independent of @xmath528 . then , equation yields @xmath529$}.\ ] ] integrating by parts , we find @xmath530,\ ] ] therefore , equations and yield @xmath531 a similar analysis for @xmath342 and @xmath532 implies @xmath533 furthermore , equations - suggest that @xmath534 and @xmath535 , where @xmath536.\label{alpha}\end{aligned}\ ] ] hence it follows from the definitions and that @xmath537 and @xmath538 . then , the @xmath515 term of equation yields @xmath539 employing the formulae , and into equation , and using the definition @xmath540,\ ] ] we recover the expression . it is shown in @xcite that there exists a much simpler derivation for the integral representation of @xmath39 for the linearised kpii equation than the one presented in section [ lkp ] . however , this simpler approach can not be `` nonlinearised '' . on the other hand , the approach presented in section [ lkp ] contains all necessary conceptual steps needed for the solution of kpii . however , the analysis of kpii is technically quite involved . in particular , in order to formulate proposition , it is necessary to compute both the d - bar derivatives @xmath541 and the jumps @xmath542 ; the latter computation is quite complicated , as was also the case for the analogous computation is dsii @xcite . we note that an additional novelty of the ibv problems in 2 + 1 , as opposed to ibv problems in 1 + 1 , is that now the global relation plays an even more important role . namely , it is necessary in order to express the jumps of @xmath17 across the real @xmath19-axis in terms of the initial and the boundary values . several problems remain open , in particular : 1 . for a well posed problem , either @xmath37 or @xmath38 are prescribed as boundary conditions ; on the other hand , the solution obtained in proposition [ qfinal ] depends on _ both _ @xmath37 and @xmath38 . thus , in order for this solution to be effective it is necessary to use the global relation to eliminate the unknown boundary value . for the linearised kpii , this is achieved in @xcite ; the analogous problem for the kpii remains open . in this connection we note that until recently , for problems in 1 + 1 , it was not possible to express the spectral functions directly in terms of the given boundary conditions , but it was necessary to first obtain the unknown boundary values , i.e. to determine the so - called dirichlet to neumann map ( see @xcite-@xcite ) . however , in a recent breakthrough , it has been shown that for the nonlinear schrdinger ( nls ) equation on the half - line it is possible to express the spectral functions directly in terms of the given initial and boundary conditions . the question of whether this new approach can be extended to equations in 2 + 1 remains open . 2 . for the initial - value problem of the kpii , the linear integral equation analogous to admits a unique solution for @xmath17 for real initial conditions . this is due to the existence of a so - called `` vanishing lemma '' , which is based on the theory of generalised analytic functions of vekua . the question of whether there exists an analogous result for equation remains open . the spectral functions are defined in terms of linear integral equations . the question of existence and uniqueness for these equations remains open . actually , it is expected that these equations do possess homogeneous solutions which will give rise to coherent structures . 4 . in spite of the fact that the representation of @xmath39 involves both @xmath37 and @xmath38 , it should still be possible to obtain effective formulae for the large @xmath64 asymptotics of the solution . ( for equations in 1 + 1 this has been achieved by the deift - zhou approach @xcite , @xcite ) 5 . the new method is particularly effective for a class of boundary conditions called linearisable . in this case , it is possible to express the spectral functions directly in terms of the given initial and boundary conditions using only _ algebraic _ manipulations , see for example @xcite-@xcite . the question of identifying lineasable boundary conditions for the kpii remains open . 6 . for kpii the formalism presented here involves the crucial assumption that several linear integral equations have a unique solution . in spite of the fact that these equations are of fredholm type , it is not difficult to establish uniqueness under the assumption of sufficiently `` small data '' . however , the elimination of the `` small norm '' assumption is a formidable task ( see also remark 3 above ) . 111 a. s. fokas , the davey - stewartson equation on the half - plane , comm . phys . * 289 * , 957 - 993 ( 2009 ) . a. s. fokas and m. j. ablowitz , on the inverse scattering of the time dependent equation and the associated kpi equation , stud . appl . math . * 69 * , 211 - 228 ( 1983 ) . m. j. ablowitz , d. baryaacov and a. s. fokas , on the inverse scattering transform for the kadomtsev - petvisvhili equation , stud . appl . math . * 69 * , 135 - 143 ( 1983 ) . c. s. gardner , j. m. greene , m. d. kruskal , and r. m. miura , method for solving the korteweg - de vries equation , phys . * 19 * , 1095 - 1097 ( 1967 ) . m. boiti , j.j .- p . leon , f. pempinelli , spectral transform and orthogonality relations for the kadomtsev - petviashvili equation , physics letters a , * 141 * , 96 - 100 ( 1989 ) . a. s. fokas and l. y. sung , on the solvability of the n - wave , the davey - stewartson and the kadomtsev - petviashvili equation , inverse problems * 8 * , 673 - 708 ( 1992 ) . a. s. fokas , the kadomtsev - petviashvili equation revisited and integrability in 4 + 2 and 3 + 1 , stud . appl . math . * 122 * , 347 - 359 ( 2009 ) . r. beals and r.r . coifman , linear spectral problem , nonlinear equations and the @xmath543 method , inverse problem * 5 * , 87 - 130 ( 1989 ) . a. s. fokas a unified transform method for solving linear and certain nonlinear pdes , proc . a * 453 * , 1411 - 1443 ( 1997 ) . a. s. fokas and b. pelloni , the solution of certain initial - boundary value problems for the linearized korteweg - de vries equation , proc . phys . and eng . sciences * 454 * , 645 - 657 ( 1998 ) . a. s. fokas , a new transform method for evolution pdes , i m a j. appl . math . * 67 * , 1 - 32 ( 2002 ) . a. s. fokas , _ a unified approach to boundary value problems _ , siam ( 2008 ) . a. boutet de monvel , a. s. fokas and d. shepelsky , the analysis of the global relation for the nonlinear schrdinger equation on the half - line , lett . math . phys . * 65 * , 199 - 212 ( 2003 ) . a. s. fokas , a generalised dirichlet to neumann map for certain nonlinear evolution pdes , comm . pure appl . math . * lviii * , 639 - 670 ( 2005 ) . p. a. treharne , a. s. fokas , the generalized dirichlet to neumann map for the kdv equation on the half - line , j. nonlinear science * 18 * , 191 - 217 ( 2008 ) . p. deift and x. zhou , a steepest descent method for oscillatory riemann - hilbert problems , bull.amer.math.soc . ( n.s . ) * 20 * , 119 - 123 ( 1992 ) . p. deift and x. zhou , a steepest descent method for oscillatory riemann - hilbert problems , asymptotics for the mkdv , ann . of math . ( 2 ) * 137 * , 245 - 338 , ( 1993 ) . a. s. fokas , integrable nonlinear evolution equations on the half - line , comm . 230 * , 1 - 39 ( 2002 ) . a. s. fokas , linearizable initial - boundary value problems for the sine - gordon equation on the half - line , nonlinearity * 17 * , 1521 - 1534 ( 2004 ) . a. s. fokas , a. r. its and l. y. sung , the nonlinear schrdinger equation on the half - line , nonlinearity * 18 * , 1771 - 1822 ( 2005 ) . a. s. fokas and j. lenells , explicit soliton asymptotics for the korteweg - de - vries equation on the half - line , nonlinearity * 23 * , 937 - 976 ( 2010 ) . a. s. fokas and b. pelloni , boundary value problems for the elliptic sine - gordon equation in a semi - strip ( submitted ) . a. s. fokas and j. lenells , linearizable boundary value problems for the elliptic sine - gordon and the elliptic ernst equations ( submitted ) . asf is grateful to guggenheim foundation for partial support and to v. zakharov for suggesting the investigation of kpii on the half - plane . d. mantzavinos was supported by an epsrc doctoral training grant .

we present the results of a program to monitor the four - image gravitational lens b1608 + 656 with the vla . the system was observed over a seven month period from 1996 october to 1997 may . the 64 epochs of observation have an average spacing of 3.6 d. the light curves of the four images of the background source show that the flux density of the background source has varied at the% level . we measure time delays in the system based on common features that are seen in all four light curves . the three independent time delays in the system are found to be d , d , and d at 95% confidence . the uncertainties on the time delays are determined by monte carlo simulations which use fake light curves that have the characteristics of the observed light curves . this is the first gravitational lens system for which three independent time delays have been measured . a companion paper presents a mass model for the lensing galaxy which correctly reproduces the observed image positions , flux density ratios , and time delay ratios . the last condition is crucial for determining with a four - image lens . we combine the time delays with the model to obtain a value for the hubble constant of km s mpc at 95% confidence ( statistical ) for . in addition , there is an estimated systematic uncertainty of km s mpc from uncertainties in modeling the radial mass profiles of the lensing galaxies . the value of presented in this paper is comparable to recent measurements of from the gravitational lenses 0957 + 561 , pg 1115 + 080 , b0218 + 357 , and pks 1830 .

astro-ph9907257

even before the discovery of the first gravitational lens system , a technique for using gravitational lenses to measure the distance scale of the universe had been developed ( @xcite ) . the technique requires a lens system in which multiple images of the background source are formed . a `` map '' of the geodesics along which the light travels to form the images is constructed and used to predict the differences in light travel times along the geodesics . if the background source is variable , these time delays can be measured as each image varies in turn . the ratios between the observed and predicted delays give the hubble constant in the assumed world model @xmath10 . the use of gravitational lenses for determining @xmath4 has major advantages over traditional `` distance ladder '' approaches . first , the technique gives a direct estimate of @xmath4 at cosmological distances , where the effects of peculiar velocities are minimal . second , this measurement of @xmath4 is obtained in one step , without the propagation of errors inherent in the distance ladder approach . with the discovery that the `` double quasar '' 0957 + 561 a , b was a lens system ( @xcite ) , the effort to use lenses to measure @xmath4 began in earnest . for many years the effort was hindered by both the paucity of known lens systems and the difficulty in measuring time delays in the systems . in fact , an unambiguous time delay has only recently been measured in 0957 + 561 in spite of over ten years of intensive monitoring ( @xcite ; @xcite ; @xcite ; @xcite ) . another of the earliest known lenses , pg 1115 + 080 ( @xcite ) , has also just produced measurable time delays ( @xcite ) . however , we may have entered a new era for time delay measurements from gravitational lenses . one reason for this is the accelerated rate of discovery of new lenses from systematic radio surveys . the jodrell - vla astrometric survey ( jvas ; @xcite ; @xcite ; @xcite ) has produced 6 new lenses , and time delays have been measured for one of them ( b0218357 ; @xcite ) . the ongoing cosmic lens all - sky survey ( class ; @xcite ) has found 12 new lenses since it began in 1994 . this paper reports the first measurement of time delays from a class lens . the class project is a large search for gravitational lenses with the vla , with an explicit goal of finding lens systems which can be used to measure @xmath4 . the gravitational lens b1608 + 656 ( ra : 16 09 13.956 , dec : + 65 32 28.971 , j2000 ) was observed in the first phase of class and was immediately recognized as a lens system . the radio discovery image shows four unresolved components in a typical lens geometry ( @xcite ; see fig . [ fig_1608map ] for a map of the system ) . the system was also discovered in a search for gigahertz - peaked spectrum sources and was found to be the lensed core of a classical radio double source ( @xcite ) . further investigations of the system have provided data crucial for using the system for measuring @xmath4 . we have measured the lens redshift ( @xmath11 ; @xcite ) and source redshift ( @xmath12 ; @xcite ) . optical and infrared observations taken with the _ hubble space telescope _ reveal that the background source is being lensed by a pair of possibly merging galaxies ( @xcite ) . the positions of the lensing galaxies relative to the lensed images are important constraints on models of the lensing potential ( @xcite ; hereafter paper 2 ) . the hst images also show arcs due to the lensing of stellar emission from the background source . these arcs could be used as further constraints of the lens model . flat - spectrum cores of radio galaxies such as the lensed object in b1608 + 656 are often variable . to test for variability in the b1608 + 656 background source , we made several observations of the system with the vla , separated by time - scales of months . these data showed that the flux density of the background source varied by up to 15% . the variability makes b1608 + 656 an excellent candidate for a dedicated monitoring program to determine time delays . this paper presents the results of vla monitoring from october 1996 to may 1997 . these observations have resulted in the measurement of the three independent time delays in the b1608 + 656 system and a subsequent determination of @xmath4 . the hubble constant is expressed as @xmath13 . throughout this paper we assume @xmath14 . the effect of varying the cosmological model is treated in paper 2 . we observed b1608 + 656 between 1996 october 10 and 1997 may 09 , during which time the vla was in the a , bna , and b configurations . the 64 epochs were separated , on average , by 3.6 d. the observations were carried out at 8.5 ghz , giving angular resolutions ranging from 025 to 07 in the different array configurations . the observations are summarized in table [ tab_obs ] . the typical observation is 60 min long and includes scans on b1608 + 656 , a flux calibrator ( 3c286 or 3c48 ) , a phase calibrator ( 1642 + 689 ) chosen from the vla calibrator list ( @xcite ) , and two secondary flux calibrators ( 1634 + 627 and 1633 + 741 ) . the secondary flux calibrators are nearby steep - spectrum sources that are not expected to vary over the time - scales of the observations . we observe these sources to determine corrections for errors in the absolute flux calibration from epoch to epoch . the basic observing pattern is : * 1642 + 689 ( 1 min on source ) * 1634 + 627 ( 1 min on source ) * b1608 + 656 ( 46 min on source ) * 1633 + 741 ( 2 min on source ) a typical 60 min observation begins with a 9 10 min scan ( including slew time ) on the flux calibrator , contains three repetitions of the basic pattern on b1608 + 656 , and ends with scans on 1642 + 689 ( 1 min ) and the flux calibrator again ( @xmath03 min ) . for the few observations that are 30 min or 90 min in length , the number of repetitions of the basic pattern is altered . rlclrrl 1996 oct 10 & 366 & d@xmath15a & 16:30 & 60 & 27 & + 1996 oct 12 & 368 & d@xmath15a & 15:30 & 30 & 5 & + 1996 oct 16 & 372 & d@xmath15a & 21:00 & 60 & 21 & + 1996 oct 18 & 374 & a & 17:30 & 60 & 21 & + 1996 oct 20 & 376 & a & 01:30 & 60 & 21 & elev . thunderstorms + 1996 oct 23 & 379 & a & 17:30 & 30 & 6 & + 1996 oct 26 & 382 & a & 14:30 & 60 & 18 & @xmath17k . + 1996 oct 29 & 385 & a & 22:30 & 30 & 5 & elev . @xmath16 . gusting winds + 1996 oct 31 & 387 & a & 20:00 & 60 & 18 & + 1996 nov 01 & 388 & a & 18:30 & 60 & 18 & + 1996 nov 03 & 390 & a & 18:00 & 60 & 21 & wind @xmath18 10 m / s + 1996 nov 07 & 394 & a & 07:00 & 30 & 6 & elev . @xmath16 + 1996 nov 08 & 395 & a & 13:00 & 60 & 21 & + 1996 nov 11 & 398 & a & 13:00 & 60 & 18 & + 1996 nov 12 & 399 & a & 07:30 & 30 & 8 & elev . @xmath16 + 1996 nov 15 & 402 & a & 17:00 & 60 & 21 & high wind gusts + 1996 nov 18 & 405 & a & 15:30 & 60 & 21 & + 1996 nov 23 & 410 & a & 17:00 & 60 & 20 & wind @xmath18 10 m / s + 1996 nov 27 & 414 & a & 21:00 & 60 & 20 & + 1996 dec 01 & 418 & a & 14:00 & 60 & 20 & + 1996 dec 05 & 422 & a & 14:30 & 30 & 7 & + 1996 dec 07 & 424 & a & 15:00 & 30 & 5 & + 1996 dec 10 & 427 & a & 17:00 & 60 & 20 & wind @xmath18 10 m / s + 1996 dec 15 & 432 & a & 17:00 & 90 & 34 & + 1996 dec 20 & 437 & a & 19:15 & 30 & 5 & + 1996 dec 23 & 440 & a & 17:30 & 90 & 34 & + 1996 dec 24 & 441 & a & 15:30 & 30 & 10 & + 1996 dec 29 & 446 & a & 15:00 & 60 & 20 & + 1997 jan 03 & 451 & a & 16:30 & 60 & 20 & wind @xmath18 10 m / s + 1997 jan 07 & 455 & a & 18:30 & 90 & 35 & + 1997 jan 11 & 459 & a & 16:00 & 60 & 20 & + 1997 jan 16 & 464 & a@xmath15bna & 16:00 & 60 & 20 & + 1997 jan 17 & 465 & a@xmath15bna & 09:30 & 30 & 8 & elev . flurries + 1997 jan 20 & 468 & a@xmath15bna & 15:30 & 60 & 20 & + 1997 jan 26 & 474 & bna & 14:00 & 60 & 20 & rain + 1997 jan 30 & 478 & bna & 13:00 & 60 & 19 & + 1997 feb 02 & 481 & bna & 20:00 & 60 & 20 & + 1997 feb 08 & 487 & bna & 19:00 & 60 & 19 & + 1997 feb 13 & 492 & bna@xmath15b & 15:30 & 60 & 18 & + 1997 feb 18 & 497 & b & 18:00 & 60 & 19 & + 1997 feb 23 & 502 & b & 18:00 & 60 & 19 & + 1997 feb 28 & 507 & b & 18:00 & 60 & 19 & snow storms . + 1997 mar 03 & 510 & b & 18:00 & 60 & 19 & + 1997 mar 08 & 515 & b & 16:00 & 60 & 18 & + 1997 mar 14 & 521 & b & 18:15 & 75 & 30 & + 1997 mar 16 & 523 & b & 17:45 & 45 & 12 & + 1997 mar 21 & 528 & b & 18:00 & 60 & 20 & + 1997 mar 25 & 532 & b & 21:30 & 60 & 19 & wind @xmath18 10 m / s . snowing . + 1997 mar 30 & 537 & b & 19:00 & 90 & 30 & + 1997 apr 01 & 539 & b & 17:30 & 60 & 20 & + 1997 apr 05 & 543 & b & 18:00 & 60 & 20 & + 1997 apr 07 & 545 & b & 18:00 & 60 & 20 & + 1997 apr 11 & 549 & b & 18:00 & 60 & 20 & + 1997 apr 15 & 553 & b & 18:00 & 60 & 20 & + 1997 apr 19 & 557 & b & 15:00 & 60 & 18 & + 1997 apr 22 & 560 & b & 18:00 & 60 & 20 & + 1997 apr 26 & 564 & b & 18:00 & 60 & 20 & + 1997 may 03 & 571 & b & 18:00 & 60 & 20 & + 1997 may 08 & 576 & b & 18:00 & 90 & 30 & + 1997 may 13 & 581 & b & 18:00 & 60 & 20 & + 1997 may 17 & 585 & b & 16:30 & 60 & 19 & + 1997 may 21 & 589 & b & 18:00 & 60 & 20 & + 1997 may 23 & 591 & b & 18:00 & 60 & 20 & + 1997 may 26 & 594 & b & 20:00 & 60 & 19 & + the data reduction is separated into two major steps , calibration and mapping . the data for each epoch are calibrated using standard routines in the nrao data reduction package aips . before calibration , the data quality for all sources is assessed and bad points are flagged with the edita and tvflg tasks . both of the flux calibrators are heavily resolved by the vla in a configuration at 8.5 ghz . hence , we can not treat the calibrators as point sources without limiting the number of baselines that can be used to calculate phase and gain solutions . in order to increase the number of baselines available for the calculations , we create models of 3c286 and 3c48 that incorporate the extended emission from the sources . we combine observations from several epochs with the dbcon task in aips . the resulting data sets have excellent @xmath19-plane coverage , from which we can make high dynamic - range maps . the mapping , which is performed in the difmap package ( @xcite ) , consists of alternating iterations of cleaning ( @xcite ) and self - calibration . the final lists of clean components are read back into aips and serve as the calibrator models . the procedures for the phase calibrator are simpler because the emission from 1642 + 689 is dominated by an unresolved component . thus , the assumption that 1642 + 689 is a point source leads to adequate phase and gain solutions . these calibration solutions are applied to the b1608 + 656 , 1633 + 741 , and 1634 + 627 data . we map the data and determine flux densities using the difmap package . we do not expect to see any structural changes over the course of the observations ; only changes in flux densities should be observed . to treat the data from each epoch in a uniform fashion and to shorten the mapping procedure , we create models of the observed source structures from high dynamic - range maps . for epochs with noisy data , these models are needed to fix the locations of the regions of low surface brightness emission , which otherwise would not be well - constrained by the data . to make the high dynamic - range maps of each source , we combine 13 high - quality data sets from a and b configuration observations with the dbcon task . the combined data sets , which have excellent @xmath19 coverage , are then mapped in difmap . all the maps are made with natural weighting . the secondary flux calibrators , which have significant emission from extended structures , are mapped by using an iterative cycle of cleaning and self - calibration . both phase and amplitude self - calibration are used . the models for these sources consist of the final lists of clean components . the emission from b1608 + 656 is dominated by the four unresolved images of the background source ( fig . [ fig_1608map ] ) . hence , instead of cleaning the data , we assign point - source model components to the four images of the background source . we then use the difmap modelfit function , which varies the component positions and flux densities to obtain the best fit to the @xmath19-plane visibilities . the model fitting iterations are alternated with phase and amplitude self - calibration . in later rounds of the model fitting , several nearby weak sources are seen in the residual maps . these sources are included in the model for the last few iterations of the model fitting . the nearby sources can be seen in fig . [ fig_1608field ] ; their locations and flux densities are listed in table [ tab_posflux ] . lrrr b1608 + 656 a & @xmath200.018 & @xmath201.001 & 34.3 + b1608 + 656 b & @xmath200.757 & @xmath90.961 & 16.8 + b1608 + 656 c & @xmath200.763 & @xmath200.548 & 17.4 + b1608 + 656 d & @xmath91.111 & @xmath90.256 & 5.9 + b1608 + 656 n & @xmath2015.0 & @xmath2023.6 & 0.6 + b1608 + 656 s & @xmath913.9 & @xmath913.0 & 0.5 + source 1 & @xmath934.7 & @xmath20167.0 & 3.4 + source 2 & @xmath942.0 & @xmath961.2 & 0.8 + source 3 & @xmath9178.0 & @xmath943.5 & 0.2 + source 4 & @xmath926.7 & @xmath2024.6 & 0.2 + the first step in the mapping procedure at each epoch is to read in the data and perform a phase - only self - calibration against the model of the source . this procedure aligns the phase center of the observation with that of the model and eliminates the need for many early steps of cleaning and self - calibration . after this point , the procedures used for b1608 + 656 differ from those used for the secondary flux calibrators , as discussed below . for each epoch , we determine the flux densities of the four lensed images in the b1608 + 656 system as follows . after the initial phase self - calibration step , the model described in the previous section is varied to find the best fit to the @xmath19-plane visibilities for that epoch . the component positions are held fixed and only the flux densities are allowed to vary . after several iterations of model fitting , another phase self - calibration is performed against the new model and more iterations of the model fitting are performed . at this point , the component flux densities and the rms noise in the residual map are recorded in a log file ( the `` phasecal '' data set ) . we then perform an overall gain calibration on the data , getting one gain correction per antenna for the observation . typical gain corrections are on the order of 1 2% . another round of model fitting is performed and the final component flux densities and rms noise in the residual map are recorded in the log file ( the `` gscale '' data set ) . we record the two separate data sets in case the gain calibration introduces any errors which may bias the subsequent analysis . all subsequent analysis is performed on both data sets and no significant differences are seen in the results . the steep - spectrum secondary flux - density calibrators contain significant extended emission and should not vary over the course of the observations . we expect that any observed variations in flux density are due to errors in the absolute flux calibration , and as such can be expressed as an overall scaling of the model clean component flux densities . that is , the clean component flux densities should not vary with respect to each other . the task of finding the overall flux density of these sources is thus simplified into finding the scale factor @xmath21 that , when multiplied by the component flux densities , gives the best fit to the data . to find the best - fit scaling for each data set , we create 11 scaled model files , based on the clean - component models described above . the 11 files have @xmath22 ranging from 0.9 to 1.1 in steps of 0.02 . each of the scaled models is compared to the @xmath19-plane visibilities and a reduced @xmath23 goodness - of - fit value is returned . we then fit a parabola to the points in the reduced-@xmath23 curve and find the value of @xmath22 which corresponds to the minimum reduced @xmath23 . this scaling gives the total flux density of the source at that epoch . moore & hewitt ( 1997 ) , in their analysis of the 15 ghz light curves of the gravitational lens mg 0414 + 0534 , developed objective criteria to flag questionable data . they deleted from their light curves all points associated with observations with the following conditions : the telescope elevations were less than 30@xmath24 , the wind speed was greater than 10 m / s , or there was precipitation . we have noted all epochs satisfying their criteria in the `` comments '' column of table [ tab_obs ] . however , we are able to include many of these points in our analysis because observations at 8.5 ghz are less sensitive to the observing conditions than are observations taken at 15 ghz . as such , we have only excluded epochs for which the data are severely affected by the observing conditions . this assessment is made by examining the light curves of the secondary calibrator sources . all epochs for which the flux densities of the calibrators deviate by more than 15% from the mean value are deleted . only two days are deleted after the application of this criterion : 376 and 382 ( mjd@xmath950000 ) . note that epoch 376 satisfies two of the flagging criteria defined by moore & hewitt . at epoch 382 , the system temperatures for all of the telescopes were in the range 100 200 k , as compared to the 30 50 k system temperatures measured for all other epochs . these high system temperatures may have resulted from the fact the subreflectors of several of the antennas had frozen prior to the observation and had just thawed . the signal - to - noise ratio of the maps for epoch 382 were so low that no useful information could be extracted from them . the final edited light curves contain 62 epochs , with an average spacing of 3.7 d. edited versions of the secondary flux calibrator and b1608 + 656 light curves are shown in figures [ fig_normcmp ] and [ fig_w08 ] . the normalized light curves of the secondary flux calibrators are shown in fig . [ fig_normcmp ] . as expected for non - variable sources , the light curves are close to constant . the scatter about the mean is small , with the exception of a few outlying points . all of these outlier points are due to bad observing conditions . the five worst points , for example , occur on days 390 , 402 , 427 , 474 , and 532 , all of which experienced high winds , precipitation , or both ( table [ tab_obs ] ) . because of these outliers , the formal calculation of @xmath25 overestimates the width of the error distribution . consequently , we have estimated @xmath25 using the interquartile range ( iqr ) , which is less affected by the outliers . for a gaussian distribution , @xmath26 . for the secondary flux calibrators , we obtain @xmath27 and @xmath28 for 1634 + 627 and 1633 + 741 , respectively . gaussian distributions with these values of @xmath25 are reasonable fits to the data ( see fig . [ fig_normcmp ] ) . the light curves of the two secondary flux calibrators track each other extremely well , suggesting that the observed variations are due to errors in the absolute flux calibration of the data rather than any intrinsic variability of the sources . because of this , we can use the light curves of the secondary flux calibrators to remove residual flux calibration errors in the b1608 + 656 component light curves . we create a calibration correction curve by first normalizing the light curves of the secondary flux calibrators and then averaging the two normalized fluxes at each epoch . we then apply the secondary flux calibration by dividing the b1608 + 656 component light curves by the calibration correction curve . the correction is effective in reducing the scatter in the b1608 + 656 light curves , as can be seen by comparing fig . [ fig_w08 ] to fig . [ fig_wflat08 ] . the corrected component light curves are listed in table [ tab_complcurve ] . an electronic version of the light curve table may also be obtained from http://www.nrao.edu/~cfassnac/1608flux.tab . ccccc 366 & 34.831 @xmath29 0.309 & 16.106 @xmath29 0.149 & 17.602 @xmath29 0.162 & 5.822 @xmath29 0.070 + 368 & 33.251 @xmath29 0.307 & 16.233 @xmath29 0.166 & 17.269 @xmath29 0.175 & 5.736 @xmath29 0.097 + 372 & 34.063 @xmath29 0.308 & 16.277 @xmath29 0.151 & 16.643 @xmath29 0.154 & 5.701 @xmath29 0.063 + 374 & 34.003 @xmath29 0.294 & 16.532 @xmath29 0.146 & 17.349 @xmath29 0.153 & 5.760 @xmath29 0.060 + 379 & 33.589 @xmath29 0.307 & 16.904 @xmath29 0.165 & 17.042 @xmath29 0.166 & 5.768 @xmath29 0.085 + 385 & 32.972 @xmath29 0.310 & 17.002 @xmath29 0.176 & 16.863 @xmath29 0.175 & 5.695 @xmath29 0.101 + 387 & 34.217 @xmath29 0.317 & 16.883 @xmath29 0.161 & 16.831 @xmath29 0.160 & 5.997 @xmath29 0.070 + 388 & 33.381 @xmath29 0.299 & 16.884 @xmath29 0.155 & 17.002 @xmath29 0.156 & 6.000 @xmath29 0.066 + 390 & 34.845 @xmath29 0.283 & 17.365 @xmath29 0.149 & 17.429 @xmath29 0.150 & 5.881 @xmath29 0.073 + 394 & 32.744 @xmath29 0.307 & 16.811 @xmath29 0.174 & 16.898 @xmath29 0.175 & 5.727 @xmath29 0.101 + 395 & 33.060 @xmath29 0.297 & 17.276 @xmath29 0.159 & 16.715 @xmath29 0.154 & 5.763 @xmath29 0.064 + 398 & 32.795 @xmath29 0.295 & 16.705 @xmath29 0.154 & 16.995 @xmath29 0.157 & 5.755 @xmath29 0.065 + 399 & 33.422 @xmath29 0.304 & 16.830 @xmath29 0.162 & 16.661 @xmath29 0.161 & 5.874 @xmath29 0.082 + 402 & 33.927 @xmath29 0.271 & 16.943 @xmath29 0.139 & 16.840 @xmath29 0.138 & 6.036 @xmath29 0.060 + 405 & 33.846 @xmath29 0.296 & 16.740 @xmath29 0.150 & 17.107 @xmath29 0.153 & 5.910 @xmath29 0.062 + 410 & 34.527 @xmath29 0.310 & 16.966 @xmath29 0.156 & 17.294 @xmath29 0.159 & 5.813 @xmath29 0.064 + 414 & 33.878 @xmath29 0.308 & 17.096 @xmath29 0.160 & 17.447 @xmath29 0.163 & 5.813 @xmath29 0.067 + 418 & 34.996 @xmath29 0.315 & 17.054 @xmath29 0.159 & 17.632 @xmath29 0.164 & 5.807 @xmath29 0.071 + 422 & 35.218 @xmath29 0.315 & 16.645 @xmath29 0.157 & 17.431 @xmath29 0.164 & 5.765 @xmath29 0.077 + 424 & 35.464 @xmath29 0.323 & 16.168 @xmath29 0.158 & 17.324 @xmath29 0.168 & 5.850 @xmath29 0.084 + 427 & 35.909 @xmath29 0.302 & 16.613 @xmath29 0.144 & 17.505 @xmath29 0.151 & 5.574 @xmath29 0.061 + 432 & 34.811 @xmath29 0.317 & 16.423 @xmath29 0.152 & 17.280 @xmath29 0.160 & 5.663 @xmath29 0.059 + 437 & 34.475 @xmath29 0.308 & 16.234 @xmath29 0.158 & 17.368 @xmath29 0.167 & 5.791 @xmath29 0.088 + 440 & 34.875 @xmath29 0.310 & 16.236 @xmath29 0.147 & 17.532 @xmath29 0.158 & 5.483 @xmath29 0.057 + 441 & 34.831 @xmath29 0.315 & 16.286 @xmath29 0.155 & 17.203 @xmath29 0.163 & 5.525 @xmath29 0.074 + 446 & 33.832 @xmath29 0.304 & 16.358 @xmath29 0.151 & 17.405 @xmath29 0.160 & 5.868 @xmath29 0.064 + 451 & 34.444 @xmath29 0.298 & 16.320 @xmath29 0.146 & 17.802 @xmath29 0.158 & 5.765 @xmath29 0.066 + 455 & 33.230 @xmath29 0.301 & 16.271 @xmath29 0.150 & 17.042 @xmath29 0.157 & 5.825 @xmath29 0.062 + 459 & 33.463 @xmath29 0.298 & 16.667 @xmath29 0.152 & 16.946 @xmath29 0.154 & 5.964 @xmath29 0.064 + 464 & 34.451 @xmath29 0.310 & 16.851 @xmath29 0.156 & 17.364 @xmath29 0.160 & 5.979 @xmath29 0.068 + 465 & 34.010 @xmath29 0.309 & 16.720 @xmath29 0.164 & 17.825 @xmath29 0.173 & 6.049 @xmath29 0.090 + 468 & 33.989 @xmath29 0.305 & 16.624 @xmath29 0.152 & 16.867 @xmath29 0.154 & 5.970 @xmath29 0.063 + 474 & 33.015 @xmath29 0.322 & 16.290 @xmath29 0.172 & 16.755 @xmath29 0.176 & 5.974 @xmath29 0.095 + 478 & 32.845 @xmath29 0.295 & 16.223 @xmath29 0.149 & 16.945 @xmath29 0.156 & 5.845 @xmath29 0.064 + 481 & 33.523 @xmath29 0.306 & 16.735 @xmath29 0.155 & 16.709 @xmath29 0.155 & 6.015 @xmath29 0.063 + 487 & 33.643 @xmath29 0.306 & 16.325 @xmath29 0.151 & 17.410 @xmath29 0.161 & 5.960 @xmath29 0.062 + 492 & 34.166 @xmath29 0.308 & 16.528 @xmath29 0.152 & 16.923 @xmath29 0.155 & 5.820 @xmath29 0.062 + 497 & 33.754 @xmath29 0.304 & 17.080 @xmath29 0.156 & 16.961 @xmath29 0.155 & 5.725 @xmath29 0.060 + 502 & 34.051 @xmath29 0.307 & 16.427 @xmath29 0.150 & 17.323 @xmath29 0.158 & 5.879 @xmath29 0.060 + 507 & 34.040 @xmath29 0.308 & 16.761 @xmath29 0.156 & 17.495 @xmath29 0.162 & 5.808 @xmath29 0.067 + 510 & 33.924 @xmath29 0.307 & 16.779 @xmath29 0.154 & 17.754 @xmath29 0.163 & 5.779 @xmath29 0.061 + 515 & 32.624 @xmath29 0.292 & 17.156 @xmath29 0.156 & 17.050 @xmath29 0.155 & 5.804 @xmath29 0.060 + 521 & 33.901 @xmath29 0.307 & 16.774 @xmath29 0.154 & 17.356 @xmath29 0.159 & 5.743 @xmath29 0.060 + 523 & 34.102 @xmath29 0.310 & 16.648 @xmath29 0.157 & 17.337 @xmath29 0.162 & 5.792 @xmath29 0.069 + 528 & 34.548 @xmath29 0.314 & 16.691 @xmath29 0.155 & 17.386 @xmath29 0.161 & 5.582 @xmath29 0.061 + 532 & 34.210 @xmath29 0.295 & 16.394 @xmath29 0.145 & 17.069 @xmath29 0.150 & 5.777 @xmath29 0.061 + 537 & 33.749 @xmath29 0.307 & 16.636 @xmath29 0.153 & 17.181 @xmath29 0.158 & 5.792 @xmath29 0.058 + 539 & 34.388 @xmath29 0.309 & 16.958 @xmath29 0.155 & 16.927 @xmath29 0.155 & 5.826 @xmath29 0.061 + 543 & 34.049 @xmath29 0.306 & 16.921 @xmath29 0.155 & 17.185 @xmath29 0.157 & 5.830 @xmath29 0.062 + 545 & 33.959 @xmath29 0.308 & 17.031 @xmath29 0.156 & 17.422 @xmath29 0.160 & 5.903 @xmath29 0.060 + 549 & 34.338 @xmath29 0.311 & 16.267 @xmath29 0.150 & 17.507 @xmath29 0.161 & 5.758 @xmath29 0.061 + 553 & 33.728 @xmath29 0.306 & 16.519 @xmath29 0.152 & 17.374 @xmath29 0.160 & 5.820 @xmath29 0.061 + 557 & 33.004 @xmath29 0.295 & 16.755 @xmath29 0.152 & 17.459 @xmath29 0.158 & 5.707 @xmath29 0.060 + 560 & 33.241 @xmath29 0.300 & 16.622 @xmath29 0.152 & 17.630 @xmath29 0.161 & 5.833 @xmath29 0.061 + 564 & 33.049 @xmath29 0.300 & 16.850 @xmath29 0.156 & 16.666 @xmath29 0.154 & 5.805 @xmath29 0.062 + 571 & 34.091 @xmath29 0.309 & 16.535 @xmath29 0.152 & 17.653 @xmath29 0.162 & 5.957 @xmath29 0.063 + 576 & 33.332 @xmath29 0.302 & 16.748 @xmath29 0.154 & 17.093 @xmath29 0.157 & 5.847 @xmath29 0.059 + 581 & 34.055 @xmath29 0.308 & 15.971 @xmath29 0.147 & 18.138 @xmath29 0.166 & 5.768 @xmath29 0.060 + 585 & 34.151 @xmath29 0.307 & 16.127 @xmath29 0.148 & 17.287 @xmath29 0.158 & 5.732 @xmath29 0.060 + 589 & 33.723 @xmath29 0.307 & 15.952 @xmath29 0.148 & 17.538 @xmath29 0.162 & 6.051 @xmath29 0.064 + 591 & 34.341 @xmath29 0.312 & 16.068 @xmath29 0.150 & 17.214 @xmath29 0.160 & 5.962 @xmath29 0.066 + 594 & 33.130 @xmath29 0.302 & 16.309 @xmath29 0.152 & 17.018 @xmath29 0.158 & 5.993 @xmath29 0.065 + the errors on the final flux densities in table [ tab_complcurve ] are a combination of additive and multiplicative terms . the additive uncertainty is well - approximated by the rms noise in the residual map at the end of the model - fitting . the multiplicative term is indicative of how well the model - fitting procedure is able to find the `` true '' total flux density of a source . we estimate this error by calculating the ratio of the flux densities of 1634 + 627 and 1633 + 741 , which we expect not to vary with respect to each other . the flux density ratio is not affected by errors in the absolute flux calibration , so any scatter in the ratio can be attributed to errors in the model fitting procedure . from the scatter in the flux density ratio , we estimate the fractional error in the component flux densities to be 0.9% ; the product of the component flux densities and this value gives the multiplicative uncertainty . the final errors on the flux densities are a combination in quadrature of the additive and multiplicative terms . for the three brightest components ( a , b , and c ) , the multiplicative terms dominate ; for component d the multiplicative and additive terms are comparable . the flux density of a lensed image of the background source is @xmath30 , where @xmath31 is the true flux density of the background source and @xmath32 is the magnification associated with the @xmath33 ( unresolved ) image . similarly , the lensed images will vary by @xmath34 if the flux density of the background source changes by an amount @xmath35 . hence , any variations in the b1608 + 656 component light curves that are due to variations in the background source must have the same _ fractional _ amplitudes in all four curves . for this reason , the component light curves presented in figs . [ fig_w08 ] and [ fig_wflat08 ] are divided by their mean flux densities to allow direct comparison of the fractional variations in the curves . the mean values used to normalize the light curves are 34.29 , 16.79 , 17.41 , and 5.884 mjy for components a , b , c , and d , respectively . all four light curves show variations in flux density at the @xmath05% level , and there are common features that can be seen in each . each light curve shows a rise of @xmath05% in flux density , followed by a plateau lasting @xmath020 d and then a drop of @xmath04% . models of the b1608 + 656 system predict that , if the background source is variable , the lensed images will vary in the order b @xmath15 a @xmath15 c @xmath15 d ( @xcite ; paper 2 ) . this is , in fact , the behavior seen in the b1608 + 656 light curves . in order to determine time delays between component b and the other three components , we have used three statistical methods : ( 1 ) smoothing and @xmath23 minimization , ( 2 ) smoothing and cross - correlation , and ( 3 ) dispersion analysis ( pelt et al . 1994 , 1996 , 1998 ) . the methods are discussed below . all of the analysis has been performed on both the `` phasecal '' and `` gscale '' data sets ( [ sec_1608maps ] ) . no significant differences in the results are seen , so we present only the results from the `` gscale '' set in the subsequent discussion . the first two methods of determining the time delays from the b1608 + 656 light curves require that the observed data be transferred onto a regular grid . some previous determinations of time delays ( e.g. , @xcite ; @xcite ) have accomplished this transfer by linear interpolation of the data . however , the flux density variations seen in the b1608 + 656 images are small compared to the noise in the curves ; therefore , linear interpolation can amplify noise spikes . in contrast , smoothing reduces the effects of noise compared to the true variations , if the variations have typical time - scales significantly longer than the sampling interval . we smooth and re - sample the data by calculating the weighted mean of points within a smoothing window that is moved from the beginning to the end of the observations in regular steps . the step size is set to one day . we smooth each light curve with several different functions to avoid biasing the results by our choice of weighting function or window size . for completeness , we also include the results obtained from interpolating the data by piecewise linear interpolation . we use the following smoothing schemes : ( 1 ) boxcar - weighted mean , ( 2 ) triangle - weighted mean , ( 3 ) gaussian - weighted mean , ( 4 ) boxcar - weighted mean with a variable - width smoothing window , and ( 5 ) triangle - weighted mean with a variable - width window . in the last two schemes , the width of the smoothing window is varied such that the same number of points are always included in the window . we use 3 , 5 , and 7 point windows for these methods . for the fixed - width window schemes , we smooth with window widths of 5 , 10 , and 15 d. the gaussian smoothing scheme uses values of @xmath25 equal to 3 , 5 , and 7 d. in addition to multiplying by the value of the smoothing function , the data points are also variance - weighted . thus , the overall weighting on a point at @xmath36 which is being used to calculate a weighted mean at @xmath37 is : @xmath38},\ ] ] where @xmath39 is the value of the smoothing function ( e.g. , g(t ) = 1 for boxcar smoothing ) , @xmath40 is the uncertainty on the flux density at @xmath36 ( calculated in [ flat ] ) , and the sum is over the points used to calculate the mean . the variance of the weighted mean @xmath41 at @xmath37 in the smoothed curve then becomes @xmath42 \right ) ^2 } \sum_i \frac{[g(t_i - t_k)]^2}{\sigma_i^2 } , \label{eqn_var_wmean}\ ] ] and @xmath43 is taken as the uncertainty in the smoothed flux density at that step . for the @xmath23 minimization technique , we compare two light curves , one of which is designated the `` control '' curve and the other of which is designated the `` comparison '' curve . both curves are smoothed , and then the comparison curve is multiplied by a scale factor @xmath22 so that its mean flux density is comparable to that of the control curve . after this scaling , the comparison curve is shifted in time by an amount @xmath44 with respect to the control curve . we form a grid of @xmath45 pairs and calculate the @xmath23 statistic at each grid point . the step size for @xmath44 used in the grid is 1 d and the minimum and maximum shifts are set to @xmath29114 d , i.e. , half of the total length of the observations . the amplitude scale factors are set as percentages of the scale factor @xmath46 that equalizes the mean flux densities of the two curves being compared . the scale factors used in the grid range from 0.9@xmath46 to 1.1@xmath46 , in steps of 0.001@xmath46 . one result of the smoothing and interpolation performed on the input light curves is that the points in the interpolated curves are no longer independent . when computing the reduced @xmath23 statistic we estimate an effective number of independent points , @xmath47 , in the overlapping region between the shifted and unshifted curves . this quantity is set to the number of points that would have been present in the overlapping region if the original light curves had been regularly sampled at the average spacing of 3.7 d. this simplifying assumption should not significantly affect the quantity in which we are interested , which is the location of the minimum on the @xmath23 surface . in fact , the @xmath23-minimization method finds delays that match closely those found from the completely non - interpolative dispersion method ( [ dispersion ] ) , indicating the robustness of the delays determined from this method . we repeat the @xmath23 minimization three times , once for each independent pair of light curves . the component b light curve is always taken as the control curve . in the three repetitions of the @xmath23 minimization , the comparison curves are a , c , and d , respectively . the minimum delay is found by fitting a parabola to the points at the minimum of the gridded @xmath23 curve . table [ tab_delays ] summarizes the results . typical goodness - of - fit curves for the three pairs of light curves are shown in figure [ fig_pchi3 ] . lrrrrrrr boxcar & 5 & 30.97 & 35.96 & 74.27 & 28 & 36 & 74 + & 10 & 29.25 & 36.24 & 75.32 & 29 & 37 & 75 + & 15 & 31.84 & 36.10 & 75.58 & 32 & 36 & 75 + triangle & 5 & 29.16 & 36.07 & 74.20 & 28 & 36 & 74 + & 10 & 29.35 & 35.63 & 75.27 & 29 & 35 & 75 + & 15 & 30.77 & 36.03 & 75.97 & 30 & 36 & 76 + gaussian & 3 & 30.65 & 35.93 & 75.93 & 30 & 36 & 76 + & 5 & 31.55 & 36.17 & 76.70 & 31 & 37 & 76 + & 7 & 31.52 & 36.37 & 77.06 & 31 & 37 & 77 + variable - width & 3 & 30.13 & 35.79 & 76.58 & 30 & 36 & 75 + boxcar & 5 & 29.92 & 36.18 & 77.32 & 30 & 37 & 78 + & 7 & 29.44 & 36.38 & 77.22 & 30 & 39 & 77 + variable - width & 3 & 30.52 & 35.84 & 76.48 & 30 & 36 & 76 + triangle & 5 & 30.74 & 36.22 & 77.03 & 30 & 37 & 77 + & 7 & 29.93 & 36.25 & 77.21 & 30 & 37 & 77 + interpolation & & 29.02 & 36.09 & 74.82 & 28 & 36 & 74 + lrrrr boxcar & 5 & 2.0387 & 1.0368 & 0.3506 + & 10 & 2.0383 & 1.0378 & 0.3508 + & 15 & 2.0411 & 1.0379 & 0.3508 + triangle & 5 & 2.0387 & 1.0368 & 0.3505 + & 10 & 2.0400 & 1.0376 & 0.3507 + & 15 & 2.0409 & 1.0379 & 0.3511 + gaussian & 3 & 2.0407 & 1.0378 & 0.3511 + & 5 & 2.0411 & 1.0379 & 0.3511 + & 7 & 2.0410 & 1.0379 & 0.3508 + variable - width & 3 & 2.0406 & 1.0383 & 0.3511 + boxcar & 5 & 2.0415 & 1.0384 & 0.3512 + & 7 & 2.0413 & 1.0382 & 0.3512 + variable - width & 3 & 2.0407 & 1.0383 & 0.3511 + triangle & 5 & 2.0414 & 1.0384 & 0.3512 + & 7 & 2.0393 & 1.0382 & 0.3512 + interpolation & & 2.0394 & 1.0374 & 0.3506 + for the cross - correlation calculations the component b light curve is once again taken as the control curve . before the cross - correlations are performed , both the comparison and control curves are smoothed and then divided by their mean values over the observations . the normalization by the mean value puts the fractional variations in the light curves at the same level . the cross - correlation functions are computed in the time domain , with the value at each lag calculated in the standard fashion : @xmath48 where @xmath49 is the covariance defined by @xmath50 } { \frac{1}{n } \sum_i \frac{1}{\sigma_{ij}^2+\sigma_{ik}^2}}\ ] ] and @xmath51 is the number of overlapping points ( e.g. , @xcite ) . note that in @xmath52 or @xmath53 , the first subscript refers to a particular observation , and the second subscript refers to the name of the variable under discussion . thus @xmath52 is the @xmath33 observation of variable @xmath54 . the variance , @xmath55 , is given by @xmath56 } { \frac{1}{n}\sum_i \frac{1}{\sigma_i^2}}\ ] ] where the means are weighted means , @xmath57 } { \sum_i \frac{1}{\sigma_i^2}}.\ ] ] the cross - correlation calculations are repeated for all of the smoothing functions described in [ smooth ] . in all cases , clear peaks in the cross - correlation curves are seen . typical curves are shown in fig . [ fig_pcorr3 ] . the lags at which the peak correlation coefficients occur are given in table [ tab_delays ] . the average displacement between the cross - correlation and @xmath23 minimization lags for each pair of curves is less than one day . the dispersion analysis methods presented by pelt et al . ( 1994 , 1996 , 1998 ) do not involve any interpolation of the component light curves . these methods thus have the advantage of avoiding effects introduced by the interpolation and smoothing associated with the methods discussed in [ smooth ] . the dispersion analysis begins with the construction of a composite curve @xmath58 from two input light curves @xmath59 and @xmath60 . curve @xmath59 is not modified , while curve @xmath60 is scaled by a flux density ratio @xmath22 and shifted in time by a delay @xmath44 , i.e. @xmath61 ( @xcite ) . we have used a grid of @xmath45 pairs to construct the @xmath62 curves . aside from changing the spacing on the delay axis to 0.5 d , the grid limits and spacings are the same as those used in the @xmath23 minimization analysis presented in [ chi2 ] . the internal dispersion in each curve is calculated , and the grid point associated with the minimum dispersion is recorded . in our analysis of the b1608 + 656 light curves we calculate dispersions using the non - parametric @xmath63 and the one - parameter @xmath64 statistics , where the notation is taken from pelt et al . ( 1996 ; note that these statistics are called @xmath65 and @xmath63 , respectively , in pelt et al . the @xmath63 dispersion is calculated using only immediately adjacent points in the composite curve , with the caveat that a pair of points only contributes to the dispersion if the two points are from different input curves . the @xmath64 dispersion is similar , but uses all pairs of points that lie within a time range @xmath66 of each other . for a detailed description of these estimates , see pelt et al . table [ tab_disp ] gives the results for the @xmath63 analysis and the @xmath64 analysis with several values of the @xmath66 parameter . the results do not depend strongly on the value of @xmath66 and are consistent within the errors ( [ mc_delay ] ) with the results from the @xmath23 minimization and cross - correlation analyses . the dispersion spectra plotted in fig . [ fig_pdisp3 ] are cuts through the @xmath45 grid at a value of @xmath22 corresponding to the minimum dispersion . crcccrrr @xmath63 & & 30.5 & 2.0388 & 36.5 & 1.0372 & 74.5 & 0.3505 + @xmath64 & 3.5 & 32.0 & 2.0429 & 35.5 & 1.0361 & 74.5 & 0.3501 + @xmath64 & 4.5 & 31.5 & 2.0429 & 36.5 & 1.0372 & 74.5 & 0.3502 + @xmath64 & 5.5 & 31.0 & 2.0429 & 37.0 & 1.0372 & 76.5 & 0.3509 + @xmath64 & 6.5 & 31.0 & 2.0429 & 34.5 & 1.0361 & 76.5 & 0.3509 + @xmath64 & 7.5 & 31.5 & 2.0429 & 35.0 & 1.0372 & 76.5 & 0.3509 + @xmath64 & 8.5 & 32.0 & 2.0429 & 35.0 & 1.0372 & 77.0 & 0.3512 + @xmath64 & 9.5 & 31.5 & 2.0449 & 34.5 & 1.0372 & 77.5 & 0.3512 + @xmath64 & 10.5 & 31.5 & 2.0449 & 35.0 & 1.0372 & 77.0 & 0.3512 + @xmath64 & 11.5 & 32.0 & 2.0449 & 35.5 & 1.0372 & 77.5 & 0.3512 + @xmath64 & 12.5 & 32.0 & 2.0449 & 36.5 & 1.0372 & 78.5 & 0.3512 + @xmath64 & 13.5 & 32.0 & 2.0449 & 35.5 & 1.0372 & 78.5 & 0.3512 + @xmath64 & 14.5 & 32.5 & 2.0449 & 35.0 & 1.0372 & 78.5 & 0.3512 + @xmath64 & 15.5 & 33.0 & 2.0449 & 34.5 & 1.0372 & 77.5 & 0.3512 + @xmath64 & 16.5 & 32.5 & 2.0449 & 35.0 & 1.0372 & 79.5 & 0.3516 + @xmath64 & 17.5 & 32.5 & 2.0449 & 35.0 & 1.0372 & 79.0 & 0.3512 + @xmath64 & 18.5 & 32.0 & 2.0429 & 35.0 & 1.0372 & 79.5 & 0.3512 + @xmath64 & 19.5 & 32.0 & 2.0429 & 35.0 & 1.0372 & 79.5 & 0.3512 + @xmath64 & 20.5 & 32.0 & 2.0429 & 35.0 & 1.0372 & 80.0 & 0.3516 + there is some scatter in the results presented in tables [ tab_delays ] and [ tab_disp ] , which is not surprising considering the low levels of variation and sparse sampling of the light curves . however , the scatter is small compared with both the length of the time delays and the uncertainties in the delays that we find from the monte carlo simulations in [ mc_delay ] . for each of the three methods used to find the delays ( @xmath23 minimization , cross - correlation , and dispersion analysis ) , we take the median values of the delays in tables [ tab_delays ] and [ tab_disp ] . the three median values for each delay are then averaged to obtain @xmath67 = 31 d , @xmath68 = 36 d , and @xmath69 = 76 d. the median flux density ratios , computed in a similar manner , are found to be @xmath70 , @xmath71 , and @xmath72 we shift the light curves by the mean delays and normalize them using the mean flux density ratios to create a composite light curve of the background source ( fig . [ fig_compos ] ) . a composite curve constructed from the smoothed and interpolated component light curves is shown in fig . [ fig_compos_smo ] . the variations seen in the b1608 + 656 light curves are not large , either in a fractional _ or _ absolute sense , compared to what has been seen in other lens systems . the fractional variations seen in b0218357 , 0957 + 561 , and pg 1115 + 080 are all two to three times larger than those seen in b1608 + 656 ( @xcite ; kundi et al.1995,1997 ; @xcite ) , and pks 1830@xmath9211 shows 50% variations in flux density ( although the time delay measurement is based on smaller variations ; @xcite ) . a sceptic might argue that the correlations between the b1608 + 656 light curves are not significant and could be duplicated by any set of light curves containing random scatter about a constant value . in theory , the value of the correlation coefficient can be used to assess the significance of the correlation . with the coefficient , @xmath73 calculated as described in [ xcorr ] , the probability of obtaining a value @xmath74 from two _ uncorrelated _ curves of gaussian - distributed random variables is @xmath75 where @xmath76}{\gamma(\nu/2)}\;(1-r^2)^{(\nu-2)/2 } \label{eqn_bev_prob}\ ] ] and @xmath77 ( e.g. , @xcite ) . however , the smoothing and interpolation performed on the sparsely sampled b1608 + 656 light curves makes the interpretation of the significance of the value of @xmath73 complicated . the difficulty lies in assessing the number of _ independent _ points in the curves at each lag . monte carlo simulations show that the number of independent points can not be estimated simply as the width of the overlap region divided by the width of the smoothing window ( i.e. , the number of smoothing windows in the overlap region ) . this calculation _ underestimates _ the number of independent points in the region . because it is difficult to determine the significance of the correlation analytically , we perform monte carlo simulations to find the significance empirically . in the simulations , we calculate correlations between light curves consisting of randomly distributed data . to create random light curves with the same distribution of flux densities as seen in the data , we simply randomize the time series for the component light curves while preserving the flux densities at their measured values . by randomizing the times at which the flux densities are measured , we destroy any possible correlations between the curves . each light curve is randomized independently to avoid correlations at zero lag which are associated with measurement errors . the simulations are conducted with 3000 sets of randomized curves . each set of light curves is processed in the manner described in [ xcorr ] and produces three sets of correlation curves ( b a , b c , and b d ) . all values of the correlation coefficient are recorded . the distributions of the cross - correlation values obtained from the 10 d boxcar smoothing scheme are shown in fig . [ fig_randplot ] . the empirical probabilities of obtaining at least the observed peak values , which are indicated by the vertical dashed lines in the figure , from uncorrelated curves are all low , with @xmath78 , @xmath79 , and @xmath80 . it is even more unlikely that three pairs of randomized curves could produce three such anomalously high cross - correlation peaks . there can thus be no significant doubt that the correlations we measure are real . the time delay measurement uncertainties contribute directly to the error budget for measuring @xmath4 with a gravitational lens . in particular , the fractional uncertainties in the time delays contribute a matching fractional uncertainty in @xmath4 , i.e. , @xmath81 we estimate the uncertainties in the time delays by performing monte carlo simulations of the observations . in the simulations we assign time delays between the pairs of light curves and then see how well we can recover the input delays . we also use the simulations to estimate the uncertainties in the flux density ratios , which are necessary for modeling the lensing potential ( paper 2 ) . we produce fake curves with the same characteristics as our real data by smoothing the composite light curve ( fig . [ fig_compos ] ) with a 10 d boxcar filter . this normalized and smoothed curve is the master light curve for the simulations , representing the assumed true behavior of the background source . the offsets between the points of the composite curve and the master curve are distributed as a zero - mean gaussian with @xmath82 . note that this is a _ value since all of the light curves have been normalized to create the master light curve . the gaussian distribution is used to generate the random offsets for the simulations . the appropriate rescaling of the random offsets is achieved through equation [ eqn_mc ] . we note that smoothing procedure used in constructing the master curve destroys information on possible short - timescale variations . thus , if such short - term variations do exist , their effects will be attributed to measurement and calibration errors in this method . however , given the low level of variability of the background source during the observations and the sparse sampling , it is difficult to distinguish any possible short - term variations from measurement error . thus , we accept @xmath82 as an indication of the measurement error , but with the caveat that the uncertainties in the time delays derived from the monte carlo simulations described below may be over - estimated . for each round of the simulation we generate four sparsely - sampled fake light curves . the flux density for component @xmath83 at a time @xmath84 in the fake curves is given by : @xmath85 where @xmath86 is the normalized master flux density , the @xmath87 are the input flux density ratios ( 2.0418 , 1.000 , 1.0376 , and 0.3512 ) , the @xmath88 are the input time delays ( 31 d , 0 d , 36 d , and 76 d ) , and @xmath89 is the random offset . the four curves are sampled with the pattern used in the observations ( see table [ tab_obs ] ) . the flux density error at each point in the sparsely - sampled curves is set to the observed flux density error for that epoch ( see [ flat ] ) . the fake curves are processed in the manner described in [ timedelay ] . the best - fit time delays and flux density ratios for each simulation are recorded . histograms of typical distributions of time delays from 10,000 repetitions of the above procedure are shown in fig . [ fig_mc_xc ] . the distributions are non - gaussian , both in the shape of the peak of the distribution and in the long tail of outliers at negative delays . thus , we determine the confidence limits by finding the range of delays inside which 95% of the simulation results lie , rather than fitting a gaussian to the distribution . the limits are chosen such that the minimum range of delays that encloses both the median value and 95% of the simulation results is found . the results for the @xmath23 minimization and cross - correlation techniques are given in table [ tab_mc ] . we conservatively take the broader distribution in each case as our estimate of the 95% confidence contours . thus , we estimate the time delays to be @xmath1 d , @xmath90 d , and @xmath3 d at 95% confidence . we have also run monte carlo simulations in which the sampling pattern is varied , following the method used in biggs et al . the distributions of time delays do not differ significantly from those presented above . cccccc @xmath91 & 31 & 1 & 28.6 33.0 & 26.7 34.9 & 24.4 37.3 + @xmath68 & 36 & 1 & 33.5 38.7 & 31.8 40.3 & 29.0 43.3 + @xmath69 & 76 & 1 & 73.2 78.9 & 71.1 81.0 & 67.4 84.7 + @xmath91 & 31 & 2 & 29 34 & 27 36 & 25 38 + @xmath68 & 36 & 2 & 33 38 & 31 40 & 29 42 + @xmath69 & 76 & 2 & 71 77 & 68 80 & 66 82 + @xmath92 & 2.0418 & 1 & 2.0386 2.0506 & 2.0344 2.0549 & 2.0322 2.0570 + @xmath93 & 1.0376 & 1 & 1.0340 1.0401 & 1.0318 1.0422 & 1.0306 1.0433 + @xmath94 & 0.3512 & 1 & 0.3500 0.3525 & 0.3492 0.3534 & 0.3487 0.3539 the goal of monitoring a gravitational lens system is to measure time delays that can then be combined with a model of the lensing potential to produce a measurement of @xmath4 . we have been successful in measuring the three independent time delays in the b1608 + 656 system . paper 2 presents a model for the b1608 + 656 system . the model is based on the time delays and flux density ratios presented here , and on positions from vlba ( @xcite ) and hst observations of the system ( @xcite ) . the lensing potential contains contributions from the two lensing objects seen in hst images of the system ( @xcite ) , each of which is modeled as an elliptical isothermal mass distribution . the effects of varying the positions of the lensing galaxies , of changing the nature of the lensing galaxy cores ( singular or non - singular ) , and of departing from an isothermal profile are all explored . the best - fit model is obtained through a simulated annealing process . it correctly reproduces the positions and flux density ratios of the lensed images ( with the exception of the d / a flux density ratio ) . most importantly , the predicted time delay _ ratios _ match the observed time delay ratios to within 1% . although the individual time delays depend on the hubble constant ( @xmath95 ) , the time delay ratios have no @xmath4 dependence . thus , the model for any lens system with more than two images _ must _ correctly reproduce the observed time delay ratios if it is to be used in the determination of @xmath4 . in this sense , gravitational lenses which produce more than two images can put stronger constraints on lens models than can two image lenses , as long as the time delays can be measured . the b1608 + 656 system is the first four - image lens for which the three independent time delays have been measured and a model correctly reproduces the time delay ratios . the best - fit isothermal model from paper 2 predicts time delays of @xmath96 d , @xmath97 d , and @xmath98 d. combining these predicted values with the observed time delays gives three individual determinations of @xmath4 : @xmath99 , @xmath100 , and @xmath101 . we combine these by calculating a weighted mean of @xmath102 , where the weights are derived from the uncertainties in the time delays from [ mc_delay ] . note that the b - d value dominates the mean since it has the smallest fractional uncertainty . in order to estimate the uncertainty on @xmath103 , we use the results of the monte - carlo simulation presented in [ mc_delay ] . the distribution of time delays from the simulations are converted into distributions of @xmath4 by dividing the predicted delays from the paper 2 model by the simulation results . the resulting @xmath4 distributions from the b - a , b - c , and b - d delays are combined into an overall distribution , which is shown in fig . [ fig_hall ] . the confidence limits are estimated by finding the ranges that enclose 68 , 90 , and 95% of the data . the resulting 95% confidence limits are @xmath104 . these confidence limits are very close to the statistical 95% limits derived from the lens model , which includes @xmath4 as a model parameter ( @xmath105 ; paper 2 ) . the above estimates of the uncertainties in the determination of @xmath4 have not included the systematic effects from the choice of the radial mass profile in the lens modeling . the estimated systematic error is @xmath106 ( paper 2 ) . it may be possible to reduce this error by modeling the extended lensed stellar emission from the background source , as has been done with the radio einstein ring mg 1654 + 1346 ( @xcite ) . this modeling approach is being conducted by surpi & blandford ( private communication ) . we have presented the results of an intensive program of monitoring the four - image lens system b1608 + 656 with the vla . the component light curves show @xmath05% variations in flux density from which we have measured the three independent time delays in this system : @xmath107 d , @xmath90 d , and @xmath108 d. these time delays are combined with the mass model of the lens presented in paper 2 to give @xmath109 at 95% confidence ( statistical ) @xmath110 ( systematic ) . the statistical uncertainties represent the 95% confidence interval . the statistical uncertainties in the time delays can be reduced if a stronger variation in the background source is observed , while the systematic uncertainties may be reduced through the inclusion of the lensed extended emission in the lens modeling process . our previous observations have shown that the background source in this system has varied by as much as 15% in the past , so we are conducting another program of monitoring . if a stronger variation is detected in the new data , the uncertainties on the time delays will be reduced and the accuracy of the measurement of @xmath4 with this system will be improved . for generously donating portions of their vla observing time to help us fill in gaps in our time coverage , we are grateful to erik leitch , brian mason , jackie hewitt , cathy trotter , gillian knapp , michael rupen , and james gunn . the observations would not have been possible without the expertise and help provided by the vla analysts and operators . we thank the anonymous referee for helpful comments . for useful discussions , we are indebted to lori lubin , andy biggs , roger blandford , geoff bower , ketan desai , debbie haarsma , phillip helbig , jackie hewitt , tomislav kundi , erik leitch , chung - pei ma , mark metzger , chris moore , gerry neugebauer , frazer owen , michael rupen , david rusin , martin shepherd , and ed turner . this work is supported in part by the nsf under grant # ast 9420018 and by the european commission , tmr program , research network contract erbfmrxct96 - 0034 `` ceres . ''

this paper considers a multiple - input multiple - output ( mimo ) receiver with very low - precision analog - to - digital convertors ( adcs ) with the goal of developing massive mimo antenna systems that require minimal cost and power . previous studies demonstrated that the training duration should be _ relatively long _ to obtain acceptable channel state information . to address this requirement , we adopt a joint channel - and - data ( jcd ) estimation method based on bayes - optimal inference . this method yields minimal mean square errors with respect to the channels and payload data . we develop a bayes - optimal jcd estimator using a recent technique based on approximate message passing . we then present an analytical framework to study the theoretical performance of the estimator in the large - system limit . simulation results confirm our analytical results , which allow the efficient evaluation of the performance of quantized massive mimo systems and provide insights into effective system design . bayes - optimal inference , joint channel - and - data estimation , low - precision adc , massive mimo , replica method .

1507.07766

fifth - generation ( 5 g ) mobile communication systems are expected to achieve a @xmath0-fold increase in capacity , a @xmath1-fold increase in spectral and energy efficiencies , and a @xmath2-fold increase in average cell throughput @xcite . such significant enhancements can be achieved with large - scale multiple - input multiple - output ( mimo ) antenna systems , which are also referred to as `` massive mimo '' systems , e.g. , @xcite . these systems employ hundreds , or even thousands , of antennas at base stations ( bss ) to serve tens or hundreds of user terminals with the same time frequency resources . as such , array gains are expected to grow infinitely with the number of antennas at the bss , in which case the radiated energy efficiency increases dramatically and multiuser interference is eliminated completely . however , the high dimensionality of massive mimo systems considerably increases hardware cost and power consumption . in particular , the hardware complexity and power consumption of analog - to - digital converters ( adcs ) increase exponentially with the number of bits per sample @xcite and thus present a major obstacle . this drawback has motivated the use of low - cost low - precision adcs ( e.g. , @xmath3 bits ) for antennas , which has resulted in _ quantized _ mimo systems . bits are used in modern communication systems to process received signals in the digital domain . in this paper , the `` quantized '' mimo system specifically represents a mimo system equipped with very low - precision adcs ( e.g. , @xmath3 bits ) . ] such coarse quantization leads to the failure of all communication theories , as well as signal processing techniques dedicated to high - resolution quantization @xcite . some aspects of quantized mimo systems have been studied in the literature on capacity analysis @xcite , energy efficiency analysis @xcite , feedback codebook design @xcite , data detection @xcite , and channel estimation @xcite . the current work is focused on data detection and channel estimation for quantized mimo systems . previous studies on these subject mainly assumed perfect channel state information ( csi ) at the receiver ( csir ) or considered problems in channel estimation . the use of coarse quantization greatly reduces the number of _ effective _ measurements and hinders the easier acquisition of csir in quantized mimo systems than in unquantized ones . as explained in @xcite , a one - bit quantized mimo system requires an extremely long training sequence ( e.g. , approximately @xmath4 times the number of users ) to achieve the same performance as that in a full csi case . this requirement motivates us to consider joint channel - and - data ( jcd ) estimation , in which the estimated payload data are utilized to aid channel estimation . a major advantage of jcd estimation is that relatively few pilot symbols are required to achieve equivalent channel and data estimation performances @xcite . although an improved performance with the jcd technique is expected , its performance in _ quantized _ mimo systems is not clearly understood . the most related work appears to be that in @xcite , which investigated the achievable throughput in a one - bit quantized single - input single - output ( siso ) channel using jcd estimation ( i.e. , least squares channel estimation jointly on pilot and data symbols ) . for the one - bit quantized mimo system in @xcite , the authors considered a pilot - only scheme with least - squares channel estimation , followed by data detection that utilizes the maximal - ratio combining . although high - order constellation , such as 16-qam , was found to be capable of being supported by the one - bit quantized mimo system , which outperforms the ones reported in @xcite for qpsk , the long training sequence is still a requirement . hence , the fundamental performance limits on quantized mimo systems imposed by the jcd estimation represents an interesting research topic . in the present work , we propose a framework for analyzing the achievable performance of quantized mimo systems with jcd estimation . unlike other jcd estimation schemes based on suboptimal criteria @xcite , the bayes - optimal inference for jcd estimation is used in this work because this approach generates minimum mean square errors ( mmses ) with respect to ( w.r.t . ) the channels and data symbols . in the conference version of this work @xcite , our simulation results indicate that the bayes - optimal jcd estimator exhibits a significant advantage over pilot - only schemes in quantized mimo systems . in addition to the derivations omitted in @xcite , the main contributions of this work are summarized as follows . * to implement the bayes - optimal jcd estimator , we use a variant of belief propagation ( bp ) in approximating the marginal distributions of each data and channel component . we modify the bilinear generalized approximate message passing ( big - amp ) algorithm in @xcite and adapt it to the quantized mimo system by providing the corresponding closed - form expressions for the nonlinear steps . we refer to this scheme as the gamp - based jcd algorithm . * by performing a large - system analysis based on the replica method from statistical physics , we show the _ decoupling principle _ for the bayes - optimal jcd estimator . that is , in the large - system regime , the input output relationship of a quantized mimo system using the bayes - optimal jcd estimator is decoupled into a bank of scalar additive white gaussian noise ( awgn ) channels w.r.t . the data symbols and channel response . this decoupling property allows the characterization of several system performances of interest in an intuitive manner . in particular , the average symbol error rate ( ser ) w.r.t . the data symbols and the average mse w.r.t . the channel estimate for the bayes - optimal jcd estimator are determined . * finally , computer simulations are performed to verify the efficiency of the proposed gamp - based jcd algorithm and the accuracy of our analysis . the high accuracy of our results ensures the quick and efficient evaluation of the performances of quantized mimo systems . several useful observations related to system design are derived from the analysis . _ notations_throughout , for any matrix @xmath5 , @xmath6 refers to the @xmath7th entry of @xmath5 , @xmath8 denotes the transpose of @xmath5 , @xmath9 is the conjugate transpose of @xmath5 , and @xmath10 denotes its trace . also , @xmath11 denotes the identity matrix , @xmath12 is the zero matrix , @xmath13 denotes the frobenius norm , @xmath14 $ ] represents the expectation operator , @xmath15 is the natural logarithm , and @xmath16 is the signum function . in addition , a random vector @xmath17 drawn from the proper complex gaussian distribution of mean @xmath18 and covariance @xmath19 is described by the probability density function : @xmath20 where @xmath21 returns the determinant . we write @xmath22 . with @xmath23 denoting the real ( or complex ) gaussian integration measure , for an @xmath24 real valued vector @xmath17 , we have @xmath25 or @xmath26 for the complex valued vector , where @xmath27 and @xmath28 extracts the real and imaginary components , respectively . finally , @xmath29 denotes the cumulative gaussian distribution function @xcite . we consider a mimo uplink system in which a bs is equipped with @xmath30 receive antennas serving @xmath31 single - antenna users . the channel is assumed to be flat block fading , wherein the channel remains constant over @xmath32 consecutive symbol intervals ( i.e. , a block ) . the received signal @xmath33 \in { { \mathbb c}}^{n \times t}$ ] over the block interval can be written in matrix form as @xmath34 where @xmath35 \in { { \mathbb c}}^{k \times t}$ ] denotes the transmit symbols in the block , @xmath36\in { { \mathbb c}}^{n \times k}$ ] denotes the channel matrix containing the fading coefficients between the transmit antennas and the receive antennas , @xmath37\ \in { { \mathbb c}}^{n \times t}$ ] represents the additive temporally and spatially white gaussian noise with zero mean and element - wise variance @xmath38 , and we define @xmath39 \triangleq \frac{1}{\sqrt{k}}{{\bf h}}{{\bf x}}\in { { \mathbb c}}^{n \times t}$ ] . on the receiver side , each received signal is down - converted into analog baseband @xmath40 and then discretized using a _ complex - valued _ quantizer @xmath41 , as illustrated in figure [ fig : quantizedmimo ] . here , each complex - valued quantizer @xmath42 is defined as @xmath43 , i.e. , the real and imaginary parts are quantized separately . in practice , a variable gain amplifier ( vga ) with an automatic gain control ( agc ) is used before the quantization to ensure that the analog baseband is within a proper range , e.g. , @xmath44 . analog baseband @xmath40 is assumed to include the agc gain and is thus in a proper range . the resulting quantized signal @xmath45 \in { { \mathbb c}}^{n \times t}$ ] is therefore given by @xmath46 where the quantization is applied element - wise . specifically , each complex - valued quantizer @xmath41 consists of two real - valued @xmath47-bit quantizers @xmath48 . each real - valued quantizer maps a real - valued input to one of the @xmath49 bins , which are characterized by the set of @xmath50 thresholds @xmath51 $ ] , such that @xmath52 . for notational convenience , we define @xmath53 and @xmath54 . the output is assigned a value in @xmath55 $ ] when the quantizer input falls in the interval @xmath55 $ ] ( namely , the @xmath56-th bin ) . for example , the threshold of a typical uniform quantizer with the quantization step size @xmath57 is given by @xmath58 and the quantization output is assigned the value @xmath59 when the input falls in the @xmath56-th bin .. if @xmath60 , the quantization output is assigned the value @xmath61 . ] figure [ fig : quantizedmimo ] shows an example of the @xmath62-bit uniform quantizer . notice that in practice , the vga gain can be adjusted to attain the desired step size @xmath57 . the channel matrix @xmath63 needs to be estimated at the receiver ; thus , the first @xmath64 symbols of the block of @xmath32 symbols serve as the pilot sequences . the remaining @xmath65 symbols are used for data transmissions . the training and data phases are referred to as @xmath66-phase and @xmath67-phase , respectively . this setting is equivalent to partitioning @xmath68 and @xmath69 as @xmath70,~\mbox{with $ { { \bf x}}_{{{\sf t } } } \in { { \mathbb c}}^{k\times t_{{{\sf t } } } } $ , $ { { \bf x}}_{{{\sf d } } } \in { { \mathbb c}}^{k \times t_{{{\sf d } } } } $ , } \\ { { \widetilde{{{\bf y}}}}}&= \big[{{\widetilde{{{\bf y}}}}}_{{{\sf t}}}~{{\widetilde{{{\bf y}}}}}_{{{\sf d } } } \big],~\mbox{with $ { { \widetilde{{{\bf y}}}}}_{{{\sf t } } } \in { { \mathbb c}}^{n \times t_{{{\sf t } } } } $ , $ { { \widetilde{{{\bf y}}}}}_{{{\sf d } } } \in { { \mathbb c}}^{n \times t_{{{\sf d } } } } $ } .\end{aligned}\ ] ] we assume that @xmath71 ( or @xmath72 ) is composed of independent and identically distributed ( i.i.d . ) random variables @xmath73 ( or @xmath74 ) drawn from the known probability distribution @xmath75 ( or @xmath76 ) , i.e. , @xmath77 given that the pilot and data symbols should appear on constellation points uniformly , the ensemble averages of @xmath78 and @xmath79 are assumed to be zero . in addition , we let @xmath80 and @xmath81 be the transmit powers in the @xmath66-phase and @xmath67-phase , respectively , i.e. , @xmath82 and @xmath83 . for ease of notation , we refer an entry of @xmath68 to @xmath84 instead of @xmath85 or @xmath86 . therefore , we use @xmath87 and @xmath88 to denote the sets of symbol indices in the @xmath66-phase and @xmath67-phase , respectively . similarly , we assume that each entry @xmath89 is drawn from a complex gaussian distribution @xmath90 , where @xmath91 is the large - scale fading coefficient . if @xmath92 , then @xmath93 to prevent the key features of our results from being obfuscated by complex notations , we consider the case in which all the users have the same large - scale fading factor in the main text . a generalized version of our main result , in which the users have different large - scale fading factors , is presented in appendix e. this generalization can be easily achieved by plugging user index @xmath94 into @xmath91 . we consider the case in which the receiver knows the distributions of @xmath63 and @xmath68 but not their realizations . in the conventional pilot - only scheme , the receiver first uses @xmath95 and @xmath71 to generate an estimate of @xmath63 and then uses the estimated channel for estimating data @xmath72 from @xmath96 @xcite . in contrast to the pilot - only scheme , we consider jcd estimation , where the bs estimates both @xmath63 and @xmath72 from @xmath97 given @xmath71 . we will treat the problem under the framework of bayesian inference , which provides a foundation for achieving the best mse estimates @xcite . we define the likelihood , i.e. , the distribution of the received signals under ( [ eq : qsys ] ) conditional on the unknown parameters , as @xmath98 where @xmath99 when @xmath100}$ ] and @xmath101}$ ] . based on the cumulative gaussian distribution function ( see the definition in notations ) , ( [ eq : lnkelihood_each ] ) becomes @xmath102 where @xmath103 the prior distributions of @xmath68 and @xmath63 are given by ( [ eq : prox ] ) and ( [ eq : proh ] ) , respectively . the posterior probability can then be computed according to bayes rule as @xmath104 where @xmath105 is the marginal likelihood . given the posterior probability , an estimator for @xmath89 can be obtained by the posterior mean @xmath106 where @xmath107 denotes the marginal posterior probability of @xmath89 . in ( [ eq : esth ] ) , the notation @xmath108 denotes the integration over all the variables in @xmath63 except for @xmath89 . similarly , the estimator for @xmath109 for @xmath110 can be obtained by the posterior mean @xmath111 where @xmath112 is the marginal posterior probability of @xmath109 . the notation @xmath113 denotes the integration over all the variables in @xmath68 except for @xmath109 . the posterior mean estimators ( [ eq : esth ] ) and ( [ eq : estx ] ) minimize the ( bayesian ) mse @xcite defined as [ eq : msehx ] @xmath114 where the expectation operator is w.r.t . @xmath115 ; moreover , @xmath116 $ ] and @xmath117 $ ] . we refer to ( [ eq : esth ] ) and ( [ eq : estx ] ) as the bayes - optimal estimator . [ remark_1 ] given a _ known _ pilot matrix , @xmath118 , which by definition is given by @xmath119 , we obtain @xmath120 from ( [ eq : estx ] ) ; therefore , @xmath121 . for the case of interest , we always have @xmath121 . the algorithm as well as the analytical results still work even if the pilots are unknown , and the mse can be expressed as ( [ eq : msex ] ) . to better understand the bayes - optimal estimator , we first explain it in a simple siso system . we consider a siso version of the system ( [ eq : sys ] ) given by @xmath122 recall that @xmath123 is the additive white gaussian noise with zero mean and variance @xmath124 . after a complex - valued quantizer , @xmath125 is obtained . based on the system model ( [ eq : sys ] ) , @xmath126 should be kept . however , to facilitate interpretation , we first let @xmath127 be a random variable with distribution @xmath128 . according to bayes rule ( [ eq : posteriorpr ] ) , the posterior probability can be computed as @xmath129 where @xmath130 is the marginal likelihood . then , from ( [ eq : esth ] ) or ( [ eq : estx ] ) , the posterior mean estimator for @xmath127 is given by @xmath131 to specify the estimator , we further assume that @xmath127 is a proper complex gaussian with mean @xmath132 and variance @xmath133 , i.e. , @xmath134 . then , we derive the estimator ( [ eq : postest_z ] ) under the two channels , unquantized and quantized , in the following examples . * example 1 * ( unquantized channel ) . in this case , we have @xmath135 and @xmath136 . using these distributions , we obtain @xmath137 where @xmath138 , and the second equality follows the _ gaussian reproduction property _ * ( a.7 ) ) . ] where @xmath139 , @xmath140 , and @xmath141 . ] substituting ( [ eq : post_z_num ] ) into ( [ eq : margpost_z ] ) , we obtain @xmath142 the estimator ( [ eq : postest_z ] ) , which is the mean of @xmath143 after _ rearranging _ is determined as @xmath144 the mse of the estimator , which is the variance of @xmath143 , is @xmath145 * example 2 * ( quantized channel ) . if @xmath100}$ ] and @xmath101}$ ] , then the likelihood of the quantized measurement @xmath146 is given by ( [ eq : lnkelihood_each_simple ] ) . the calculation of the posterior mean and variance in the quantized channel is technical , but it basically follows a procedure similar to that in the unquantized channel . a derivation is given in appendix a , which turns out to yield @xmath147 where [ eq : eta_def ] @xmath148 the real and imaginary parts are quantized separately , and each complex - valued channel can be decoupled into two real - valued channels . the expressions ( [ eq : hatz_realgaussian ] ) and ( [ eq : msez_realgaussian ] ) are the estimators only for the real part of @xmath127 . to facilitate notation , we have abused @xmath146 and @xmath149 in ( [ eq : hatz_realgaussian ] ) and ( [ eq : msez_realgaussian ] ) to denote @xmath150 and @xmath151 , respectively . the estimator for the imaginary part @xmath152 can be obtained analogously as ( [ eq : hatz_realgaussian ] ) and ( [ eq : msez_realgaussian ] ) , while @xmath146 and @xmath56 should be replaced by @xmath153 and @xmath154 , respectively . recall @xmath155 and @xmath156 . therefore , if @xmath157 or @xmath158 , we obtain @xmath159 , @xmath160 , and @xmath161 . additionally , for the special case of @xmath162 ( i.e. , one - bit quantizer ) , the expressions of ( [ eq : hatz_realgaussian ] ) and ( [ eq : msez_realgaussian ] ) agree with those reported in @xcite . for another extreme case of @xmath163 and @xmath164 , we return to the unquantized channel in example 1 . instead of using the procedure in example 1 , we show how the expressions ( [ eq : postest_z ] ) and ( [ eq : postest_z_var ] ) can be obtained from ( [ eq : hatz_realgaussian ] ) and ( [ eq : msez_realgaussian ] ) . recall that @xmath165 and @xmath166 are the upper and lower bin boundary positions w.r.t . the @xmath56-th bin , respectively . let @xmath167 and @xmath168 . as @xmath163 and @xmath164 , we obtain @xmath169 , which results in @xmath170 and @xmath171 . furthermore , we obtain @xmath172 , @xmath173 , and @xmath174 . by substituting these relationships into ( [ eq : hatz_realgaussian])([eq : msez_realgaussian ] ) and applying the facts that @xmath175 , @xmath176 , and @xmath177 , we recover the same expressions as those given in ( [ eq : postest_z ] ) and ( [ eq : postest_z_var ] ) for the real part of @xmath149 . the imaginary part for @xmath149 can be obtained analogously . the aforementioned example is the estimator for @xmath127 . the same concept can be easily applied to the estimate of @xmath178 or @xmath179 , if @xmath127 is replaced by @xmath178 or @xmath179 in ( [ eq : example_sys ] ) . however , if @xmath126 and both @xmath178 and @xmath179 are unknown , the complexity of the bayes - optimal estimator increases . in this case , the posterior probability in ( [ eq : margpost_z ] ) becomes @xmath180 , which involves two prior distributions for @xmath178 and @xmath179 as that in ( [ eq : posteriorpr ] ) . to implement the posterior mean estimator for @xmath178 and @xmath179 , we need the marginal posterior probabilities @xmath181 and @xmath182 , respectively . a closed form for the posterior probability @xmath183 does not appear possible . although we can resort to numerical integration to implement the estimator , the computational complexity is high . therefore , one might consider an alternative technique ; that is , the estimate of @xmath178 is performed with fixed @xmath179 and vice versa . in the next subsection , we develop a practical algorithm for the bayes - optimal estimator in the quantized mimo system . before proceeding , we intend to provide an intuition on the algorithm . a representation of the algorithm is shown in fig . [ fig : gamp - basedjcd ] , which seems to operate in the alternative manner . conceptually , when the posterior mean and variance of @xmath127 are obtained from the quantized observation @xmath146 , we can _ reconstruct _ @xmath184 and then approximate @xmath185 as a gaussian distribution . then , the posterior mean estimator for @xmath178 ( or @xmath179 ) can be conducted through @xmath184 , which is an _ awgn _ channel rather than a quantized channel , in an alternative manner . this representation is merely for an intuition . the accurate algorithm development takes a different route . from the discussions above , the direct computations of ( [ eq : esth ] ) and ( [ eq : estx ] ) are intractable because of the high - dimensional integrals in the marginal posteriors @xmath186 and @xmath187 . to make these tractable , we first note that by combining ( [ eq : prox])([eq : lnkelihood ] ) , the posterior probability ( [ eq : posteriorpr ] ) can be factored into @xmath188 an example factor graph for ( [ eq : postfg ] ) is shown in figure [ fig : factorgraph ] , where a square represents a factor node associated with the sub - constraint function @xmath189 in ( [ eq : postfg ] ) , whereas a circle shows a variable node associated with @xmath89 , @xmath86 , or @xmath85 . the factor graph suggests the use of the canonical sum - product algorithm to compute the marginal posterior probabilities . the algorithm uses a set of message - passing equations that go from factor nodes to variable nodes and vice versa . however , the computational complexity of the sum - product algorithm remains infeasible in the case of interest because it still involves high - dimensional integration and summation . thus , we resort to recently developed approximation algorithms : the so - called amp ( approximate message - passing ) algorithm @xcite and the generalized amp ( gamp ) algorithm @xcite . in particular , the amp algorithm , which is a variant of the sum - product algorithm , was initially proposed by donoho _ et al . _ @xcite to solve a linear inverse problem in the context of compressive sensing . the use of gamp in our mimo system means that given @xmath63 is known ; thus , gamp can provide a tractable method to approximate the marginal posteriors @xmath186 s . this part corresponds to addressing the message - passing equations between @xmath190 and @xmath84 , i.e. , the left - hand side of figure [ fig : factorgraph ] . for the study , see @xcite . more recently , et al . _ in @xcite applied a similar gamp strategy , referred to as big - amp , to the problem of reconstructing matrices from bilinear noisy observations ( i.e. , reconstructing @xmath63 and @xmath68 from @xmath191 ) . [ ago : bigamp - jcd ] big - amp for jcd estimation is presented in algorithm [ ago : bigamp - jcd ] for a given instantiation of the quantized observations @xmath69 , the pilot matrix @xmath71 , as well as the likelihood @xmath192 , and the variable distributions @xmath193 and @xmath194 . we refer to this scheme as the gamp - based jcd algorithm , which follows the same structure as big - amp @xcite except for the steps in dealing with the known pilots , i.e. , @xmath195 in algorithm [ ago : bigamp - jcd ] . reference @xcite presents the derivation details of big - amp . to better understand the algorithm , we provide some intuition on each step of algorithm [ ago : bigamp - jcd ] . [ fig : gamp - basedjcd ] illustrates the structure of the algorithm . in lines 36 , an estimate @xmath196}$ ] of the matrix product @xmath197 and the corresponding variances @xmath198 are estimated . @xmath199 } $ ] and @xmath200 in lines 3 and 4 can be regarded as auxiliary variables . is a plug - in estimate of @xmath201 , whereas @xmath202 $ ] provides a refinement by introducing the `` onsager '' correction into the context of amp . for details , see @xcite . ] the same procedure is followed in lines 1 and 2 but for the matrix product @xmath203 . given that the pilot matrix @xmath71 is known , the corresponding variances for @xmath71 are zero , i.e. , @xmath204 for @xmath205 . with @xmath204 , we thus have plugged @xmath206 and @xmath207 into @xmath208 and @xmath209 for @xmath210 to obtain lines 1 and 2 . the posterior means @xmath211 $ ] and variances @xmath212 of @xmath213 are obtained in lines 7 and 8 using @xmath214 . subsequently , the posterior moments are used in lines 9 and 10 to compute the residual @xmath215 $ ] and the inverse residual variances @xmath216 . in lines 11 and 12 , these residual terms are used to compute @xmath217}$ ] and @xmath218 , where @xmath219 can be interpreted as an observation of @xmath86 under an awgn channel with zero mean and a variance of @xmath220 . similarly , @xmath221 $ ] and @xmath222 , where @xmath223 can be interpreted as an observation of @xmath89 under an awgn channel with noise variance of @xmath222 , are evaluated in lines 13 and 14 . finally , the posterior mean @xmath224 $ ] and variances @xmath225 are estimated in lines 15 and 16 by taking into account the prior @xmath76 ; the same is performed for @xmath89 in lines 17 and 18 . algorithm [ ago : bigamp - jcd ] provides a high - level description of big - amp to perform jcd estimation . lines 78 , 1516 , and 1718 of algorithm [ ago : bigamp - jcd ] perform the posterior mean and variance estimators for @xmath226 , @xmath84 , and @xmath89 , respectively . a remarkable feature of the algorithm is that at each iteration , the estimates of @xmath226 , @xmath84 , and @xmath89 can separately serve as the estimators over a bank of scalar channels . next , we describe these nonlinear steps in detail . for brevity , we omit the subscript indexes @xmath227 hereafter . first , we notice that lines 78 compute the posterior mean and variance of @xmath127 ; in this computation , the expectation operator is w.r.t . @xmath228 where @xmath229 is given by ( [ eq : lnkelihood_each_simple ] ) , and @xmath230 . this process is identical to that implemented in example 2 . as a result , lines 78 of algorithm [ ago : bigamp - jcd ] for each real - valued channel can be computed using the expressions in ( [ eq : hatz_realgaussian])([eq : msez_realgaussian ] ) . next , we discuss the nonlinear steps used to compute @xmath231 and @xmath232 in lines 1516 and 1718 of algorithm [ ago : bigamp - jcd ] . specifically , the expectations and variances in lines 15 - 16 and 17 - 18 are taken w.r.t . the marginal posterior @xmath233 these posterior probabilities are similar to those of @xmath234 , except that @xmath235 is replaced with a gaussian distribution and the corresponding priors @xmath76 and @xmath236 are used in place of @xmath128 . in fact , the former change results in an estimator that is fundamentally different from that in the case of @xmath127 . in example 1 , if @xmath235 is a gaussian distribution , then the estimator is operated in an _ unquantized _ channel . that is , the estimates of @xmath178 and @xmath179 in algorithm [ ago : bigamp - jcd ] are based on awgn channels . to specify @xmath231 , we consider the square qam constellation with @xmath237 points @xmath238 where @xmath239 is the power normalization factor . if @xmath179 is drawn from the constellation points uniformly , i.e. , @xmath240 for @xmath241 , then lines 1516 of algorithm [ ago : bigamp - jcd ] can be computed using @xmath242 where @xmath243 finally , recall that @xmath244 . then lines 1718 of algorithm [ ago : bigamp - jcd ] can be computed using @xmath245 the derivation of ( [ eq : hath_realgaussian ] ) is identical to that in example 2 . using the above nonlinear steps ( [ eq : hatz_realgaussian])([eq : msez_realgaussian ] ) and ( [ eq : hatx_realgaussian])([eq : hath_realgaussian ] ) , we implement the gamp - based jcd algorithm based on the open - source `` gampmatlab '' software suite . in this section , we present a framework to analyze the bayes - optimal jcd estimator . the key strategy for analyzing @xmath246 and @xmath247 is through the average free entropy @xmath248 where @xmath249 denotes the marginal likelihood in ( [ eq : marginalpr ] ) , that is , the partition function . aligned with the argument in @xcite , @xmath247 and @xmath246 are saddle points of the average free entropy . thus , the goal is reduced to finding ( [ eq : freeen ] ) . the analysis is based on a large - system limit , that is , when @xmath250 but the ratios @xmath251 are fixed and finite . for convenience , we simply use @xmath252 to denote this large - system limit . even in the large - system limit , the computation of ( [ eq : freeen ] ) is difficult . the major difficulty in computing ( [ eq : freeen ] ) is the expectation of the logarithm of @xmath249 , which , nevertheless , can be facilitated by rewriting @xmath253 as ] where @xmath254 is any positive random variable . ] @xmath255 the expectation operator is then transformed inside the log - function . we first evaluate @xmath256 for an integer - valued @xmath257 , and then generalize the result to any positive real number @xmath257 . this technique , called _ the replica method _ , is from the field of statistical physics @xcite , which is not mathematically rigorous . nevertheless , the replica method has proved successful in a number of highly difficult problems in statistical physics @xcite and information theory @xcite . some of the results originally obtained by the replica method have been subsequently validated by other approaches ( e.g. , @xcite ) . under the assumption of @xmath252 and replica symmetry ( rs ) , an asymptotic free entropy can be obtained later in proposition [ pro_freeentropy ] . we check the accuracy of the replica - based analysis via simulations . proposition [ pro_freeentropy ] involves several new parameters . most parameters ( except for some auxiliary parameters ) of proposition [ pro_freeentropy ] can be illustrated systematically by the scalar awgn channels : [ eq : scalch ] @xmath258 where @xmath259 , @xmath260 , and @xmath261 . we shall specify how the parameters @xmath262 and @xmath263 are related to the asymptotic free entropy later in proposition [ pro_freeentropy ] . thus far , we know that the parameters @xmath263 and @xmath262 serve as the signal - to - noise ratios ( snrs ) of the above awgn channels . the likelihoods under ( [ eq : scalch_x ] ) and ( [ eq : scalch_h ] ) are respectively given by @xmath264 and then we get the posteriors @xmath265 some important quantities are obtained with the posteriors . for example , the posterior mean estimators for @xmath266 and @xmath178 read [ eq : est_xh ] @xmath267 the mses of the estimators are thus given by [ eq : mse_xh ] @xmath268 where the expectations are taken over @xmath269 and @xmath270 , respectively . equations ( [ eq : hatx_realgaussian])([eq : hath_realgaussian ] ) are explicit expressions of the above quantities . in addition , the mutual information between @xmath271 and @xmath272 reads @xcite @xmath273 and the mutual information between @xmath274 and @xmath275 is @xmath276 from ( [ eq : scalch ] ) , an inference that another scalar awgn channel w.r.t . the @xmath66-phase exists can be made , i.e. , @xmath277 where @xmath278 and @xmath279 . as the pilot is known , @xmath280 can be obtained easily following the argument in remark [ remark_1 ] ; and the mutual information between @xmath281 and @xmath282 is @xmath283 . as all the performances relating to ( [ eq : scalch_xt ] ) are trivial , we will not use ( [ eq : scalch_xt ] ) in the following discussions . [ pro_freeentropy ] as @xmath284 , the asymptotic free entropy is @xmath285 where @xmath286 @xmath287 for @xmath288 ; @xmath289 s are given by ( [ eq : mix_scalch ] ) and ( [ eq : mih_scalch ] ) ; and @xmath290 , @xmath291 . in ( [ eq : freeentropyfinal ] ) , the other parameters @xmath292 are obtained from the solutions to the fixed - point equations [ eq : fxiedpoints ] @xmath293 where @xmath294 , and @xmath295 and @xmath296 are given by ( [ eq : mse_xh ] ) . in ( [ eq : fxiedpoints ] ) , we have defined @xmath297 with @xmath298 given by ( [ eq : psi_def ] ) and @xmath299 see appendix b. as previously mentioned , the asymptotic mses of @xmath72 and @xmath63 are the saddle points of the free entropy . clearly , from proposition [ pro_freeentropy ] , they are @xmath296 and @xmath295 , respectively . notably , the mses are associated with the scalar awgn channels ( [ eq : scalch_x ] ) and ( [ eq : scalch_h ] ) . therefore , the performances of the quantized mimo system seem to be fully characterized by the scalar awgn channels ( [ eq : scalch ] ) . the following proposition formulates such intuition . [ pro_decoupling ] let @xmath86 , @xmath89 , @xmath300 , and @xmath301 denote the @xmath302-th and @xmath303-th entries of @xmath72 , @xmath63 , @xmath304 , and @xmath305 , respectively . as @xmath284 , the joint distribution of @xmath306 of channels ( [ eq : qsys ] ) , ( [ eq : esth ] ) , and ( [ eq : estx ] ) converges to the joint distribution of @xmath307 of the scalar channels ( [ eq : scalch_x ] ) and ( [ eq : scalch_x ] ) . see appendix c. proposition [ pro_decoupling ] shows that , in the large - system limit , the input output of the quantized mimo system employing the bayes - optimal jcd estimator is equivalently decoupled into a bank of the scalar awgn channels ( [ eq : scalch_x ] ) and ( [ eq : scalch_h ] ) . this characteristic is known as the decoupling principle , which was introduced by @xcite for treading an _ unquantized _ mimo system with _ perfect _ csir . if perfect csir is available , then we will not need ( [ eq : scalch_h ] ) for treating the channel estimation quality . notably , proposition [ pro_decoupling ] extends the decoupling principle to a general setting . in particular , we employ the jcd estimator so that the decoupled awgn channels involve not only the data symbol [ i.e , ( [ eq : scalch_x ] ) ] but also the channel response [ i.e. , ( [ eq : scalch_h ] ) ] as well . the equivalent channels ( [ eq : scalch ] ) are the scalar awgn channels , with @xmath308 and @xmath262 being the equivalent snrs . as shown in ( [ eq : fxiedpoints ] ) and ( [ eq : chi_def ] ) , the quantization effect is included in @xmath308 and @xmath262 through @xmath309 for @xmath310 . consider the extreme case of @xmath163 and @xmath164 , i.e. , the unquantized channel . in this case , the riemann sums @xmath311 in ( [ eq : chi_def ] ) becomes the riemann integral over the interval @xmath312 . applying the technique in remark 3 to ( [ eq : chi_def ] ) and evaluating the integrals , @xmath309 can be simplified to @xmath313 substituting ( [ eq : fxiedpoints ] ) for @xmath314 and @xmath315 in the denominator of ( [ eq : chi_unq ] ) , we obtain @xmath316 . the quantity in this form can be understood as the noise plus the residual interference resulting from the estimation errors of the data symbol and channel response . we focus our interest on several special cases in the following examples to obtain more insight into proposition [ pro_decoupling ] . * example 3 * ( constellation - like inputs ) . based on proposition [ pro_decoupling ] , the asymptotic mses w.r.t . @xmath72 and @xmath63 can be determined by using the mses of the scalar awgn channels ( [ eq : scalch_x ] ) and ( [ eq : scalch_h ] ) , respectively . thus , if the data symbol is drawn from a quadrature phase - shift keying ( qpsk ) constellation , then we will derive @xmath317 the ser w.r.t . @xmath72 can also be evaluated through the scalar awgn channel ( [ eq : scalch_x ] ) , which is given by @xcite @xmath318 ^ 2,\ ] ] where @xmath319 is the q - function . in fact , all these performances w.r.t . @xmath72 can be determined on the basis of knowledge of the scalar awgn channel with snr @xmath320 . thus , if the data symbol is drawn from other square qam constellations , then the corresponding ser can be easily obtained by using the closed - form ser expression in @xcite . * example 4 * ( perfect csir ) . if the channel matrix @xmath63 is perfectly known , then the @xmath66-phase will not be required so that @xmath321 given that @xmath63 is perfectly known , @xmath322 . integrating this into ( [ eq : asyvarh ] ) , we immediately obtain @xmath323 , which yields @xmath324 in which ( [ eq : ccpqq_pcsi ] ) follows the result of @xmath325 and ( [ eq : asyvarxd ] ) . substituting ( [ eq : beta_pcsi])([eq : ccpqq_pcsi ] ) into ( [ eq : psi_def ] ) , ( [ eq : chi_def ] ) and ( [ eq : psip_def ] ) , we derive more concise expressions for @xmath326 , @xmath327 , and @xmath328 . notably , when particularizing our results to the case with the qpsk inputs , we recover the same asymptotic mse expression as given in ( * ? ? ? * ( 7 ) and ( 8) ) . more precisely , in @xcite , the real - valued system with bpsk signal was considered . in such case , @xmath329 in our study should be replaced with @xmath166 . * example 5 * ( pilot - only scheme ) . in the conventional pilot - only scheme , the receiver solely uses @xmath95 and @xmath71 to generate an estimate of @xmath63 and subsequently uses the estimated channel for estimating the data @xmath72 from @xmath96 @xcite . the analysis of the asymptotic mse w.r.t . @xmath63 is the same as that in example 4 , but the roles of @xmath63 and @xmath71 are exchanged . in particular , during the @xmath66-phase , we derive @xmath330 and @xmath294 because no data symbol is involved and the pilot matrix @xmath71 is known . after substituting these parameters into ( [ eq : fxiedpoints ] ) and simplification , we obtain the _ self - contained _ fixed - point equations @xmath331 with @xmath332 in ( [ eq : mse_h_pilotonly ] ) , @xmath295 represents the asymptotic mse w.r.t . @xmath63 for the pilot - only scheme , which is also the mse w.r.t . @xmath178 for the scalar awgn channel ( [ eq : scalch_h ] ) . notably , @xmath262 serves as the snr of the awgn channel ( [ eq : scalch_h ] ) . comparing @xmath262 in ( [ eq : q_h_pilotonly ] ) with that in ( [ eq : asyvarh ] ) , we determine that the second term of @xmath262 in ( [ eq : asyvarh ] ) is the gain achieved by data - aided channel estimation . [ cols="^,^,^,^,^,^,^,^",options="header " , ] [ table : coefficientsofdelta ] we fit the optimal step sizes for different input signals by using a first - degree polynomial equation @xmath333 where @xmath334 represents the snr in db scale to derive a general expression . the ( least squares fit ) coefficients @xmath335 are listed in table [ table : coefficientsofdelta ] . the optimal step sizes determined by using @xmath336 are also indicated by the shaded area in figure [ fig : stepsize_qpsk16qam_ser ] . we observe that , although @xmath336 is nonoptimal for each specific input , their corresponding performances remain acceptable . following the same argument , we derive the corresponding polynomial equation @xmath337 for different quantization bits , with their coefficients listed in table [ table : coefficientsofdelta ] . comparing figures [ fig : ser_qpsk](a ) and [ fig : ser_qpsk](b ) , we observe that the loss due to no csir is small for the bayes - optimal jcd estimator . then , we discuss the performances of the bayes - optimal jcd estimator _ with _ and _ without _ the perfect csir under various system settings to obtain a better understanding on the effects of csir over the quantized mimo system . in contrast to the qpsk signals used in previous simulations , we focus on the gaussian inputs , that is , @xmath338 , in the subsequent experiments . the other system parameters are the same as that in the previous experiment , that is , @xmath339 , @xmath340 , and @xmath341 . figure [ fig : msex_qvsunq ] shows the asymptotic mse @xmath296 for the bayes - optimal jcd estimator with and without perfect csir . the mse for the pilot - only scheme is also shown . notably , the bayes - optimal jcd estimator shows a significant improvement over the pilot - only scheme in the @xmath342 bit and unquantized cases . the gap between the bayes - optimal jcd estimator with and without perfect csir is small in the unquantized case , whereas the gap is large in the case of the @xmath342 bit quantizer . by observing the @xmath342 bit and unquantized cases , we can expect that the gap decreases with the increase in the adc resolution . a straightforward method to reduce the gap of the @xmath342 bit case is increasing the training length . we provide the mse results for @xmath343 and @xmath344 in figure [ fig : msex_all_jcd ] to verify this argument . however , the improvement achieved by increasing the training length is limited even if @xmath345 , leaving only @xmath346 for data . alternatively , we may consider a larger @xmath347 , but the maximum @xmath347 is limited by the coherence time . if @xmath348 and @xmath349 , then the performance of the bayes - optimal jcd estimator without perfect csir is approximately similar to the ( fundamental limit ) case with perfect csir . nevertheless , such a long coherence time is usually unavailable in practice . we developed a framework for studying the _ best _ possible estimation performance of the quantized mimo system , namely , the massive mimo system with very low - precision adcs . in particular , we used the bayes - optimal inference for the jcd estimation and achieve this estimation by applying the big - amp technique . the asymptotic performances ( e.g. , mses and sers ) w.r.t . the channels and the payload data were derived and shown as simply characterized by scalar awgn channels . monte - carlo simulations were conducted to demonstrate the accuracy of our analytical results . the high accuracy of the analytical expressions enable us to quickly and efficiently assess the performance of the quantized mimo system . thus , we obtained the following useful guidelines for the system design : * we showed that the asymptotic mse of the channel estimate in the conventional pilot - only scheme decreases by approximately @xmath350db for each bit added to the adcs or each doubling of training length . this finding supports the importance of the jcd technique , especially in the _ quantized _ mimo system . * the optimal step size for minimizing sers of the bayes - optimal estimator were shown to be highly different from that for minimizing the distortion of a gaussian signal and , fortunately , can be quickly determined by our analytical expressions . * the bayes - optimal estimator already exhibits the best possible estimation performance . even so , we showed that the performance gap between the bayes - optimal jcd estimator with and without perfect csir still can not be negligible in the _ quantized _ mimo system . we also discussed the ways to reduce the gap and then concluded that achieving the same performance as the full csir case in the quantized mimo system is very difficult . many potential directions for future work are available . the gamp - based jcd algorithm presented in this paper is a first step toward achieving the _ optimal _ jcd estimate under the quantized mimo system . the computational complexity of the gamp - based jcd algorithm may still be too high to be affordable in a commercial system . one possible solution is to adopt other suboptimal schemes such as linear estimators . another feasible solution is using mixed - adc receiver architecture @xcite wherein a small number of high - resolution adcs is available . thus , csir gains high accuracy and facilitates the jcd procedure . for a development in this direction , see @xcite . in this appendix , we derive the expressions ( [ eq : hatz_realgaussian ] ) and ( [ eq : msez_realgaussian ] ) , by applying the techniques in ( * ? ? ? * chapter 3.9 ) . the derivations below are only dedicated for the real part of the estimator because the imaginary part of the estimator can be obtained analogously . note that the signal power and noise power are @xmath351 and @xmath352 , respectively , per real and imaginary part . for ease of notation , we have abused @xmath146 , @xmath127 , and @xmath132 to denote @xmath150 , @xmath353 , and @xmath354 , respectively . to get ( [ eq : hatz_realgaussian ] ) , we begin by deriving the denominator of ( [ eq : margpost_z ] ) . first , recall from ( [ eq : lnkelihood_each ] ) that if @xmath355 $ ] and @xmath356 , the likelihood is given by @xmath357 note that for the special case @xmath157 , we have @xmath53 , and the second term of ( [ eq : pout_app ] ) will disappear . substituting ( [ eq : pout_app ] ) into the denominator of ( [ eq : margpost_z ] ) , it can be shown that @xmath358 differentiating w.r.t . @xmath132 on both sides of ( [ eq : normconst ] ) yields @xmath359 where we have used @xmath360 . using ( [ eq : normconst ] ) , ( [ eq : difnormconst ] ) can be rearranged as @xmath361 where @xmath362 and @xmath363 are given by ( [ eq : eta_def ] ) . multiplying both sizes by @xmath364 , we obtain the marginal posterior mean given in ( [ eq : hatz_realgaussian ] ) . similarly , ( [ eq : msez_realgaussian ] ) can be calculated by differentiating ( [ eq : normconst ] ) twice as @xmath365 which then can be rearranged as @xmath366 we also note that @xmath367 after plugging ( [ eq : dif2normconst2 ] ) and ( [ eq : hatz_realgaussian ] ) into ( [ eq : varz_def ] ) , we obtain ( [ eq : msez_realgaussian ] ) . in the above derivations , we have assumed @xmath356 . for @xmath368 , the above derivations can be used in the same way . using the replica method , we first compute the replicate partition function @xmath369 in ( [ eq : limf ] ) , which with the definition of ( [ eq : posteriorpr ] ) can be expressed as @xmath370 where we define @xmath371 with @xmath372 and @xmath373 being the @xmath374-th replica of @xmath63 and @xmath68 , respectively , and @xmath375 and @xmath376 . here , @xmath377 are random matrices taken from the distribution @xmath378 for @xmath379 . in addition , @xmath380 denotes the integral w.r.t . a discrete measure because the quantized output @xmath69 is a finite set . we will calculate the right - hand side of ( [ eq : sf_e1 ] ) , by applying the techniques in @xcite after additional manipulations . to average over @xmath381 , we introduce two @xmath382@xmath383 matrices @xmath384 $ ] and @xmath385 $ ] for @xmath110 whose elements are defined by @xmath386 and @xmath387 for @xmath388 , where , @xmath389 is the @xmath390th row vector of @xmath372 , and @xmath391 is the @xmath392th column vector of @xmath373 for @xmath393 or @xmath394 . the definitions of @xmath395 and @xmath396 are equivalent to @xmath397 where @xmath398 denotes dirac s delta . let @xmath399 and @xmath400 . inserting the above into ( [ eq : sf_e1 ] ) yields @xmath401 where @xmath402 using the fourier representation of the @xmath403 function via auxiliary matrices @xmath404 } \in { { \mathbb c}}^{(\tau+1)\times(\tau+1)}$ ] , @xmath405 \in { { \mathbb c}}^{(\tau+1)\times(\tau+1 ) } , \forall o \ } $ ] and performing the saddle point method for the integration over @xmath406 , we attain @xmath407 with [ eq : saddlepoint ] @xmath408 where @xmath409 denotes the extreme value of @xmath410 w.r.t . @xmath411 ; @xmath412 @xmath413^{t}$ ] , @xmath414 $ ] . according to ( [ eq : limf ] ) , the average free entropy turns out to be @xmath415 the saddle points of @xmath416 can be found by the point of zero gradient w.r.t . @xmath417 but it is still prohibitive to get explicit expressions about the saddle points . thus , we assume that the saddle points follow the rs form @xcite as @xmath418 in addition , the application of the central limit theorem suggests that @xmath419^t$ ] are gaussian random vectors with @xmath420 covariance matrix @xmath421 . if @xmath422 , then the @xmath423th entry of @xmath424 is given by @xmath425 as such , we set @xmath426 , which is equivalent to introducing to the gaussian random variable @xmath427 for @xmath422 as @xmath428 where @xmath429 and @xmath430 are independent standard complex gaussian random variables . with rs , the problem of seeking the extremum w.r.t . @xmath417 is reduced to seeking the extremum over @xmath431 , which can be obtained by equating the corresponding partial derivatives of the rs expression @xmath416 to zero . to this end , we calculate the rs expression of @xmath416 by substituting these rs expressions into ( [ eq : saddlepoint1])([eq : saddlepoint3 ] ) . first , for ( [ eq : saddlepoint1 ] ) , we substitute ( [ eq : eqz ] ) and perform the expectation w.r.t . @xmath213 and integration over @xmath432 , to yield @xmath433 where we define @xmath434 and @xmath435 with @xmath436 , and @xmath437 and @xmath438 being independent _ real _ standard gaussian random variables . and @xmath430 in ( [ eq : eqz ] ) are standard `` complex '' gaussian random variables . in this paper , we process the real and imaginary parts separately . therefore , for ease of notation , we have rescaled all the observation outputs @xmath439 and @xmath440 by @xmath441 so that the real and imaginary parts of these random variables can be regarded as the standard `` real '' gaussian random variables . ] performing the expectation w.r.t . @xmath437 , ( [ eq : psib1 ] ) can be expressed as ( [ eq : psi_def ] ) . next , we move to the rs calculation of ( [ eq : saddlepoint2 ] ) . under the rs assumption , the first term of ( [ eq : saddlepoint2 ] ) can be written as @xmath442 then we decouple the first quadratic term in the exponent using the hubbard - stratonovich transformation and introducing the auxiliary vector @xmath443 , to rewrite ( [ eq : lt_rate_fun1 ] ) as @xmath444 with rs , the second term of ( [ eq : saddlepoint2 ] ) can be expressed as @xmath445 similarly , for the first and second terms of ( [ eq : saddlepoint2 ] ) , we have @xmath446 putting together ( [ eq : lt_rate_fun2])([eq : lt_rate2_fun2 ] ) , we have the rs expression of @xmath416 . the parameters @xmath431 are determined by setting the partial derivatives of @xmath416 to zeros . in doing so , as @xmath447 , it is easy to get that @xmath448 , @xmath449 , @xmath450 , and @xmath451 . in order to obtain the more meaningful expressions for the other parameters , we introduce two scalar awgn channels given in ( [ eq : scalch ] ) and their associated parameters in section iv - a . equating the partial derivatives of @xmath416 w.r.t . @xmath452 to zeros , we obtain the fixed - point equations given in ( [ eq : fxiedpoints ] ) . finally , taking the partial derivatives of @xmath416 at @xmath453 , and applying the parameters introduced in section iv - a , we obtain ( [ eq : freeentropyfinal ] ) . consider the @xmath454-th and @xmath455-th entries of @xmath63 and @xmath72 , respectively . we will show that the joint moments of the joint distribution of @xmath456 for some indices @xmath454 and @xmath455 converges to the joint distribution of @xmath457 independent of @xmath454 and @xmath455 . following @xcite , we proceed to calculate the joint moments @xmath458 for arbitrary non - negative integers @xmath459 , @xmath460 , @xmath461 , @xmath462 , @xmath463 , @xmath464 , @xmath465 , @xmath465 . to proceed , we define @xmath466 with @xmath467 , @xmath468 and @xmath469 , @xmath470 . if we define the _ generalized _ free entropy as @xmath471 it exactly provides the joint moments of interest . as @xmath472 and @xmath473 , @xmath474 reduces to @xmath256 given in ( [ eq : sf_e1 ] ) . therefore , proceeding with the same steps as in appendix b from ( [ eq : sf_e1 ] ) to ( [ eq : sf_e3 ] ) , we get @xmath475 where @xmath476 is exactly identical to ( [ eq : saddlepoint ] ) while @xmath477 and @xmath478 should be replaced by @xmath479 thus , except for @xmath480 and @xmath481 , the rs expressions for the other parts of @xmath476 can be obtained as in appendix b. we now only need to obtain the rs expressions for @xmath482 and @xmath483 . the generalized free energy in ( [ eq : genfreeentropy ] ) becomes @xmath484 which is the joint moments of @xmath485 . consequently , the joint moment of interest is thus uniquely determined by ( [ eq : finalf_jmoment ] ) due to the carleman theorem . in this derivation , we consider the case at infinity snr , i.e. @xmath486 , and we let @xmath487 and @xmath488 without loss of generality . from ( [ eq : mse_h_pilotonly ] ) , as @xmath489 , we have @xmath490 . an application of the taylor expansion yields @xmath491 , and thus we have @xmath492 let @xmath493 . we then evaluate @xmath494 in ( [ eq : chi_pilotonly ] ) by changing the integration variable from @xmath438 to @xmath437 , which yields @xmath495 where @xmath496 as @xmath497 , @xmath498 can be approximated by @xmath499 which is a quantizer - dependent constant . using @xmath500 given in ( [ eq : chi_pilotonly ] ) and combining ( [ eq : approx_1smseh ] ) and ( [ eq : chi_pilotonly_2 ] ) , we obtain @xmath501 or ( [ eq : mse_highsnr ] ) in db scale , wherein @xmath502 . the values of @xmath503 in table [ table : coefficient_cb ] are obtained from ( [ eq : cb_val ] ) numerically . in this appendix , we extend proposition [ pro_freeentropy ] into the case where users have different large - scale fading factors @xmath504 . this task can be performed by proceeding with the same steps as in appendix a , and the proof is omitted . [ pro_freeentropy_mu ] as @xmath284 , the asymptotic free entropy is @xmath511 where @xmath512 @xmath513 for @xmath288 ; @xmath514 is the mutual information between @xmath515 and @xmath516 ; @xmath517 is the mutual information between @xmath518 and @xmath519 ; and @xmath520 , @xmath521 . in ( [ eq : freeentropyfinal_mu ] ) , the other parameters @xmath522 are obtained from the solutions to the fixed - point equations where @xmath524 , and @xmath525 and @xmath526 are the mses of the bayes - optimal estimators over ( [ eq : scalch_h_mu ] ) and ( [ eq : scalch_x_mu ] ) , respectively . in ( [ eq : fxiedpoints ] ) , we have defined @xmath527 with @xmath298 given by ( [ eq : psi_def_mu ] ) and @xmath528 . , `` low power analog - to - digital conversion in millimeter wave systems : impact of resolution and bandwidth on performance , '' in _ information theory and applications workshop ( ita ) _ , san diego , ca , usa , feb . 2015 , pp . 191198 . , `` performance analysis of signal detection using quantized received signals of linear vector channel , '' in _ proc . inform . theory and its applications ( isita ) _ , auckland , new zealand , dec . 2008 . h. nishimori , _ statistical physics of spin glasses and information processing : an introduction_.1em plus 0.5em minus 0.4emser . number 111 in int . series on monographs on physics . oxford u.k . : oxford univ . press , 2001 . , `` asymptotic analysis of spatially correlated mimo multiple - access channels with arbitrary signaling inputs for joint and separate decoding , '' _ ieee trans . inf . theory _ , 53 , no . 1 , 252268 , jan .

we present new _ very large array _ h i spectral line , archival _ sloan digital sky survey _ , and archival _ spitzer space telescope _ imaging of eight star - forming blue compact dwarf galaxies that were selected to be optically compact ( optical radii < 1 kpc ) . these systems have faint blue absolute magnitudes ( m ) , ongoing star formation ( based on emission - line selection by the h or [ ] lines ) , and are nearby ( mean velocity 3315 kms 45 mpc ) . one galaxy in the sample , adbs 113845 + 2008 , is found to have an h i halo that extends 58 r - band scale lengths from its stellar body . in contrast , the rest of the sample galaxies have h i radii to optical - scale - length ratios ranging from 9.3 to 26 . the size of the h i disk in the `` giant disk '' dwarf galaxy adbs 113845 + 2008 appears to be unusual as compared to similarly compact stellar populations .

1304.3171

blue compact dwarf galaxies ( hereafter , bcds ) have emerged as a particularly intriguing subset of dwarf galaxies that have been studied in the local universe ( [email protected] ) . since the early studies and characterizations by @xcite and @xcite , many investigations have classified bcds as systems that harbor spatially and temporally concentrated ongoing star formation in a low - mass galaxy potential well . while definitions in the literature vary , bcds are instrinsically faint systems ( m@xmath617 ) that exhibit ongoing star formation over the bulk of an otherwise physically compact stellar population . detailed studies of individual , nearby bcds have provided important insights into their physical properties and into the mechanisms that initiate and govern their star formation . @xcite found that five compact dwarf galaxies that are currently experiencing strong bursts of star formation exhibit central h i surface densities that are significantly higher than those of dwarf irregular ( dirr ) galaxies . similarly , some well - studied bcds harbor very high column density neutral hydrogen gas in the regions of active star formation ( e.g. , n@xmath7 @xmath4 2.4@xmath810@xmath9 @xmath10 in izw18 ; see brown etal . 2002 and van zee etal . other recent studies have verified that many nearby and well - resolved systems have high neutral gas surface densities in the regions of ongoing star formation ( see , e.g. , thuan etal . 2004 , elson etal . 2010 , and references therein ) . the physical mechanism(s ) that create this central concentration of neutral gas remain(s ) ambiguous . one commonly proposed solution is that the current burst events in bcds are the result of a gravitational interaction with a nearby companion ( e.g. , * ? ? ? * ) . while there are examples of apparently interaction - driven burst events ( e.g. , the sbs1129@xmath11576 system studied by ekta etal . 2006 ) , there are also systems that have been shown to be quite isolated ( see , for example , van zee etal . 1998 , bureau etal . 1999 , and further discussion below ) . while undiscovered , very low - mass companions may reside near some of these isolated systems ( consider , for example , the discovery of an interaction with a low surface brightness companion to the magellanic irregular ngc4449 by martnez - delgado etal . 2012 ) , the gravitational interaction trigger does not appear to be universal in bcd galaxies . the origin of the concentrated star formation in at least some of the isolated bcds appears to require a different initiation mechanism . while the properties of isolated bcds have been studied in some detail ( e.g. , the studies of ngc2915 by meurer etal . 1994 , 1996 ) , a particularly intriguing system was discovered in the _ arecibo dual beam survey _ ( adbs ) by @xcite : adbs113845 + 2008 . this galaxy was subsequently cataloged as a uv - excess source in the 2nd kiso survey ( miyauchi - isobe & maehara , 2000 ) . its extremely compact nature was first recognized during _ wiyn _ 0.9-meter broadband imaging of selected adbs galaxies . the optical component was found to be extremely compact ( b - band exponential scale length @xmath20.57 kpc ) ; h@xmath1 narrowband imaging confirmed a high ongoing star formation rate ( 0.06 m@xmath12yr@xmath3 ) within this optically compact stellar component . subsequent _ very large array _ ( _ vla _ ) h i spectral line imaging of the source revealed that the neutral gas disk was extremely extended compared to the compact optical structure . as discussed in detail in @xcite , h i gas was detected to a maximum radial distance of @xmath1325 kpc at the 10@xmath14 @xmath10 level ; it thus possesses one of the largest h i - to - optical size ratios of any known dwarf galaxy . in contrast to some detailed h i studies of other bcd systems that have found complex velocity structure ( e.g. , * ? ? ? * ; * ? ? ? * ) , adbs113845 + 2008 is undergoing well - ordered rotation throughout most of the neutral gas disk . perhaps the most mysterious quality of adbs113845 + 2008 is that its neutral gas morphology is dominated by a broken ring structure that is roughly 15 kpc in diameter . this ring contains the highest column density neutral gas in the galaxy and is not coincident with the optical component . rather , the star - forming component resides in a relative h i under - density . @xcite interpret this result to be a signature of an extremely inefficient mode of star formation in a very isolated galaxy . we hereafter refer to this striking collection of physical properties as the `` giant disk '' phenomenon in dwarf galaxies . the remarkable characteristics of adbs113845 + 2008 are the origin of the present investigation . stated most simply , we seek to understand if this `` giant disk '' scenario is common or rare in star - forming dwarf galaxies that have stellar components that are as strongly centrally concentrated as the one in adbs113845 + 2008 . to this end , we present new optical , infrared , and h i imaging of a sample of extremely optically compact dwarf galaxies . the new observations presented here confirm the rarity of the `` giant disk '' found in the adbs113845 + 2008 system . the sample of optically compact galaxies studied in this paper was produced using specific selection criteria designed to identify star - forming galaxies with compact stellar components ; adbs113845 + 2008 and ic2271 were already known to have such characteristics from previous imaging . starting with the full _ kpno international spectroscopic survey _ ( _ kiss _ , an objective - prism survey for extragalactic emission - line objects ; salzer etal . 2000 , 2001 , 2002 ) database , we searched for emission - line objects that were spectroscopically - confirmed . to maximize physical resolution an upper velocity constraint was set at v@xmath15 @xmath16 5,000 kms@xmath3 . photometric parameters were obtained for these objects from the _ sdss _ using the _ skytools _ interface . objects with optical radii less than 3 were selected ; these were then individually inspected to ensure a compact optical structure . finally , we searched for archival single - dish h i observations of these remaining systems in order to determine their suitability for _ vla _ observing . this resulted in a sample of six galaxies selected from the _ kiss _ catalogs . we stress that this sample is meant to be representative of the compact dwarf class , but it is not complete in the sense of being a volume - limited sample of such sources . all of these galaxies share the following characteristics : 1 ) optical compactness ( optical radii at 50% peak brightness of @xmath51 kpc ) ; 2 ) faint blue absolute magnitudes ( m@xmath1717 ) ; 3 ) ongoing star formation ( based on emission - line selection by the h@xmath1 or [ ] lines ) ; and 4 ) proximity for spatially resolved work with the _ vla _ c - array ( 15 beam @xmath55 kpc for all systems ; mean velocity @xmath2 3315 kms@xmath3 @xmath4 45 mpc ) . note that in the work of @xcite , adbs113845 + 2008 was selected using these same criteria . it is important to emphasize that the six _ kiss _ galaxies in the present study ( see table [ t1 ] for salient properties ) were selected based on their optical properties alone ; the h i flux integrals were used only to determine observational feasibility . h i spectral line imaging of each of the sample galaxies was obtained with the _ vla _ in the c configuration via observing programs ac 841 and ac 919 ( pi : cannon ) . the _ vla _ correlator was used in a standard single polarization mode , providing 64 or 128 channels over a 1.56 or 3.12 mhz bandwidth . the velocity resolution ( 2.56 , 5.15 , or 10.3 kms@xmath3ch@xmath3 ) was selected based on the ( single dish ) line width of each source . combinations of bandwidth and frequency resolution were selected to guarantee sufficient line - free channels to allow an accurate removal of the continuum emission . table [ t2 ] provides details about each observing session from which data are presented in this work . all reductions were performed using the astronomical image processing ( aips ) package . due to the array being in a `` hybrid '' state , containing both _ vla _ and _ evla _ receivers , standard reduction procedures were followed with necessary modifications to account for the `` aliasing '' effect that arose due to the digital - to - analog conversion required to correlate _ evla _ signals using the old _ vla _ correlator . this effect was most egregious for narrow bandwidth observations . we thus followed the nrao instructions of flagging _ baselines before and unflagging them after the tasks bpass and calib . the resulting continuum subtraction was performed using a first - order fit in uvlsf to channels where the bandpass was demonstrably flat . we found no measurable differences between this calibration approach and one where we disregarded the _ evla _ antennas completely and followed standard reduction procedures . we note that there were only 5 _ evla _ antennas in the array during the c configuration observation of adbs113845 + 2008 ( which would contain the most measurable aliasing effect , given the bandwidths listed in table [ t2 ] ) . two data cubes were produced for each galaxy using the task imagr and each one was cleaned to 2 times the rms noise ( hgbom 1974 ; clark 1980 ) . first , `` low - resolution '' cubes were created using natural weighting ( robust = 5 in the aips imagr task ) . these cubes were convolved to a circular beam size of 20 x 20 . second , `` high - resolution '' cubes were created using robust weighting ( robust = 0.5 ; see briggs 1995 ) . these cubes were convolved to a circular beam size of 15 x 15 . to create blanked cubes and moment maps , the naturally weighted cubes were convolved to a beam size of 30 . these cubes were blanked at the 2 @xmath18 level found in line - free channels . the cubes were then inspected by hand , and emission in three or more consecutive channels was considered to be real . these cubes were then used to blank both the `` low- '' ( 20 ) and `` high - resolution '' ( 15 ) cubes . through this procedure , we assure that the same pixels contribute to both the `` low- '' and the `` high - resolution '' cubes . moment maps showing integrated h i emission and velocity structures were created using the techniques discussed in detail in @xcite . as seen in table [ t3 ] , the total h i flux integrals derived for the sample galaxies are each lower than the _ arecibo observatory _ single dish observations @xcite . the cause of these differences is at least partly due to the missing short _ uv _ spacings of the _ vla _ compared to a single dish measurement . for example , the observed h i flux integral of ic2271 of 1.22 jy kms@xmath3 is significantly lower than the value found by @xcite using the _ vla _ d array ( 4.7 jy kms@xmath3 ) . similarly , the total h i flux integral derived for adbs113845 + 2008 ( 1.58 jy kms@xmath3 ) is lower than the 2.58 jy kms@xmath3 found by @xcite using combined c and d configuration data . a re - reduction of the d configuration data sets as part of the present analysis confirms a higher flux integral using the more compact configuration ( s@xmath7 = [email protected] jykms@xmath3 for adbs 113845 + 2008 and s@xmath7 = [email protected] jykms@xmath3 for ic 2271 , both with 55 circular beam sizes ) . since the kissr galaxies ( which constitute the bulk of the present sample ) have only c configuration observations , we chose to intentionally omit the d configuration data for ic2271 and adbs113845 + 2008 from the present analysis ; this assures that all eight galaxies have similar thresholds for surface brightness sensitivity at different spatial scales . in an effort to recover as much flux from each of the target galaxies as possible , we performed various tests with the _ vla _ _ uv _ datasets . long baselines were downweighted in imagr via tapering in the _ uv _ plane . circular beam size cubes with resolution elements of 15 , 30 , and 60 were created for each galaxy using appropriate uvtaper values . the results of these tests are summarized in table [ t4 ] . there are only modest gains in the recovered flux integrals for most sources . kissr278 shows the most pronounced increase in s@xmath7 when moving from 15 to 30 resolution . note that ic2271 and adbs113845 + 2008 , which have the largest single dish flux integrals ( see table [ t3 ] ) , show only modest increases in s@xmath7 when downweighting long baselines . this is strong evidence that the missing shortest spacing baselines are important for the recovery of all of the h i flux from these systems . we also experimented with calibrating our _ uv _ data using only certain baselines in the array . using kissr561 as a template , we independently calibrated the _ uv _ data in five different ways : by removing the _ vla_-_evla _ baselines ; by removing the _ vla_-_vla _ baselines ; by retaining only the _ evla_- _ evla _ baselines ; by retaining only the _ vla_-_vla _ baselines ; and by the previously discussed method in which the _ evla_-_evla _ baselines were flagged and unflagged around specific tasks . we find no significant differences in the galaxy s morphology amongst these tests . the last method ( i.e. , the flagging and unflagging of _ evla_-_evla _ baselines ) proved to recover the most flux , and so was used for all datasets in the sample . as we discuss in detail in [ s3.2 ] below , five of the eight sample galaxies have nearby companions , some of which are very h i - bright . a third possible explanation for the lower flux integrals derived for some of the systems studied here is that these high - surface - brightness companions introduce sidelobes , the imperfect cleaning of which will prevent a full recovery of the flux integral from the dim target galaxies . in fact , this effect can lead to the single dish flux measurements being erroneously high . the system that is most egregiously affected in this way is kissr396 ; as we discuss below , this source is a companion to the massive spiral ngc5375 . it is possible that the single dish measurement of kissr396 is an over - estimate , since ngc5375 is in the first sidelobe of the single - dish primary beam . in an attempt to image this complex field with high fidelity , we implemented cleaning algorithms using both multi - faceted cleans around the detected galaxies and multi - scale clean algorithms @xcite . neither approach produced significant increases in the flux integral recovered from kissr396 . we thus use the _ vla _ s@xmath7 values quoted in table [ t3 ] throughout the remainder of this work . optical and infrared imaging of the sample galaxies was obtained from public archival databases to facilitate detailed comparisons with the h i imaging presented here . this was necessary because the imaging data obtained as part of _ kiss _ @xcite had an image scale of 2.0pixel@xmath3 ; none of the _ kiss _ galaxies studied here are resolved in the survey data . optical r - band images were obtained from the _ sdss _ data release eight @xcite . 3.6 @xmath20 m and 4.5 @xmath20 m _ spitzer _ images of the six kissr galaxies were obtained from the _ spitzer heritage archive _ ( _ spitzer _ imaging of ic2271 and adbs113845 + 2008 is not available ) . no further treatments of the pipeline - processed archival data were required . these images are discussed in further detail in the sections below . throughout this paper we will discuss the physical sizes of the galaxies in this sample . we thus draw specific attention to the angular resolution of the optical and infrared imaging used in our analysis . the fwhm of the _ spitzer_/irac images is 1.66 at 3.6 @xmath20 m and 1.72 at 4.5 @xmath20 m . similarly , the median psf fwhm of _ sdss _ r - band images is quoted as 1.3 , and we adopt this value here ; using the individual images we analyze in this work , we measure empirical psf fwhm values ranging from 0.94 ( ic2271 ) to 1.22 ( kissr245 ) . using the blanked datacubes discussed in [ s2.2 ] , real emission in each channel was summed in order to create the global h i profiles shown in figures [ figcap1 ] and [ figcap2 ] . it is immediately evident that the neutral gas disks of the sample galaxies span a wide range of physical characteristics . two systems show broad h i profiles ( @xmath21v @xmath22100 kms@xmath3 ) that are indicative of bulk rotation ( e.g , adbs113845 + 2008 , kissr561 ) . three systems show what may be double - horned or broad profiles that are measured with poor fidelity ( e.g. , kissr245 , kissr396 , kissr1021 ) . three systems show simple single - peak profiles indicative of either low rotation velocities or nearly face - on inclination angles ( e.g , ic2271 , kissr278 , kissr572 ) . the profiles shown in figures [ figcap1 ] and [ figcap2 ] should be compared with the recovered flux integrals presented in table [ t3 ] and discussed in the three systems with the smallest recovered s@xmath7 values in table [ t3 ] ( kissr278 , kissr396 , kissr572 ) are those systems with the narrowest h i profiles . note specifically that kissr278 , kissr396 , and kissr572 are each only detected in 4 , 7 , and 3 consecutive channels of their datacubes , respectively . it is important to keep the low signal to noise ratio of some of these galaxies in mind during the discussion that follows . single dish profiles of the six kissr galaxies in figures [ figcap1 ] and [ figcap2 ] were presented in @xcite . a comparison of these two sets of profiles shows similar morphologies in some cases ; the profiles of kissr278 , kissr396 , and kissr561 shown in figures [ figcap1 ] and [ figcap2 ] are similar in shape as those shown in @xcite , although with slightly weaker peak flux densities in the interferometric global profiles . conversely , the other three kissr systems ( kissr245 , kissr572 , and kissr1021 ) show significant differences in the shapes of their profiles ; the single - dish profiles not only recover more flux ( see detailed discussion in [ s2.2 ] above ) , but they also recover different morphologies ( note , for example , that the single dish profile of kissr1021 shows a pronounced double - horn morphology that is only poorly recovered in the interferometric global profile ) . figures [ figcap3 ] through [ figcap10 ] compare the h i , optical , and infrared ( available for six of the eight systems , excluding ic2271 and adbs113845 + 2008 ) images of the eight sample galaxies . using the 2 kpc physical length denoted in panel ( b ) of each figure , it is immediately evident that the high surface brightness stellar component of each of these eight galaxies is extremely compact . in fact , some of these galaxies are only slightly more extended than foreground stars found in the fields [ note the close proximity of foreground stars to adbs113845 + 2008 ( figure [ figcap4 ] ) and kissr561 ( figure [ figcap8 ] ) ] ; we discuss the effects of resolution on our results in more detail below . while the h i surface brightness varies from one system to the next , each of these eight galaxies is detected at high significance in both the _ sdss _ and in the _ spitzer _ images . an examination of the h i column density images and contours in figures [ figcap3 ] through [ figcap10 ] reveals that all of the galaxies in this sample , with the exception of adbs113845 + 2008 , harbor centrally concentrated neutral gas distributions at these spatial resolutions . in most cases the optical and infrared surface brightness maxima are spatially coincident with the highest detected column density region ( e.g. , ic2271 in figure [ figcap3 ] , kissr572 in figure [ figcap9 ] ) ; in some systems there is a small offset between these two maxima , but most of these offsets are smaller than the synthesized beam sizes and are thus not significant . the most significant discrepancy between the stellar and gaseous components is found in adbs113845 + 2008 : as discussed in detail in @xcite , the h i morphology is dominated by a broken ring structure that surrounds the optical body of the system . the stellar component is many kpc away from the highest column density neutral gas . similarly , an examination of the optical and infrared images of these sources reveals that most members of the sample are extremely compact . however , it is important to stress that each of the systems is spatially resolved in the _ sdss _ images , the _ spitzer _ images , or both ; we discuss this in more detail in [ s4.1 ] below . a few of the galaxies do have low surface brightness emission in the optical and infrared that may be indicative of the compact burst being only the highest surface brightness component of a larger stellar disk . the most obvious systems with these properties are kissr278 ( figure [ figcap6 ] ) , kissr561 ( figure [ figcap8 ] , but note that the compact and high surface brightness object in the northern part of the stellar disk is a foreground star ) , and kissr572 ( figure [ figcap9 ] ) . the images shown in figures [ figcap3 ] through [ figcap10 ] reveal that five of the eight neutral gas disks extend well beyond the beam size at this sensitivity level ; the h i disks of kissr278 , kissr396 , and kissr572 are the same size as the beam . there are low surface brightness extensions of h i gas in some systems : ic2271 ( figure [ figcap3 ] ) and kissr1021 ( figure [ figcap10 ] ) harbor h i disks that are roughly twice the size of the stellar body ; kissr245 ( figure [ figcap5 ] ) and kissr561 ( figure [ figcap8 ] ) show tenuous h i filaments extending toward the southwest . however , the low signal to noise ratio of most of these datasets ( see discussion in [ s2 ] ) precludes a robust kinematic study of seven of the eight systems in this work . the detailed dynamical analysis of adbs113845 + 2008 is already presented in @xcite , to which we refer the interested reader for details and results . while dynamical analyses are not possible with most of the data presented here , the derived surface densities are robustly measured ( at this modest sensitivity level ) . stated differently , an analysis of the ( projected ) surface densities in each system does not require a detailed knowledge of the gas dynamics . we thus limit our interpretation to those qualities and quantities of the neutral gas that do not depend on dynamical properties . we focus on quantifying the relative sizes of the neutral gas and stellar disks , as a measure of the frequency of the `` giant disk '' phenomenon ; most of [ s4 ] is devoted to addressing this science goal . one such well - studied metric is the empirical relationship between h i column density and ongoing star formation . while we do not have h@xmath1 imaging of all of these sources , each is a spectroscopically confirmed emission line source from kiss @xcite . we make the implicit assumption that the majority of the ongoing star formation in these systems is spatially coincident with the high surface brightness stellar components shown in the optical images in figures [ figcap3 ] through [ figcap10 ] . we note explicitly that the most extreme member of this sample , adbs113845 + 2008 , has an h@xmath1 morphology that is essentially indistinguishable from that of the r - band image shown in figure [ figcap4 ] ( see , e.g. , figure 5 of cannon etal . 2009 ) . a gas column density of 10@xmath23 @xmath10 has been shown to trace locations of active star formation ( e.g. , skillman 1987 ; kennicutt 1989 ; kennicutt 1998 ; and references therein ) . interestingly , in all of the galaxies , except ic 2271 and kissr561 , the peak observed h i column densities are below this 10@xmath23 @xmath10 level , even when the h i column density and stellar surface brightness maxima are spatially coincident . this is consistent with the results of other studies of dirr systems ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? it is important to note that beam smearing may be removing the signatures of high column density gas at our current resolutions of 15 and 20 , so we are cautious not to over - interpret this signature at the present time . higher resolution and sensitivity observations of these galaxies , while requiring extensive amounts of additional observing time , would be very insightful in revealing the small - scale structure of the neutral gas disks . the enigmatic properties of adbs113845 + 2008 led @xcite to argue for an evolutionary scenario where isolation allows for a very inefficient mode of star formation or , alternatively , very inefficient gas consumption . one might naively expect that systems with similar optical properties might also be similarly isolated . we were thus somewhat surprised to discover that the majority of the sample members ( five of eight systems ) are in relative physical proximity to other galaxies ; we hereafter refer to these systems as loose associations . h i and optical images of these loose associations are shown in figure [ figcap11 ] ( ic 2271 ) , figure [ figcap12 ] ( kissr245 ) , figure [ figcap13 ] ( kissr396 ) , figure [ figcap14 ] ( kissr561 ) , and figure [ figcap15 ] ( kissr572 ) . using the 10 kpc physical length denoted in panel ( b ) of each of these figures , one can discern the ( projected ) separation of the galaxies in each loose association . we draw attention to the ( projected ) separation of each association member from the compact dwarf system studied here in the caption of each figure and in table [ t5 ] ; the minimum and maximum separations are 26 kpc ( ic2271 and sdssj081808.31 + 243006.4 ) and 230 kpc ( kissr572 and sdss j144754.68 + 291929.4 ) . a commonly posited mechanism to initiate a burst of star formation ( i.e. , a starburst ) is a tidal interaction . dynamical instabilities can drive gas deep into the potential well , initiating spatially and temporally concentrated star formation ( e.g. , * ? ? ? * ; * ? ? ? while a direct signature of a tidally triggered star formation event has been difficult to produce observationally , one might suspect that extended neutral gas in the form of tidal tails might be considered to be an indirect signature of interaction - driven evolution ( e.g. , the extended tidal structure in the m81 group of galaxies ; * ? ? ? * ) . to search for such signs of interactions amongst the loose associations studied here , we examined the moment zero ( integrated h i intensity ) maps and full datacubes at a variety of spatial resolutions ( from 15 to 60 ) for signatures of tidal material between the compact dwarf systems and other association members . some systems have asymmetric , low surface brightness extensions of h i gas away from the central disk ( see discussion of ic2271 , kissr1021 , kissr245 , and kissr561 above ) that can be interpreted as weak evidence for gravitational interactions playing a role in their recent evolution . none of the loose associations shown in figures [ figcap11 ] through [ figcap15 ] show direct evidence of strong tidal interaction in the form of tails or diffuse gas ( at the n@xmath7 @xmath22 10@xmath14 @xmath10 level ) connecting one system to another . the most promising candidate for a strong interaction scenario might be the loose association containing ic2271 ; there is some low surface brightness h i gas surrounding multiple galaxies in this field . given the lack of very short _ uv _ spacings in the data presented here , the surface brightness sensitivity to large - scale structure in these images is only modest . we conclude that the membership of the majority of these star - forming systems in loose associations is intriguing but weak evidence for tidally triggered star formation events ; very deep observations of these systems with the compact _ vla _ d configuration could be very fruitful in further quantifying the role of interactions in initiating starburst events . the primary goal of the present investigation is to quantify the relative sizes of the neutral gas and stellar disks of each of the optically compact dwarf galaxies in our sample , thus estimating the frequency of the `` giant disk '' phenomenon exemplified by adbs113845 + 2008 and discussed in detail in @xcite . to make this comparison , we create radially averaged surface brightness profiles of each system in the optical , infrared , and h i spectral line . we measure the exponential scale lengths at optical and infrared wavelengths , and compare these to both the exponential scale length of the neutral gas disks ( when possible ) and to the sizes of the gaseous disks at the 10@xmath14 @xmath10 level . it is important to emphasize that while this comparison does not include dynamical parameters for each system ( e.g. , an inclination derived from dynamical modeling of an individual galaxy s velocity field ) , the comparisons discussed below do use the same corrections ( when applicable ) for galaxy inclination at each wavelength . similarly , our analysis uses matched apertures between wavelengths to assure self - similar measurements . we define a galaxy as being `` resolved '' in a certain image if the derived scale length or size exceeds the fwhm or beam size . therefore , we classify systems as `` resolved '' in the _ sdss _ r - band , the _ 3.6 @xmath20 m band , or the _ spitzer _ 4.5 @xmath20 m band , if the derived scale lengths exceed 1.3 , 1.66 , or 1.72 , respectively . similarly , we classify systems as `` resolved '' in h i if the size of the h i disk at the 10@xmath14 @xmath10 column density level exceeds the 15 beam size . we discuss the details of each of these metrics in the subsections that follow . optical and infrared surface brightness profiles for each of the galaxies were produced by fitting elliptical isophotes to the _ sdss _ r - band , _ spitzer _ 3.6 @xmath20 m , and _ spitzer _ 4.5 @xmath20 m images using the iraf task ellipse . as noted above , some of these systems have easily identifiable foreground stars that can contaminate the surface brightness profiles [ e.g. , adbs113845 + 2008 ( figure [ figcap4 ] ) and kissr561 ( figure [ figcap8 ] ) ] . such foreground objects were masked by hand using the iraf task fixpix . for each system , the position angle and inclination were determined from the _ sdss _ r - band image . when a system was extended in one dimension , the position angle was defined as that angle east of north that aligns with the larger projected optical axis . the inclination was defined by @xmath24 where _ a _ and _ b _ are the major and minor axes , respectively . the ellipse parameters used to derive the isophotal photometry are given in table [ t6 ] . surface photometry was measured in elliptical isophotes incremented in one - pixel steps along the major axis ( 0.4 per ellipse for the _ sdss _ images and 0.6 per ellipse for the _ spitzer _ images ) . the r - band ellipses are shown in red in figures [ figcap3 ] though [ figcap10 ] , while the infrared ellipses are shown in green in figures [ figcap5 ] though [ figcap10 ] . figures [ figcap16 ] and [ figcap17 ] show the resulting surface brightness plots for all eight of the sample galaxies . as expected based on the selection criteria used to create the sample ( two of which were an optical radius less than 3 and an upper limit to the recessional velocity or distance ; see discussion in [ s2 ] above ) , the profiles are all centrally peaked and show little structure throughout the optical disks of the galaxies . while there is some variation in the steepness of the profile in the inner regions of the systems ( i.e. , within the inner few arcseconds ) , it is immediately obvious that each system harbors an exceptionally compact stellar component . to quantify the sizes of these compact stellar populations , we next derived the scale lengths of each system in the _ sdss _ r - band and in the _ spitzer _ 3.6 @xmath20 m and 4.5 @xmath20 m bands . in deriving these profiles , we follow the methodology in @xcite . using logarithmic units of mag arcsec@xmath25 , a very good approximation for the exponential disk component of the optical radial profile can be made using : @xmath26 where @xmath27 is the central surface brightness in units of mag arcsec@xmath25 , _ r _ is the radius , @xmath28 is the surface brightness in units of mag arcsec@xmath25 as a function of radius , and _ h _ is the scale length . table [ t7 ] provides the calculated scale lengths ( in both angular and in physical units ) at each wavelength , derived from the profiles shown in figures [ figcap16 ] and [ figcap17 ] , with upper limit errors of 25% at each ellipse location . note that we use the nomenclature of _ h_@xmath29 , _ h_@xmath30 , and _ h_@xmath31 to denote the scale lengths ( in angular units and in physical units of kpc ) of each galaxy in the _ sdss _ r - band , the _ 3.6 @xmath20 m band , and the _ spitzer _ 4.5 @xmath20 m band , respectively . the scale lengths presented in table [ t7 ] show that all of the galaxies are resolved in at least one optical or infrared image , and that most are resolved in both . recall that the fwhm of the _ sdss _ r - band , _ spitzer _ 3.6 @xmath20 m , and _ 4.5 @xmath20 m images are 1.3 , 1.66 , and 1.72 , respectively . values in table [ t7 ] show that all systems except for kissr245 are resolved in the _ sdss _ images ( that is , _ h_@xmath29 exceeds the 1.3 fwhm of the _ sdss _ r - band images ) . h_@xmath29 value for kissr245 is formally smaller than the fwhm , although it is consistent within the errorbar . similarly , _ spitzer _ resolves each galaxy ; note , however , that the scale lengths of kissr245 at 3.6 @xmath20 m and of kissr1021 at 4.5 @xmath20 m are consistent with the fwhm , given the errorbars . the optical and infrared scale lengths in table [ t7 ] quantify the sizes of the compact stellar components of the sample galaxies . at the adopted distances , the stellar component of each system has an r - band scale length much less than 1 kpc . all systems have _ @xmath320.5 kpc , except for kissr561 . moving to the infrared , the scale lengths increase . this is exactly the behavior expected of an older and non - star forming stellar population . while each of the infrared scale lengths exceeds the r - band value , it is remarkable to note that each system is still exceptionally compact even when considering the older stellar populations . for example , at 4.5 @xmath20 m , some systems have scale lengths slightly larger than 1 kpc ( kissr278 , kissr561 , kissr572 ) , but others show very little increase in scale length with increasing wavelength ( consider kissr1021 , whose scale length does not change between the optical and the infrared , within the measurement errors ) . having quantitatively established the compact nature of the stellar population of each of the sample galaxies , we now seek to quantify the sizes of the h i neutral gas disks via two methods . first , we derive radially averaged surface brightness profiles for each system ; when possible , we use these profiles to derive an exponential scale length . second , we quantify the size of the disks at the 10@xmath14 @xmath10 levels . we discuss each method in detail below . we begin by creating radial h i surface brightness profiles for each of the galaxies using the 15 beam images and the gipsy task ellint . as an inspection of figures [ figcap3 ] through [ figcap10 ] shows , the neutral gas disks studied here are larger than the beam size for five of the eight galaxies ( excluding kissr278 , kissr396 , and kissr572 ) . a detailed dynamical analysis is only possible for adbs113845 + 2008 ; this precludes the derivation of kinematic position angles and inclinations for most galaxies . in order to assure an accurate comparison of stellar and gaseous properties , we thus use the same position angles and inclinations as derived using the _ sdss _ r - band images ( see table [ t6 ] and the detailed discussion in [ s4.1 ] above ) . these ellipses are shown in blue in figures [ figcap3 ] through [ figcap10 ] . a semi - major axis and step size of 5 was used to create each of the h i profiles ; technically this means that the profiles we discuss below are oversampled ( compared to the 15 beam size ) by a factor of 3 . however , this oversampling does allow us to extract meaningful radially averaged surface brightness profiles . figures 18 and 19 show these surface brightness profiles for each of the eight sample galaxies . as was evident from the surface density images presented in [ s3.1 ] , most of the sample galaxies have h i surface brightness profiles that are centrally peaked at the resolution of these observations and that fall off with radius in a more or less smooth way . note that low surface brightness features in some of the galaxies moment zero ( integrated h i intensity ) maps do cause departures from smooth profiles in the outer disks ; for example , the low surface brightness feature extending to the southwest of kissr245 in figure [ figcap5 ] is easily identified as the slight increase in h i surface density between @xmath1325 and @xmath1350 in figure [ figcap18 ] . the obvious exception to this trend is adbs113845 + 2008 , whose radially averaged h i surface density profile ( created with the optically derived inclination and position angle discussed above , and shown in red in figure [ figcap18 ] ) rises with increasing galactocentric radius , peaks , and then falls off in the outer disk . the peak surface densities are located in the broken ring structure @xmath137.5 kpc from the stellar component . a cursory comparison of the profiles in figures [ figcap18 ] and [ figcap19 ] already reveals the striking differences between adbs113845 + 2008 and the other compact dwarf galaxies in the present sample ; we discuss these differences in more detail below , and refer the interested reader to @xcite for a more complete discussion of the remarkable characteristics of the adbs113845 + 2008 system . the radial profiles shown in figures [ figcap18 ] and [ figcap19 ] ( specifically , the red profile for adbs 113845 + 2008 ) were created using exactly the same inclinations and position angles as used to derive the optical and infrared surface brightness profiles in figures [ figcap16 ] and [ figcap17 ] . using these neutral gas radial profiles , exponential fits were made using a version of equation [ eq1 ] converted to linear units : @xmath33 where @xmath34 is the central surface brightness of the disk in m@xmath12 pc@xmath25 , _ r _ is the radius , @xmath35 is the surface brightness of the disk in m@xmath12 pc@xmath25 as a function of radius , and _ h _ is the scale length @xcite . we hereafter refer to the h i scale length derived using this equation as _ h_@xmath7 . we tabulate the derived _ values in table [ t7 ] . for all of the systems except adbs113845 + 2008 , the calculated values of _ h_@xmath7 are smaller than the beam size . formally , this metric to measure the size of the h i disks only provides meaningful results when it can be fit over multiple independent beams . as noted above , our radially averaged profiles required over - sampling by a factor of three . while the sizes of the h i disks based on the _ h_@xmath7 metric are interesting , and we discuss these briefly below , we seek an alternative measure of the size of the neutral gas disks of these galaxies . it is obvious from the neutral gas morphology ( figure [ figcap3 ] ) and surface brightness profile ( figure [ figcap18 ] ) of adbs113845 + 2008 that an exponential scale length is undefined for a disk whose morphology is not centrally concentrated . while we considered fitting the outer disk of adbs113845 + 2008 ( i.e. , that region at radii larger than that of the broken h i ring ) with an exponential function , the striking disparity between the neutral gas and stellar disks of this system warrants caution against such an approach . due to the remarkable qualities of this galaxy s neutral gas disk , we thus do not calculate an exponential h i scale length . the properties of adbs113845 + 2008 naturally require a second metric for the size of the neutral gas disks of these systems . to address these issues , we thus use the radially averaged h i surface density profiles in figures [ figcap18 ] and [ figcap19 ] to determine the ( radially averaged ) size of the h i disk at the 10@xmath14 @xmath10 column density level . this column density corresponds to a mass surface density of 0.08 m@xmath12 pc@xmath25 ; recall from the discussion in [ s2.2 ] that this emission is above the 2@xmath18 level in at least 3 consecutive channels . we extract the radii of the sample galaxies at this level ( hereafter referred to as r@xmath7 ) from figures [ figcap18 ] and [ figcap19 ] and tabulate the values in table [ t7 ] ; note that @xcite quantified the physical size of the adbs113845 + 2008 disk in the same manner . as expected , r@xmath7 h_@xmath7 in all systems where both quantities are defined . it is important to stress that for kissr245 , kissr278 , kissr396 , and kissr572 , r@xmath7 is equal ( within errors ) to the beam size ; this means that the r@xmath7 metric for these galaxies is formally an upper limit ( but note that kissr245 does have h i gas extending to the southwest , well beyond a single beam size ) . these four systems are `` unresolved '' by the present observations ( that is , r@xmath7 @xmath5 beam size , thus the r@xmath7 and related values being listed as upper limits in tables [ t7 ] and [ t8 ] ) ; they may harbor smaller neutral gas components than calculated here . we discuss the ratios of both _ h_@xmath7/_h_@xmath29 and r@xmath7/_h_@xmath29 for all eight sample galaxies in [ s4.3 ] . we again note that no kinematic constraints are used to derive the scale lengths or disk sizes discussed here ; the profiles shown in figures [ figcap16 ] through [ figcap19 ] use only optically - derived position angles and inclinations . since seven of the eight galaxies are not sufficiently resolved to undertake a robust dynamical analysis , we can only test these dependences for the one system that does have a full kinematic model , adbs113845 + 2008 . to test how using the position angles and inclinations derived from the optical images affect our calculations ( as opposed to using those derived from the h i dynamics ) , we show three radial mass surface density profiles for adbs113845 + 2008 in figure [ figcap18 ] . the red profile is derived using the optical inclination ( 0@xmath37 ) and position angle ( 0@xmath37 ) and is discussed above . the blue profile uses the dynamically - derived position angle ( 155@xmath37 ) and inclination ( 30@xmath37 ) from @xcite . the green profile uses the same position angle ( 155@xmath37 ) as in @xcite , but assumes a larger value for the galaxy s inclination angle ( 60@xmath37 ) . since _ h_@xmath7 is undefined for adbs113845 + 2008 , it is difficult to use this test to assess the dependence of h i scale length on inclination or position angle . the r@xmath7 values , on the other hand , do change slightly for these different fits ( r@xmath7 @xmath2 17.3 @xmath19 3.5 kpc , 18.1 @xmath19 3.6 kpc , and 22.3 @xmath19 4.5 kpc , for the red , blue , and green profiles in figure [ figcap18 ] , respectively ) , but the resulting h i disk radii are equal within the ( substantial ) errorbars . while differences from one system to another are expected , this simple test gives a baseline for how to interpret the measurements made using the position angles and inclinations derived from the _ sdss _ r - band images within the rest of the sample . we consider the r@xmath7 values for ic2271 , adbs113845 + 2008 , kissr561 , and kissr1021 to be accurate representations of their ( average ) physical size at the 10@xmath14 @xmath10 level . the r@xmath7 values in table [ t7 ] for the other four systems should be considered upper limits on the physical size of each neutral gas disk . when available , we also tested how the disk size measurements change when using c configuration versus d configuration data . this comparison can only be made for adbs113845 + 2008 and ic2271 . the scale length measurement of ic2271 with the lower resolution d configuration data is _ kpc ( [email protected] ) ; this can be compared with the _ h_@[email protected] kpc ( [email protected] ) value derived from the c configuration data only ( see table [ t7 ] ) . the radii of the galaxy disks , when using the c versus d configuration datasets , change from r@[email protected] kpc to [email protected] kpc ( ic2271 ) and from [email protected] kpc to [email protected] kpc ( adbs113845 + 2008 ) . while these increases are slightly larger than the sizes of the errorbars , we conclude that there is only a modest dependency of the size estimates of these galaxy disks on the use of only the c configuration data . given that our analysis uses the same imaging strategy for all systems , the surface brightness sensitivity of each dataset is very comparable . we can now compare , in a quantitative sense , the relative sizes of the stellar and gaseous disks of the eight galaxies in this sample . table [ t8 ] contains the ( dimensionless ) ratios of _ h_@xmath7 and r@xmath7 versus _ h_@xmath29 , _ h_@xmath30 , and _ h_@xmath31 , for all galaxies with available data ( recall that adbs113845 + 2008 has an undefined _ h_@xmath7 , and that no _ spitzer _ imaging is available for adbs113845 + 2008 or for ic2271 ) . from the contents of this table , we note the following trends and behaviors . as discussed in [ s4.2 ] above , four of the eight systems have h i disks that exceed the sizes of the stellar bodies based on the r@xmath7 metric ( and are thus considered `` resolved '' in h i ) . hence , the dimensionless ratios in table [ t8 ] for kissr245 , kissr278 , kissr396 , and kissr572 should be considered upper limits ; the actual size ratios of neutral to stellar disks may in fact be smaller for these systems . for the entire sample , the ratios of _ h_@xmath7 to r - band or infrared scale lengths ( spanning a range of @xmath132 to @xmath137 ) are systematically smaller than the ratios of r@xmath7 to the same r - band or infrared values ( which span a range of @xmath134 to @xmath1358 ) . similarly , for each individual galaxy , the relative size ratios measured using either r@xmath7 or _ h_@xmath7 decrease with increasing wavelength used to measure the stellar disk size . these general trends hold across the entire sample , regardless of which tracer is used to measure the size of either the stellar or the gaseous disk . the r@xmath7/_h_@xmath29 ratio for adbs113845 + 2008 ( 58@xmath1921 ) is more than a factor of two larger than the same ratio for any of the other systems ; if the ratios in table [ t7 ] for the other sample members are smaller , then this reinforces the enormity of the h i disk of adbs113845 + 2008 . using the other seven sample galaxies as benchmarks , it is physically intuitive that the infrared scale length of adbs113845 + 2008 would be larger than the measured _ h_@xmath29 value . however , the enormous r@xmath7/_h_@xmath29 ratio of this system , and the relative size of the broken h i ring ( @xmath137.5 kpc in radius ) versus the optical scale length ( [email protected] kpc ) , are challenging parameters to explain unless the optical body is only a high surface brightness knot in an otherwise extremely low surface brightness underlying stellar disk . we again refer the interested reader to @xcite for more detailed discussion ; for the purposes of this work , we interpret the remarkable properties of adbs113845 + 2008 as the signature of a `` giant '' h i disk . the present sample of compact dwarf galaxies was created using consistent selection criteria ; the observations and analysis have produced a uniform set of results . while the sample is small , it appears that the frequency of `` giant disk '' dwarf galaxies such as adbs113845 + 2008 is low . a few other examples of such systems have been identified ( e.g. , ddo154 , carignan & beaulieu 1989 , carignan & purton 1998 ; ngc2915 , meurer etal . 1996 ; ugc5288 , van zee 2004 ; ngc3741 , bremnes etal . 2000 , begum etal . this small collection of galaxies suggests that systems with qualities similar to adbs113845 + 2008 are comparatively rare in the local volume . a larger sample of h i observations of local compact dwarf systems could be fruitful in improving the statistics of the trends identified in table [ t8 ] . we have presented _ vla _ c configuration imaging of eight optically compact dwarf galaxies . these sources were selected to share the following common characteristics : 1 ) compact stellar populations ; 2 ) faint blue absolute magnitudes ( i.e. , low masses ) ; 3 ) emission lines indicative of recent star formation ; and 4 ) proximity for spatially resolved work . the sample includes the `` giant disk '' dwarf galaxy adbs113845 + 2008 studied in @xcite . the _ vla _ data are supplemented by archival optical ( _ sdss _ r - band ) and infrared ( _ spitzer _ 3.6 @xmath20 m and 4.5 @xmath20 m ) imaging . we use this data suite to quantitatively address the relative sizes of the stellar and gaseous components of these galaxies . a comparison of the morphology of the stellar and gaseous disks of each galaxy in the sample reveals that all eight systems have extremely compact ( though in almost all cases resolved ) stellar components ; this is of course expected given the selection criteria . seven of the eight systems in the sample harbor centrally concentrated neutral gas distributions . adbs113845 + 2008 differs from the other systems in that its neutral gas disk is structured in a broken ring that is well outside of the stellar component . in most of the other cases , the stellar and gaseous disks have surface brightness maxima that are in close physical proximity to one another . five of the eight systems are located in loose associations , with neighboring systems with projected separations of less than a few hundred kpc and with recessional velocities within a few hundred kms@xmath3 . some of the loose association galaxies have low surface brightness neutral gas in the outer disks that might be construed as evidence of tidal interactions . however , given the lack of very short _ uv _ spacings in our data , we conservatively conclude that the membership of these star - forming dwarfs in such loose associations is only weak evidence for a tidal burst trigger . while each of the eight sample galaxies is detected in h i at meaningful significance , these _ vla _ data do not allow a robust dynamical analysis in any of the systems except for adbs113845 + 2008 . thus , we concentrate on the morphological properties of the galaxies . we use the optical , infrared , and h i data to derive radially averaged surface brightness profiles that can be used to measure the sizes of both the stellar and the gaseous components . we quantify the sizes of the h i disks via both an exponential scale length ( available for all systems except for adbs113845 + 2008 ) and a ( radially averaged ) limit of the gaseous disk at the 10@xmath14 @xmath10 column density ( 0.08 m@xmath12 pc@xmath25 ) level . we find that four of the eight systems ( specifically , ic2271 , adbs113845 + 2008 , kissr561 , and kissr1021 ) have larger gaseous disks than stellar disks ; the remaining four ( kissr245 , kissr278 , kissr396 , and kissr572 ) have only upper limits on the physical size of their h i disks ( but note that kissr245 has low - surface brightness h i gas extending well beyond a single beam size ; see figure [ figcap5 ] ) . when parameterized by an h i disk scale length , the neutral gas disks of the seven available galaxies ( excluding adbs113845 + 2008 ) are larger than the stellar disks by factors of @xmath1327 . when parameterized by the size of the h i disk at the 10@xmath14 @xmath10 column density level , the neutral gas disks of the eight sample galaxies ( including adbs113845 + 2008 ) are larger than the stellar disks by factors of @xmath13458 . for each galaxy , the relative size ratios measured using either r@xmath7 or _ h_@xmath7 decrease with increasing wavelength used to measure the stellar disk size ( i.e. , the optical scale lengths are slightly shorter than the infrared scale lengths ) . if the values of r@xmath7 and _ h_@xmath7 are upper limits , then these ratios decrease . adbs113845 + 2008 stands out among the members of this sample as harboring a `` giant '' h i disk . while the h i - to - optical size ratios of some other members of the sample are substantial ( e.g. , ic2271 , kissr245 ; see table [ t8 ] ) , this ratio for adbs113845 + 2008 is more than twice as large as any other measured here . as discussed in @xcite , the physical properties of this galaxy are extreme and enigmatic ; adbs113845 + 2008 has one of the largest known h i - to - optical size ratios in the local universe . given that the present sample was selected _ only _ based on optically - derived properties , and that the observations of the present sample were acquired and analyzed in a uniform way , suggests that this `` giant disk '' dwarf galaxy is rare in the local volume . the present sample is small and is limited in number by available ( single dish h i ) data ; the statistics behind the present results could be significantly strengthened by a larger study of additional extremely compact , star - forming dwarf galaxies in the local volume . this research was supported by a cottrell college science award from research corporation . jlr acknowledges support from nsf ast-000167932 . the authors thank the national radio astronomy observatory and macalester college for partial support of this work , and dr . gustaaf van moorsel for helpful discussions . this work is based in part on archival data obtained with the _ spitzer space telescope _ , which is operated by the jet propulsion laboratory , california institute of technology under a contract with nasa . support for this work was provided by an award issued by jpl / caltech . this work has made use of data from the _ sloan digital sky survey_. funding for the _ sdss _ and _ sdss - ii _ has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the _ sdss _ web site is http://www.sdss.org/. the _ sdss _ is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , university of cambridge , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . 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 , and nasa s astrophysics data system . lcccccc ic 2271 & 08:18:19.7 & 24:31:37 & 2184 @xmath19 10 & 29.9 & 25.1 @xmath19 6.3 & 101 + adbs 113845 + 2008 & 11:38:45.2 & 20:08:24 & 3054 @xmath19 3 & 41.8 & 40.2 @xmath19 10.1 & 84.6 + kissr245 & 13:16:27.9 & 29:25:11 & 4762 @xmath19 5 & 65.2 & 13.0 @xmath19 3.3 & 28.4 + kissr278 & 13:23:37.7 & 29:17:17 & 4022 @xmath19 5 & 55.1 & 4.4 @xmath19 1.1 & 12.0 + kissr396 & 13:57:10.0 & 29:13:10 & 2238 @xmath19 5 & 30.7 & 1.0 @xmath19 0.3 & 11.6 + kissr561 & 14:29:53.6 & 29:20:11 & 3750 @xmath19 5 & 51.4 & 34.3 @xmath19 8.6 & 56.5 + kissr572 & 14:46:48.2 & 29:25:18 & 3693 @xmath19 5 & 50.6 & 2.3 @xmath19 0.6 & 16.6 + kissr1021 & 16:19:02.5 & 29:10:22 & 2519 @xmath19 5 & 34.5 & 7.0 @xmath19 1.8 & 14.5 + [ t1 ] lccccc ic2271 & 2006 oct . 31 & 241 & 3.12 & 64 & 10.30 + adbs113845 + 2008 & 2006 oct . 31 & 245 & 1.56 & 128 & 2.58 + kissr245 & 2008 apr . 23 & 252 & 3.12 & 128 & 5.15 + kissr278 & 2008 apr . 29 & 253 & 3.12 & 128 & 5.15 + kissr396 & 2008 may 1 & 254 & 3.12 & 128 & 5.15 + kissr561 & 2008 may 13 & 252 & 3.12 & 128 & 5.15 + kissr572 & 2008 may 8 & 251 & 3.12 & 128 & 5.15 + kissr1021 & 2008 may 15 & 255 & 3.12 & 128 & 5.15 + [ t2 ] lcccc ic2271 & 2121 - 2246 & 0.39 & 1.22 @xmath19 0.12 & 4.702 + adbs113845 + 2008 & 2987 - 3117 & 0.73 & 0.98 @xmath19 0.10 & 1.982 + kissr245 & 4744 - 4779 & 0.52 & 0.13 @xmath19 0.01 & 0.28 + kissr278 & 4014 - 4030 & 0.48 & 0.062 @xmath19 0.006 & 0.17 + kissr396 & 2221 - 2253 & 0.53 & 0.043 @xmath19 0.004 & 0.51 + kissr561 & 3695 - 3809 & 0.50 & 0.55 @xmath19 0.06 & 0.89 + kissr572 & 3688 - 3699 & 0.48 & 0.038 @xmath19 0.004 & 0.27 + kissr1021 & 2473 - 2562 & 0.46 & 0.25 @xmath19 0.03 & 0.51 + [ t3 ] lccc ic2271 & 1.2 & 1.5 & 1.5 + adbs113845 + 2008 & 0.98 & 0.90 & 0.90 + kissr245 & 0.13 & 0.14 & 0.15 + kissr278 & 0.062 & 0.30 & - + kissr396 & 0.043 & 0.038 & 0.039 + kissr561 & 0.55 & 0.65 & 0.66 + kissr572 & 0.038 & 0.028 & - + kissr1021 & 0.25 & 0.35 & 0.23 + [ t4 ] lcccc ic2271 & adbs j081707 + 2433 & 995 & 121 & 144 + & ic 2267 & 790 & 146 & 115 + & sdss j081808.31 + 243006.4 & 180 & 105 & 26 + kissr245 & maps - ngp_o_323_0594797 & 477 & 12 & 151 + kissr396 & ngc 5375 & 270 & 137 & 40 + & ngc 5375a & 633 & 18 & 94 + kissr561 & ngc 5657 & 861 & 110 & 215 + kissr572 & sdss j144754.68 + 291929.4 & 936 & 33 & 230 + [ t5 ] lcc ic2271 & 143@xmath37 & 45@xmath37 + adbs113845 + 2008 & 0@xmath37 & 0@xmath37 + kissr245 & 0@xmath37 & 0@xmath37 + kissr278 & 40@xmath37 & 73@xmath37 + kissr396 & 175@xmath37 & 43@xmath37 + kissr561 & 0@xmath37 & 57@xmath37 + kissr572 & 60@xmath37 & 70@xmath37 + kissr1021 & 130@xmath37 & 48@xmath37 + [ t6 ] lccccc ic2271 & 0.33 @xmath19 0.08 & n / a & n / a & 1.72 @xmath19 0.43 & 8.74 @xmath19 2.19 + adbs113845 + 2008 & 0.30 @xmath19 0.08 & n / a & n / a & - & 17.3 @xmath19 4.3 + kissr245 & 0.38 @xmath19 0.10 & 0.55 @xmath19 0.14 & 0.70 @xmath19 0.18 & 2.48 @xmath19 0.62 & @xmath168.28 @xmath19 2.07 + kissr278 & 0.41 @xmath19 0.10 & 0.80 @xmath19 0.20 & 1.36 @xmath19 0.34 & 2.64 @xmath19 0.66 & @xmath166.19 @xmath19 1.55 + kissr396 & 0.19 @xmath19 0.05 & 0.31 @xmath19 0.08 & 0.36 @xmath19 0.09 & 0.78 @xmath19 0.20 & @xmath162.65 @xmath19 0.66 + kissr561 & 0.72 @xmath19 0.18 & 1.30 @xmath19 0.33 & 1.20 @xmath19 0.30 & 2.59 @xmath19 0.65 & 11.8 @xmath19 3.0 + kissr572 & 0.50 @xmath19 0.13 & 0.77 @xmath19 0.19 & 1.18 @xmath19 0.30 & 2.19 @xmath19 0.55 & @xmath164.64 @xmath19 1.16 + kissr1021 & 0.29 @xmath19 0.07 & 0.37 @xmath19 0.09 & 0.31 @xmath19 0.08 & 1.18 @xmath19 0.30 & 5.39 @xmath19 1.35 + & & & & & + & & & & & + ic2271 & 2.29 @xmath19 0.57 & n / a & n / a & 11.9 @xmath19 3.0 & 60.3 @xmath19 15.1 + adbs113845 + 2008 & 1.44 @xmath19 0.36 & n / a & n / a & - & 85.3 @xmath19 21.3 + kissr245 & 1.18 @xmath19 0.30 & 1.71 @xmath19 0.43 & 2.21 @xmath19 0.55 & 7.84 @xmath19 1.96 & @xmath1626.2 @xmath19 6.6 + kissr278 & 1.51 @xmath19 0.38 & 2.99 @xmath19 0.75 & 5.12 @xmath19 1.28 & 9.88 @xmath19 2.47 & @xmath1623.2 @xmath19 5.8 + kissr396 & 1.33 @xmath19 0.33 & 2.13 @xmath19 0.53 & 2.50 @xmath19 0.63 & 5.25 @xmath19 1.31 & @xmath1617.8 @xmath19 4.5 + kissr561 & 2.93 @xmath19 0.73 & 5.27 @xmath19 1.32 & 4.86 @xmath19 1.22 & 10.4 @xmath19 2.6 & 47.4 @xmath19 11.9 + kissr572 & 2.07 @xmath19 0.52 & 3.18 @xmath19 0.80 & 4.87 @xmath19 1.22 & 8.93 @xmath19 2.23 & @xmath1618.9 @xmath19 4.7 + kissr1021 & 1.74 @xmath19 0.44 & 2.23 @xmath19 0.56 & 1.86 @xmath19 0.47 & 7.05 @xmath19 1.76 & 32.2 @xmath19 8.1 + [ t7 ] lcccccc ic2271 & 5.2 @xmath19 1.8 & 26 @xmath19 9 & n / a & n / a & n / a & n / a + adbs113845 + 2008 & - & 58 @xmath19 21 & n / a & n / a & n / a & n / a + kissr245 & 6.5 @xmath19 2.4 & @xmath1622 @xmath19 8 & 4.5 @xmath19 1.6 & @xmath1615 @xmath19 5 & 3.5 @xmath19 1.3 & @xmath1612 @xmath19 4.2 + kissr278 & 6.4 @xmath19 2.2 & @xmath1615 @xmath19 5 & 3.3 @xmath19 1.2 & @xmath167.7 @xmath19 2.7 & 1.9 @xmath19 0.7 & @xmath164.6 @xmath19 1.6 + kissr396 & 4.1 @xmath19 1.5 & @xmath1614 @xmath19 5 & 2.5 @xmath19 0.9 & @xmath168.5 @xmath19 3.0 & 2.2 @xmath19 0.8 & @xmath167.4 @xmath19 2.6 + kissr561 & 3.6 @xmath19 1.3 & 16 @xmath19 6 & 2.0 @xmath19 0.7 & 9.1 @xmath19 3.2 & 2.2 @xmath19 0.8 & 9.8 @xmath19 3.5 + kissr572 & 4.4 @xmath19 1.6 & @xmath169.3 @xmath19 3.3 & 2.8 @xmath19 1.0 & @xmath166.0 @xmath19 2.1 & 1.9 @xmath19 0.7 & @xmath163.9 @xmath19 1.4 + kissr1021 & 4.1 @xmath19 1.4 & 19 @xmath19 6 & 3.2 @xmath19 1.1 & 14.6 @xmath19 5.2 & 3.8 @xmath19 1.4 & 17 @xmath19 6.2 + [ t8 ]

this preprint contains the english translation of the paper `` relaxation dynamics of a quantum chain of harmonic oscillators '' , which was published by the author in 1980 in the ukrainian physical journal . comments describing its motivations , background ideas and results are presented as well . this paper addressed to the problem of approach to the thermal equilibrium in an isolated macroscopic quantum system , which was studied on the example of the quantum chain of weakly interacting harmonic oscillators . in the initial state , macroscopic energy was supplied to one oscillator ( atom ) in the chain . subsequent evolution of the quantum state of the whole chain was determined due to the model integrability . the main subject of interest was the time evolution of the reduced density operators characterizing the quantum state of a particular atom . on the short time - scale , the energy perturbation expands along the chain with the velocity of the fastest phonon mode . on the long - time scale , the single - atom density operators display strong fluctuations around the canonical gibbs distribution . these fluctuations are caused by the degeneracy of the energy level differences ( presence of resonances ) in the model of coupled harmonic oscillators . after lifting this degeneracy , fluctuations are suppressed providing , that the reduced density operator of each atom in the chain becomes very close to the gibbs distribution at almost any time moment .

1201.0578

the claim that an isolated ( closed ) macroscopic system evolves to the thermal equilibrium state lies in the very basis of the statistical mechanics . the question about the precise meaning of this statement for classical and quantum systems arises for every student studying the university course of statistical mechanics . with me this happens in kharkov university in 1978 . two years later , i published two papers @xcite in russian - language journals , in which relaxation of the isolated macroscopic quantum system to the thermal equilibrium was studied on the example of the chain of harmonic oscillators . it turns out , that the set of problems considered in these papers and underlying ideas became the subject of considerable theoretical interest in subsequent years , to much extent , due to experiments on quantum dynamics in the ultracold atomic gases @xcite . this relates , in particular , to the quantum quench problem @xcite , and to the equilibration scenario in quantum systems based on the _ eigenstate thermalization hypothesis _ ( eth ) @xcite and on the idea of canonical typicality @xcite . while the second paper @xcite was later translated into english , the first one @xcite remained without translation . the aim of the present retrospective publication is to present the english translation of paper @xcite , complemented with comments relating to its motivations and results . in the both papers @xcite , the isolated linear chain of @xmath0 weakly coupled quantum harmonic oscillators ( the atoms ) was considered . the model hamiltonian reads as @xmath1 , \quad \omega\ll\omega_0.\ ] ] here index @xmath2 @xmath3 enumerates the atoms in the chain , @xmath4 and @xmath5 denote the special coordinate and momentum of the @xmath6-th atom , periodical boundary conditions are implied . at @xmath7 , the chain was taken in the inhomogeneous pure state @xmath8 , in which the macroscopic energy @xmath9 , with @xmath10 was supplied to one atom with @xmath11 . at @xmath12 , the chain evolved into the state @xmath13 , which was determined due to exact integrability of the model ( [ hc ] ) . in the first paper @xcite , the time evolution of the reduced density operator @xmath14 of the atom located at the @xmath6-th site of the chain was calculated and studied , @xmath15 where @xmath16 , @xmath17 is the projector - type density operator of the whole chain , and @xmath18 denotes the trace over the space corresponding to the rest @xmath19 atoms . on the short - time scale , the slow spreading of the energy from the initially excited atom along the chain with the velocity @xmath20 was found . on the other hand , the time average of the reduced density operators @xmath14 of each atom in the chain over the infinite time interval was found to approach to the gibbs distribution in the thermodynamic limit @xmath21 . it turns out , however , that the time - fluctuations of the single - atom reduced density operators @xmath14 remain considerable on the long - time scale . such atypical behavior was associated with the strong degeneracy of the energy level differences @xmath22 in model ( [ hc ] ) . it was shown in @xcite , that lifting this degeneracy by arbitrary weak nonlinear interaction provides suppression of these fluctuations and guaranties approach of the single - atom density operators @xmath14 to the gibbs distribution at large @xmath23 . though this result was proved only for the case @xmath24 , it is very likely , that it holds also for @xmath25 . it was shown in @xcite as well , that for a huge majority of eigenstates @xmath26 of the total hamiltonian ( [ hc ] ) , the reduced density operators @xmath27 corresponding to each individual @xmath6-th atom in the chain are equal to the gibbs distribution in the thermodynamic limit . in other words , a typical exact pure stationary state @xmath26 of the chain hamiltonian would be perceived as the thermal equilibrium state by the observer , which can see just one atom in the chain . this result agrees with the eth , which was put forward later by deutsch @xcite and srednicki @xcite . in paper @xcite the results of @xcite were partly extended to the subsystems consisting of @xmath28 neighbouring atoms in the chain , @xmath29 . starting from the same initial pure state @xmath8 , the time evolution of the reduced density operator @xmath30 of the subsystem containing @xmath28 atoms was calculated . qualitative behavior of @xmath30 at large @xmath23 was shown to be similar to that of the single - atom reduced density operator @xmath14 described in @xcite . it should be noted , however , that calculation of the time - average in @xcite was not completely rigorous , but applied a certain reasonable conjecture , see equation ( 10 ) in @xcite . the following note is in order here . the quantum chain of harmonic oscillators determined by hamiltonian ( [ hc ] ) has the complete set of @xmath3 commuting integrals of motion @xmath31 which characterize the number of phonons with given quasimomenta @xmath32 , @xmath33 . integrability of model ( [ hc ] ) is the direct consequence of existence of these integrals of motion . the another side of the coin is that conservation of @xmath3 integrals of motion makes impossible the complete thermalization of a typical initial state in this model . however : * an observer , which can see _ just one atom in the chain _ , can not distinguish phonon modes with different quasimomenta @xmath34 . by this reason , conservation of the phonon numbers @xmath31 in the whole quantum chain does not conflict with the chain _ partial thermalization _ , which is registered by the observer watching at a single atom . * anyway , complete thermalization in model ( [ hc ] ) could be still possible for some particular initial states of the chain . the rest of this preprint is organized as follows . section [ sec1 ] describes in detail the background ideas on thermalization in quantum systems , which stimulated me to do the model calculations described in @xcite . in a very brief form , these ideas were present already in the introductory parts of these two papers . section [ sec2 ] contains comments on the results and some calculation details of paper @xcite , which english translation is given in the appendix . section [ s3 ] contains new results illustrating the effect of the nonlinear interaction on the thermalization of the atom energies in the chain . the statistical mechanics , besides the dynamical laws , is based on certain additional probability hypothesis ( the micro - canonical distribution postulate , the correlation weakness principle , etc . ) which are justified finally by the good agreement of the theory with experiment . one can guess , that such additional hypothesis are in fact the consequences of the dynamical laws . this idea was realized in the boltzmann s approach , for which ergodicity of the hamiltonian flows plays an important role . verification of the latter for real systems is a difficult problem . in @xcite ergodicity of a system of @xmath3 hard spheres was proved . however , this proof can not be extended to oscillatory systems . in many papers @xcite the ergodicity problem is analyzed analytically and numerically . for quantum systems , justification of the micro - canonical distribution is even more problematic , since the matrix elements of the density operator calculated in the basis of the stationary states only change their phases and do not approach to the equilibrium distribution . nevertheless , von neumann @xcite managed to prove some important statements , which could be treated as quantum basis of the classical statistical mechanics . in the von neumann s approach , it is allowed to measure not all dynamical quantities , but only classical ( macroscopic ) variables which quantum operators mutually commute . one can guess , that in a non - isolated system being in contact with the thermal bath , all dynamical variables are characterized by statistical properties . in other words , the density operator of such a system at almost all time moments is the gibbs distribution . in the present paper this assumption is verified for the linear chain of harmonic oscillators - the system of interacting particles , which allows exact solution of the schrdinger equation . the dynamics of relaxation in this system is studied as well . the fundamental question about the meaning and the mechanism of thermalization in a macroscopic isolated dynamical system can be divided into two parts : * what is the precise meaning ( microscopic definition ) of the term the thermal equilibrium state for the isolated macroscopic system ? * in what sense the isolated macroscopic system approaches to the equilibrium state due to its dynamical evolution ? though for classical systems , at least , partial answers to these questions could be found in the ergodic theory , the situation with the quantum systems was much less certain . in the statistical mechanics textbooks ( for example , see @xcite ) , the equilibrium state of an isolated quantum system with the hamiltonian @xmath35 and the energy lying in the interval between @xmath36 and @xmath37 , ( @xmath38 ) is usually defined through the equal _ a priori _ probability postulate , leading to the microcanonical density operator @xmath39 , @xmath40 where @xmath41 denotes the trace of the operator @xmath42 , and @xmath43 is the heaviside step function . however , it is not easy to answer to the question ( ii ) with this conventional definition of the thermal equilibrium state . really , the matrix elements of the density operator @xmath44 of the isolated system calculated in the basis of its stationary states only change their phases with time and do not approach to the matrix elements of the microcanonical distribution ( [ mc ] ) . furthermore , evolution of the isolated system from some non - equilibrium pure state @xmath8 is described by the unitary operator @xmath45 , @xmath46 . corresponding density operator @xmath47 being the orthogonal projector operator at @xmath7 , @xmath48 remains , of course , to be the orthogonal projector operator at all @xmath49 , @xmath50 since the microcanonical density operator ( [ mc ] ) is not a projector operator , the projector - type density operator @xmath51 can not approach to it in some simple sense , say , in the strong or weak operator topology . the common way used in the textbooks to avoid this problem , is to remind that the notion of the isolated system is a physical idealization , and in reality , all systems interact with their surrounding . however , though this statement is undoubtedly correct , one could hardly believe , that it is the true origin of thermalization , and that the latter can not be understood within the concept of the isolated quantum system . if only isolated systems are concerned , the possible way to answer question ( ii ) is to restrict the set physically observable operators @xmath52 , to some family @xmath53 and to claim , that only for these operators @xmath54 one should expect approaching of the quantum expectation values @xmath55 to the their ( microcanonical ) equilibrium values : @xmath56 \leadsto { \rm tr } \ , [ { a}_i\,\cdot \rho_{\delta}(e , { h})],\ ] ] with subsequent fluctuations . notation @xmath57 will be used to indicate that the value of the time - dependent quantity @xmath58 is close ( equals in the thermodynamic limit ) to @xmath59 at almost any moment @xmath23 . this idea was realized by von neumann in his quantum ergodic theorem @xcite . in this theorem , operators @xmath60 were treated by von neumann as operators of the macroscopic observables , and were assumed to commute with each other , @xmath61=0 , \quad { \rm for } \ ; a_i , a_j\in { \mathcal a}. $ ] a deep analysis of foundations of the statistical mechanics was given by landau and lifshitz in their course of theoretical physics . they define the thermal equilibrium in the following way ( see pages 6 and 2 in @xcite ) : [ ll ] . if a closed macroscopic system is in a state such that in any subsystem the physical quantities are to a high degree of accuracy equal to their mean values , the system is said to be in a state of statistical equilibrium ( or thermodynamic or thermal equilibrium ) . the subsystems here are implied to be small compared with the whole system and weakly interacting with surrounding . such subsystems will be called as allowed subsystems in the sequel . in the above definition , thermalization behavior is required for the reduced family of the macroscopic observables , which relate to the macroscopic subsystems of the isolated system . taking definition [ ll ] as a starting point , one could arrive to a nice physical scenario , in which thermalization in the isolated macroscopic quantum system could result solely from its quantum mechanical evolution . the first step was to drop two underlined words in the above definition . this extended the family @xmath62 of thermalizable physical observable to _ all _ physical observables , which characterize the state of _ allowed subsystems _ , either macroscopic or not . then , definition [ ll ] becomes equivalent to the following one : [ q ] @xcite . a closed macroscopic system with hamiltonian h at the time moment t is in the state of equilibrium if any sufficiently small , possibly macroscopic subsystem of it that is weakly coupled to the surroundings has the gibbs distribution . let us denote by @xmath63 some subdivision of our isolated macroscopic quantum system into the allowed subsystem @xmath64 with the hamiltonian @xmath65 , and the rest of the system @xmath66 , which plays the role of the thermal bath . the hilbert space @xmath67 of the whole system can be decomposed into the tensor product @xmath68 of the hilbert spaces @xmath69 and @xmath70 corresponding to the subsystem , and its surrounding , respectively . [ part ] an isolated macroscopic quantum system will be called to be in the ( partial ) thermal equilibrium with respect to the allowed subsystem @xmath64 , if the reduced density operators @xmath71 of this subsystem is to a high accuracy equal to the canonical density operators @xmath72 given by the gibbs formula @xmath73 } , \\ z_{s,\alpha } = { { \rm tr } ^{(s,\alpha ) } \,\exp [ -\beta(\alpha ) \ , h_{s,\alpha } ] } , \nonumber\end{aligned}\ ] ] where @xmath74 denotes the trace over the hilbert space corresponding to the quantum states of the subsystem @xmath64 , and @xmath75 is the inverse temperature which could depend on the subsystem @xmath64 . note , that the reduced density operator of the subsystem @xmath76 is related with the density operator @xmath77 of the whole system as @xmath78 where @xmath79 denotes the trace over the thermal bath space @xmath70 . it is clear , that the macroscopic quantum system stays in the global thermal equilibrium in the sense of definition [ q ] , if and only if it is in the partial thermal equilibrium with respect to all its allowed subsystems @xmath64 with the same inverse temperature @xmath80 . definition [ q ] has several important advantages . first , it does not refer to the notion of the microcanonical ensemble and to the equal _ a priori _ probability postulate . second , with definition [ q ] , the answer to the question ( ii ) becomes straightforward and leads to the natural treatment of the thermalization phenomenon in quantum systems . [ t ] thermalization in the macroscopic isolated quantum system means , that after the relaxation time , the matrix elements of the reduced density operators @xmath81 of all its allowed subsystems @xmath64 in the basis of their stationary states @xmath82 should approach ( up to further fluctuations ) to those of the canonical density operators ( [ g ] ) , @xmath83 and @xmath84 , @xmath85 is the energy spectrum of the subsystem @xmath64 . third , one can easily see , that the problem mentioned in the beginning of this section does not arise with definition [ t ] . really , if the whole system is in a pure state described by some projector density operator ( [ nes ] ) , the subsystem reduced density operator ( [ rd ] ) can correspond to a mixed state due to the quantum entanglement , and this mixed state can approach to the equilibrium gibbs state ( [ g ] ) . if a macroscopic isolated quantum system possesses the property of thermalization , one should expect , that a typical initial state @xmath86 having macroscopically well determined total energy would approach in its unitary evolution to the equilibrium state , understood according to definition [ q ] and then remain in this state almost at any time . it is tempting to take the initial state @xmath87 of the whole system as a single stationary state @xmath88 of its hamiltonian , @xmath89 . then , the previous statement naturally leads to the eth @xcite , which can be formulated as follows : [ eth ] in a thermalizable isolated quantum system , the huge majority of individual stationary states @xmath88 of the whole system represent the thermal equilibrium state . this means , that @xmath90 for all allowed subsystems @xmath64 , and almost all stationary states @xmath88 . though the described above heuristic scenario of thermalization looked rather reasonable and attractive for me in 1979 , i had no idea how to put it on a more firm basis for a generic isolated quantum system . so , it was natural to try to test this scenario on some simple model of interacting particles . such a test was partly realized in @xcite for the model of the quantum chain of harmonic oscillators . it should be noted , that definition [ q ] provided the basis of the described above treatment of thermalization . it implies that the family @xmath62 of thermalizable observables can be associated with all observables which characterize the states of allowed subsystems . approach to the thermalization problem emphasizing the role of the canonical ensemble for the subsystem state as the main indication of thermal equilibrium in quantum systems , the origin of this approach comes back to erwin schrdinger @xcite , see discussion in @xcite . was developed later by several groups . important results along this direction were obtained by tasaki @xcite , popescu , short , and winter @xcite , linden _ et al . _ @xcite , and by goldstein _ in the normal mode representation , the quantum chain of harmonic oscillators reduces to the free phonon gas . as it was mentioned in the introduction , conservation of phonon numbers @xmath91 makes impossible complete thermalization of a typical initial state @xmath86 in this model . however , the quantum measurements performed on one atom in the chain does not allow one to distinguish different phonon modes . so , it is natural to expect , that for model ( [ hc ] ) , the weak versions of the thermalization property ( proposition [ t ] ) and of the eth ( proposition [ eth ] ) are still valid , in which equations ( [ g ] ) and ( [ g1 ] ) are required only for the subsystems @xmath92 consisting of one @xmath6-th atom in the chain . note , that due to the requirement @xmath93 , each atom in the chain weakly interacts with its surrounding and , therefore , represents an allowed subsystem by itself . paper @xcite , which english translation is given in appendix , was dedicated to verification of this statement . the eigenstate thermalization hypothesis ( see proposition [ eth ] ) was the important part of the described above heuristic scenario of thermalization , which provided motivation for @xcite . note , that the eth was not formulated explicitly in @xcite , this has been done later by deutsch @xcite and srednicki @xcite . however , a weak form of the eth for model ( [ hc ] ) was proved in @xcite . namely , equation validates the eth ( proposition [ eth ] ) in model ( [ hc ] ) for the _ single - atom subsystems_. it was conjectured also in @xcite , that , `` perhaps , this result holds to some extent in other dynamical systems '' . parameter @xmath94 in is related with the inverse temperature @xmath80 by the plank s formula @xmath95 and @xmath96 . in the case @xmath24 , equation can be easily obtained from which , in turn , follows from the results of appendix 1 of paper @xcite . to prove , let us rewrite for @xmath97 in the explicit form @xmath98 where @xmath99 denotes the set of integer numbers @xmath100 , which lie in the interval @xmath101 . denote by @xmath102 the number of nonzero integers in the set @xmath31 . for fixed @xmath103 and macroscopic @xmath102 , the leading contribution to the sum @xmath104 comes from configurations @xmath100 , in which @xmath105 takes values either @xmath106 , or @xmath107 . contribution of all the rest terms is smaller by the factor @xmath108 . the same is valid also for the sum @xmath109 therefore , @xmath110 = \\ \nonumber \frac { { \mathfrak n } ! } { p ! ( { \mathfrak n}-p ) ! } [ 1+o(n_1^{-1 } ) ] = \frac { { \mathfrak n}^p } { p ! } [ 1+o(n_1^{-1})],\end{aligned}\ ] ] at fixed @xmath111 and @xmath112 . formal substitution of into yields @xmath113 = \\\nonumber \,[1+o(n_1^{-1})]\ , \sum_{j=0}^\infty \frac{{\bar n}^j}{(1+{\bar n})^{j+1}}\ , |j\rangle\langle j|,\end{aligned}\ ] ] in agreement with . if @xmath114 , the above formal manipulations can be justified , since the infinite series in @xmath111 in converges uniformly in @xmath21 . at @xmath115 , the series in @xmath111 in diverges . nevertheless , it is also possible to prove at @xmath115 directly from , though the proof is more complicated , than at @xmath116 . on the other hand , a much easer indirect way to confirm exists , which is suitable for all positive @xmath117 . namely , one can check that the quantum expectation values of the operator @xmath118 calculated with the both sides of this equation lead to the same results for any integer @xmath119 . let @xmath120 and @xmath121 are the two sets of @xmath3 bosonic annihilation operators , which are related by the linear transformation @xmath122 with the unitary @xmath3-dependent matrix @xmath123 , and @xmath124 denotes the basis associated with the first set of the bosonic operators . for each @xmath3 , let us choose a basis vector @xmath125 in such a way , that @xmath126 , \textrm { with some fixed } c>0,\ ] ] and the limits @xmath127 exist for given @xmath6 . then @xmath128^{m_1}\ , \delta_{m_1 m_2}.\ ] ] first , it is clear , that operator @xmath129 with @xmath130 has zero diagonal matrix elements in the basis , since this operator changes the number of bosons . so , it is sufficient to put @xmath131 in . then , substitution of into the matrix element in the left - hand side of this equation followed by straightforward calculations yields @xmath132 where @xmath133 is the laguerre polynomial , @xmath134 the explicit form of read as @xmath135 ^ 2- \sum_{l=1}^n |a_{kl}|^4 ( n_l + n_l^2 ) , \nonumber\\ & & \langle n_n,\ldots , n_1|b_k^{\dagger 3}b_k^3|n_1,\ldots , n_n\rangle = 3!\left [ \sum_{l=1}^n |a_{kl}|^2 n_l\right]^3 + \sum_{l=1}^n |a_{kl}|^6 ( 2 n_l+6 n_l^2 + 4 n_l^3)- \\ & & 9 \sum_{l=1}^n |a_{kl}|^4 \sum_{m=1}^n ( n_l\,n_m + n_l^2),\\ & & \langle n_n,\ldots , n_1|b_k^{\dagger m}b_k^m|n_1,\ldots , n_n\rangle = m!\left [ \sum_{l=1}^n |a_{kl}|^2 n_l\right]^m+\ldots\end{aligned}\ ] ] it is easy to understand , that only the first term in the right - hand side of the above relations survives in the thermodynamic limit @xmath21 due to and . let us associate with each pair of operators @xmath136 the single - atom subsystems @xmath92 with the hamiltonians @xmath137 . suppose now , that each such atom is in the thermal equilibrium state characterized by the @xmath6-dependent inverse temperature @xmath138 this means , that the reduced density operator of the @xmath6-th atom has the gibbs form , @xmath139}}{z_k^{(s ) } } = \sum_{j=0}^\infty \frac { [ { \bar{n}}(k ) ] ^j}{[1 + { \bar{n}}(k ) ] ^{j+1}}\,\frac{1}{j!}\,\ , b_k^{\dagger j}|0\rangle\langle0|b_k^j.\ ] ] then , for arbitrary natural @xmath140 , one can easily obtain the expectation value of the product @xmath129 in this state @xmath141 = m_1 ! \ , [ { \bar{n}}(k ) ] ^{m_1}\ , \delta_{m_1 m_2},\ ] ] which coincides with the right - hand side of . thus , at large @xmath3 , the huge majority of basis states correspond to the partial equilibrium states with respect to the single - atom subsystems @xmath92 in the sense of definition [ part ] . the inverse temperature @xmath142 of the @xmath6-th atom , generally speaking , depends on @xmath6 , and is given by equations , . let us return to the chain of harmonic oscillator . for the bosonic annihilation operators @xmath143 and @xmath144 determined according to , the matrix @xmath145 introduced in reduces at @xmath146 to the simple form @xmath147 , and @xmath148 . parameter @xmath149 defined by reduces to the average number of phonons in the chain @xmath150 and becomes @xmath6-independent , together with the inverse temperature . this proves for all @xmath151 . relaxation of the macroscopic chain of harmonic oscillators from a nonequilibrium nonuniform initial state @xmath152 was studied in section 2 of @xcite . first , the time evolution of the chain quantum state @xmath13 was determined in the leading order in the interaction constant @xmath153 . then , representation ( [ rho ] ) was obtained for the density operators @xmath14 of the @xmath6-th atom at time @xmath23 , which is asymptotically exact at @xmath154 . the full form of this expression reads as @xmath155^{2j}\left\{1-\left[\frac{r(k , t)}{n}\right]^{2}\right\}^{{\mathfrak n}-j},\ ] ] with @xmath156 } \bigg| .\ ] ] remind , that @xmath157 and @xmath158 are the phonon quasimomentum and the energy , respectively , @xmath159^{1/2}=\omega_0+\frac { \omega^2}{\omega_0}(1-\cos \tau)+o(\omega^4 ) . \end{aligned}\ ] ] parameter @xmath160 is proportional to the energy @xmath161 , which is located at the @xmath6-th atom at the time moment @xmath23 : @xmath162= { \mathfrak n}\ , \omega_0 \,r^2(k , t ) /{n^2 } .\ ] ] at small enough @xmath23 , the sum in ( [ r2 ] ) can be replaced by the integral . this allows one to express the energy @xmath163 normalized to the total energy in the chain in terms of the cylindric bessel function @xmath164 , @xmath165 \bigg|^2= \\ \left [ j_k(t\,\omega^2/\omega_0)\right]^2 . \label{bes}\end{aligned}\ ] ] the plot of this normalized energy versus @xmath6 and @xmath23 is shown in figure [ rel ] . space and time dependence of the energy localized on the @xmath6-th atom normalized to the total energy in the chain , @xmath166 ^ 2 $ ] . ] it is clear from ( [ en ] ) and figure [ rel ] , that the energy perturbation spreads along the chain with the velocity @xmath167 , which is just the maximum group velocity of phonons @xmath168 this feature of the relaxation dynamics known as the horizon effect is typical in the quantum quench phenomenon @xcite . now let us turn to the dynamics on the long time scale . when time @xmath23 becomes larger then the relaxation time @xmath169 , the energy perturbation reaches all atoms in the chain , all exponential terms in the sum in right - hand side of ( [ r2 ] ) completely dephase from each other , and this sum can not be approximated by the integral , as it was done in ( [ bes ] ) . this leads to the qualitative change of of the chain evolution character from regular at @xmath170 , to the stochastic regime at @xmath171 , which is illustrated in figure [ r2 ] . the red line shows the time dependence of the quantity @xmath172 defined by ( [ r2 ] ) , which is proportional to the energy @xmath173 localized on the @xmath106-th atom , @xmath174 . the blue line plots @xmath175 ^ 2 $ ] . the model parameter values are taken as @xmath176 . change of the evolution regime takes place at @xmath177 . ] for the atom reduced density operators averaged over the infinite time interval , the canonical gibbs distribution ( [ r6 ] ) with the site - number independent temperature was obtained in @xcite . however , derivation of this result was incorrect , since it was based there on the erroneous equation ( a.14a ) in page , see the footnote in this page . the fallacy in this equation came from the implicit and mistaken assumption , that the energy levels @xmath178 of the whole system are non - degenerate , @xmath179 really , the energy levels ( measured from the ground state energy ) in model ( [ hc ] ) are given by @xmath180 where summation in the quasimomentum @xmath181 runs over the brillouin zone , @xmath182 ) . due to the the mirror symmetry @xmath183 , of the phonon dispersion law , this energy spectrum can be rewritten in the form @xmath184 + \sum_{\substack{\tau\in { \mathcal b}\\ 0<\tau<\pi } } ( n_\tau+n_{-\tau})\,\omega_\tau,\ ] ] which indicates the strong degeneracy of the energy levels @xmath178 in model , in contrast to . accordingly , one gets for the time average @xmath185}$ ] : @xmath186 } = \begin{cases } \delta_{n_0 , n_0 ' } \,\delta_{n_\pi , n_\pi'}\prod_{\substack{\tau\in { \mathcal b}\\ 0<\tau<\pi } } \delta _ { n_\tau+n_{-\tau } , n_\tau'+n_{-\tau } ' } , & { \rm for\ ; even}\ ; n , \\ \delta_{n_0 , n_0'}\prod_{\substack{\tau\in { \mathcal b}\\ 0<\tau<\pi } } \delta _ { n_\tau+n_{-\tau } , n_\tau'+n_{-\tau}'},&\ , { \rm for\ ; odd}\ ; n , \end{cases}\ ] ] instead of equation ( a.14a ) . it turns out , however , that this correction does not change the final result ( [ r6 ] ) for the time averages of the reduced density operators almost for all atoms in the chain . [ av ] in the thermodynamic limit , the time averages of the atom reduced density operators @xmath14 determined by equation [ or , eqivalently , by equations ( [ rhoa ] ) , ] are given by the canonical gibbs distributions with two different temperatures , @xmath187 where the time average is defined as @xmath188 equation ( [ avr ] ) corrects equation ( [ r6 ] ) for @xmath189 , and @xmath190 . will be given only for the case of even @xmath3 , since extension to the case of odd @xmath3 is straightforward . let us introduce the @xmath191-dimensional phase space with the points @xmath192 , where @xmath193 , and @xmath194 , with @xmath195 . let us define the operator - valued function @xmath196 on this space which is obtained from ( [ r1 ] ) by the replacement @xmath197 and averaging over the phases @xmath198 of the @xmath199-periodical functions @xmath200 , @xmath201 then , it follows from ( [ avv ] ) , that @xmath202}=\langle\exp[i(\phi(n'-n))]\rangle_\phi.\ ] ] applying this equality to ( [ r1 ] ) , one can conclude , that the time average of the operator @xmath14 is equal to the phase average of the operator @xmath196 . therefore , the time average of the operator ( [ rhoa ] ) can be written as @xmath203^{2 j}\left[1-\left[r(k,\phi)/n\right]^2\right]^{{\mathfrak n}-j } \big\rangle_\phi,\ ] ] where @xmath204\right|,\\ w_l(k)= \begin{cases } 1 , & { \rm for } \ ; l=0,\\ 2 \cos[k\ , \tau(l ) ] , & { \rm for } \ ; l=1,\ldots,\frac{n}{2}-1,\\ ( -1)^k , & { \rm for } \ l=\frac{n}{2}. \end{cases}\end{aligned}\ ] ] accordingly , the matrix elements of the operator ( [ r5 ] ) in the basis @xmath205 take the form @xmath206^{{\mathfrak n}-j } p_n(k , r ) , \label{43}\\ & & p_n(k , r ) = \langle \delta[r - r(k,\phi)]\rangle_{\phi}. \label{pn}\end{aligned}\ ] ] after rescaling of the integration variable @xmath207 in the right - hand side of ( [ 43 ] ) , the large-@xmath3 asymptotics of this equation can be written as @xmath208 function @xmath209 defined by ( [ pn ] ) is the probability density to find the total length @xmath210 of the sum of @xmath211 vectors in the plane which have arbitrary directions and different lengths @xmath212 . straightforward calculations lead to the exact representation of this probability density in terms of the cylindrical bessel function @xmath213 [ cf . equation in appendix ] , @xmath214.\ ] ] at large @xmath3 , the main contribution to the integral in the right - hand side of comes from small @xmath215 , since the integrand vanishes at @xmath216 . so , in the thermodynamic limit we get from ( [ pnk ] ) : @xmath217[1+o(n^{-1 } ) ] = \\ & & \frac{2 x}{y(k , n ) } \exp[-x^2/y(k , n)]+o(n^{-1}),\label{pnk1}\end{aligned}\ ] ] where @xmath218 substitution of ( [ pnk1 ] ) and ( [ sk ] ) into yields finally to the result @xmath219^j } { [ 1+\bar{n}\,y(k , n)]^{j+1}}+ o(n^{-1 } ) = \begin{cases } \frac { [ 2\bar{n } ] ^j}{(1 + 2\bar{n})^{j+1 } } + o(n^{-1 } ) , & { \rm if \:}k=0,n/2,\\ \frac { \bar{n}^j}{(1+\bar{n})^{j+1 } } + o(n^{-1 } ) , & { \rm if \:}k=1,\ldots,\frac{n}{2}-1 , \end{cases}\ ] ] which , together with ( [ jj ] ) proves . let us return now to the time evolution of the atom reduced density operators ( [ rho ] ) at large @xmath171 . at a fixed time moment @xmath171 one can treat @xmath220 as the absolute value of the sum of @xmath3 unit vectors on the plane which have stochastic phases . accordingly , it is again natural to rescale the length of the random walk path @xmath221 with @xmath222 for almost any @xmath171 . substitution of ( [ st ] ) into leads in the thermodynamic limit @xmath21 , @xmath223 , @xmath224 to the poisson distribution @xmath225^j}{j!}\exp{[-\bar{n}\ , x^2(k , t)]}\ ] ] for the atom density operators at almost any time moments @xmath226 . on the other hand , the time average ( [ timeav ] ) of the density operator @xmath227 over the infinite time interval was proved to approach to the canonical gibbs distributions in the thermodynamic limit , see . it follows from and , that the reduced density operator @xmath228 of the @xmath6-th atom never approach to its time average at any time moment . similarly , the quantum averages @xmath229 $ ] of the local observables @xmath230 relating to the @xmath6-th atom should also strongly fluctuate in time . for the 0-th atom energy operator @xmath231 , the time fluctuations of its quantum expectation values @xmath173 are clearly seen in figure [ r2 ] at @xmath232 . this means , in turn , that the time dispersion @xmath233 of the density operator matrix elements @xmath234 remains considerable and does not vanish in the thermodynamic limit . such type of evolution indicating luck of partial thermalization in model ( [ hc ] ) was associated in @xcite with the strong degeneracy of the energy level differences , see equation . really , the dispersion can be written as @xmath235| j\rangle \,\ , \langle j'|{\rm tr}^{(r , k)}\left[|n_2'\rangle\langle n_2|\right]| j'\rangle \cdot \\ \nonumber & & \langle n_1|\psi(0)\rangle \langle \psi(0)|n_1'\rangle \langle n_2'|\psi(0)\rangle \langle \psi(0)|n_2\rangle \cdot\\ & & \left\ { \overline{\exp[i t(e_{n_2}-e_{n_2'}-e_{n_1}+e_{n_1 ' } ) ] } - \overline{\exp[i t(e_{n_2}-e_{n_2 ' } ) ] } \,\,\,\overline{\exp[i t(e_{n_1'}-e_{n_1 } ) ] } \label{dis1 } \right\},\end{aligned}\ ] ] where @xmath178 are the energy levels ( [ ele ] ) in model ( [ hc ] ) . the last line in has the structure @xmath236 } - \overline{\exp[i t(e_{n_2}-e_{n_2 ' } ) ] } \,\,\,\overline{\exp[i t(e_{n_1'}-e_{n_1})]}\right\ } = \\ & & \delta_{n_1 n_2 } \delta _ { n_1 ' n_2 ' } - \delta_{n_1 n_2 n_1 ' n_2 ' } + c_{n_1 n_2 n_1 ' n_2 ' } , \label{rt}\end{aligned}\ ] ] where @xmath237 is the kronecker delta , and @xmath238 the _ resonance term _ @xmath239 in the right - hand side of is nonzero , in particular , at @xmath240 , and as well at @xmath241 due to . it is precisely this term which is responsible for the strong time fluctuations of the density operator @xmath14 in model ( [ hc ] ) . on the other hand , the resonance term @xmath239 vanishes in ( [ rt ] ) , if a weak anharmonic interaction is applied providing for the shifted energy levels that @xmath242 in this case , one should expect , that the time dispersion @xmath243 would also vanish in the thermodynamic limit @xmath244 . for @xmath24 , this was indeed proved in @xcite . the crucial role of the non - resonance condition for ability of an isolated quantum system to thermalize was first established by von neumann @xcite , and confirmed later by many authors @xcite . it is interesting to illustrate the dramatic effect of the nonlinear interaction on the time evolution of the quantum expectation value of the energy @xmath163 of a particular atom , @xmath245 where @xmath246 is the hamiltonian of the @xmath6-th atom , and @xmath13 is the quantum state of the chain at time @xmath23 which evolves from the initial state ( [ in ] ) . in the purely harmonic chain ( [ hc ] ) , the energy @xmath163 is determined by ( [ en ] ) : @xmath247 taking into account ( [ r2 ] ) , one obtains from for the time averages : @xmath248 where @xmath249 ^ 2.\ ] ] thus , the time dispersion @xmath250 of the quantum expectation value of the atom hamiltonian remains finite in the thermodynamic limit in the purely harmonic chain . on the other hand , in the chain perturbed by the extremely weak anharmonic interaction , the @xmath6-th atom energy ( [ en0 ] ) can be determined as @xmath251,\ ] ] and the atom reduced density operator @xmath252\}}{(n - l ) ! } \sum_{\sum_\tau n'_\tau={\mathfrak n}}\frac{\exp \{i [ -k(n'\tau ) + t\,e_{n'}]\}}{(n'-l)!},\label{r1a}\end{aligned}\ ] ] can be obtained from by replacement @xmath253 corresponding to the small shift of the energy levels . if the perturbed energy levels @xmath178 are not degenerate , averaging in time of ( [ en3 ] ) with the atom density operator @xmath254 given by ( [ r1a ] ) provides @xmath255 instead of ( [ eav ] ) . in turn , if the energy levels @xmath178 satisfy the non - resonance condition ( [ er ] ) , the time dispersion ( [ dis4 ] ) can be written as @xmath256 where @xmath243 is given by ( [ s ] ) . due to the equality @xmath257 several summations can be explicitly performed in ( [ sj ] ) . the result reads as @xmath258^{n } , \label{sb}\end{aligned}\ ] ] where integration is performed in the complex @xmath259-plane along the circle centered at origin and going in the positive direction , and @xmath260 is the modified bessel function of the first kind , @xmath261 function @xmath262 determined by . ] in the thermodynamic limit @xmath21 , @xmath263 , the integral in the right - hand side of is determined by contributions of two saddle points at @xmath264 , where @xmath265 is the positive solution of the equation @xmath266_{x = x_0(\bar{n})}=0.\end{aligned}\ ] ] the saddle - point asymptotics of in the thermodynamic limit reads as @xmath267^{1/2 } [ w(\bar{n})]^n \,\left[1+o(n^{-1 } ) \right],\ ] ] where @xmath268 - 2\bar{n } \log x,\ ] ] and @xmath269 } { [ x_0(\bar{n})]^{2\bar{n}}}.\ ] ] the plot of function @xmath270 is shown in figure [ ww ] . this function has the following asymptotic behavior @xmath271 , & \quad \bar{n}\to+0 , \\ ( 4\pi \bar{n})^{-1/2}\,[1+o(\bar{n}^{-1 } ) ] , & \quad\bar{n}\to+\infty , \end{cases}\end{aligned}\ ] ] and monotonically decreases between values @xmath272 and @xmath273 in the half - axis @xmath274 . therefore , it follows from ( [ spd ] ) , that the time dispersion ( [ dis4 ] ) of the quantum expectation value of the atom energy vanishes in the thermodynamic limit @xmath21 for all fixed @xmath151 . i am thankful to p. calabrese , j. cardy , h. w. diehl , d. huse , m. olshanii and a. silva for interesting discussions in the recent years . i am also grateful to m. olshanii and a. silva for their advise to publish the english translation of @xcite . 10 s. b. rutkevich . relaxation dynamics of a quantum chain of harmonic oscillators . , 25(7):11351141 , 1980 . s. b. rutkevich . statistical equilibrium setting in the quantum chain of harmonic oscillators . , 44(2):263270 , 1980 . english translation : , 44:732737 , 1980 . m. greiner , o. mandel , t. esslinger , t. w. hnsch , and i. bloch . quantum phase transition from a superfluid to a mott insulator in a gas of ultracold atoms . , 415(3):3944 , 2002 . i. bloch , j. dalibard , and w. zwerger . many - body physics with ultracold gases . , 80:885964 , 2008 . t. kinoshita , t. wenger , and d. s. weiss . a quantum newton s cradle . , 440:900903 , 2006 . p. calabrese and j. cardy . time dependence of correlation functions following a quantum quench . , 96:136801 , 2006 . p calabrese and j. cardy . quantum quenches in extended systems . , p06008 , 2007 . m. rigol , v. dunjko , and m. olshanii . thermalization and its mechanism for generic isolated quantum systems . , 452(7189):854858 , 2008 . d. rossini , a. silva , g. mussardo , and g. e. santoro . effective thermal dynamics following a quantum quench in a spin chain . , 102:127204 , 2009 . a. iucci and m. a. cazalilla . quantum quench dynamics of the luttinger model . , 80:063619 , 2009 . a. polkovnikov , k. sengupta , a. silva , and m. vengalattore . nonequilibrium dynamics of closed interacting quantum systems . , 83:863883 , 2011 . p. calabrese , f. h. l. essler , and m. fagotti . quantum quench in the transverse - field ising chain . , 106:227203 , 2011 . j. m. deutsch . quantum statistical mechanics in a closed system . , 43:20462049 , 1991 . m. srednicki . chaos and quantum thermalization . , 50:888901 , 1994 . s. popescu , a. j. short , and a. winter . entanglement and foundations of statistical mechanics . , 2:754758 , 2006 . n. linden , s. popescu , a. j. short , and a. winter . quantum mechanical evolution towards thermal equilibrium . , 79:061103 , 2009 . s. goldstein , j. l. lebowitz , r. tumulka , and n. zangh . canonical typicality . , 96:050403 , 2006 . s. goldstein , j. l. lebowitz , c. mastrodonato , r. tumulka , and n. zangh . normal typicality and von neumann s quantum ergodic theorem . , 466(2123):32033224 , 2010 . s. goldstein , j. l. lebowitz , r. tumulka , and n. zangh . long - time behavior of macroscopic quantum systems . commentary accompanying the english translation of john von neumann s 1929 article on the quantum ergodic theorem . , 35(2):173200 , 2010 . h. tasaki . from quantum dynamics to the canonical distribution : general picture and a rigorous example . , 80:13731376 , 1998 . k. huang . . science , 1963 . j. von neumann . beweis des ergodensatzes und des h - theorems in der neuen mechanik . , 57(2):3070 , 1929 . j. von neumann . . springer , 1932 . l. d. landau and e. m. lifshitz . , volume 5 of _ course of theoretical physics_. pergamon press , third edition , 1980 . part 1 . e. schrdinger . . cambridge university press , second edition , 1952 . appendix . ukrainian physical journal , * 25 * , n 7 , 1135 - 1141 ( 1980 ) . + s. b. rutkevich + + 4em the aim of the present paper is to trace the appearance of some statistical properties in an exactly solvable dynamic system - a quantum chain with weak harmonic interaction . the time dependence of the statistical single particle matrices is derived when the system relaxes from the non - equilibrium pure state . it is shown that the thermodynamic limit of the matrices in question is the gibbs distribution . moreover , the presence of any weak anharmonic interaction in the system causes each oscillator to have a canonical distribution almost at any moment . we consider the periodical chain of harmonic oscillators . each of them is coupled with two neighbours and with its equilibrium point @xcite . the hamiltonian has the form @xmath275 , \quad q_n\equiv q_0.\ ] ] here @xmath276 are the coordinate and momentum of the @xmath6-th oscillator , @xmath277 @xmath3 is the numbers of oscillators , @xmath278 is the interaction potential of the neighbouring oscillators , @xmath279 is their own frequency , the plank constant and the oscillator mass are put to the unit value , periodical boundary conditions are chosen . the @xmath6-th oscillator will be called the @xmath6-th atom. model has been studied in @xcite in order to calculate correlation functions and to estimate the poincare periods in the thermodynamics limit . the dynamical problem can be solved exactly if @xmath280 . in this case the normal coordinate @xmath281 , @xmath282 are introduced as @xmath283 with integer @xmath284 . two sets of secondary quantization operators are related with @xmath285 : @xmath286 where @xmath287 . the hamiltonian can be rewritten in the form @xmath288 which shows that the normal modes ( phonons ) do not interact . the eigenstate of the hamiltonian reads as @xmath289 where @xmath290 is the ground state , @xmath291 is the number of phonons with the quasi - momentum @xmath292 , @xmath293 . if @xmath294 , each atom can be considered as a subsystem which interacts weakly with surrounding ( the thermal bath ) . if @xmath295 is small , one can follow the relaxation dynamics of the single - atom density operator for the system prepared in some realistic non - equilibrium initial state . let us introduce the following notations ; @xmath296 if the whole system is described by the density operator @xmath297 , than the atom with the number @xmath11 in the chain has the density operator @xmath298 , where @xmath299 only at @xmath146 : @xmath300 where @xmath301 , @xmath302 . the state @xmath26 has the energy @xmath303 , where @xmath304 is the ground state energy . using ( [ tr ] ) it is possible to show that the atom s density operator which corresponds to the stationary state @xmath26 of the system , has the form of the gibbs distribution in the thermodynamic limit ( @xmath21 , @xmath305 ) at least for @xmath24 , if the number @xmath306 of nonzero integers in the set @xmath307 is macroscopic ( i.e. @xmath308 ) : @xmath309.\ ] ] perhaps , this result holds to some extent in other dynamical systems . let us show , how the evolution of the system from some non - equilibrium initial state can be described . let @xmath310 and @xmath295 is so small , that one can neglect the terms of order @xmath311 in @xmath312|\,n[\omega(t)=0]\,\rangle=\delta_{nn'}+\epsilon_{nn'},\ ] ] i.e. interaction almost does not change the stationary states @xmath26 . at @xmath313 one can use expression ( [ tr ] ) , and at @xmath314 the system relaxes due to the weak interaction . if at @xmath7 the system is in the stationary state @xmath26 , then at time @xmath315 its state vector is given by @xmath316 $ ] ( the terms of order @xmath311 are omitted ) , where @xmath317 . let us choose the non - equilibrium initial state as @xmath318 the 0-th atom is exited to the energy @xmath319 , while all the rest atoms remain unexcited . then one gets from ( [ tr ] ) , ( [ in ] ) @xmath320\}}{(n - l ) ! } \sum_{\sum_\tau n'_\tau={\mathfrak n}}\frac{\exp\ { i [ -k(n'\tau ) + t(n'\omega ) ] \}}{(n'-l)!},\label{r1}\end{aligned}\ ] ] where @xmath14 is the density operator of the @xmath6-th atom ( @xmath321 at the time @xmath23 , and @xmath322 . straightforward calculations yield @xmath323^{{\mathfrak n}-j},\ ] ] where @xmath324 keeping the linear terms in @xmath325 in the expansion of @xmath326 , and replacing the sum in @xmath181 in @xmath327 by the integral , we come to the integral representation of the bessel function of the @xmath328-th order : @xmath329\right|.\ ] ] one can safely replace the sum by the integral at large @xmath3 and if @xmath330 $ ] ( the period of the integrand is much large than @xmath331 ) . one can easily see , that the energy of the @xmath6-th atom at the time @xmath23 exceeds the ground state energy by the quantity @xmath332.\ ] ] perturbation reaches the @xmath6-th atom after the time of order @xmath333 ^ 2 $ ] , and after that the energy of this atom oscillates slowly with decreasing amplitude . when the excitation energy distributes throughout the whole chain , one can not any more replace the sum ( [ rho ] ) by the integral . for the classical harmonic chain , formula ( [ en ] ) was obtained in @xcite . let us introduce the following notations for the average quantities : @xmath334 where @xmath335 are an operator a function of time , respectively . in order to calculate @xmath336 , it is sufficient to know @xmath337 . consider the space of phases with points @xmath338 , @xmath339 ( we use the same notations for indexes as in formula ( [ n ] ) , and the measure @xmath340 . let us define the operator - valued function @xmath341 on this space which is obtained from ( [ r1 ] ) by the replacement @xmath342 [ s.r . 2012 ] the first equality in equation ( a.14a ) is wrong , since it does not take into account that the energy spectrum @xmath343 of model ( [ h ] ) is strongly degenerate . the corrected form of this equation is given by formula in subsection [ ddo ] . this correction induces also certain modifications into the calculation of @xmath337 in this page and into the final result . these modifications are described in proposition [ av ] and equation ( [ avr ] ) in subsection [ ddo ] . fortunately , the above mentioned modifications are minimal , and equation holds for all but one or two atoms in the chain , depending on the parity of @xmath3 . on the other hand , application of a weak anharmonic interaction breaking the energy spectrum degeneracy restores the result : the time average of the reduced density operators of all atoms in the chain is given by the gibbs distribution with _ the same temperature_. @xmath202}=\langle\exp[i(\varphi(n'-n))]\rangle_\varphi = \delta_{nn ' } , \quad\quad \quad\quad\quad\quad \quad\quad\quad\quad \quad\quad \quad\quad \quad\quad(a.14a)\ ] ] one can conclude from ( [ r1 ] ) , that the time average of the operator @xmath14 equals to the phase average of the operator @xmath341 . this allows us to write the time average of the operator ( [ rho ] ) in the form @xmath344^{2 j}\left[1-\left[r(\varphi)/n\right]^2\right]^{{\mathfrak n}-j } \big\rangle_\varphi,\ ] ] where @xmath345 the problem is reduced to the random walk problem , which was studied in @xcite . using formulas ( 51)-(53 ) , ( 103 ) , ( 104 ) of reference @xcite we find @xmath346 using the above large-@xmath3 asymptotics for @xmath347 and for the asymptotic formula @xmath348^{{\mathfrak n}-j}\to \exp[-r^2 { \mathfrak n}/n^2 ] , \quad { \rm at } \quad n\to\infty,\ ] ] we get latexmath:[\[\label{r6 } \lim_{n\to\infty}\bar{\rho}_n=\sum_j\frac{1}{1+\bar{n } } \left(\frac{\bar{n}}{1+\bar{n}}\right)^j it is clear from ( [ rho ] ) that the time fluctuations of the density operator do not vanish in the thermodynamic limit . this behavior is caused by the degeneracy of the system in the differences of the energy levels of the stationary states : @xmath350 a weak anharmonic interaction can be described as a perturbation leading to a small shift of each energy level @xmath352 therefore , in equation ( [ r1 ] ) , we should replace now @xmath353 by @xmath178 , and @xmath354 by @xmath355 , and the statement ( a ) becomes evident . let us prove ( b ) . one obtains from ( [ rho ] ) @xmath356 in the sum ( [ s ] ) , the following inequality holds @xmath357 where @xmath358 . since @xmath359 the series in ( [ s ] ) uniformly converges at @xmath24 . therefore , one can treat @xmath111 and @xmath360 as finite numbers . let @xmath361 . the number of terms in the sum @xmath362 is smaller than @xmath363 ^ 2,\ ] ] therefore @xmath364 ^ 2\le \frac{\bar{n}^{\mathfrak n}n^{2p}}{p!^2},\end{aligned}\ ] ] which proves statement ( b ) . it is quite possible that ( [ tr1 ] ) and ( b ) hold also for @xmath25 . in conclusion let us note that the considered model can be useful for many other applications . since in the case of the harmonic interaction , its dynamics is integrable , the model can be used to verify different approximate methods of studying relaxation processes . in the case of the anharmonic interaction , one can study analytically or numerically the relaxation dynamics in a nonlinear system . the model could be useful also for analyzing the arising statistical properties in a non - macroscopic systems , like three - atom molecules . dan sssr , ser . math . * 153 * , n6 , 1261 - 1264 ( 1963 ) . j. ford , and g. h. lunsford . stochastic behaviour of resonant nearly linear oscillator system in the limit of zero nonlinear coupling . phys . rev . a * 1 * , n 1 , 57 - 59 ( 1970 ) . f. m. izrailev , and b. v. chirikov . statistical properties of the nonlinear string , dan sssr , ser . fiz . , * 166 * , n1 , 57 - 59 . g. m. zaslavsky . _ statistical irreversibility in nonlinear systems _ , `` nauka '' , moscow , 1970 . j. von neumann . _ mathematical foundations of quantum mechanics _ , princeton univ , 1996 e. m. henley , and w. thirring . _ elementary quantum field theory _ , mcgraw hill book company inc . , ny , 1962 . p. mazur , and e. montroll . poincare cycles , ergodicity and irreversibility in assemblies of coupled harmonic oscillators , j. math . , * 1 * , n 1 , 70 - 84 ( 1960 ) . s. chandrasekar . _ stochastic problems in physics and astronomy _ , rev . phys . , * 15 * , 1 , ( 1943 ) .

we investigate the transport through a few - level quantum system described by a markovian master equation with temperature and particle - density - dependent chemical potentials . from the corresponding onsager relations we extract linear response transport coefficients in analogy to the electronic conductance , thermal conductance , and thermopower . considering ideal fermi and bose gas reservoirs , we observe steady - state currents against the thermal bias as a result of the nonlinearities introduced by the constraint of a constant particle density in the reservoirs . most importantly , we find signatures of the on - set of bose - einstein condensation in the transport coefficients .

1309.3488

transport processes are widespread and commonly occur in many fields of physics , chemistry , and biology . in electronic and photonic systems , the chemical potential as well as the temperature are normally treated as independently controllable parameters @xcite . in equilibrium thermodynamics , the chemical potential is in general a function of the intensive properties of the gas such as , e.g. , temperature and particle density . in most setups studied so far , assuming an independent , constant chemical potential is a valid conjecture , since the transport setup is usually embedded in a much larger environment . this surrounding environment acts as a particle and temperature reservoir providing the necessary resources to fix the chemical potential at a constant value . however , there are also transport setups which can not be treated in this manner . in particular , there has been a lot of progress in the production and manipulation of ultracold quantum gases . this includes the production of bose - einstein condensates in electromagnetic traps @xcite or even in standing light fields @xcite that allow for an experimental implementation of hubbard - type quantum models @xcite . these ultracold atomic gases have been studied in equilibrium situations for quite a while and with huge success . nowadays , the focus shifts to investigating the nonequilibrium properties of such systems @xcite . recently , there have been the first real transport experiments with ultracold atoms @xcite . in these setups , the transport processes are driven by at least two reservoirs which are initialized in different equilibrium states and attached to the system of interest . accompanying these experimental advances , there has been also theoretical research involving atomic reservoirs coupled to , e.g. , a lattice system @xcite , a potential trap @xcite , or even quantum dot systems @xcite . since the ultracold atoms are trapped in a high - vacuum chamber , the gas is well separated from its environment such that no additional particle reservoir is present . therefore , in these experiments the particle density is constant and the chemical potential can not be treated as free accessible parameter but becomes a function of the temperature and particle number in the trap . this strongly motivates us to investigate the influence of such a temperature- and particle - density - dependent chemical potential on the transport properties of a two - terminal open quantum system setup . this analysis is especially interesting for systems that undergo quantum phase transitions . for example , transport through such systems contains information about such transitions even in extreme nonequilibrium setups @xcite . it should be noted that such transitions only occur in the thermodynamic limit ( infinite system size ) , which suggests to investigate the role of criticality within the reservoirs . in fact , criticality is normally associated with a characteristic change in the chemical potential . hence , it is crucial to describe the chemical potential in dependence of the thermodynamic state variables of the gas in order to correctly describe these critical phenomena . within this paper , we particularly investigate the difference between the transport of fermionic and bosonic particles through such systems . since an ideal bose gas shows a quantum phase transition from a normal phase to a bose - einstein condensate in thermodynamic equilibrium , we expect to find signatures of this critical reservoir behavior in the transport coefficients . in sec . [ s : thoereticalframework ] , we present the general theoretical framework that we use throughout this paper . we review the properties of ideal quantum gases in sec . [ s : atomicreservoirs ] , derive the master - equation formalism with which we describe the transport through an open quantum system in sec . [ s : masterequation ] , and analyze the steady - state entropy production in sec . [ s : entropyproduction ] . finally , we apply a linear response theory to extract the linear transport coefficients in sec . [ s : transportcoefficients ] . subsequently , we apply this formalism to ideal fermi gases in sec . [ s : fermigas ] and to ideal bose gases in sec . [ s : bosegas ] , respectively , and summarize our results in sec . [ s : summary ] . the main difference of our setup compared to the usual schemes to describe transport through nanostructures lies in the utilization of massive ultracold atoms with a temperature- and density - dependent chemical potential . consequently , the reservoirs are modeled as ideal quantum gases in the thermodynamic limit . due to their statistical properties , bosonic reservoirs are fundamentally different from their electronic counterpart . in particular , the bosonic reservoirs undergo a quantum phase transition from a normal phase to a bose - einstein condensate @xcite . therefore , criticality is induced in the transport setup via the reservoirs . this critical behavior of the reservoirs is discussed in more detail in the following . the atomic baths are modeled as ideal gases of massive particles with a bath hamiltonian given by @xmath0 with operators @xmath1 and @xmath2 creating and annihilating a particle with momentum @xmath3 and energy @xmath4 in reservoir @xmath5 . note that we use natural units throughout this paper , i.e. , @xmath6 . in the weak coupling limit the reservoirs enter additively , such that it suffices to consider here just a single reservoir . therefore , we will drop the reservoir index @xmath5 in this section . the mean occupation of the @xmath3-th energy level of an ideal quantum gas is given by @xmath7 } $ ] , where @xmath8 corresponds to a bose gas and @xmath9 corresponds to a fermi gas . here , we have introduced the inverse temperature @xmath10 and the chemical potential @xmath11 . since the lowest energy level of a free quantum gas is given by @xmath12 the positivity of the mean occupation demands that for bosons , in contrast to the electronic case , the chemical potential is restricted to negative values @xmath13 . the exact value of the chemical potential in the grand canonical ensemble is determined by the condition that the mean total number of particles @xmath14 is constant . we assume that the gas is confined in a three - dimensional ( @xmath15d ) cuboid of volume @xmath16 with periodic boundary conditions . in the thermodynamic limit where @xmath17 , @xmath18 with @xmath19const , the summation is replaced by an integral @xmath20 with the density of states for an ideal , non - degenerate quantum gas given by @xmath21 . here , @xmath22 is the spin degeneracy coefficient . however , when replacing the sum by an integral , we need to take extra care of the ground - state occupation in the bosonic case , since in the regime where @xmath23 it can be macroscopically occupied . this phenomenon does not occur in a fermi gas due to the pauli principle and is known as bose - einstein condensation @xcite . keeping this in mind the mean total particle density is given by @xmath24 where we introduced the thermal de broglie wavelength @xmath25 , the fugacity @xmath26 , and the poly - logarithm @xmath27 @xcite . the explicit ground - state contribution @xmath28 is only present in bose gases . this equation implicitly defines the chemical potential @xmath29 as a function of temperature and mean particle density . due to the complexity of the poly - logarithm , the chemical potential can not be determined analytically and one has to use numerical methods . the results for the chemical potential of a fermi and a bose gas are depicted in fig . [ f : chemicalpotential ] . fortunately , all other quantities of interest can be expressed in terms of the chemical potential , leaving it to be the only numerical problem . of special interest for the further calculations are the first derivatives of the chemical potential in the thermal phase which can be calculated from eq . to @xmath30 in a real experiment with cold atoms , the chemical potential can not be tuned directly by applying an external voltage as usually considered for electronic transport . instead one can introduce a thermal or density bias which causes a bias in the chemical potentials of the reservoirs . from the left panel in fig . [ f : chemicalpotential ] we can see that applying a positive temperature bias at constant density where @xmath31 results in an opposite chemical potential bias @xmath32 . the same effect occurs for a density bias at constant temperature . furthermore , if the particle density is kept constant , one finds characteristic temperatures for the respective ideal quantum gases . for a fermi gas , one defines the fermi temperature @xmath33 that relates to the fermi energy @xmath34 which equals the chemical potential at absolute zero , i.e. , @xmath35 . for a bose gas one finds a critical temperature @xmath36 where the chemical potential vanishes and all particles start to condense in the same ground state . thus , for temperatures below the critical point the bose gas is in a mixed phase consisting of a normal thermal phase and a condensate fraction . when the temperature is absolute zero all particles occupy the ground state and the gas forms a pure bose - einstein condensate . these characteristic temperatures are defined by @xmath37 where @xmath38 is the riemann zeta function . analogous to the situation with constant density , we can also consider the case that the temperature is constant . this allows us to define a critical density @xmath39 for bosons and a fermi - density @xmath40 for fermions according to @xmath41 in both cases , the chemical potential can be treated as a function of a single dimensionless variable @xmath42 , @xmath43 or @xmath44 , @xmath45 for ideal bose and fermi gases , respectively . ( dashed line ) , @xmath46 ( solid line ) and @xmath47 ( dotted - dashed line ) . using a constant chemical potential , e.g. , @xmath48 ( thin , dotted line ) the occupation increases exponentially with the temperature . _ right _ : for an ideal fermi gas at @xmath49 ( dashed line ) , @xmath50 ( solid line ) and @xmath51 ( dotted - dashed line ) . the fermionic occupation for a constant chemical potential approaches @xmath52 for large temperature . this value is approached either from above if @xmath53 , e.g. , @xmath54 ( orange , dotted line ) or from below if @xmath55 , e.g. , @xmath56 ( green , dotted line ) . the circles indicate a set of temperatures with the same occupation for a given transition frequency [ compare with figs . ( [ f : fermicurrent ] ) and ( [ f : bosecurrent ] ) ] . ] inserting the temperature- and density - dependent chemical potential into the definition of the bosonic and fermionic mean occupations , we find a temperature dependence as shown in fig . [ f : meanoccupation ] . in the bosonic case , we see that the mean occupation of a given energy level is peaked around a corresponding characteristic temperature . for low temperatures close to zero , the particles occupy lower energy levels and , thus , the mean occupation of the considered energy level decreases . analogously , for high temperatures the particles are excited to higher energy levels and the mean occupation of the considered energy level decreases . in the case of ideal fermi gases we find a similar behavior if the considered energy level lies above the fermi energy ( dotted - dashed line ) however , the situation changes if one considers the occupation of an energy level below the fermi energy ( dashed line ) . here , the occupation becomes constant if the temperature is decreased due to the pauli principle . contrary , in the case of an independent chemical potential , the bosonic mean occupation increases with increasing temperature ( dotted line ) and the fermionic mean occupation approaches the value @xmath52 as the temperature is increased ( dotted line ) . we therefore find that the temperature - dependent chemical potential strongly affects the high - temperature behavior of the mean occupations . are in thermal equilibrium and characterized by a chemical potential @xmath57 that depends on the respective temperature @xmath58 and particle density @xmath59 . the system dynamics is governed by the hamiltonian @xmath60 and the weak system - bath coupling is mediated by tunneling rates @xmath61 . ] we investigate a transport setup as depicted in fig . [ f : generaltransportscheme ] with two reservoirs in thermal equilibrium described by @xmath62 and @xmath63 . these reservoirs coupled to the transport system are denoted by the labels @xmath64 and @xmath65 . we assume that the system - bath coupling is weak such that we can use the born - markov - secular approximation ( bms ) @xcite . starting from the von - neumann equation this formalism allows to extract a quantum master equation that for non - degenerate system - energy eigenvalues assumes the form of a rate equation for the reduced system density matrix @xmath66 in the system - energy eigenbasis @xcite . for sequential particle tunneling we can uniquely identify the jump terms in the master equation which enables one to convert it into a conditional master equation this master equation is conditioned on the number @xmath63 of particles tunneled via one reservoir into or out of the system and the amount of energy @xmath67 transfered from this reservoir into the system . due to conservation laws we just need to consider one transport channel . therefore , without loss of generality we will focus on the left reservoir only . the corresponding conditional master equation reads as @xmath68 here , the super - operators @xmath69 , @xmath70 and @xmath71 are acting on the reduced system density matrix with @xmath69 describing the internal dynamics and @xmath70 and @xmath71 describing jumps out of and into the system with transferred energy @xmath72 , respectively . this particle - number and energy resolved master equation can also be established using virtual detectors as bookkeeping operators @xcite . subsequently , we perform a fourier transformation @xmath73 $ ] which introduces a particle counting field @xmath74 @xcite and an energy counting field @xmath75 @xcite for the left reservoir . the resulting liouville super operator for the left reservoir becomes a function of these counting fields and reads as @xmath76 from this liouvillian together with the normalization condition @xmath77 , one can uniquely determine the steady - state reduced system density matrix @xmath78 by solving the equation @xmath79 . subsequently , the steady - state particle current @xmath80 and energy current @xmath81 for the left reservoir are obtained by @xmath82 the particle and energy conservation implies the relations @xmath83 and @xmath84 for the currents measured at the left and right reservoirs , respectively . note , that the currents @xmath80 and @xmath81 are defined as positive if the corresponding flow is directed from the left reservoir to the right reservoir . furthermore , we would like to point out that in real experiments with ultracold atoms the reservoirs contain a finite number of particles and energy only . therefore , a flow through the system would on longer time scales ( that may , however , be experimentally relevant ) lead to an equilibration of the reservoirs . assuming that the change of the temperature and density in each reservoir are both linearly related to the respective heat current @xmath85 and particle current @xmath80 , i.e. , @xmath86 where @xmath87 denotes the heat capacity , one may establish a closed set of equations for @xmath88 , @xmath89 , and @xmath90 . these also determine the chemical potentials of the reservoirs @xmath91 . for the case of temperature - independent chemical potentials , this method has already been applied @xcite . for the considered transport setup we calculate the entropy production following the approach outlined in ref . we rewrite the time derivative of the shannon entropy of the system @xmath92 as a sum over an internal entropy production @xmath93 and an entropy flow @xmath94 from the system to the environment . if the system is in its steady state , @xmath95 vanishes and the internal entropy production of the system is given by the negative entropy flow to the environment @xmath96 . the steady - state entropy flow is given by @xmath97 where @xmath98 and @xmath99 denote the energy- and particle - currents from reservoir @xmath5 into the system , respectively . using the conservation of particle number and energy , we obtain the entropy production @xmath100 where we introduced the discrete affinities @xmath101 and @xmath102 . in the following , we assume a small thermal- and chemical potential bias between the reservoirs such that @xmath103 , @xmath104 and @xmath105 , @xmath106 . linearizing eq . around the equilibrium , i.e. , @xmath107 , and collecting the terms associated to the affinity of the same intensive parameter yields the linear response entropy production as a function of the temperature and chemical potential @xmath108 here , the quantity @xmath109 is the usual heat current . furthermore , we assume that the chemical potential is not an independent parameter but a function @xmath110 of the temperature and the particle density in the reservoir . hence , the chemical potential bias in eq . has to be replaced by the linearized expression @xmath111 and we find the linear response entropy production as a function of the temperature and particle density @xmath112 here , we introduced the generalized heat current @xmath113 which corresponds to the conventional heat current @xmath85 with a modified chemical potential . this modified chemical potential correctly describes the classical limit : for high temperatures , it assumes the value @xmath114 , which is the classical amount of heat per particle in three dimensions . thus , we find that the constraint of a constant particle density in the reservoirs leads to a modification of the chemical potential in the heat current and the emergence of a density driven particle current . both of these contributions arise from the linearized affinity . they can be interpreted as the work one needs to perform on a particle which is traveling through the transport setup , in order to overcome the chemical potential bias caused by the thermal or density bias , respectively . it is well known that the onsager theorem @xcite is very useful for describing linear , purely resistive systems . this theorem has been analyzed and proven to be also valid for open quantum systems @xcite . in particular , the onsager theorem holds for open quantum systems which can be described by markovian master equations @xcite . these are the quantum mechanical analog to purely resistive classical systems , i.e. , systems without memory . within this section we demonstrate the validity of this theorem and extract the linear response transport coefficients . in order to appropriately describe an irreversible transport process one rewrites the entropy production as a sum @xmath115 over generalized fluxes @xmath116 and their corresponding affinities @xmath117 @xcite . the linear response entropy production and , hence , the fluxes and affinities , will be different depending on whether we assume a constant particle density or not . to compare these two situations , we first consider the case without density constraint , i.e. , for an independent chemical potential , and afterwards analyze the case with constant particle density . first , we review the usual electronic transport approach with an independent chemical potential . here , the entropy production is given by eq . and , thus , the generalized currents are given by @xmath118 with the corresponding affinities @xmath119 linearizing these currents with respect to their respective affinities around the equilibrium ( @xmath120 , @xmath121 ) yields an onsager system in the form @xmath122 where the entries of the onsager matrix @xmath123 with constant chemical potential are defined as derivatives evaluated at the equilibrium values , i.e. , @xmath124 . these so - called kinetic coefficients fulfill the onsager reciprocal relation @xmath125 which is related to the time - reversal symmetry of physical laws @xcite . furthermore , the onsager matrix is positive definite which guarantees the positivity of the entropy production in accordance with the second law of thermodynamics . from the onsager system , one can subsequently extract the linear transport relations for different setups . if no thermal bias is present , i.e. , @xmath126 , one finds ohm s law @xmath127 with the electronic conductance @xmath128 . similarly , one finds fourier s law @xmath129 for a thermocouple under the constraint of vanishing particle current @xmath130 . this defines the linear heat conductance @xmath131 where @xmath132 is the determinant of the onsager matrix . additionally , such a system can produce a potential bias @xmath133 as a response to a thermal bias at vanishing particle current . this so - called seebeck effect is characterized by the thermopower @xmath134 . the reverse process where a thermal bias is created by applying a bias in the chemical potentials is known as peltier effect which is characterized by the peltier coefficient @xmath135 . the efficiency of these processes can be characterized by the dimensionless figure - of - merit @xmath136 with the lorenz number @xmath137 defined by the wiedemann - franz law @xcite . in analogy to the discussion in the previous paragraph , we now focus on the situation where @xmath62 and @xmath63 are independent variables . when the temperature and particle - density in the reservoirs are held at constant differences , the entropy production is given by eq . with generalized currents @xmath138 and their respective affinities @xcite @xmath139 the corresponding linearized onsager system reads as @xmath140 where the kinetic coefficients @xmath141 are now functionals of the chemical potential @xmath110 . due to the linearity of the system of equations , we can find a linear mapping which transforms the onsager matrices in eqs . ( [ e : usualonsagersystem ] ) and ( [ e : onsagersystem ] ) into each other ( see appendix [ a : onsagermatrices ] ) . hence , we can rewrite the matrix @xmath142 using the kinetic coefficients defined in eq . which now become functionals of the temperature- and density - dependent chemical potential , i.e. , @xmath143 $ ] . this yields @xmath144 where the onsager reciprocal relation is preserved , i.e. , @xmath145 . from the above equation , we derive linear transport coefficients analogous to the electronic case . we see that in the absence of a thermal bias @xmath146 , the particle current @xmath147 becomes proportional to the applied density bias . this yields an equation similar to ohm s law @xmath148 with an isothermal matter conductance @xmath149 given by @xmath150 , \;\textrm{for}\;\;\delta_t = 0.\label{e : matterconductance } } \end{aligned}\ ] ] in a similar way to the matter conductance , we can extract the analog of the thermal conductance @xmath151 from the modified fourier s law @xmath152 under the constraint @xmath153 which yields @xmath154.\label{e : thermoconductance } } \end{aligned}\ ] ] here , @xmath155 is the determinant of the onsager matrix @xmath142 . note that this transport coefficient vanishes if the determinant is zero . in general , this happens in the so - called tight - coupling limit where the energy current @xmath81 becomes proportional to the particle current @xmath80 and , hence , the generalized fluxes @xmath156 and @xmath157 are not independent of each other @xcite . furthermore , we find that a vanishing particle current @xmath158 for finite thermal and density bias implies @xmath159 . therefore , such a thermodynamic device produces a density - induced chemical potential bias as a response to a thermal bias . this allows us to define @xmath160 analogous to the thermopower by @xmath161 + \beta^2 \frac{\partial\mu}{\partial\beta}.\label{e : thermopower}}\end{aligned}\ ] ] this coefficient characterizes the linear density response to a temperature difference at vanishing particle current . it is related to the analog of the peltier coefficient @xmath162 by the thomson relation @xmath163 . using these transport coefficients , we can calculate the dimensionless figure - of - merit @xmath164 @xcite which characterizes the efficiency of the thermodynamic device . it is given by @xmath165 + \frac{\beta^2 \frac{\partial\mu } { \partial \beta } \left(\beta^2 \frac{\partial\mu } { \partial \beta } + 2 \sigma[\mu(t , n ) ] \right)}{l[\mu(t , n)]},\label{e : figureofmerit } } \end{aligned}\ ] ] with the modified lorenz number @xmath166 . from the definitions ( [ e : matterconductance ] ) - ( [ e : figureofmerit ] ) we see that for an independent chemical potential , where the derivative with respect to temperature vanishes , i.e. , @xmath167 , we recover the usual linear response transport coefficients . are in thermal equilibrium characterized by temperature @xmath58 and chemical potential @xmath168 for fixed particle number @xmath169 . the mean occupation @xmath170 of the transition energy @xmath72 is given by the fermi - dirac distribution . the system is composed of a double quantum dot in the coulomb blockade regime with two transition energies @xmath171 and @xmath172 only . ] as a first example , we consider a fermionic system as shown in fig . [ f : fermimodel ] which is composed of a double quantum dot in the coulomb blockade regime and two fermionic terminals . the system is described by the hamiltonian @xmath173 here , the operators @xmath174 and @xmath175 which obey the fermionic anti - commutation relation @xmath176 annihilate and create a fermion particle in quantum dot @xmath5 , respectively . these two quantum dots are labeled by @xmath64 and @xmath65 and they are coupled via a coherent tunneling process with amplitude @xmath177 . each dot can be empty or occupied by a single particle increasing the system energy by @xmath178 . in the coulomb blockade limit the coulomb repulsion @xmath179 is the dominating energy scale . hence , the state corresponding to a doubly occupied double quantum dot does not take part in the long - time dynamics and can be safely neglected . the remaining energy eigenstates of the system are the vacuum state @xmath180 and the superposition states @xmath181 and @xmath182 with eigenvalues @xmath183 , @xmath171 and @xmath172 , respectively . the system is coupled to the reservoirs by the system - bath interaction hamiltonian @xmath184 where the tunneling amplitude of a particle hopping from the reservoir @xmath5 into the respective quantum dot or vice versa is proportional to @xmath185 and @xmath186 , respectively . we start by calculating the steady - state particle and energy currents according to eq . ( see appendix [ a : fermionicliouvillian ] ) . this yields for the steady - state current measured at reservoir @xmath5 the relations @xmath187 where we defined the abbreviation @xmath188 } \bar\rho_{n}\right\},}\end{aligned}\ ] ] with the steady - state density vector @xmath189 . since the complete expression is too long we state here the particle current in the limit of a single transition frequency only . this limit can be obtained by shifting the second transition energy to high values such that transport through this level is strongly suppressed . in consequence , we find a current involving the lowest transition energy only which reads as @xmath190,\label{e : fermisinglefrequencycurrent } } \end{aligned}\ ] ] with the effective coupling rate @xmath191 . thus , the particle current through a system with transition energy @xmath192 is proportional to the difference of the mean occupations of the corresponding energy level in the reservoirs . the particle current through a double - dot system with two transition frequencies is shown in fig . [ f : fermicurrent ] . ( a ) , @xmath193 ( b ) , and @xmath194 ( d ) and used the same tunneling amplitude @xmath195 . in plot ( c ) , we used @xmath196 and set the tunneling amplitude to @xmath197 . for all plots , the rates are set to @xmath198 . the circles in the plots ( b ) and ( c ) correspond to the set of temperatures marked in fig . [ f : meanoccupation ] . the dashed curves indicate a vanishing of the corresponding energy current @xmath81 . ] we observe two different regimes reflecting the different behavior of the mean occupations for energies below and above the fermi energy as shown in fig . [ f : meanoccupation ] . if at least one transition energy lies below the fermi energy , as shown in fig . [ f : fermicurrent](a ) , we observe a finite steady - state current against the thermal bias . this is due to the fact that in the hotter reservoir the particles are excited to higher energy - states . since the density is fixed , there are not enough particles to refill the depleted energy levels . the occupation in these levels decreases leading to a flow from the colder reservoir , where the energy levels are occupied , to the hotter reservoir . this behavior is a consequence of the mean occupation under the restraint of constant particle density . if all transition energies are above the fermi energy , as shown in figs . [ f : fermicurrent](b ) and [ f : fermicurrent](c ) , this behavior changes such that for a small thermal bias the steady - state current flows with the bias . however , if the thermal bias is increased above a critical value the current flows against the bias again . moreover , there is always a finite steady - state current for an arbitrary high thermal bias although it is exponentially suppressed away from the optimal temperature . taking a look at eq . we see that the critical lines where the current vanishes are defined by the relation @xmath199 . this is trivially fulfilled in equilibrium where @xmath200 . away from equilibrium we find that the particle current only vanishes if the mean occupation of a given transition energy in the reservoirs takes on the same value for different temperatures . comparing with the result presented in fig . [ f : meanoccupation ] we can immediately deduce that this condition can only be satisfied for transition frequencies above the fermi energy . in this case one always finds a set of two different temperatures for the left and right reservoirs where the current vanishes . as an example , we indicated such a set of temperatures in fig . [ f : meanoccupation ] and show the corresponding points in the current plots in figs . [ f : fermicurrent](b ) and [ f : fermicurrent](c ) . in fig . [ f : fermicurrent](c ) , we chose the dot energy @xmath178 and the tunneling amplitude @xmath177 in such a way that the transition energy @xmath201 is shifted to high energies . thus , the corresponding particle current can be approximately described by the single - level limit given in eq . . comparing the currents plotted in figs . [ f : fermicurrent](b ) and [ f : fermicurrent](c ) , we see that for a system with two transition frequencies [ fig . [ f : fermicurrent](b ) ] the line of vanishing particle current it shifted to higher temperatures compared to the effective single - level result [ fig . [ f : fermicurrent](c ) ] . this effect results from the additional transport channel which modifies the condition for a vanishing particle current . in fact , depending on the number and values of the transition frequencies in the system , there can also by more lines where the particle current vanishes [ see panel fig . [ f : fermicurrent](d ) ] . finally , we note that the energy current @xmath81 in general vanishes ( dashed lines ) for different parameters than the particle current @xmath80 . thus , we can observe a finite energy - current even for a vanishing particle - current in a fermionic system with two transition frequencies . moreover , in the upper right corner of fig . [ f : fermicurrent](c ) we even find a regime for high temperatures where the energy current flows against the particle current . in this section we calculate the linear transport coefficients for the fermionic transport setup . for reasons of brevity , we use the wide - band limit with energy independent rates @xmath202 . , conductance @xmath203 _ ( left ) _ and the thermal conductance @xmath151 _ ( right ) _ for ideal fermi gas reservoirs with different on - site energies @xmath204 ( dashed line ) , @xmath205 ( dotted - dashed line ) and @xmath194 ( solid line ) versus the normalized temperature and density , respectively . in both plots we assume equal tunneling rates @xmath206 and a constant coherent coupling strength of @xmath195 . for the conductance with constant chemical potential we use @xmath194 , @xmath195 and additionally set @xmath207 ( dotted line ) . ] we plot some results for these transport coefficients in fig . [ f : fermiconductance ] and fig . [ f : fermiseebeck ] . in all these plots we analyze three different transport channel configurations . the first configuration corresponds to a system with transition energies below the fermi energy of the reservoirs ( dashed line ) . the second configuration corresponds to a system with transition energies above the fermi energy of the reservoirs ( dotted - dashed line ) . in the third configuration we analyze a system whose lowest transition energy is exactly equal to the fermi energy of the reservoirs ( solid line ) . the results in these situations are discussed in more detail within the following subsections . calculating the matter conductance in the wide - band limit according to eq . yields the relation @xmath208 \left[1-\bar{n}\left(\omega _ + \right)\right ] \left[\bar{n}\left(\omega _ -\right)+ \bar{n}\left(\omega _ + \right)\right]}{2 t \left[1-\bar{n}\left(\omega _ -\right ) \bar{n}\left(\omega _ + \right)\right]}.\label{e : fermidiffusion } } \end{aligned}\ ] ] for @xmath209 , i.e. , in the limit of a single transition frequency @xmath210 , this expression coincides with @xmath211 } \right ) } $ ] , the well - known coulomb blockade conductance peak for a single resonant level @xcite . similarly , in the limit @xmath212 , i.e. , @xmath213 the matter conductance approaches a limiting value . however , at @xmath214 the energies @xmath215 are degenerate and our rate equation approach can not be applied . furthermore , the whole temperature dependence of the chemical potential enters implicitly via the mean occupations of the energy levels in the reservoirs in compliance with eq . . in the left panel of fig . [ f : fermiconductance ] we plot the matter conductance as a function of the normalized temperature for different on - site energies and a constant tunneling amplitude . for a configuration where both transition energies lie below the fermi energy of the reservoirs ( dashed line ) we observe a maximal conductance at a specific temperature which basically depends on the frequency @xmath178 . decreasing the temperature further diminishes the conductance as the respective energy levels in the reservoirs become occupied and the current decreases . we observe a similar behavior for a configuration where both transition energies lie above the fermi energy of the reservoirs ( dotted - dashed line ) . in this situation the conductance vanishes for decreasing temperature due to the fact that the transition energy in the reservoirs gets exponentially depleted . finally , we show the result for a configuration where the lower transition energy equals the fermi energy of the reservoirs ( solid line ) . only in this case we observe a non - vanishing conductance as the temperature approaches @xmath216 . this is due to the fact that the fermi energy level in the reservoirs is at most half filled whereas all other energy levels are either completely filled or empty . therefore , particle transport is possible even for low temperatures . however , the observation that in this case the matter conductance @xmath149 diverges like @xmath217 is unphysical . this behavior stems from the fact that the born - markov - secular master equation breaks down if @xmath218 . in all three situations the conductance vanishes for increasing temperature due to the reduction of the occupation of the energy levels in the reservoirs . for comparison , we additionally plotted the conductance @xmath203 for a constant chemical potential ( dotted line ) . in this situation , we find qualitatively the same behavior as for the modified matter conductance . for low temperatures , the respective energy levels in the reservoirs are depleted and , hence , the conductance vanishes . contrary for high temperatures they are equally filled which leads to a vanishing net current . thus , the modified matter conductance @xmath219 basically follows @xmath203 . however , the high - temperature behavior is changed due to the temperature dependence of the chemical potential . from eq . we find that the heat conductance in the wide - band limit is given by @xmath220 ^ 2}\tilde\sigma.\label{e : fermithermalconductance } } \end{aligned}\ ] ] this expression has no explicit dependence on the chemical potential and , thus , formally corresponds to the thermal conductance for an independent chemical potential as shown in sec . [ s : onsagerdependent ] . furthermore , we immediately see that the above equation obeys the wiedemann - franz law , i.e. , @xmath221 with the dimensionless lorenz number @xmath222 . in the limit of a single transition - frequency , i.e. , @xmath223 , the heat conductance @xmath224 vanishes trivially because there is no pure heat flow through a single level without particle flow . in the right panel of fig . [ f : fermiconductance ] we plot the modified thermal conductance versus the inverse normalized particle density for different on - site energies and different tunneling amplitudes according to eq . . for all considered configurations , we observe qualitatively the same behavior . the heat transport is maximal at a characteristic density . this maximum is shifted to higher densities with increasing transition energies . for low densities , the thermal conductance vanishes as the reservoir energy levels become less occupied . for high densities , the heat conductance vanishes because the transition energies in both reservoirs become maximally occupied and the matter conductance vanishes . and conventional thermopower @xmath225 ( _ left _ ) and the figure - of - merit for dependent and constant chemical potential ( _ right _ ) versus the normalized temperature for an ideal fermi gas and different symmetric on - site energies @xmath204 ( dashed line ) , @xmath194 ( solid line ) and @xmath205 ( dotted - dashed line ) . for @xmath194 and a constant chemical potential @xmath207 ( thin , dotted line ) the thermopower approaches @xmath216 for large t and the figure - of - merit takes on a constant value depending on the system transition frequencies . in both plots , we fixed the coherent tunneling amplitude at @xmath195 . ] next , we determine the analog to the thermopower defined in eq . . for our setup we find the expression @xmath226 where we defined the average energy @xmath227 here , the expression @xmath228 describes the amount of energy one particle traveling through a transport system with only a single transition - energy @xmath229 takes from one reservoir to the other . we can formally recover the result for the conventional thermopower with an independent chemical potential by setting @xmath230 . in the left panel of fig . [ f : fermiseebeck ] we plot the temperature dependence of the modified thermopower @xmath231 for different transition energies in the case of fermi reservoirs according to eq . . when the temperature decreases we observe different behavior for the modified thermopower depending on the transport system . if the lower transition energy is below the fermi energy the modified thermopower remains positive for all temperatures . as the temperature approaches zero the average energy approaches a constant positive value and , thus , the modified thermopower diverges like @xmath232 . if the transition energies are above the fermi energy we observe a similar behavior but the coefficient becomes negative since below some critical temperature the transition frequencies exceed the chemical potentials . only for the case when the lower transition energy equals the fermi energy the average energy and hence the modified thermopower vanishes at @xmath233 . for high temperatures the average energy @xmath234 is dominated by the classical thermal energy contribution of @xmath235 per particle . therefore , we observe that the modified thermopower assumes a constant positive value of @xmath236 for high temperature independent of the respective transition energies . this behavior can not be predicted if the chemical potential is treated as an independent parameter . in this conventional approach the thermopower vanishes for high temperatures independent of the transition frequencies ( dotted line ) . furthermore , the conventional thermopower never changes its sign as a function of the temperature . in this situation the sign is fixed by the choice of the constant chemical potential . finally , we analyze the figure - of - merit for the thermodynamic device . this coefficient is defined in eq . which relates to the efficiency of the device . for the considered fermionic setup the figure - of - merit reads as @xmath237 ^ 2 { \left<\phi_f(\omega_-,\omega_+)\right>}^2}{\bar{n}\left(\omega _ -\right)\bar{n}\left(\omega _ + \right ) \left(\omega _ --\omega _ + \right)^2}.\label{e : fermizt } } \end{aligned}\ ] ] in the right panel of fig . [ f : fermiseebeck ] we plot the linear response figure - of - merit for different on - site energies . we find that the figure - of - merit increases exponentially for high temperatures independent of the transition frequencies . in contrast , for a constant chemical potential the figure - of - merit approaches a constant value for high temperatures . additionally , we see that there are specific temperatures where the figure - of - merit vanishes . these are the temperatures for which the linear response particle current vanishes and , hence , no power can be extracted . if the temperature approaches zero , we observe that the figure - of - merit increases again except for the situation where the transition frequencies lie below the fermi energy of the reservoir . in this case , the figure - of - merit assumes a finite value as the temperature approaches zero . this is again caused by the fact that the relevant energy levels in the reservoirs are occupied for low temperatures . however , although the conversion of energy into particle currents seems to be very efficient for low and high temperatures , this picture might be misleading since in these regimes the actually generated particle current is exponentially suppressed and so is the power . thus , one has to look out for a high efficiency at maximum power @xcite . it turns out that for the specific situations presented in fig . [ f : fermiseebeck ] the currents become maximal in the interval @xmath238 $ ] ( not shown ) and , thus , we find that at maximum power the figure - of - merit for the configuration with the transition energies above the fermi energy is largest with @xmath239 . least efficient is the configuration with the lowest transition frequency equal to the fermi energy . here , only a figure - of - merit at maximum power of @xmath240 is reached . whereas for the configuration with both energies below the fermi energy we find @xmath241 at maximum power . usually , in experiments figure - of - merits of about @xmath242 and higher are considered as efficient . of course , in our model we assume ideal quantum gases and , thus , the calculated figure - of - merit is probably overestimated . however , from our results we argue that by optimizing the parameters efficiencies at maximum power close or even equal to the optimum are possible . the optimum efficiency in linear response theory is given by half the linear response carnot efficiency @xmath243 @xcite . and chemical potential @xmath168 for fixed particle number @xmath169 . the mean occupation @xmath170 of the energy level @xmath72 is given by the bose - einstein distribution . the system is described by a harmonic oscillator of frequency @xmath244 with an additional interaction energy @xmath245 if two particles are present . ] motivated by the results presented in sec . [ s : atomicreservoirs ] we now focus on a transport setup involving bosonic reservoirs . contrary to the fermionic setup , here , we expect that the critical behavior of the reservoirs leads to characteristic signatures in the transport properties . therefore , we analyze the transport characteristics of a bosonic transport system as shown in fig . [ f : bosemodel ] . the bosonic transport system is described by the hamiltonian @xmath246 each particle that is added to or removed from the system changes the system energy at least by a constant amount @xmath244 . if more than one particle is present in the system , these particles interact with a two - body interaction strength @xmath245 and , thus , increase the total energy of the system . this system hamiltonian is diagonal in the fock basis , i.e. , @xmath247 , with energy eigenvalues @xmath248 . due to the interaction this energy spectrum is nonlinear and the system generates many non - equivalent transport channels . the system - bath interaction hamiltonian reads as @xmath249 where the tunneling amplitude of an atom hopping from the reservoir @xmath250 into the system or vice versa is proportional to @xmath185 and @xmath186 , respectively . in the thermodynamic limit , the steady - state currents through this system measured at reservoir @xmath5 are given by a sum over all possible system occupations @xmath251 where we defined the abbreviation @xmath252 } \bar\rho_{n+1}\notag\\ & -\gamma_\alpha(\omega_n-\omega_{n+1})\bar n_\alpha(\omega_n-\omega_{n+1 } ) \bar\rho_n . } \end{aligned}\ ] ] because of the infinite summation , the above expression can not be solved in general . therefore , we truncate the bosonic hilbert space at low particle numbers ( see appendix [ a : bosonicliouvillian ] ) . taking the limit @xmath253 restricts the hilbert space to at most one bosonic particle in the system . thus , the system can be either empty or singly occupied which gives rise to a single transition frequency @xmath244 . in this case the steady - state currents in the wide - band limit @xmath254 can be evaluated to @xmath255 } } { \gamma_{\rm{l } } { \left [ 1 + 2 \bar n_{\rm{l}}(\omega ) \right ] } + \gamma_{\rm{r } } { \left [ 1 + 2 \bar n_{\rm{r}}(\omega ) \right ] } } , \label{e : singlewsscurrentbose } } \end{aligned}\ ] ] which coincides with the result found in ref . @xcite . ( _ left _ ) and @xmath256 ( _ right _ ) . in both plots , we set @xmath257 and @xmath198 . the dashed lines indicate a vanishing of the corresponding energy current @xmath81 . the solid lines signal the phase transition to a bose - einstein condensate in the left and right reservoirs . the circle insets correspond to fig . [ f : meanoccupation ] ] analogous to the fermionic case we find that the current through a transport channel with energy @xmath244 is proportional to the difference in the corresponding mean occupations in the left and right reservoir . as an example we plotted the steady - state particle current for different transition energies in fig . [ f : bosecurrent ] . first , we observe that the current is strongest if the interaction strength is weak . increasing the interaction strength shifts the corresponding transport channel to higher energies which are less occupied in the reservoirs . therefore , the contribution of these transport channels to the current is diminished . additionally , we see two lines where the steady - state particle current vanishes . the diagonal line reflects the thermodynamic equilibrium , i.e.if @xmath258 . the reason for the emergence of the second line lies in the temperature dependence of the mean occupations as discussed for the fermionic setup in sec . [ s : fermisteadystatecurrent ] . for the energy current ( not shown ) , we find qualitatively the same behavior as for the particle current . however , depending on the system parameters the energy current can be finite even for vanishing particle current . we indicated the temperatures where the energy current vanishes by dashed lines in both plots of fig . [ f : bosecurrent ] . we observe that the nonequilibrium lines where the energy current vanishes are shifted to higher temperatures compared to the vanishing particle current . contrary to the fermionic steady - state current plotted in fig . [ f : fermicurrent ] , we do not observe a qualitative change in the bosonic particle current in dependence of the transition energies . this behavior stems from the fact that there is no equivalent of the fermi energy and no pauli exclusion principle in bosonic systems . hence , the bosonic mean occupations look qualitatively the same for all energy levels ( see fig . [ f : meanoccupation ] ) . within the condensate phase , we observe a finite particle - current which results from the thermal fraction of the bose gas . this thermal fraction decreases with temperature like @xmath259 and , therefore , the current exactly vanishes at @xmath233 . in our bms master - equation approach , the coherences decouple from the occupations and , thus , can be neglected . however , if one enters the condensate phase the coherences between the particles become stronger with decreasing temperature . therefore , the decoupling between coherences and occupations is not expected to hold and the coherences can not be neglected anymore . hence , we do not expect that our results remain valid deep in the condensate phase . in the limit of a single transition - frequency in the system , i.e. , @xmath260 or @xmath261 , the energy current is proportional to the particle current . as shown in sec . [ s : transportcoefficients ] in this situation not all transport coefficients can be calculated . therefore , we consider the case of two transport channels with different energies . this situation is established by truncating the hilbert space at two particles leading to two transitions in the system from zero to one particle , i.e.@xmath262 and from one to two particles , i.e. , @xmath263 . in the following paragraphs , we present the resulting bosonic transport coefficients in the wide - band limit . some results for the bosonic transport coefficients are plotted in figs . [ f : boseconductance ] and [ f : boseseebeck ] . the bosonic matter conductance for the considered transport setup reads as @xmath264 } { \left [ 1+\bar{n}(\omega _ 1)+2 \bar{n}(\omega _ 2 ) \right ] } } { t { \left\ { 1+\bar{n}\left(\omega _ 2\right)+\bar{n}\left(\omega _ 1\right ) { \left [ 2 + 3\bar{n}\left(\omega _ 2\right ) \right ] } \right\ } } } .\label{e : bosediffusion } } \end{aligned}\ ] ] ( dashed line ) , @xmath265 ( solid line ) and @xmath266 ( dotted - dashed line ) versus the inverse normalized temperature . the transition energy is fixed to @xmath267 . for a constant chemical potential @xmath268 ( thin , dotted line ) where we use @xmath269 , the matter conductance becomes constant for high temperatures . _ right _ : thermal conductance for the ideal bose gas reservoirs plotted for different on - site interaction strength @xmath270 ( dashed line ) , @xmath265 ( solid line ) and @xmath271 ( dotted - dashed line ) versus the inverse normalized density . the single - particle energy is @xmath272 and the rates are fixed to @xmath198 . ] in the left panel of fig . [ f : boseconductance ] , we plot this transport coefficient for different transition energies versus the normalized temperature . depending mostly on the lowest transition frequency @xmath244 , the matter conductance has a maximum value at a finite temperature above the critical value @xmath273 . decreasing the lower transition frequency shifts the maximum closer to the critical temperature whereas increasing the transition frequency shifts the maximum away from the critical temperature . the maximum can never lie below the critical temperature since there the chemical potential vanishes and the number of thermal particles which can contribute to the particle current decreases . increasing the temperature leads to a decrease of the matter conductance since the occupation of the transition energy level in the reservoirs is reduced . the influence of the second transport channel is mainly reflected in the maximum value of the transport coefficient . this value is increased if the transport channels are close together , i.e. , if the interaction strength @xmath245 is small . if the interaction strength is increased the respective transport channel is shifted to higher energies and contributes less to the current because of the lower occupations in the reservoir . thus , the maximum conductance decreases to a minimum value resulting from the lower transport channel ( dotted - dashed line ) . for comparison we also included a plot for the conductance with constant chemical potential @xmath268 ( dotted line ) . here , the conductance takes on a constant finite value @xmath274 in the limit of high temperature . this is caused by the fact that for a constant chemical potential the occupations of energy levels in the reservoirs increase linearly with the temperature in the high - temperature limit . if the temperature approaches zero , the conductance vanishes due to the depletion of the transition energy - levels in the reservoirs . the bosonic heat conductance for the considered transport setup reads as @xmath275 \bar{n}\left(\omega _ 2\right ) \left(\omega _ 1-\omega _ 2\right)^2}{t \left[1+\bar{n}\left(\omega _ 1\right)+2 \bar{n}\left(\omega _ 2\right)\right]^2 } \tilde\sigma.\label{e : bosethermalconductance } } \end{aligned}\ ] ] in the right panel of fig . [ f : boseconductance ] , we plot the thermal conductance for different transition energies versus the normalized density . we observe that this transport coefficient increases with increasing density and reaches a maximum value at the critical density when bose - einstein condensation sets in . the value of the maximum depends on the transition energies of the system . in general , there is a finite interaction strength that maximizes the heat conductance . for a low interaction - strength , the heat conductance is strongly decreased since it is proportional to @xmath276 . for a high interaction - strength , the heat conductance is also diminished because the occupation of the upper transition energy is decreased . for densities above the critical value , the thermal conductance remains constant since the thermal gas fraction in this phase is independent of the density and depends on the temperature only . all additional particles occupy the reservoir ground state and , thus , do not contribute to the currents . the analog to the thermopower for the considered bosonic transport setup in the wide - band limit looks formally the same as for the fermionic case @xmath277 however , here appears the bosonic average energy which we define as @xmath278 \phi\left(\omega _ 1\right)+2 \bar{n}\left(\omega _ 2\right ) \phi\left(\omega _ 2\right)}{1+\bar{n}\left(\omega _ 1\right)+2 \bar{n}\left(\omega _ 2\right)},\label{e : bosemeantransportedenergy } } \end{aligned}\ ] ] where we used the expression defined in eq . . in the left panel of fig . [ f : boseseebeck ] , we plot the temperature dependence of this transport coefficient for different values of the interaction strength . analogously to the fermionic case , we find that the modified seebeck coefficient takes on a finite positive value in the high temperature limit where the average transported energy become @xmath235 . when the temperature is lowered , the modified thermopower decreases . at a temperature where the chemical potential contribution starts to dominate over the transport - channel energies , the modified thermopower changes its sign . when the temperature is decreased further , the modified thermopower crosses the critical temperature of the phase transition continuously . however , at the critical temperature the modified thermopower is not analytic . thus , the second derivative with respect to temperature shows a jump when the condensate phase is entered . this behavior is also well known for the heat capacity of the ideal bose gas @xcite . in the condensate phase , the modified thermopower decreases further and diverges like @xmath279 when the temperature is close to absolute zero . in general , the particle current is mainly influenced by the lower transport channel . hence , the modified thermopower just weakly depends on the interaction strength @xmath245 . for high and low values of the interaction strength , the modified thermopower approaches the single - frequency limit result ( dashed line ) . in - between , there is a finite interaction - strength that maximizes the modified thermopower at the critical temperature ( solid line ) . however , the relative increase in the thermopower output is still small . on the contrary , the approach with a constant chemical potential predicts a vanishing thermopower for high temperatures ( dotted line ) . there is no change of sign of the thermopower in dependence of the reservoir temperature . additionally , the conventional thermopower is continuous and differentiable at the critical temperature and , thus , it is not sensitive to the quantum phase transition of the ideal bose gas . finally , we analyze the efficiency of the bosonic thermodynamic device characterized by the figure - of - merit which for the bosonic system reads as ( solid line ) and @xmath266 ( dashed line ) . _ right _ : plot of the figure - of - merit versus the normalized temperature for different interaction strengths @xmath280 ( dashed line ) , @xmath281 ( solid line ) and @xmath282 ( dotted - dashed line ) . for @xmath283 and a constant chemical potential @xmath268 ( thin , dotted line ) , the thermopower vanishes for high temperatures and the figure - of - merit becomes constant . the transition energy is fixed to @xmath267 in all plots . ] @xmath284 ^ 2 { \left<\phi_b(\omega_1,\omega_2)\right>}^2}{2 \left[1+\bar{n}\left(\omega _ 1\right)\right ] \bar{n}\left(\omega _ 2\right ) \left(\omega _ 1-\omega _ 2\right){}^2}.\label{e : bosezt } } \end{aligned}\ ] ] in the right panel of fig . [ f : boseseebeck ] , we plot the temperature dependence of the figure - of - merit for different values of the interaction strength . we observe that the figure - of - merit vanishes for vanishing linear response particle current . at the critical temperature of the phase transition , the figure - of merit is nondifferentiable but continuous . in the limit of high and low temperatures , the figure - of - merit increases exponentially . however , if we compare it with the linear response particle current ( not shown ) , we find that the figure - of - merit at maximum negative current around @xmath285 takes on the values @xmath286 ( dashed line ) , @xmath287 ( solid line ) and @xmath288 ( dotted - dashed line ) in the thermal phase . additionally , the linear response current shows a positive maximum around @xmath289 in the condensate phase , where we find that the figure - of - merit takes on the values @xmath286 ( dashed line ) , @xmath290 ( solid line ) and @xmath291 ( dotted - dashed line ) . for comparison , we also plot the figure - of - merit for a constant chemical potential ( thin , dotted line ) . again , we find that in this case the high - temperature behavior is modified as the figure - of - merit becomes constant . we calculated the steady - state fluxes and affinities from a master equation in born - markov - secular approximation for a general two - terminal transport setup . there , we took into account that in transport experiments with ultracold atoms the chemical potential in general depends on the temperature and the particle density of the reservoirs . we found that the nonlinearity introduced by the temperature- and density - dependent chemical potential strongly influences the steady - state particle and energy currents through the system . depending on the energy spectrum of the system , we could observe multiple regimes where the steady - state currents flow with or against an externally applied thermal bias . this counterintuitive result stems from the temperature- and density - dependencies of the mean occupations in the reservoirs induced by the chemical potential . subsequently , we derived the corresponding onsager system of equations from which we calculated the linear response transport coefficients . comparing the results for fermionic and bosonic transport , we found clear signatures of criticality in the bosonic transport coefficients . thus , transport measurements provide new tools to study critical phenomena in nonequilibrium setups . finally , we investigated the figure - of - merit for the bosonic and fermionic setups . in correspondence with experimental results @xcite , we found that high figures - of - merit at maximum power can be obtained in both systems . this suggests to further investigate transport setups using ultracold atomic gases in view of efficient thermopower devices . financial support by the dfg ( sfb 910 , scha 1646/2 - 1 and grk 1558 ) is gratefully acknowledged . depending on the experimental setup , different intensive state parameters can be held constant . thus , the experimentally controllable affinities change accordingly . for the setups compared within this paper , we focus on situations with constant chemical potential where the affinities are given by @xmath292 and @xmath293 . on the other hand , we analyze a setup with constant particle density where the affinities are given by @xmath294 and @xmath293 . these constraints yield different affinities which in linear response theory can be related to each other according to the linear transformation @xmath295 here , we used the linear expansion of the potential bias @xmath292 with respect to the new affinities @xmath294 and @xmath296 which reads as @xmath297 analogously , we find that the generalized linear fluxes can be transformed according to @xmath298 due to the linearity of the system of equations we can also find a transformation for the onsager matrices themselves . inserting the onsager system together with we find @xmath299 which yields the relation stated in eq . . with this , results obtained for a setup with constant chemical potential can be transformed to the corresponding result for the case of constant particle density . the reduced density matrix of an electronic double quantum dot in the coulomb blockade regime has the diagonal matrix elements @xmath300 , @xmath301 and @xmath302 . using the wide - band limit @xmath202 the conditioned liouvillian in the energy eigenbasis which obeys @xmath303 is given by @xmath304 & - e^{{{\rm{i}}\,}\left(\chi _ { \alpha } -\eta _ { \alpha } \omega _ + \right ) } \left [ 1-\bar{n}_{\alpha } \left(\omega _ + \right)\right ] \\ - e^{-{{\rm{i}}\,}\left(\chi _ { \alpha } -\eta _ { \alpha } \omega _ -\right ) } \bar{n}_{\alpha } \left(\omega _ -\right ) & 1-\bar{n}_{\alpha } \left(\omega _ -\right ) & 0 \\ - e^{-{{\rm{i}}\,}\left(\chi _ { \alpha } -\eta _ { \alpha } \omega _ + \right ) } \bar{n}_{\alpha } \left(\omega _ + \right ) & 0 & 1-\bar{n}_{\alpha } \left(\omega _ + \right ) \end{array } \right ) . \end{aligned}\ ] ] the steady - state vector @xmath189 of this liouvillian is defined by @xmath305 and reads as @xmath306 } { \left [ 1-\bar n_\beta ( \omega_+ ) \right ] } \\ \bar n_\alpha ( \omega_- ) { \left [ 1-\bar n_\beta ( \omega_+ ) \right ] } \\ { \left [ 1-\bar n_\alpha ( \omega_- ) \right ] } \bar n_\beta ( \omega_+ ) \end{pmatrix},\ ] ] where the normalization factor @xmath307 is given by @xmath308 with this result , we can calculate the fermionic energy and particle currents according to eqs . ( [ e : staticcurrentdefinition ] ) and ( [ e : staticenergycurrentdefinition ] ) . in order to solve the master equation for the bosonic system , we have to truncate the respective hilbert space at low particle numbers . for a bosonic system with two transition frequencies only , we truncate the bosonic hilbert space described by the hamiltonian such that at most two particles at a time can be present in the system . then the eigenstates are given by the bosonic fock states @xmath180 , @xmath309 and @xmath310 with the corresponding eigenvalues @xmath183 , @xmath311 and @xmath312 . @xmath314 & 0 \\ e^{-{{\rm{i}}\,}\left(\chi _ { \alpha } -\eta _ { \alpha } \omega _ 1\right ) } \bar{n}_{\alpha } \left(\omega _ 1\right ) & -2 \bar n_\alpha ( \omega_2 ) - \left [ 1+\bar{n}_{\alpha } ( \omega _ 1)\right ] & 2 e^{{{\rm{i}}\,}\left(\chi _ { \alpha } -\eta _ { \alpha } \omega _ 2\right ) } { \left [ 1+\bar{n}_{\alpha } \left ( \omega _ 2 \right ) \right ] } \\ 0 & 2 e^{-{{\rm{i}}\,}\left(\chi _ { \alpha } -\eta _ { \alpha } \omega _ 2\right ) } \bar n_\alpha ( \omega_2 ) & -2 { \left [ 1+\bar{n}_{\alpha } \left ( \omega _ 2 \right ) \right ] } \end{array } \right ) . \end{aligned}\ ] ] the steady - state vector @xmath315 of this liouvillian is defined by @xmath305 and reads as @xmath316 } { \left [ 1+\bar n_\beta ( \omega_2 ) \right ] } \\ \bar n_\alpha ( \omega_1 ) { \left [ 1+\bar n_\beta ( \omega_2 ) \right ] } \\ \bar n_\alpha ( \omega_1 ) \bar n_\beta ( \omega_2 ) \end{pmatrix},\ ] ] where the normalization factor @xmath307 is given by @xmath317 } \right\ } } .}\end{aligned}\ ] ] with this result we can calculate the bosonic energy and particle currents according to eqs . ( [ e : staticcurrentdefinition ] ) and ( [ e : staticenergycurrentdefinition ] ) .

we have made an extensive survey of emission - line stars in the ic 1396 h ii region to investigate the low - mass population of pre - main sequence ( pms ) stars . a total of 639 h emission - line stars were detected in an area of 4.2 deg and their-photometry was measured . their spatial distribution exhibits several aggregates near the elephant trunk globule ( rim a ) and bright - rimmed clouds at the edge of the h ii region ( rim b and sfo 37 , 38 , 39 , 41 ) , and near hd 206267 , which is the main exciting star of the h ii region . based on the extinction estimated from the near - infrared ( nir ) color color diagram , we have selected pre - main sequence star candidates associated with ic 1396 . the age and mass were derived from the extinction corrected color - magnitude diagram and theoretical pre - main sequence tracks . most of our pms candidates have ages of 3 myr and masses of 0.20.6 m . although it appears that only a few stars were formed in the last 1 myr in the east region of the exciting star , the age difference among subregions in our surveyed area is not clear from the statistical test . our results may suggest that massive stars were born after the continuous formation of low - mass stars for 10 myr . the birth of the exciting star could be the late stage of slow but contiguous star formation in the natal molecular cloud . it may have triggered to form many low - mass stars at the dense inhomogeneity in and around the h ii region by a radiation - driven implosion .

1201.2761

as observational instruments have developed , systematic studies of stellar clusters have revealed many low - mass populations of pre - main sequence ( pms ) stars in massive star - forming regions . it has been found that most stars are born in groups or clusters , not in isolated regions ( e.g. , * ? ? ? * ; * ? ? ? stellar clusters provide a tool for investigating stellar evolution , because they are considered to constitute a natural sample of coeval stars at the same distance and with the same chemical composition . young clusters appear to be a snapshot of stellar evolution over a wide range of masses . however , a careful study of the orion nebula cluster @xcite has revealed that the younger stars ( @xmath3 0.3 myr ) are clustered towards the center of the trapezium cluster ( a projected distance of @xmath3 0.3 pc from @xmath5c ori ) , while the older stars are widely distributed . this indicates that star formation has recently occurred in the center of the cluster and that star formation with a low rate has taken place over a long time scale over the whole region . @xcite has suggested that there is no physical reason to separate the orion nebula cluster and the trapezium cluster , and the entire region is a single contiguous star forming event . there are also examples of stellar clusters still producing massive stars . cep a hw2 is the brightest radio source in cep a @xcite , an active star forming region within the molecular cloud associated with the cluster with an age of 5.5 myr ( cep ob3b ; * ? ? ? @xcite found a hot core internally heated by a massive protostar , and a circumstellar rotating disk around hw2 . this suggests that a massive star is still forming in the young cluster . furthermore , x - ray stars with a mid - infrared ( mir ) spectral energy distribution of young stellar objects ( ysos ) around hw2 have been reported by @xcite . it is most likely that a low - mass pms population was already formed before the formation of high - mass stars . once a high - mass star starts to radiate significant amounts of uv radiation , the surrounding material is affected dramatically and the natal molecular cloud is disrupted and the birth of the next generation of stars might be triggered . the apparent age spread observed in the color - magnitude diagram of young stellar groups has been investigated by @xcite . they found that the spread could not be explained adequately by the combination of the binarity , photometric uncertainty and variability of the members . in the case of the @xmath6 ori young stellar group , they suggested an age spread of 24 myr . a possible explanation is that accretion from a disk causes scatter of the stellar continuum light @xcite . however , another possibility is that stars in a cluster are formed over a period of time . three - dimensional numerical simulations including protostellar outflow feedback show that the cluster - forming clumps can maintain a quasi - virial equilibrium state for several periods of free - fall time @xcite , suggesting continuous star formation . @xcite also demonstrated outflow - regulated massive star formation . the episodic star formation causes an intrinsic spread in age and hence in the position in the color - magnitude diagram . we have reported that young stars near the exciting star of the w5e h ii region are systematically older than those near the edge of the h ii region @xcite . we further suggested that the birth of the most massive star occurs in the late stage of star formation in the cluster . although the structure of the h ii region is simple to investigate , the distance of w5 is large , at 2.3 kpc , and therefore it is not easily accessible to the wide range of stellar masses . ic 1396 is a ring - shaped giant h ii region in cep ob2 with a size of about 3@xmath7 , located at the edge of the cepheus bubble , which is a large , far - infrared shell with a diameter of @xmath810@xmath7 @xcite . the structure of ic 1396 is rather simple and is powered by the multiple o star system hd 206267 ( ra = 21@xmath938@xmath1057@xmath1162 , dec = + 57@xmath1229@xmath1320@xmath145 ) , which is located near the geometrical center of ic 1396 in the young cluster trumpler 37 . the bright members of trumpler 37 have been investigated by many authors ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? there are a number of bright - rimmed globules @xcite at its periphery . its modest distance ( @xmath3 1 kpc ) and low foreground extinction ( @xmath15v @xmath3 1 mag ) make ic 1396 a good subject for our study of low - mass population associated with ob stars . the mean distance of 76 members of cep ob2 derived from the hipparcos parallax is 615 pc @xcite . @xcite made a color - magnitude diagram of the intermediate - mass members and fitted them to main - sequence stars at 870 pc . in this paper , we adopt 870 pc , which is frequently quoted by the recent works ( e.g. , * ? ? ? * ) . h@xmath0 emission , a characteristic of t tauri stars ( tts ) , is radiated from the solar - type magnetic activity in weak - lined t tauri stars ( wtts ) or the chromospheric activity and magnetospheric accretion shock in classical t tauri stars ( ctts ) @xcite . thus , a survey of h@xmath0 emission - line stars is a powerful method for searching for young low - mass populations ( e.g. , * ? ? ? recently , nir observations with array detectors and mir observations from space ( e.g. with the _ spitzer _ space telescope ) have advanced the protostar survey in the dense parts of giant molecular clouds . however , the h@xmath0 survey is still useful in areas with moderate or low extinction , and it is also a complementary method to search for pms populations in a wide and rather dispersed region . here , we report an extensive survey of emission - line stars in the ic 1396 h ii region and discuss the formation sequence of low and high mass populations in ic 1396 . slitless grism spectroscopic searches for h@xmath0 emission objects in ic 1396 were conducted in 2004 and between 2006 and 2009 using the wide field grism spectrograph 2 ( wfgs2 ; * ? ? ? * ) at the f/10 cassegrain focus of the university of hawaii ( uh ) 2.2-m telescope on mauna kea . the log of the observations is given in table 1 . a 300-line mm@xmath16 grism was used in conjunction with a wide h@xmath0 filter to isolate the first - order spectra between 62656765 . a tektronix 2048@xmath172048 ccd was used as the detector , yielding a dispersion of 3.8 pixel@xmath16 , a pixel scale of 0@xmath1434 and an instrumental field of view of 11@xmath185 @xmath17 11@xmath185 the whole ic 1396 region was covered by 157 field positions including overlapping areas . about 4.2 deg@xmath1 were observed , both in the grism and the direct imaging mode of wfgs2 . figure 1 shows the boundary of the observed fields and indicates the names of bright - rims , dark clouds , and two o stars in the ic 1396 h ii region . for each field , we have taken a single 30-s exposure in the direct mode with a wide h@xmath0 filter . then , we inserted a grism and obtained three dithered frames of 300-s exposures for slitless spectroscopy . direct images with an @xmath2-band filter were also taken with wfgs2 in 2008 august and 2009 august . three 30-s exposures were made for each field . the @xmath2-band flux is less affected by excess emission of the circumstellar disk , h@xmath0 emission , and continuum veiling of ctts . thus , @xmath2-band photometry is useful as a measure of the underlying photosphere of ctts ( e.g. , * ? ? ? 35 standards for the sloan digital sky survey ( sdss ) system @xcite were observed each night at nearly the same airmass ( difference of @xmath3 0.1 ) as the targets . we used a dome flat for flat fielding . the data were reduced using the iraf software package . we applied the standard procedure for bias subtraction and flat fielding . for each field , three dithered grism frames are combined into one frame by referring the position of the h@xmath0 emission or absorption line of the star(s ) in the field . we picked all stars in each wide h@xmath0 image using the daofind package and extracted their one - dimensional spectra from the grism image . we visually investigated the spectra for the presence of the h@xmath0 emission line and found 639 h@xmath0 emission - line objects in the whole surveyed area . we used the splot package to measure the equivalent width of the h@xmath0 line in the spectra . their coordinates were determined by the astrometric calibration tool of the starlink program _ gaia _ using usno - b catalog stars as the reference in the field . all of the photometric measurements were made with the apphot package for aperture photometry . the typical seeing of the image was 2.73.0 pixels . we used a 4-pixel diameter aperture for the aperture photometry . we then calibrated the photometric values to the sdss system via standards . in general , the run was of high quality , but thin cloud affected a few fields . we checked the stability of the magnitude for three direct images in each field and examined the quality of the measurement . excluding stars with saturated images or images with low photometric quality , we measured 586 stars in 13.6 @xmath3 @xmath2 @xmath3 19.3 . we list the 639 h@xmath0 stars in table 2 . the table shows the ( 1 ) identification number , ( 2 , 3 ) j2000 coordinates , ( 4 ) @xmath2-band magnitudes and ( 5 ) equivalent widths of h@xmath0 emission ( ) . the h@xmath0 equivalent widths were measured for 608 stars . we compared our list of emission stars with the two micron all sky survey ( 2mass ) point source catalog with 12 stars supplemented by the 2mass 6x point source catalog . we found matches for 617 stars with 2mass stars within 2@xmath19 and determined that 548 of these have good quality @xmath20 photometry ( quality flag aaa ) . due to small separation angles or positional uncertainties , two pairs of h@xmath0 stars are identified with single 2mass sources ( no . 250/251 with j21381703 + 5739265 , and no . 600/601 with j21451578 + 5717392 ) . we have included these stars in the above number . we list the 2mass designation of the nir counterpart , photometric quality , @xmath21 magnitude , and @xmath21@xmath22 , @xmath22@xmath23 colors in columns ( 6)(10 ) of table 2 . @xmath15v estimated from the nir color color diagram ( see 4.1 ) is also listed in column ( 11 ) . about 70% of our h@xmath0 stars can not be identified with known sources in the simbad database . the identification in the literature or comments are listed in column ( 12 ) . figure 2a shows the spatial distribution of emission - line objects while figure 2b represents the contour map of the number density of the detected emission - line objects with the kernel method @xcite . we adopted a gaussian shape for the density distribution of the kernel with a smoothing parameter of h@xmath242@xmath184 . the isolines are drawn at intervals of 150 stars per deg@xmath1 , which is about three times the average number density in the field region ( see table 4 ) , starting from 100 stars per deg@xmath1 . it is clear from figure 2 that the distribution of emission - line stars is far from uniform . to demonstrate the different properties in the different areas , we divided the surveyed field into the following subregions according to the number density of h@xmath0 stars . the primary peak of the most rich aggregate ( the west region ) is located between the main exciting star hd 206267 and the western elephant trunk globule rim a ( ic 1396a ) . the secondary peak is at the opposite side , to the east of hd 206267 ( the east region ) , and the third peak is at the south of the bright - rimmed globule sfo 38 ( sfo : @xcite , the sfo38 region ) . other small aggregates of emission - line stars are located near sfo 37 ( the sfo37 region ) , sfo 39 and sfo 41 ( the sfo39/41 region ) , rim b ( the rim b region ) , and at the north of the b 365 dark cloud ( b : @xcite ) or around ( 21@xmath2535@xmath26 , @xmath2757@xmath7 ) . see figure 1 for references . the west and the east regions appear to be surrounded by a halo - like structure ( the halo region ) . some h@xmath0 stars appear to be associated with other small bright rimmed clouds such as rim c and sfo 34 ( rim d ) . we defined the rest of the region as the field area ( the field region ) . our sample of h@xmath0 stars contains two carbon stars cgcs 5454 ( no . 575 ) and cgcs 5401 ( no . 287 ) , and one be / x - ray binary cep x-4 ( no . 340 ) , whose distance was confirmed to be 3.8 kpc by @xcite . our sample of h@xmath0 stars also contains a candidate of the exciting source of the herbig haro object hh 588 ( no . 436 ) , which is deeply embedded in sfo 37 @xcite , one herbig ae / be star ( haebe ) mva 426 ( no . 236 ) @xcite , and @xmath8 100 tts including class - ii , iii , and transient objects @xcite . an objective prism survey using the schmidt telescope @xcite detected 155 h@xmath0 emission objects in ic 1396 and an additional 65 emission stars were reported in a subsequent paper @xcite . their spatial distribution shows no noticeable concentration or clustering . out of 220 kun h@xmath0 stars , 150 stars are located in our observed field ; however , only 11 stars ( no . 287 = kun 80 , no . 13 = kun 171 , no . 340 = kun 193 , no . 618 = kun 215 , no . 49 = kun 308 , no . 80 = kun 310 , no . 97 = kun 314 s , no . 252 = kun 321 , no . 318 = kun 325 , no . 410 = kun 330 , no . 487 = kun 332 ) are identified among our h@xmath0 stars . although kun 315 ( lkh@xmath0 349 ) is not included in table 2 , our grism spectrum appears saturated but indicates a p cyg - type h@xmath0 profile . kun 320 ( gl cep ) and kun 337 ( hd 239745 ) are also too bright and saturated in our grism spectra , and the positions of nos . 77 , 230 , 12 , and 266 are offset from kun 53 , 73 , 170 , and 322 by around 10@xmath19 . even if we include these seven stars , only 12% of stars can be identified with kun stars . the rest of the kun stars show obvious h@xmath0 absorption or no emission , and 10 of them show the spectra of m - type stars in our grism spectra . medium resolution spectroscopy by @xcite show that kun 193 ( no . 340 ; cepx-4 ) exhibits conspicuous hydrogen emission , but 34 other kun stars do not . they suggested that the lack of emission is due to the time variability of h@xmath0 stars . @xcite failed to detect h@xmath0 emission in spectroscopic observations of 20 selected kun stars , only finding emission in kun 314 s. thus , they suggest that the original identification of emission was incorrect . our result is consistent with their suggestion . @xcite extensively studied the low - mass population of the two clusters in cep ob2 , trumpler 37 and ngc 7160 , to investigate the disk evolution at ages 110 myr . for trumpler 37 , they have performed optical photometry and multifiber spectroscopy in a square about 45@xmath13 per side , centered at hd 206267 . approximately 45% of the low - mass population indicate nir excesses , suggesting heated dust in the circumstellar disk . finally , @xcite examined the membership using h@xmath0 emission , li @xmath286707 absorption lines , and radial velocities as membership criteria . they identified accretion properties in 170 members in the central part of ic 1396 . 53% of the members of this list are also detected in our survey . the equivalent width of the h@xmath0 emission - line of our candidates is consistent with that of @xcite , except for only a few stars . we have detected @xmath8 90% and @xmath29 30% of class - ii and class - iii objects in their list as h@xmath0 stars . this result is naturally explained by the higher detectability of weak emissions in their high dispersion slit spectroscopy . using archival data of _ chandra _ and _ spitzer _ , @xcite have identified 25 x - ray sources as cluster members , which have mir colors of ctts or wtts in the 10@xmath13 @xmath17 8@xmath13 region around hd 206267 . their sources consist of five ctts and 20 wtts . although four of the ctts were detected in our observations , only two wtts match our h@xmath0 stars . @xcite obtained x - ray images of the bright rimmed cloud sfo 38 ( ic 1396n ) using the _ chandra _ acis detector . they identified x - ray sources in the area of a square of 5@xmath13 with the _ spitzer _ irac sources and concentrated their study on 25 sources associated with the globule sfo 38 . of the 117 x - ray sources , 17 sources are identified with our h@xmath0 stars and five of these x - ray sources are associated with h@xmath0 stars in the sfo 38 cloud . in the recent paper by @xcite , they identified 158 pms candidates in a 7 deg@xmath1 area towards ic 1396 by using the data of iphas ( the isaac newton telescope / wide field camera photometric h@xmath0 survey ; * ? ? ? * ) . out of 143 pms candidates identified by @xcite in our field of view , 119 stars are in common with our h@xmath0 sample . we examined the common h@xmath0 stars in iphas to confirm our photometric quality . the iphas magnitude is based on the vega magnitude and hence we converted it to ab magnitude @xcite . @xcite restricted their sample to sources with @xmath30 @xmath3 20 mag , which corresponds to @xmath2 @xmath3 19 mag if we adopt the colors of known t tauri stars in their figure 4 . for most of the stars in common , our @xmath2-magnitudes are consistent with those in the iphas catalog within 0.1 mag . although a large scatter ( @xmath6 = 0.3 mag ) of the magnitude differences is found , it may be due to the photometric variability of young stars . @xcite compared their objects with the members confirmed by @xcite , and estimate the completeness . 89% of pms candidates with the h@xmath0 ew @xmath31 30 identified by @xcite are recovered by @xcite . we also estimated the completeness of our survey with the same manner . for the pms candidates including smaller ew ( @xmath31 10 ) , the fraction of recovered sources by @xcite falls off to 50% . on the other hand , the fraction of recovered sources with ew @xmath31 10 maintains 86% ( 51 out of 59 ) in our sample . we confirmed that none of their rejected candidates ( 30 stars ) by @xcite were included in our list . thus , our observations are more sensitive than their survey for the emission - line stars with small h@xmath0 equivalent width . we identified 27 sources among our h@xmath0 stars with the sources of the point source catalog of the mir all - sky survey obtained with the infrared camera ( irc ) on board the _ akari _ satellite @xcite , with position differences of less than 5@xmath19 . table 3 gives the _ akari _ 9-@xmath32 ( @xmath33 ) and 18-@xmath32 ( @xmath34 ) magnitudes and their 2mass photometric data . their positions on ( @xmath21@xmath23 ) vs. ( @xmath33@xmath34 ) , ( @xmath21@xmath23 ) vs. ( @xmath23@xmath33 ) , and ( @xmath21@xmath23 ) vs. ( @xmath23@xmath34 ) diagrams are shown in figure 3 . the criteria for extracting tts from these diagrams , proposed by @xcite , are also shown as a dashed line ( figure 3a ) or areas ( figure 3b , c ) . the colors of three stars ( nos . 287 , 309 , 575 ) are consistent with cool carbon stars , and about 20 stars are pre - main sequence stars . the bright source no . 436 is the candidate for the exciting source of the hh 588 object . figure 4 shows the contour map of the number of emission - line stars overlaid on a false - color map using the irc on _ blue is for 9 @xmath32 and red is for 18 @xmath32 . probable members of 66 b f stars from @xcite , supplemented by 8 ob stars from _ hipparcos _ results @xcite , are also shown as filled blue circles . as for h@xmath0 stars in the west and east regions , high or intermediate - mass stars are concentrated near the center of the large dust cavity , which is presumably swept up by the expansion of the h ii region . we note that their distribution is slightly shifted to the east of hd 206267 . our list of h@xmath0 stars in table 2 may contain foreground or background , magnetically active field stars . we estimated the number of unrelated main sequence stars with enhanced chromospheric and coronal activity . within the 4.2 deg@xmath1 field towards the ic 1396 , the besancon model of the galactic stellar populations @xcite predicts @xmath84000 and @xmath8 1700 dm stars down to @xmath35 20 mag in the foreground and the background ( the distance range between 870 and 1500 pc ) field , respectively . although 44% of the foreground dm stars are in the range of @xmath36 19 mag , 3% of the background dm stars are in the same magnitude range . following the same approach as @xcite , we used the the stellar luminosity function derived by _ hipparcos _ @xcite and the dme ( with an h@xmath0 ew @xmath31 1 ) incidence from the nearby star survey of @xcite . we adopted a completeness limit for the present observations of @xmath35 20 mag . we assumed the scale height of 325 pc for an exponential falloff in the space density , and 0.7 mag kpc@xmath16 for the diffuse absorption . the estimated number of dme stars in the foreground and the background ( up to 1500 pc ) is 193 and 33 in our observed area , respectively . the total number of the field dme stars , 54 per square degree , is consistent with the average number density of h@xmath0 stars in the field area , 58 per square degree . it is also possible that there is contamination by pre - main sequence stars associated with the foreground cloud @xcite . as the akari 9 @xmath32 image , which traces the photodissociation region in the periphery of the h ii region ( figure 4 ) , is likely to be dominated by polycyclic aromatic hydrocarbons ( pahs ) emission features , h@xmath0 stars associated with 9-@xmath32 emission are not likely to be foreground objects . the dark clouds b 163 and b 163sw do not show any sign of bright rims and their radial velocities of the molecular emission line are 6 km s@xmath16 @xcite . the radial velocities of the other associated globules are between @xmath37 km s@xmath16 and @xmath38 km s@xmath16 . from the radial velocity , @xcite suggested 150 pc for the distance of b 161 , which also shows no bright rims and no indication of heating from behind @xcite . the broad , curved dark lane kh 161 @xcite and inverse s - shaped b 365 , both of which show weak and diffuse emission at the northern and southern part of figure 1 , appear in foreground clouds . figure 5 illustrates the nir color color diagram of our emission - line stars . the solid lines yield the locus of the main - sequence and of the giant branch according to @xcite , while the dash - dotted straight line indicates the locus of the unreddened ctts following @xcite . we used the color transformation equations of @xcite to convert their loci into the 2mass system . the reddening vector adopted from @xcite is also shown . we selected nearly 40% of the h@xmath0 stars whose nir colors are consistent with the colors of the reddened ctts in figure 5 and calculated the source extinction by tracing back along the reddening line to the ctts locus . in the above procedure , we have not separated between wtts and ctts candidates , and treated all h@xmath0 stars as the same way . although generally an unreddened wtts does not show the color of the ctts but rather of the normal main sequence , most wtts have nir colors of k- to early m - type stars , i.e. , ( @xmath22@xmath23 , @xmath21@xmath22 ) @xmath8 ( 0.2 , 0.64 ) , that is close to the locus of unreddened cttss . thus , we have not lost many wtts in the above procedure . three carbon stars , discussed in the previous section and shown as open circles in figure 5 , are excluded . figure 6 shows a histogram of the measured extinction of 252 stars excluding three carbon star candidates ( see 3.3 ) and no . the median value of the source extinction is @[email protected] mag . this value is consistent with the average extinction @[email protected] mag obtained by @xcite for trumpler 37 . the high mass stars in our surveyed area are two o - type stars , hd 206267 ( o6 ) and hd 206183 ( o9.5v ) , and six early b stars ( b0b5 ) @xcite . when [email protected] , the foreground extinction of these ob stars should be 1.32.1 mag @xcite . @xcite excluded stars with @xmath15v @xmath3 1.0 mag as foreground stars and excluded stars with @xmath15v @xmath31 2.3 mag as background objects . as the background contamination by field dme stars is not so high , many of the emission - line stars with @xmath15v @xmath31 2.3 mag in our survey appear to be associated with dark clouds or globules . therefore , we selected 189 stars with @xmath15v @xmath31 1.0 mag as likely members associated with ic 1396 . most known t tauri stars ( e.g. , * ? ? ? * ) in table 2 have @xmath15v @xmath31 1.0 mag , but 12 t tauri stars have low extinction , and more than half of them are located near hd 206267 . figure 7 shows a histogram of the measured equivalent width . the fraction of weak ( ew @xmath3 10 ) emission - line stars among the ic 1396 pms candidates is low ( 27% ) relative to that over the entire sample ( 44% ) . it appears that the fraction of ctts candidates among the pms candidates in the ic 1396 region is higher than that among the candidates proposed as foreground ( or background ) objects . we could not find any evidence of the difference between the spatial distribution of ctts and wtts . to identify the haebe stars , which have larger infrared excess than ctts , we adopted the extended criteria in the nir color color diagram proposed by @xcite . they used the region between two parallel reddening lines on their figure 4 and above ( @xmath21@xmath22 ) @xmath31 0.2 mag to pick up haebe stars , excluding b[e ] and classical be stars from the young star sample as contaminants . according to their criteria , we selected five stars , redder than the reddening line passing through the truncated point of the ctts locus and @xmath21@xmath22 @xmath31 0.2 mag , including a known haebe , mva-426 @xcite . it is suggested that from their @xmath21 magnitudes nos . 13 , 236 , and 606 are haebe stars and no . 479 and 545 are class - i objects . in the following discussion , we have added these stars and 12 t tauri stars with low extinction to the probable members totaling 205 stars . the number of new unpublished pms candidates identified in the current survey is 72 including three class - i objects or haebe stars . figure 8 illustrates the contour map of the number density of emission - line stars associated with ic 1396 . dashed boxes represent the eight subregions , namely , the east , west , sfo38 , sfo37 , sfo39/41 , rim b , halo , and field regions . their distribution is not very different from figure 2 , though the peak of the distribution at the east region is slightly shifted westward . while the aggregate at the north of b 365 ( see figure 1 ) , which is shown in figure 2b , disappears in figure 8 and therefore could be a foreground object , other aggregates should be associated with ic 1396 . table 4 summarizes the number of emission - line stars in each region . the h@xmath0 stars , whose @xmath15v values we could not obtain , appear to be spread over the observed field rather uniformly . thus , we suggest a large number of them are field stars . @xcite determined that the time after the first stars reach the main - sequence is 7 myr in trumpler 37 and @xcite confirmed 4 myr for the age of trumpler 37 from optical photometry and theoretical isochrones . the expansion of the h ii region creates a swept - up shell and results in a compressed molecular ring around its periphery . from the size of the ring , @xcite suggested a dynamical age of the system of globules of 23 myr . figure 9 shows the extinction - corrected @xmath2 vs. ( @xmath2@xmath21 ) diagram of the 194 probable members with @xmath2-band data , using the @xmath2-band extinction correction from @xcite . here , we adopted the pre - main sequence model of @xcite . their effective temperature and luminosity are converted into our system using the table of @xcite and the equation of @xcite . in figure 9 , the 0.1 , 1 , 3 , 5 , 10 , and 100 myr isochrones are overlapped as well as the evolutionary tracks for masses from 0.1 to 2.0 m@xmath39 , at an assumed distance of 870 pc . as the reddening vector runs almost parallel to the isochrones , the extinction value would not severely affect our age estimate . the uncertainty of the distance modulus ( 8.949.70 ) and the choice of pms models affects the estimation of ages and masses of h@xmath0 stars from figure 9 . we used the pre - main sequence models of @xcite , which are converted to our photometric system by a table of @xcite and the equation of @xcite , for comparison . by assuming the nearest distance , 615 pc @xcite , the evolutionary tracks shift upwards , and it makes the age @xmath80.2 dex older in both models . the age and mass histograms of the probable members in different choices of pms models adopting 870 pc are shown in figure 10 . they indicate that the numbers of young stars ( @xmath3 1 myr ) predicted from the siess models ( fig 10a ) are generally smaller than those predicted from the palla & stahler models ( fig 10b ) . on the other hand , the number of low mass stars ( @xmath3 0.4 m@xmath39 ) derived from the siess models ( fig 10c ) are slightly large compared to those derived from the palla & stahler models ( fig 10d ) . we should notice the incompleteness of our pms candidate sample along with possible effects on the distribution of masses and ages . @xcite carefully investigated the stellar population of the orion nebula cluster by using these two pms models . the completeness function in the mass - age plane , derived from their statistical simulation for two evolutionary models , shows that a large incompleteness correction is needed in the low mass ( @xmath3 0.2 m@xmath39 ) range . they attributed the difference of results to the different shapes of tracks and isochrones in this mass range . although we adopted the siess models , the uncertainties would be larger for stars with ages @xmath3 1 myr and masses @xmath3 0.2 m@xmath39 . furthermore , we could not detect emission - line stars near the region of higher obscuration such as the globules , and within 1@xmath13 of the bright o star hd 206267 , due to saturation of the detectors , light contamination from this star , and diffraction spikes from this star . table 5 shows the number of h@xmath0 stars in each age and mass bin in figure 9 . the age and mass histograms of h@xmath0 stars in each subregion are shown in figure 11 . we combined the sfo37/38/39/41/rim b regions in the brc ( bright - rimmed cloud ) region . most of the h@xmath0 stars have ages of @xmath3 3 myr and masses of 0.20.6 m@xmath39 . the number of stars with an age less than 1 myr and with an age of 13 myr are comparable in the region associated with bright rims , i.e. , the brc region . we confirmed that the above result is also true for the age assigned from the palla and stahler model , and is not dependent on the choice of distance . the histogram in the west region is similar to that of the brc region , probably because most of the h@xmath0 stars in the west region are associated with rim a. from the extensive survey of young stars by _ spitzer _ , @xcite found two distinct generations of star formation in w5 . they considered two triggered star formation mechanisms , radiative driven implosion ( rdi ; e.g. , * ? ? ? * ) and collect - and - collapse , to explain their findings and concluded that both mechanisms are at work in the w5 h ii region . @xcite extensively studied the nature of infrared bubbles which enclose h ii regions and discussed star formation triggered by the expanding h ii region at their edges . they show that most of these bubbles enclose h ii regions ionized by o b2 stars , many are formed to be surrounded by cold dust shells ( good candidates for c&c processes ) , and many have bright - rimmed dust condensations protruding inside the h ii region ( good candidates for rdi process ) . in the ic 1396 region , the presence of aggregates associated with the bright rims immediately inside the ic 1396 h ii region or the near side of the ionizing star hd 206267 , such as the west ( rim a ) , rim b , and sfo38 regions , suggest that the star formation in these regions has advanced away from hd 206267 in @xmath3 3 myr . such a configuration is consistent with squeezing of the pre - existing condensations compressed by the pressure of the ionized gas . the kinematical structure of sfo 37 @xcite and an x - ray study of sfo 38 @xcite also favored the rdi model . in our previous results for the w5e h ii region @xcite , we have found three aggregates of h@xmath0 emission - line stars in the w5e h ii region . two of them are located at the edge of the h ii region , and one is near the central exciting o7 star . the former two aggregates are systematically younger than the latter . based on a comparison of the stellar ages in the brc region with those in the east region of ic 1396 ( figure 11 ) , it might appear that the low - mass population in the east region is older . we only find two stars younger than 1 myr in the east region , in contrast to the west region . it may suggest that the star formation activity in the east region has ceased recently . as there remains no molecular material associated with the population in the east region , the exciting star hd 206267 has already dissipated the natal molecular cloud . however , a kolmogorov - smirnov ( k - s ) test gives a probability of @xmath31 15 % that the two samples have the same distribution in age , even if we choose either pms model . the difference between the ages of h@xmath0 stars in the brcs region and the east region is not statistically significant . @xcite estimated ages for six members near rim a or in the globule . they found that the ages of five stars are younger than 1 myr , which is significantly lower than the average age of trumpler 37 . figure 12 plots the spatial distribution of h@xmath0 stars with derived ages . at a glance , younger stars in figure 12a are preferentially located at the west region compared to those at the east region . again we used k - s test to compare the distribution in right ascension and declination derived from different choice of pms models . a probability that two samples of different age ( fig 12a and 12b ) have the same distribution in two coordinate axis is @xmath31 25 % and @xmath31 54 % . thus , the age difference of young stars in the west and the east region is not statistically clear from the spatial distribution of our sample . in addition to the west region , figure 12b shows a number of stars with an intermediate age ( 13 myr ) in the east and sfo38 regions , extending to the halo region . a linear sequence of young stars across the center is noticeable . a filamentary dark cloud extends from rim a to the western rim of the h ii region . it is noteworthy that h@xmath0 stars are distributed along this structure and further extend to the east of hd 206267 . this spatial distribution strongly suggests a fossil record of a star forming cloud including rim a and b. @xcite argued a spatial east west asymmetry in the young low - mass members of trumpler 37 , compared with more massive ones @xcite . the same spatial asymmetry with respect to the central ionizing o star , hd 206267 , was seen in the distribution of the _ spitzer _ irac excess objects @xcite . the asymmetry of h@xmath0 stars with high disk fraction may be explained by photo - evaporation of the disk material around the forming stars by strong uv radiation . @xcite suggested that the interstellar material , lying between powerful o6 stars and young stars with disks , shields the disk from uv radiation . however , there is no evidence of such an asymmetry of interstellar material from molecular line @xcite and millimeter dust continuum observations @xcite . actually , significant destruction of disks occurs for stars located out to distances of @xmath3 0.5 pc , which corresponds to only a few arcmin from hd 206267 , as found in ngc 2244 @xcite . @xcite additionally suggested probable age differences for the explanation for the asymmetry in the ctts distribution . instead of disk photo - evaporation , we prefer that this spatial asymmetry represents the star formation sequence . if most of the young low - mass population in the west region was formed as a result of a triggered event , those in the east region could be formed along with massive stars . there are 116 probable members of emission - line star in the brc region including the west region , and the total mass is about 6070 m@xmath4 . if we add 35 emission - line stars in the east region , the total mass reaches 70 - 90 m@xmath4 . since our survey is biased toward the region of low extinction , the above estimates of the number and mass of the low - mass population in the brc region including the west and the east region should be considered the lower limits . for example , @xcite found @xmath845 ysos in and around sfo 38 , where we could not detect any h@xmath0 stars in the globule . of these , 26 sources are class-0/i , class - i , and class - i / ii , but there are no such objects situated outside the ionized rim . it follows that the birth of low - mass stars associated with bright rims appears to be the primary mode of star formation in ic 1396 at present . the higher mass o f stars of trumpler 37 @xcite total @xmath8 200 m@xmath4 in our observed fields ( figure 4 ) . assuming the initial mass function of @xcite between 0.01 m@xmath4 and 50 m@xmath4 , we estimate the total mass of stars in trumpler 37 to be @xmath8 500 m@xmath4 . young low - mass stars , which probably formed by a triggered mechanism in the brc region , contribute a fraction of the mass of stars in ic 1396 . @xcite has determined the alignment of ysos and their age gradient at the tip of sfo 37 . they argued that this alignment and the elongation of the head of the globule are attributed to uv radiation from the o9.5v star , hd206183 , at their formation stage . the dominant exciting source , however , is currently hd 206267 . consequently , hd 206267 formed at the late stage of star formation in the central part of the h ii region and disrupted the natal molecular cloud . @xcite reported the age gradient of the pms candidates along the distance from this hot star . their results also favors the triggered formation of stars in ic 1396 . hd 206267 is known as the trapezium system of ob stars with a maximum separation of 15000 au , and the main component a is a triple @xcite . @xcite confirmed the observed masses and evolutionary model masses and suggested that the two bright components of the triple are o6.5 v((f ) ) ( a@xmath40 ) with 3861 m@xmath4 and o9.5 : v ( a@xmath41 ) with 1728 m@xmath4 . therefore , the age of hd 206267 is estimated to be @xmath3 3 myr from the main sequence lifetime @xcite . numerical simulation considering protostellar outflow - driven turbulence has been discussed by @xcite and @xcite . following their study , we suggested that the massive exciting star of the w5 e h ii region was possibly formed after the continuous production of stars @xcite . the birth of the most massive star should be the last stage of star formation . the contiguous formation of stars for 10 myr in trumpler 37 could be supported by figure 12 . many of the low - mass stars near rim a and sfo 38 might be formed in @xmath29 3 myr . as there is no conspicuous molecular material except the elephant trunk globule , rim a , in the evacuated dust cavity ( figure 4 ) , hd 206267 was likely born in the central part after the continuous low mass star formation . it is reasonable to suppose that the formation of massive stars and the associated low - mass population in trumpler 37 , continued for a few myr . although it is difficult to separate the low - mass stars formed as the same generation with hd 206267 and formed by the trigger mechanism of hd 206267 , we believe that many of h @xmath0 stars at the rim were triggered to form by a radiation driven implosion mechanism after the formation of hd 206267 . we have made an extensive survey of emission - line stars in the ic 1396 h ii region to investigate the low - mass pms population near the ob stars . a total of 639 h@xmath0 stars were detected over a 4-deg@xmath1 area and their @xmath2-photometry was performed . their spatial distribution shows several aggregates near the bright - rimmed clouds at the center and the edge of the h ii region ( rim a / b and sfo 37 , 38 , 39 , 41 ) and one near the main exciting star of the h ii region . based on the source extinction estimated from the nir color color diagram , we have selected pms candidates associated with ic 1396 . the age and mass were derived from the extinction corrected color - magnitude diagram and theoretical evolutionary tracks . most of the h@xmath0 stars have ages of @xmath3 3 myr and masses of 0.20.6 m@xmath4 . the spatial distribution of the pms candidates indicates that the stellar population in the east region is spatially distinct from that in the west region , associated with rim a globule . the age difference between two populations was also suggested by @xcite . the spatio - temporal gradients of the stellar age were also demonstrated near the bright - rimmed clouds by @xcite . our k - s test does not support the significant difference between ages of two populations . also , the spatial distribution of @xmath3 1 myr stars is not significantly different from the distribution of 13 myr stars . although the star formation activity in the east region could be low in the last 1 myr , we need further confirmation . the linear sequence of stars including the west ( rim a ) region and rim b region , which is conspicuous in fig 12b of 13 myr age across the center of the h ii region , suggests a fossil record of the star forming cloud . the formation of massive stars and associated low - mass population of trumpler 37 continuing for a few myr , and the most massive star in trumpler 37 , hd 206267 , was likely born @xmath29 3 myr ago . then , the ic 1396 h ii region was formed by strong uv radiation . in the last stage , the h ii region presumably triggered the formation of low - mass stars in the dense inhomogeneity . our results may suggest that massive stars were born after the continuous formation of low - mass stars for 10 myr . the birth of the exciting star as the most massive one could be the late stage of slow but contiguous star formation in the natal molecular cloud . it presumably triggered the formation of many low - mass stars at the dense inhomogeneity in and around the h ii region by a radiation - driven implosion . spectroscopic followup of our pms candidates to establish their membership correctly and identify their nature is needed to clarify the star formation history of this region . the authors were supported by naoj for the use of the uh 2.2-m telescope for the observations . this research made use of the nasa / ipac infrared science archive , operated by the jet propulsion laboratory , california institute of technology , under contact with the national aeronautics and space administration . this work was supported in part by a grant - in - aid for scientific research b ( 20403003 ) from the ministry of education , culture , sports , science and technology . allen , l. et al . 2007 , in protostars and planets v , ed . b. reipurth , d. jewitt , & k. keil ( tucson , univ . arizona press ) , 361 alksnis , a. et al . 2001 , baltic astronomy , 10 , 1 balzs , l.g . 1996 , , 311 , 145 bessell , m. , & brett , j. 1988 , , 100 , 1134 balog , z. et al . 2007 , , 660 , 1532 barentsen , g. et al . 2011 , , 415 , 103 barnard , e.e . 1927 , a photographic atlas of selected regions of the milky way , carnegie institution , washington bonnet - bidaud , j.m . & mouchet , m. 1998 , , 332 , l9 briceo , c. et al . 2007 , in protostars and planets v. ed . b. reipurth , d. jewitt , & k. keill ( tucson : univ . arizona press ) , p.345 burkholder , v. , massay , p. , & morrell , n. 1997 , , 490 , 328 burningham , b. , naylor , t. , littlefair , s.p . , & jeffries , r.d . 2005 , , 363 , 1389 cardelli , j.a . , clayton , g.c . , & mathis , j.s . 1989 , , 345 , 245 choudhury , r. , mookerjea , b. , & bhatt , h.c . 2010 , , 717 , 1067 cieza , l.a . , kessker - silacci , j.e . , jaffe , d.t . , harvey , p.m. , & evans , n.j . , ii 2005 , , 635 , 422 . cohen , j. , frogel , j.a . , persson , s.e . , & elias , j.a . 1981 , , 249 , 481 contreras , m.e . , sicilia - afguilar , a. , muzerolle , j. , calvet , n. , berlind , p. , & hartmann , l. 2002 , , 124 , 1585 coyne , g.v . & macconnell , d.j . 1983 , vatican obs . publ . , 2 , 73 cutri , r.m . 2008 , explanatory supplement to the 2mass all sky data release and extended mission products , http://www.ipac.caltech.edu/2mass/releases/allsky/doc/sec6_4b.html dahm , s.e . & simon , t. 2005 , , 129 , 829 da rio , n. et al . 2010 , , 722 , 1092 deharveng , l. et al . 2010 , , 523 , 35 de zeeuw , p.t . , hoogerwerf , r. , de bruijne , j.h.j . , brown , a.g.a . , & blaauw , a. 1999 , , 117 , 354 drew , j.e . 2005 , , 362 , 753 garrison , r.f . & kormendy , j. 1976 , , 88 , 865 getman , k.v . , feigelson , e.d . , garmire , g. , broos , p. , and wang , j. 2007 , 654 , 316 [ gfg ] gomez , m. , hartmann , l. , kenyon , s.j . , hewett , r. 1993 , , 105 , 1927 gonzlez - solares , e.a . 2008 , , 388 , 89 hawley , s. , gizis , j.e . , & reid , i.n . 1996 , aj , 112 , 2799 hillenbrand , l.a . 1997 , , 113 , 1733 hughes , v.a . & wouterloot , j.g.a . 1984 , , 276 , 204 ikeda , h. et al . 2008 , , 135 , 2323 [ isw ] ishihara , d. et al . 2010 , , 514 , a1 jahrei , h. & wielen , r. 1997 , esa - sp-402 , 675 jeffries , r.d . , maxted , p.f.l . , oliveira , j.m . , & naylor , t. 2006 , , 371 , l6 jimnez - serra , i. , martn - pintado , j. , rodrguez - franco , a. , chandler , c. , comito , c. , schilke , p. 2007 , , 661 , l187 jordi , c . , trullols , e. , & galad - enrquez , d. 1996 , , 312 , 499 jordi , k. , grebel , e.k . , & ammon , k. 2006 , , 460 , 339 kenyon , s.j . & hartmann , l. 1995 , , 101 , 117 khavtassi , j.sh . 1960 , atlas of galactic dark nebulae , abastumani astrophys . obs . , tbilisi koenig , x.p . 2008 , , 688 , 1142 kroupa , p. 2001 , , 322 , 231 kun , m. 1986 , , 125 , 13 kun , m. , balzs , k.g . , & toth , i. 1987 , , 134 , 211 kun , m. , kiss , z.t . , & balog , z. 2008 , in handbook of star forming regions 1 , ed . b. reipurth ( astronomical society of the pacific ) , p.136 kun , m. & psztor , l. 1990 , , 174 , 13 lada , c.j . , & lada , e.a . 2003 , , 41 , 57 lee , h .- t . , & lim , j. 2008 , , 679 , 1352 leung , c.m . , kutner , m.l . , & mead , k.n . 1982 , , 262 , 583 li , z .,- y . , & nakamura , f. 2006 , , 640 , l187 marschall , l.a . , & van altena , w.f . 1987 , , 94 , 71 marschall , l.a . , comins , n.f . , & karshner , g.b . 1990 , , 99 , 1538 mercer , e.p . 2009 , , 138,7 [ mmc ] meyer , m. , calvet , n. , & hillenbrand , l.a . 1997 , , 114 , 288 miao , j. , white , g.j . , thompson , m. , & nelson , r. 2009 , , 692 , 382 muench , a. , getman , k. , hillenbrand , l. , & preibisch , t. 2008 , in handbook of star forming regions 1 , ed . b. reipurth ( astronomical society of the pacific ) , p.483 nakamura , f. , & li , z .- y . 2007 , , 662 , 395 nakano , m. , wiramihardja , s.d . , & kogure , t. 1995 , , 47 , 889 nakano , m. , sugitani , k. , niwa , t. , itoh , y. , & watanabe , m. 2008 , , 60 , 739 ogura , k. , & hasegawa , t. 1983 , , 35 , 299 ogura , k. , sugitani , k. , & pickles , a. 2002 , , 123 , 2597 [ osp ] palla , f. , & stahler , s.w . 1999 , , 525 , 772 patel , n.a . , goldsmith , p.f . , snell , r.l . , hezel , t. , & xie , t. 1995 , , 447,721 pottasch , s. 1956 , bull . netherlands , 13 , 77 pravdo , s.h . , tsuboi , y. , uzawa , a. , & ezoe , y. 2009 , , 704 , 1495 reach , w.t . 2004 , , 154 , 385 [ rry ] robin , a.c . , reyle , c. , derriere , s. , & picaud , s. 2003 , , 409 , 523 rosolowsky , e. et al . 2010 , , 188 , 123 schaerer . d. , & de koter , a. 1997 , , 322 , 598 schulz , n.s . , berghfer , t.w . , & zinnecker , h . 1997 , , 325 , 1001 sicilia - aguilar , a. , hartmann , l. , briceo , c. , muzerolle , j. , & calvet , n. 2004 , , 128 , 805 sicilia - aguilar , a. , hartmann , l. , hernandez , j. , briceo , c. , & calvet , n. 2005 , , 130 , 188 sicilia - aguilar , a. et al . 2006a , , 638 , 897 sicilia - aguilar , a. et al . 2006b , , 132 , 2135 [ shf ] sicilia - aguilar , a. , hartmann , l.w . , watson , d. , & bohac , c. 2007 , , 659 , 1637 siess , l. , dufour , e. , & forestini , m. 2000 , , 358 , 593 smith , j.a . 2002 , , 123 , 2121 stickland , d.j . 1995 , observatory , 115 , 180 sugitani , k. , fukui , y. , & ogura , k. 1991 , , 77 , 59 sugitani , k. , morita , k.i . , nakano , m. , tamura , m. , & ogura , k. 1997 , , 486 , l141 takita , s. et al . 2010 , , 519 , a83 tout , c.a . , livio , m. , & bonnell , i.a . 1999 , , 310 , 360 uehara , m. et al . 2004 , proc . spie , 5492 , 661 wang , p. , li , z .- y . , abel , t. , & nakamura , f. 2010 , , 709 , 27 weikard , h. , wouterloot , j.g.a . , castets , a. , winnewisser , g. , & sugitani , k. 1996 , , 309 , 581 lrr ut date & number of fields & + & grism & @xmath2-band + 2004 nov 12 & 1 & + 2006 sep 6 & 2 & + 2006 nov 18 , 20 & 5 & + 2007 may 7 & 2 & + 2008 aug 1 , 2 , 4 , 6 , 7 & 52 & 54 + 2008 aug 8,9,11 & 38 & + 2008 aug 21 , 22 , 24 , 26 & 34 & 80 + 2008 nov 2 & 2 & + 2009 aug 13 , 15 , 17 , 18 & 21 & 54 + total & 157 & 188 + ccrrrrrrrrrl 1 & 21:30:53.9 & 58:05:20.2 & 14.59 & 3 & 21305396 + 5805201 & aaa & 13.17 & 0.41 & 0.33 & & + 2 & 21:32:07.1 & 57:17:01.7 & 15.85 & 8 & 21320707 + 5717018 & aaa & 14.13 & 0.49 & 0.12 & & + 3 & 21:32:08.9 & 57:44:22.3 & & 5 & 21320901 + 5744224 & aaa & 14.42 & 0.72 & 0.34 & 0.14 & + 4 & 21:32:12.2 & 57:05:40.4 & 15.33 & 5 & 21321218 + 5705406 & aaa & 14.13 & 0.23 & 0.19 & & + 5 & 21:32:18.4 & 56:54:51.8 & 16.49 & 9 & 21321834 + 5654517 & aaa & 14.65 & 0.57 & 0.10 & & + 6 & 21:32:25.6 & 57:49:40.3 & 17.02 & 46 & 21322559 + 5749403 & aaa & 14.38 & 1.08 & 0.60 & 3.02 & + 7 & 21:32:36.9 & 57:57:27.8 & 15.30 & 3 & 21323682 + 5757276 & aaa & 13.84 & 0.54 & 0.16 & & + 8 & 21:32:49.1 & 57:13:48.3 & 15.06 & 3 & 21324900 + 5713478 & aaa & 13.18 & 0.66 & 0.15 & & + 9 & 21:32:52.4 & 57:19:08.5 & 18.54 & 2 & & & & & & & + 10 & 21:33:01.5 & 57:48:34.4 & 15.05 & 12 & 21330146 + 5748338 & aaa & 12.69 & 0.91 & 0.29 & & + rrrrrl 13 & 4.44 & 2.88 & 2.42 & 9.83 & lt cep , iras 21317 + 5734 + 21 & 5.72 & 3.91 & 1.20 & 10.64 & + 29 & 4.62 & 2.69 & 2.21 & 9.75 & + 59 & 6.29 & 4.24 & 1.44 & 10.22 & v488 cep + 60 & 5.81 & & 1.94 & 12.49 & + 81 & 6.99 & & 1.66 & 10.54 & + 115 & 5.88 & 3.85 & 2.53 & 9.39 & [ rry ] tet + 121 & 6.58 & 4.50 & 0.46 & 14.42 & + 127 & 6.77 & 4.89 & 1.58 & 10.03 & + 203 & 5.87 & 3.81 & 0.73 & 9.84 & + 236 & 5.93 & 3.79 & 1.74 & 8.77 & mva 426 + 252 & 4.99 & 3.04 & 1.69 & 8.59 & gm cep + 287 & 4.64 & 4.06 & 2.79 & 6.96 & kun 80 , cgcs 5401 + 309 & 2.94 & 2.20 & 3.56 & 6.94 & + 356 & 6.59 & 4.83 & 0.99 & 10.08 & + 377 & 6.47 & 4.73 & & 10.76 & [ shf]93 - 361(ii ) + 436 & 4.29 & 0.11 & 3.50 & 12.80 & iras 21388 + 5622 , hh 588 + 456 & 3.80 & & 2.29 & 10.67 & [ osp ] brc38 9 + 487 & 7.43 & & 0.72 & 8.07 & kun 332 + 575 & 1.17 & 0.47 & 3.68 & 4.83 & cgcs 5434 + 596 & 6.80 & 5.11 & 2.07 & 11.24 & + 606 & 7.07 & & 2.72 & 11.67 & + 611 & 6.61 & 3.81 & 1.64 & 9.66 & iras 21440 + 5648 + 617 & 7.07 & & 1.49 & 10.18 & kun 215 + 618 & & 5.03 & 1.76 & 11.32 & + 621 & 6.26 & 3.62 & 2.02 & 10.63 & + 622 & 6.63 & 3.68 & 1.59 & 10.87 & + llrr & sky area & number of h@xmath0 stars & number density + & deg@xmath42 & & per deg@xmath42 + east & 0.11 & 89 ( 34 ) & 8.1 @xmath17 10@xmath42 + west(rim a ) & 0.11 & 116(56 ) & 1.1 @xmath17 10@xmath43 + sfo38 & 0.13 & 51 ( 25 ) & 3.9 @xmath17 10@xmath42 + sfo37 & 0.10 & 15 ( 7 ) & 1.6 @xmath17 10@xmath42 + sfo39/41 & 0.34 & 43 ( 16 ) & 1.3 @xmath17 10@xmath42 + rim b & 0.09 & 23 ( 12 ) & 2.8 @xmath17 10@xmath42 + halo & 0.99 & 167 ( 38 ) & 1.7 @xmath17 10@xmath42 + field & 2.33 & 135 ( 17 ) & 5.8 @xmath17 10@xmath44 + total & 4.2 & 639 ( 205 ) & + crrrrrr age / mass & east & west & brc & halo & field & total + @xmath3 1 myr & 2 & 15 & 17 & 7 & 2 & 43 + 13 myr & 18 & 26 & 23 & 16 & 4 & 87 + 35 myr & 3 & 4 & 5 & 6 & 0 & 18 + 510 myr & 5 & 4 & 5 & 2 & 1 & 17 + 10 myr @xmath3 & 2 & 2 & 2 & 5 & 5 & 16 + 0.10.2 m@xmath4 & 1 & 5 & 6 & 2 & 0 & 14 + 0.20.4 m@xmath4 & 1 & 0 & 0 & 2 & 3 & 6 + 0.40.6 m@xmath4 & 5 & 9 & 9 & 9 & 1 & 33 + 0.60.8 m@xmath4 & 7 & 7 & 15 & 11 & 3 & 43 + 0.81.0 m@xmath4 & 15 & 29 & 22 & 12 & 4 & 82 + @xmath31 1.0 m@xmath4 & 1 & 1 & 0 & 0 & 1 & 3 + total & 30 & 51 & 52 & 36 & 12 & 181 +

riemannian geometry , topology and dynamics permit to introduce partially defined holomorphic functions on the variety of representations of the fundamental group of a manifold . the functions we consider are the complex valued ray singer torsion , the milnor turaev torsion , and the dynamical torsion . they are associated essentially to a closed smooth manifold equipped with a ( co)euler structure and a riemannian metric in the first case , a smooth triangulation in the second case , and a smooth flow of type described in section [ s:2 ] in the third case . in this paper we define these functions , describe some of their properties and calculate them in some case . we conjecture that they are essentially equal and have analytic continuation to rational functions on the variety of representations . we discuss the case of one dimensional representations and other relevant situations when the conjecture is true . as particular cases of our torsions , we recognize familiar rational functions in topology such as the lefschetz zeta function of a diffeomorphism , the dynamical zeta function of closed trajectories , and the alexander polynomial of a knot . a numerical invariant derived from ray singer torsion and associated to two homotopic acyclic representations is discussed in the last section .

math0507587

for a finitely presented group @xmath0 denote by @xmath1 the algebraic set of all complex representations of @xmath0 on the complex vector space @xmath2 . for a closed base pointed manifold @xmath3 with @xmath4 denote by @xmath5 the algebraic closure of @xmath6 , the zariski open set of representations @xmath7 so that @xmath8 . the manifold @xmath9 is called @xmath2-acyclic iff @xmath5 , or equivalently @xmath6 , is non - empty . if @xmath9 is @xmath2-acyclic then the euler poincar characteristic @xmath10 vanishes . there are plenty of @xmath2-acyclic manifolds . if @xmath11 then @xmath12 , where @xmath13 denotes the first betty number of @xmath9 , and @xmath14 is a finite abelian group . if in addition @xmath9 is @xmath2-acyclic and @xmath15 is torsion free , then @xmath16 . there are plenty of @xmath2-acyclic ( @xmath11 ) manifolds @xmath9 with @xmath15 torsion free . in this paper , to a @xmath2-acyclic manifold and an euler or coeuler structure we associate three partially defined holomorphic functions on @xmath5 , the complex valued ray singer torsion , the milnor turaev torsion , and the dynamical torsion , and describe some of their properties . they are defined with the help of a riemannian metric , resp . smooth triangulation resp . a vector field with the properties listed in section [ ss:2.3 ] , but are independent of these data . we conjecture that they are essentially equal and have analytic continuation to rational functions on @xmath5 and discuss the cases when we know that this is true . if @xmath11 they are genuine rational functions of @xmath13 variables . we calculate them in some cases and recognize familiar rational functions in topology ( lefschetz zeta function of a diffeomorphism , dynamical zeta function of some flows , alexander polynomial of a knot ) as particular cases of our torsions , cf . section [ s:7 ] . the results answer the question * _ _ is the ray singer torsion the absolute value of a holomorphic function on the space of representations ? _ _ ( for a related result consult @xcite ) and establish the analytic continuation of the dynamical torsion . both issues are subtle when formulated inside the field of spectral geometry or of dynamical systems and can hardly be decided using internal technologies in these fields . there are interesting dynamical implications on the growth of the number of instantons and of closed trajectories , some of them improving on a conjecture formulated by s.p . novikov about the gradients of closed morse one form , cf . section [ s:8 ] . .1 in this paper surveys results from @xcite , @xcite , @xcite and reports on additional work in progress on these lines . its contents is the following . in section [ s:2 ] , for the reader s convenience , we recall some less familiar characteristic forms used in this paper and describe the class of vector fields we use to define the dynamical torsion . these vector fields have finitely many rest points but infinitely many instantons and closed trajectories . however , despite their infiniteness , they can be counted by appropriate counting functions which can be related to the topology and the geometry of the underlying manifold cf . the dynamical torsion is derived from them . all torsion functions referred to above involve some additional topological data ; the milnor turaev and dynamical torsion involve an euler structure while the complex ray singer torsion a coeuler structure , a sort of poincar dual of the first . in section [ s:3 ] we define euler and coeuler structures and discuss some of their properties . although they can be defined for arbitrary base pointed manifolds @xmath3 we present the theory only in the case @xmath17 when the base point is irrelevant . while the complex ray singer torsion and dynamical torsion are new concepts the milnor turaev torsion is not , however our presentation is somehow different from the traditional one . in section [ s:4 ] we discuss the algebraic variety of cochain complexes of finite dimensional vector spaces and introduce the milnor torsion as a rational function on this variety . the milnor turaev torsion is obtained as a pull back by a characteristic map of this rational function . section [ s:5 ] is about analytic torsions . in section [ ss:5.1 ] , we recall the familiar ray singer torsion slightly modified with the help of a coeuler structure . this is a positive real valued function defined on @xmath6 , the variety of the acyclic representations . we show that this function is independent of the riemannian metric , and that it is the absolute value of a rational function , provided the coeuler structure is integral . in section [ ss:5.3 ] we introduce the complex valued ray singer torsion , and show the relation to the first . the complex ray singer torsion , denoted @xmath18 , is a meromorphic function on a finite cover of the space of representations and is defined analytically using regularized determinants of elliptic operators but not self adjoint . the milnor turaev torsion , defined in section [ ss:6.1 ] , is associated with a smooth manifold , a given euler structure and a homology orientation and is constructed using a smooth triangulation . its square is conjecturally equal to the complex ray singer torsion as defined in section [ ss:5.3 ] , when the coeuler structure for ray singer corresponds , by poincar duality map , to the euler structure for milnor turaev . the conjecture is true in many relevant cases , in particular for @xmath11 . up to sign the dynamical torsion , introduced in section [ ss:6.2 ] , is associated to a smooth manifold and a given euler structure and is constructed using a smooth vector field in the class described in section [ ss:2.3 ] . the sign can be fixed with the help of an equivalence class of orderings of the rest points of @xmath19 , cf . section [ ss:6.2 ] . a priori the dynamical torsion is only a partially defined holomorphic function on @xmath5 and is defined using the instantons and the closed trajectories of @xmath19 . for a representation @xmath20 the dynamical torsion is expressed as a series which might not be convergent for each @xmath20 but is certainly convergent for @xmath20 in a subset @xmath21 of @xmath5 with non - empty interior . at present this convergence was established only in the case of rank one representations . the existence of @xmath21 is guaranteed by the exponential growth property ( eg ) ( cf . section [ ss:2.3 ] for the definition ) required from the vector field . the main results , theorems [ t:1 ] , [ t:2 ] and [ t:3 ] , establish the relationship between these torsion functions , at least in the case @xmath11 , and a few other relevant cases . the same relationship is expected to hold for @xmath2 of arbitrary dimension . one can calculate the milnor turaev torsion when @xmath9 has a structure of mapping torus of a diffeomorphism @xmath22 as the `` twisted lefschetz zeta function '' of the diffeomorphism @xmath22 , cf . section [ ss:7.1 ] . the alexander polynomial as well as the twisted alexander polynomials of a knot can also be recovered from these torsions cf . section [ ss:7.3 ] . if the vector field has no rest points but admits a closed lyapunov cohomology class , cf . section [ ss:7.2 ] , the dynamical torsion can be expressed in terms of closed trajectories only , and the dynamical zeta function of the vector field ( including all its twisted versions ) can be recovered from the dynamical torsion described here . in section [ ss:8.1 ] we express the phase difference of the milnor turaev torsion of two representations in the same connected component of @xmath6 in terms of the ray singer torsion . this invariant is analogous to the atiyah singer spectral flow but has not been investigated so far . section [ s:8 ] discusses some progress towards a conjecture of novikov which came out from the work on dynamical torsion . let @xmath9 be smooth closed manifold of dimension @xmath23 . let @xmath24 denote the tangent bundle , and @xmath25 the orientation bundle , which is a flat real line bundle over @xmath9 . for a riemannian metric @xmath26 denote by @xmath27 its euler form , and for two riemannian metrics @xmath28 and @xmath29 by @xmath30 their chern simons class . the following properties follow from and below . @xmath31 if the dimension of @xmath9 is odd both @xmath32 and @xmath33 vanish . denote by @xmath34 the euler vector field on @xmath35 which assigns to a point @xmath36 the vertical vector @xmath37 . a riemannian metric @xmath26 determines the levi civita connection in the bundle @xmath24 . there is a canonic @xmath23-form @xmath38 , which assigns to an @xmath23-tuple of vertical vectors _ their volume times their orientation _ and vanishes when contracted with horizontal vectors and a global angular form , see for instance @xcite , is the differential form @xmath39 in @xcite mathai and quillen have introduced a differential form @xmath40 when pulled back to the fibers of @xmath41 the form @xmath42 coincides with @xmath43 . if @xmath44 is an open subset on which the curvature of @xmath26 vanishes , then @xmath42 coincides with @xmath43 on @xmath45 . in general we have the equalities @xmath46 for a vector field @xmath19 with non - degenerate rest points we have the singular @xmath47-chain @xmath48 defined by @xmath49 , with @xmath50 the hopf index . for two vector fields @xmath51 and @xmath52 with non - degenerate rest points we have the singular @xmath53-chain rel . boundaries @xmath54 defined from the zero set of a homotopy from @xmath51 to @xmath52 cf . they are related by the formulas , see @xcite , @xmath55 let @xmath56 be a real or complex vector bundle over @xmath9 . for a connection @xmath57 and a hermitian structure @xmath58 on @xmath56 define a real valued one form @xmath59 by @xmath60 here we consider @xmath58 as an element in @xmath61 , where @xmath62 denotes the dual of the complex conjugate bundle . with respect to the induced connection on @xmath63 we have @xmath64 and therefore @xmath65 actually the latter one form has values in the endomorphisms of @xmath56 which are symmetric with respect to @xmath58 , and thus the ( complex ) trace , see , will indeed be real . since any two hermitian structures @xmath66 and @xmath67 are homotopic , the difference @xmath68 will be exact . if @xmath57 is flat then @xmath69 is closed and its cohomology class independent of @xmath58 . replacing the hermitian structure by a non - degenerate symmetric bilinear form @xmath70 , we define a complex valued one form @xmath71 by a similar formula @xmath72 here we regard @xmath70 as an element in @xmath73 . if two non - degenerate symmetric bilinear forms @xmath74 and @xmath75 are homotopic , then @xmath76 is exact . if @xmath57 is flat , then @xmath77 is closed . note that @xmath78 depends holomorphically on @xmath57 . consider a vector field @xmath19 which satisfies the following properties : * all rest points are of hyperbolic type . * the vector field has exponential growth at all rest points . * the vector field is of lyapunov type . * the vector field satisfies morse smale condition . * the vector field has all closed trajectories non - degenerate . precisely this means that : * in the neighborhood of each rest point the differential of @xmath19 has all eigenvalues with non - trivial real part ; the number of eigenvalues with negative real part is called the index and denoted by @xmath79 ; as a consequence the stable and unstable stable sets are images of one - to - one immersions @xmath80 with @xmath81 diffeomorphic to @xmath82 resp . @xmath83 . * with respect to one and then any riemannian metric @xmath26 on @xmath9 , the volume of the disk of radius @xmath84 in @xmath85 ( w.r . to the induced riemannian metric ) is @xmath86 , for some constant @xmath87 . * there exists a real valued closed one form @xmath88 so that @xmath89 for @xmath90 not a rest point . has nothing in common with @xmath77 notation used in the previous section . ] * for any two rest points @xmath90 and @xmath91 the mappings @xmath92 and @xmath93 are transversal and therefore the space of non - parameterized trajectories form @xmath90 to @xmath91 , @xmath94 , is a smooth manifold of dimension @xmath95 . instantons are exactly the elements of @xmath94 when this is a smooth manifold of dimension zero , i.e. @xmath96 . * any closed trajectory is non - degenerate , i.e. the differential of the return map in normal direction at one and then any point of a closed trajectory does not have non - zero fixed points . recall that a trajectory @xmath97 is an equivalence class of parameterized trajectories and two parameterized trajectories @xmath98 and @xmath99 are equivalent iff @xmath100 for some real number @xmath101 recall that a closed trajectory @xmath102 is a pair consisting of a trajectory @xmath97 and a positive real number @xmath103 so that @xmath104 property ( l ) , ( h ) , ( ms ) imply that for any real number @xmath105 the set of instantons @xmath97 from @xmath90 to @xmath91 with @xmath106)\leq r$ ] is finite and properties ( l ) , ( h ) , ( ms ) , ( nct ) imply that for any real number @xmath105 the set of the closed trajectory @xmath102 with @xmath107)\leq r$ ] is finite . here @xmath108 $ ] resp @xmath109 $ ] denote the homotopy class of instantons resp . closed trajectories . ] denote by @xmath110 the set of homotopy classes of paths from @xmath90 to @xmath91 and by @xmath111 the set of rest points of index @xmath112 . suppose a collection @xmath113 of orientations of the unstable manifolds is given and ( ms ) is satisfied . then any instanton @xmath97 has a sign @xmath114 and therefore , if ( l ) is also satisfied , for any two rest points @xmath115 and @xmath116 we have the counting function of instantons @xmath117 defined by @xmath118 under the hypothesis ( nct ) any closed trajectory @xmath102 has a sign @xmath119 and a period @xmath120 , cf . @xcite . if ( h ) , ( l ) , ( ms ) , ( nct ) are satisfied , as the set of closed trajectories in a fixed homotopy class @xmath121 $ ] is compact , we have the counting function of closed trajectories @xmath122\to\mathbb q$ ] defined by @xmath123 here are a few properties about vector fields which satisfy ( h ) and ( l ) . [ p:1 ] 1 . given a vector field @xmath19 which satisfies ( h ) and ( l ) arbitrary close in the @xmath124-topology for any @xmath125 there exists a vector field @xmath126 which agrees with @xmath19 on a neighborhood of the rest points and satisfies ( h ) , ( l ) , ( ms ) and ( nct ) . \2 . given a vector field @xmath19 which satisfies ( h ) and ( l ) arbitrary close in the @xmath127-topology there exists a vector field @xmath126 which agrees with @xmath19 on a neighborhood of the rest points and satisfies ( h ) , ( eg ) , ( l ) , ( ms ) and ( nct ) . if @xmath19 satisfies ( h ) , ( l ) and ( ms ) and a collection @xmath128 of orientations is given then for any @xmath129 , @xmath130 and @xmath131 one has denotes juxtaposition . ] @xmath132 this proposition is a recollection of some of the main results in @xcite , see proposition 3 , theorem 1 and theorem 5 in there . although not always necessary in this section as in fact always in this paper @xmath9 is supposed to be closed connected smooth manifold . euler structures have been introduced by turaev @xcite for manifolds @xmath9 with @xmath17 . if one removes the hypothesis @xmath17 the concept of euler structure can still be considered for any connected base pointed manifold @xmath3 cf . @xcite and @xcite . here we will consider only the case @xmath17 . the set of euler structures , denoted by @xmath133 , is equipped with a free and transitive action @xmath134 which makes @xmath133 an affine version of @xmath135 . if @xmath136 are two euler structure we write @xmath137 for the unique element in @xmath135 with @xmath138 . to define the set @xmath133 we consider pairs @xmath139 with @xmath19 a vector field with non - degenerate zeros and @xmath140 so that @xmath141 . we make @xmath142 and @xmath143 equivalent iff @xmath144 and write @xmath145 $ ] for the equivalence class represented by @xmath139 . the action @xmath146 is defined by @xmath147,[x , c]):=[x , c'+c]$ ] . [ o:1 ] suppose @xmath19 is a vector field with non - degenerate zeros , and assume its zero set @xmath148 is non - empty . moreover , let @xmath149 be an euler structure and @xmath150 . then there exists a collection of paths @xmath151 with @xmath152 , @xmath153 and such that @xmath154 $ ] where @xmath155 . a remarkable source of euler structures is the set of homotopy classes of nowhere vanishing vector fields . any nowhere vanishing vector field @xmath19 provides an euler structure @xmath156 $ ] which only depends on the homotopy class of @xmath19 . still assuming @xmath17 , every euler structure can be obtained in this way provided @xmath157 . be aware , however , that different homotopy classes may give rise to the same euler structure . to construct such a homotopy class one can proceed as follows . represent the euler structure @xmath158 by a vector field @xmath19 and a collection of paths @xmath151 as in observation [ o:1 ] . since @xmath157 we may assume that the interiors of the paths are mutually disjoint . then the set @xmath159 is contractible . a smooth regular neighborhood of it is the image by a smooth embedding @xmath160 . since @xmath17 , the restriction of the vector field @xmath19 to @xmath161 can be extended to a non - vanishing vector field @xmath162 on @xmath9 . it is readily checked that @xmath163={\mathfrak e}$ ] . for details see @xcite . if @xmath9 dimension larger than 2 an alternative description of @xmath133 with respect to a base point @xmath164 is @xmath165 , where @xmath166 denotes the space of vector fields of class @xmath124 , @xmath167 , which vanish at @xmath164 and are non - zero elsewhere . we equip this space with the @xmath124-topology and note that the result @xmath168 is the same for all @xmath84 , and since @xmath17 , canonically identified for different base points . let @xmath169 be a smooth triangulation of @xmath9 and consider the function @xmath170 linear on any simplex of the first barycentric subdivision and taking the value @xmath171 on the barycenter @xmath172 of the simplex @xmath173 . a smooth vector field @xmath19 on @xmath9 with the barycenters as the only rest points all of them hyperbolic and @xmath174 strictly decreasing on non - constant trajectories is called an euler vector field of @xmath169 . by an argument of convexity two euler vector fields are homotopic by a homotopy of euler vector fields . satisfies ( h ) , ( eg ) , ( l ) and has no closed trajectory , hence also satisfies ( nct ) . the counting functions of instantons are exactly the same as the incidence numbers of the triangulation hence take the values @xmath53 , @xmath175 or @xmath47 . ] therefore , a triangulation @xmath169 , a base point @xmath164 and a collection of paths @xmath176 with @xmath177 and @xmath178 define an euler structure @xmath179 $ ] , where @xmath180 , @xmath181 is any euler vector field for @xmath169 , and this euler structure does not depend on the choice of @xmath181 . clearly , for fixed @xmath169 and @xmath164 , every euler structure can be realized in this way by an appropriate choice of @xmath176 , cf . observation [ o:1 ] . again , suppose @xmath17 . , cf . @xcite . ] consider pairs @xmath182 where @xmath26 is a riemannian metric on @xmath9 and @xmath183 with @xmath184 where @xmath185 denotes the euler form of @xmath26 , see section [ ss:2.1 ] . we call two pairs @xmath186 and @xmath187 equivalent if @xmath188 we will write @xmath189 for the set of equivalence classes and @xmath190 $ ] for the equivalence class represented by the pair @xmath182 . elements of @xmath189 are called _ coeuler structures . _ there is a natural action @xmath191 given by @xmath192,[g,\alpha]):=[g,\alpha-\beta]\ ] ] for @xmath193\in h^{n-1}(m;\mathcal o_m)$ ] . this action is obviously free and transitive . in this sense @xmath189 is an affine version of @xmath194 . if @xmath195 and @xmath196 are two coeuler structures we write @xmath197 for the unique element in @xmath198 with @xmath199 . [ o:2 ] given a riemannian metric @xmath26 on @xmath9 any coeuler structure can be represented as a pair @xmath182 for some @xmath200 with @xmath184 . there is a natural map @xmath201 which combined with the poincar duality map @xmath202 , the composition of the coefficient homomorphism for @xmath203 with the poincar duality isomorphism , . ] makes the diagram below commutative : @xmath204_{d\times\operatorname{pd}}\ar[r]^-{m } & { \mathfrak{eul}}(m;{\mathbb z})\ar[d]^{\operatorname{pd } } \\ h^{n-1}(m;{\mathcal o}_m)\times{\mathfrak{eul}}^*(m;{\mathbb r } ) \ar[r]^-{m^ * } & { \mathfrak{eul}}^*(m;{\mathbb r } ) } \ ] ] there are many ways to define the map @xmath205 , cf . for example , assuming @xmath17 and @xmath206 one can proceed as follows . represent the euler structure by a nowhere vanishing vector field @xmath207 $ ] . choose a riemannian metric @xmath26 , regard @xmath19 as mapping @xmath208 , set @xmath209 , put @xmath210 $ ] and check that this does indeed only depend on @xmath158 . a coeuler structure @xmath211 is called _ integral _ if it belongs to the image of @xmath205 . integral coeuler structures constitute a lattice in the affine space @xmath189 . if @xmath212 is odd , then there is a canonical coeuler structure @xmath213 ; it is represented by the pair @xmath214 $ ] , with any @xmath26 riemannian metric . in general this coeuler structure is not integral . let @xmath0 be a finitely presented group with generators @xmath215 and relations @xmath216 and @xmath2 be a complex vector space of dimension @xmath217 . let @xmath1 be the set of linear representations of @xmath0 on @xmath2 , i.e. group homomorphisms @xmath218 . by identifying @xmath2 to @xmath219 this set is , in a natural way , an algebraic set inside the space @xmath220 given by @xmath221 equations . precisely if @xmath222 represent the coordinates in @xmath220 with @xmath223 , @xmath224 , so @xmath225 and @xmath226 , then the equations defining @xmath1 are @xmath227 with each of the equalities @xmath228 representing @xmath229 polynomial equations . suppose @xmath4 , @xmath9 a closed manifold . denote by @xmath6 the set of representations @xmath20 with @xmath8 and notice that they form a zariski open set in @xmath1 . denote the closure of this set by @xmath5 . this is an algebraic set which depends only on the homotopy type of @xmath9 , and is a union of irreducible components of @xmath1 . recall that every representation @xmath7 induces a canonical vector bundle @xmath230 equipped with a canonical flat connection @xmath231 . they are obtained from the trivial bundle @xmath232 and the trivial connection by passing to the @xmath0 quotient spaces . here @xmath233 is the canonical universal covering provided by the base point @xmath164 . the @xmath0-action is the diagonal action of deck transformations on @xmath233 and of the action @xmath20 on @xmath2 . the fiber of @xmath230 over @xmath164 identifies canonically with @xmath2 . the holonomy representation determines a right @xmath0-action on the fiber of @xmath230 over @xmath164 , i.e. an anti homomorphism @xmath234 . when composed with the inversion in @xmath235 we get back the representation @xmath20 . the pair @xmath236 will be denoted by @xmath237 . if @xmath238 is a representation in the connected component @xmath239 one can identify @xmath239 to the connected component of @xmath240 in the complex analytic space of flat connections of the bundle @xmath241 modulo the group of bundle isomorphisms of @xmath241 which fix the fiber above @xmath164 . an element @xmath242 defines a holomorphic function @xmath243 the complex number @xmath244 is the evaluation on @xmath242 of @xmath245 which factors through @xmath135 . note that for @xmath246 we have @xmath247 . if @xmath248 is a torsion element , then @xmath249 is constant equal to a root of unity of order , the order of @xmath248 . let @xmath250 be a string of non - negative integers . the string is called admissible , and will write @xmath251 in this case , if the following requirements are satisfied @xmath252 denote by @xmath253 the collection of cochain complexes of the form @xmath254 with @xmath255 , and by @xmath256 the subset of acyclic complexes . note that @xmath257 is non - empty iff @xmath251 . the cochain complex @xmath258 is determined by the collection @xmath259 of linear maps @xmath260 . if regarded as the subset of those @xmath261 , with @xmath262 the space of linear maps from @xmath2 to @xmath263 , which satisfy the quadratic equations @xmath264 , the set @xmath265 is an affine algebraic set given by degree two homogeneous polynomials and @xmath257 is a zariski open set . the map @xmath266 which associates to @xmath267 the linear map @xmath268 , is a bundle whose fiber is isomorphic to @xmath269 . this can be easily generalized as follows . consider a string @xmath270 . we will write @xmath271 if @xmath272 is admissible , i.e. @xmath273 . denote by @xmath274 the subset of cochain complexes @xmath275 with @xmath276 . note that @xmath277 is non - empty iff @xmath271 . the obvious map @xmath278 , @xmath279 the space of linear maps in @xmath280 whose kernel has dimension @xmath281 , is a bundle whose fiber is isomorphic to @xmath282 . note that @xmath283 is the total space of a bundle @xmath284 with @xmath285 the tautological bundle over @xmath286 and @xmath287 resp . @xmath288 the trivial bundles over @xmath286 with fibers of dimension @xmath289 resp . @xmath290 . as a consequence we have [ p:2 ] 1 . @xmath257 and @xmath277 are connected smooth quasi affine algebraic sets whose dimension is @xmath291 \2 . the closures @xmath292 and @xmath293 are irreducible algebraic sets , hence affine algebraic varieties , and @xmath294 . for any cochain complex in @xmath267 denote by @xmath295 and consider the short exact sequence @xmath296 . choose a base @xmath297 for each @xmath298 , and choose lifts @xmath299 of @xmath300 in @xmath301 using @xmath302 , i.e. @xmath303 . clearly @xmath304 is a base of @xmath301 . consider the base @xmath305 as a collection of vectors in @xmath306 and write them as columns of a matrix @xmath307 $ ] . define the torsion of the acyclic complex @xmath258 , by @xmath308^{(-1)^i}\ ] ] where @xmath309 is turaev s sign , see @xcite . the result is independent of the choice of the bases @xmath297 and of the lifts @xmath310 cf . @xcite @xcite , and leads to the function @xmath311 turaev provided a simple formula for this function , cf . @xcite , which permits to recognize @xmath169 as the restriction of a rational function on @xmath292 . for @xmath312 denote by @xmath313 the transpose of @xmath260 , and define @xmath314 . define @xmath315 as the subset of cochain complexes in @xmath292 where @xmath316 , and consider @xmath317 defined by @xmath318 one can verify suppose @xmath319 is admissible . @xmath315 is a proper subvariety containing the singular set of @xmath292 . \2 . @xmath320 and implicitly @xmath321 has an analytic continuation to @xmath322 . in particular @xmath169 defines a square root of @xmath321 . we will not use explicitly @xmath323 in this writing however it justifies the definition of complex ray singer torsion . let @xmath9 be a closed manifold , @xmath26 riemannian metric and @xmath182 a representative of a coeuler structure @xmath211 . suppose @xmath324 is a complex vector bundle and denote by @xmath325 the space of connections and by @xmath326 the subset of flat connections . @xmath325 is a complex affine ( frchet ) space while @xmath326 a closed complex analytic subset ( stein space ) of @xmath325 . let @xmath70 be a non - degenerate symmetric bilinear form and @xmath58 a hermitian ( fiber metric ) structure on @xmath56 . while hermitian structures always exist , non - degenerate symmetric bilinear forms exist iff the bundle is the complexification of some real vector bundle , and in this case @xmath327 . the connection @xmath328 can be interpreted as a first order differential operator @xmath329 and @xmath26 and @xmath70 resp . @xmath26 and @xmath58 can be used to define the formal @xmath70-adjoint resp . @xmath58-adjoint @xmath330 resp . @xmath331 and therefore the laplacians @xmath332 they are elliptic second order differential operators with principal symbol @xmath333 . therefore they have a unique well defined zeta regularized determinant ( modified determinant ) @xmath334 ( @xmath335 ) resp . @xmath336 ( @xmath337 ) calculated with respect to a non - zero agmon angle avoiding the spectrum cf . recall that the zeta regularized determinant ( modified determinant ) is the zeta regularized product of all ( non - zero ) eigenvalues . denote by @xmath338 and by @xmath339 note that @xmath340 for any @xmath58 , and @xmath341 . both , @xmath342 and @xmath343 , are closed complex analytic subsets of @xmath326 , and @xmath344 on @xmath345 . we consider the real analytic functions : @xmath346 , @xmath347 , @xmath348 and the holomorphic functions @xmath349 , @xmath350 , @xmath351 defined by : @xmath352 we also write @xmath353 resp . @xmath354 for the same formulas with @xmath355 instead of @xmath356 . these functions are discontinuous on @xmath357 and coincide with @xmath358 resp . @xmath359 on @xmath360 . here @xmath361 stands for either @xmath70 or @xmath58 . for the definition of real or complex analytic space / set , holomorphic / meromorphic function / map in infinite dimension the reader can consult @xcite and @xcite , although the definitions used here are rather straightforward . let @xmath362 be a smooth real vector bundle equipped with a non - degenerate symmetric positive definite bilinear form @xmath363 . let @xmath364 resp . @xmath365 the space of connections resp . flat connections in @xmath366 . denote by @xmath324 the complexification of @xmath366 , @xmath367 , and by @xmath70 resp . @xmath58 the complexification of @xmath363 resp . the hermitian structure extension of @xmath363 . we continue to denote by @xmath364 resp . @xmath365 the subspace of @xmath325 resp . @xmath326 consisting of connections which are complexification of connections resp . flat connections in @xmath366 , and by @xmath57 the complexification of the connection @xmath368 . if @xmath368 , then @xmath369 and therefore @xmath370 observe that @xmath371 the ( generalized ) eigen space of @xmath372 corresponding to the eigen value zero is a finite dimensional vector space of dimension the multiplicity of @xmath47 . the restriction of the symmetric bilinear form induced by @xmath70 remains non - degenerate and defines for each component @xmath373 an equivalence class of bases . since @xmath374 commutes with @xmath372 , @xmath375 is a finite dimensional complex . when acyclic , i.e. @xmath376 , denote by @xmath377 the milnor torsion associated to the equivalence class of bases induced by @xmath70 . let @xmath324 be a complex vector bundle , and let @xmath378 be a coeuler structure . choose a hermitian structure ( fiber metric ) @xmath58 on @xmath56 , a riemannian metric @xmath26 on @xmath9 and @xmath183 so that @xmath190={\mathfrak e}^*$ ] , see section [ ss:3.2 ] . for @xmath376 consider the quantity @xmath379 referred to as the _ modified ray singer torsion . _ the following proposition is a reformulation of one of the main theorems in @xcite , cf . also @xcite and @xcite . [ p:3 ] if @xmath376 , then @xmath380 is gauge invariant and independent of @xmath381 . when applied to @xmath237 the number @xmath382 defines a real analytic function @xmath383 . it is natural to ask if @xmath384 is the absolute value of a holomorphic function . the answer is no as one can see on the simplest possible example @xmath385 equipped with the the canonical coeuler structure @xmath386 . in this case @xmath387 , and @xmath388 , cf . however , theorem [ t:2 ] in section [ ss:6.1 ] below provides the following answer to the question ( q ) from the introduction . if @xmath389 is an integral coeuler structure , then @xmath384 is the absolute value of a holomorphic function on @xmath6 which is the restriction of a rational function on @xmath5 . for a general coeuler structure @xmath390 still locally is the absolute value of a holomorphic function . let @xmath56 be a complex vector bundle equipped with a non - degenerate symmetric bilinear form @xmath70 . suppose @xmath182 is a pair consisting of a riemannian metric @xmath26 and a differential form @xmath183 with @xmath184 . for any @xmath376 consider the complex number @xmath391 referred to as the _ _ complex valued ray singer torsion.__-laplacians for torsion was brought to the attention of the first author by w. mller @xcite . the second author came to it independently . ] it is possible to provide an alternative definition of @xmath392 . suppose @xmath393 is a positive real number so that the laplacians @xmath394 have no eigen values of absolute value @xmath105 . in this case denote by @xmath395 the regularized product of all eigen values larger than @xmath105 w.r . to a non - zero agmon angle disjoint from the spectrum @xmath396 resp . @xmath396 the quantities defined by the formulae with @xmath397 instead of @xmath398 . consider @xmath399 to be the sum of generalized eigen spaces of @xmath372 corresponding to eigen values smaller in absolute value than @xmath105 . @xmath400 is a finite dimensional complex . as before @xmath70 remains non - degenerate and when acyclic ( and this is the case iff @xmath401 is acyclic ) denote by @xmath402 the milnor torsion associated to the equivalence class of bases induced by @xmath70 . it is easy to check that @xmath403 [ p:6 ] 1 . @xmath392 is a holomorphic function on @xmath404 and the restriction of a meromorphic function on @xmath326 with poles and zeros in @xmath342 . \2 . if @xmath74 and @xmath75 are two non - degenerate symmetric bilinear forms which are homotopic , then @xmath405 . \3 . if @xmath186 and @xmath187 are two pairs representing the same coeuler structure , then @xmath406 . \4 . we have @xmath407 for every gauge transformation @xmath408 of @xmath56 . @xmath409 . to check the first part of this proposition , one shows that for @xmath410 one can find @xmath393 and an open neighborhood @xmath21 of @xmath411 such that no eigen value of @xmath394 , @xmath412 , has absolute value @xmath105 . the function @xmath413 is holomorphic in @xmath412 . moreover , on @xmath21 the function @xmath414 is meromorphic in @xmath57 , and holomorphic when restricted to @xmath415 . the statement thus follows from . the second and third part of proposition [ p:6 ] are derived from formulas for @xmath416 resp . @xmath417 which are similar to such formulas for ray singer torsion in the case of a hermitian structure instead of a non - degenerate symmetric bilinear form , cf . the proof of 4 ) and 5 ) require a careful inspection of the definitions . the full arguments are contained in @xcite . as a consequence to each homotopy class of non - degenerate symmetric bilinear forms @xmath418 $ ] and coeuler structure @xmath389 we can associate a meromorphic function on @xmath326 . the reader unfamiliar with the basic concepts of complex analytic geometry on banach/ frechet manifolds can consult @xcite and @xcite . changing the coeuler structure our function changes by multiplication with a non - vanishing holomorphic function as one can see from . changing the homotopy class @xmath418 $ ] is actually more subtle . we expect however that @xmath419 remains unchanged when the coeuler structure is integral . denote by @xmath420 the union of components of @xmath5 which consists of representations equivalent to holonomy representations of flat connections in the bundle @xmath56 . suppose @xmath56 admits non - degenerate symmetric bilinear forms and let @xmath418 $ ] be a homotopy class of such forms . let @xmath150 be a base point and denote by @xmath421}$ ] the group of gauge transformations which leave fixed @xmath422 and the class @xmath418 $ ] . in view of proposition [ p:6 ] , @xmath423 defines a meromorphic function @xmath424}$ ] on @xmath425}$ ] . note that @xmath426}\to\operatorname{rep}(\gamma;v)$ ] is an principal holomorphic covering of its image which contains @xmath420 . we expect that the absolute value of this function is the square of modified ray singer torsion . the expectation is true when @xmath427 satisfies @xmath428 below . the pair @xmath427 satisfies _ property @xmath428 _ if it is the complexification of a pair @xmath429 consisting of a real vector bundle @xmath366 and a non - degenerate symmetric positive definite @xmath430-bilinear form @xmath363 and the space of flat connections @xmath365 is a real form of the space @xmath326 . we summarize this in the following theorem . [ t:1 ] with the hypotheses above we have . \1 . if @xmath195 and @xmath196 are two coeuler structures then @xmath431}=\mathcal{st}_{\text{\rm an}}^{{\mathfrak e}^*_2,[b]}\cdot e^{2([\omega(\nabla , b)],d^{-1}({\mathfrak e}^*_1-{\mathfrak e}^*_2))}\ ] ] with @xmath432 the poincar duality isomorphism . suppose that @xmath427 satisfies property @xmath433 . then : \2 . if @xmath434 is integral then @xmath424}$ ] is independent of @xmath418 $ ] and descends to a rational function on @xmath420 denoted @xmath435 . we have @xmath436}\bigr|=(t_{\text{\rm an}}^{{\mathfrak e}^*}\cdot\pi)^2.\ ] ] we expect that both 2 ) and 3 ) remain true for an arbitrary pair @xmath427 . property 5 ) in proposition [ p:6 ] shows that up to multiplication with a root of unity the complex ray singer torsion can be defined on all components of @xmath5 , since @xmath437 is trivial for sufficiently large @xmath13 . consider a smooth triangulation @xmath169 of @xmath9 , and choose a collection of orientations @xmath128 of the simplices of @xmath169 . let @xmath150 be a base point , and set @xmath438 . let @xmath2 be a finite dimensional complex vector space . for a representation @xmath7 , consider the chain complex @xmath439 associated with the triangulation @xmath169 which computes the cohomology @xmath440 . denote the set of simplexes of dimension @xmath112 by @xmath111 , and set @xmath441 . choose a collection of paths @xmath442 from @xmath164 to the barycenters of @xmath169 as in section [ ss:3.1 ] . choose an ordering @xmath443 of the barycenters and a framing @xmath444 of @xmath2 . using @xmath445 , @xmath443 and @xmath444 one can identify @xmath446 with @xmath447 . we obtain in this way a map @xmath448 which sends @xmath6 to @xmath449 . a look at the explicit definition of @xmath450 implies that @xmath451 is actually a regular map between two algebraic sets . change of @xmath452 changes the map @xmath451 . recall that the triangulation @xmath169 determines euler vector fields @xmath181 which together with @xmath445 determine an euler structure @xmath149 , see section [ ss:3.1 ] . note that the ordering @xmath443 induces a cohomology orientation @xmath453 in @xmath454 . in view of the arguments of @xcite or @xcite one can conclude ( cf . @xcite ) : [ p:7 ] if @xmath455 different choices of @xmath456 provide the same composition @xmath457 provided they define the same euler structure @xmath158 and homology orientation @xmath453 . in view of proposition [ p:7 ] we obtain a well defined complex valued rational function on @xmath5 called the milnor turaev torsion and denoted from now on by @xmath458 . [ t:2 ] 1 . the poles and zeros of @xmath458 are contained in @xmath459 , the subvariety of representations @xmath20 with @xmath460 . . the absolute value of @xmath461 calculated on @xmath455 is the modified ray singer torsion @xmath462 , where @xmath463 . \3 . if @xmath464 and @xmath465 are two euler structures then @xmath466 and @xmath467 where @xmath468 is the regular function on @xmath5 defined in section [ ss:4.1 ] . \4 . when restricted to @xmath420 , @xmath56 a complex vector bundle equipped with a non - degenerate symmetric bilinear form @xmath70 so that @xmath427 satisfies property @xmath428 , @xmath469 , where @xmath463 . we expect that 4 ) remains true without any hypothesis . parts 1 ) and 3 ) follow from the definition and the general properties of @xmath169 , part 2 ) can be derived from the work of bismut zhang @xcite cf . also @xcite , and part 4 ) is discussed in @xcite , remark 5.11 . let @xmath19 be a vector field on @xmath9 satisfying ( h ) , ( eg ) , ( l ) , ( ms ) and ( nct ) from section [ ss:2.3 ] . choose orientations @xmath128 of the unstable manifolds . let @xmath150 be a base point and set @xmath438 . let @xmath2 be a finite dimensional complex vector space . for a representation @xmath7 consider the associated flat bundle @xmath236 , and set @xmath470 , where @xmath111 denotes the set of zeros of index @xmath112 . recall that for @xmath471 , @xmath472 and every homotopy class @xmath473 parallel transport provides an isomorphism @xmath474 . for @xmath129 and @xmath475 consider the expression : @xmath476 if the right hand side of is absolutely convergent for all @xmath90 and @xmath91 they provide a linear mapping @xmath477 which , in view of proposition [ p:1](3 ) , makes @xmath478 a cochain complex . there is an integration homomorphism @xmath479 which does not always induce an isomorphism in cohomology . recall that for every @xmath7 the composition @xmath480 factors through conjugacy classes to a function @xmath481\to{\mathbb c}$ ] . let us also consider the expression @xmath482}\mathbb z_x(\gamma)(\operatorname{tr}\cdot\rho^{-1})(\gamma).\ ] ] again , the right hand side of will in general not converge . [ p:8 ] there exists an open set @xmath21 in @xmath5 , intersecting every irreducible component , s.t . for any representation @xmath483 we have : \a ) the differentials @xmath484 converge absolutely . \b ) the integration @xmath485 converges absolutely . \c ) the integration @xmath485 induces an isomorphism in cohomology . \d ) if in addition @xmath11 , then @xmath486\in\sigma } \mathbb z_x(\gamma)(\operatorname{tr}\cdot\rho^{-1})(\gamma)\biggr|\ ] ] converges , cf . . here the inner ( finite ) sum is over all @xmath121 $ ] which give rise to @xmath487 . this proposition is a consequence of exponential growth property ( eg ) and requires ( for d ) ) hutchings lee or pajitnov results . a proof in the case @xmath11 is presented in @xcite . the convergence of is derived from the interpretation of this sum as the laplace transform of a dirichlet series with a positive abscissa of convergence . we expect d ) to remain true for @xmath2 of arbitrary dimension . as in proposition [ p:8 ] . ] in this case we make precise by setting @xmath488\in\sigma } \mathbb z_x(\gamma)(\operatorname{tr}\cdot\rho^{-1})(\gamma).\ ] ] a lyapunov closed one form @xmath88 for @xmath19 permits to consider the family of regular functions @xmath489 , @xmath490 , on the variety @xmath1 defined by : @xmath491 if converges then @xmath492 exists for @xmath20 in an open set of representations . we expect that by analytic continuation this can be defined for all representations except ones in a proper algebraic subvariety . this is the case when @xmath11 or , for @xmath2 of arbitrary dimension , when the vector field @xmath19 has only finitely many simple closed trajectories . in this case @xmath492 has an analytic continuation to a rational function on @xmath1 , see section [ s:8 ] below . as in section [ ss:6.1 ] , we choose a collection of paths @xmath493 from @xmath164 to the zeros of @xmath19 , an ordering @xmath494 of @xmath148 , and a framing @xmath444 of @xmath2 . using @xmath495 we can identify @xmath496 with @xmath497 , where @xmath498 . as in the previous section we obtain in this way a holomorphic map @xmath499 an ordering @xmath494 of @xmath148 is given by orderings @xmath500 of @xmath111 , @xmath501 . two orderings @xmath502 and @xmath503 are equivalent if @xmath504 is obtained from @xmath505 by a permutation @xmath506 so that @xmath507 . we call an equivalence class of such orderings a _ rest point orientation . _ let us write @xmath508 for the rest point orientation determined by @xmath494 . moreover , let @xmath158 denote the euler structure represented by @xmath19 and @xmath445 , see observation [ o:1 ] . as in the previous section , the composition @xmath509 is a holomorphic map which only depends on @xmath158 and @xmath508 , and will be denoted by @xmath510 . consider the holomorphic map @xmath511 defined by formula . the _ dynamical torsion _ is the partially defined holomorphic function @xmath512 the following result is based on a theorem of hutchings lee and pajitnov @xcite cf . @xcite . [ t:3 ] if @xmath11 the partially defined holomorphic function @xmath513 has an analytic continuation to a rational function equal to @xmath514 . it is hoped that a generalization of hutchings lee and pajitnov results which will be elaborated in subsequent work @xcite might led to the proof of the above result for @xmath2 of arbitrary dimension . let @xmath515 be a group , @xmath516 an isomorphism and @xmath2 a complex vector space . denote by @xmath517 the group whose underlying set is @xmath518 and group operation @xmath519 . a representation @xmath520 determines a representation @xmath521 the restriction of @xmath20 to @xmath522 and an isomorphism of @xmath2 , @xmath523 . let @xmath524 be a based point compact space with @xmath525 and @xmath526 a homotopy equivalence . for any integer @xmath13 the map @xmath527 induces the linear isomorphism @xmath528 and then the standard lefschetz zeta function @xmath529 more general if @xmath20 is a representation of @xmath0 then @xmath527 and @xmath530 induce the linear isomorphisms @xmath531 and then the @xmath20-twisted lefschetz zeta function @xmath532 let @xmath217 be a closed connected manifold and @xmath533 a diffeomorphism . without loss of generality one can suppose that @xmath534 is a fixed point of @xmath535 . define the mapping torus @xmath536 , the manifold obtained from @xmath537 identifying @xmath538 with @xmath539 . let @xmath540 be a base point of @xmath9 . set @xmath541 and denote by @xmath542 the isomorphism induced by @xmath535 . we are in the situation considered above with @xmath4 . the mapping torus structure on @xmath9 equips @xmath9 with a canonical euler structure @xmath158 and canonical homology orientation @xmath453 . the euler structure @xmath158 is defined by any vector field @xmath19 with @xmath543 where @xmath544 ; all are homotopic . the wang sequence @xmath545 implies @xmath546 iff @xmath547 for all @xmath13 . the cohomology orientation is derived from the wang long exact sequence for the trivial one dimensional real representation . for details . we have [ p:11 ] with these notations @xmath548 . this result is known cf . a proof can be also derived easily from @xcite . let @xmath19 be a vector field without rest points , and suppose @xmath19 satisfies ( l ) and ( nct ) . as in the previous section @xmath19 defines an euler structure @xmath158 . consider the expression . by theorem [ t:3 ] we have : [ o:5 ] with the hypothesis above there exists an open set @xmath549 so that converges , and @xmath550 is a well defined holomorphic function on @xmath21 . the function @xmath550 has an analytic continuation to a rational function on @xmath5 equal to @xmath551 . the set @xmath21 intersects non - trivially each connected component of @xmath5 . if @xmath9 is obtained by surgery on a framed knot , and @xmath11 , then @xmath552 , and the function @xmath553 equals the alexander polynomial of the knot , see @xcite . any twisted alexander polynomial of the knot can be also recovered from @xmath458 for @xmath2 of higher dimension . one expects that passing to higher dimensional representations @xmath458 captures even more subtle knot invariants . let @xmath9 be a @xmath2-acyclic manifold and @xmath389 a coeuler structure . using the modified ray singer torsion we define a @xmath555 valued invariant ( which resembles the atiyah singer spectral flow ) for two representations @xmath556 , @xmath557 in the same component of @xmath6 . by a holomorphic path in @xmath6 we understand a holomorphic map @xmath558 where @xmath21 is an open neighborhood of the segment of real numbers @xmath559\times\{0\}\subset{\mathbb c}$ ] in the complex plane . for a coeuler structure @xmath389 and a holomorphic path @xmath560 in @xmath561 define @xmath562 here , for a smooth function @xmath535 of complex variable @xmath563 , @xmath564 denotes the complex valued @xmath53-form @xmath565 and the integration is along the path @xmath559 \times0\subset u$ ] . note that [ o:4 ] 1 . suppose @xmath56 is a complex vector bundle with a non - degenerate bilinear form @xmath70 , and suppose @xmath560 is a holomorphic path in @xmath566 . then @xmath567}(\tilde\rho(2 ) ) \big/\mathcal{st}_{\text{\rm an}}^{{\mathfrak e}^*,[b]}(\tilde\rho(1))\bigr ) \mod\pi.\ ] ] as consequence observation [ o:4 ] permits to define a @xmath555 valued numerical invariant @xmath554 associated to a coeuler structure @xmath389 and two representations @xmath578 in the same connected component of @xmath6 . if there exists a holomorphic path with @xmath579 and @xmath580 we set @xmath581 given any two representations @xmath556 and @xmath557 in the same component of @xmath6 one can always find a finite collection of holomorphic paths @xmath582 , @xmath583 , in @xmath6 so that @xmath584 for all @xmath585 , and such that @xmath586 and @xmath587 . then take @xmath588 in view of observation [ o:4 ] the invariant is well defined , and if @xmath389 is integral it is actually well defined in @xmath589 . this invariant was first introduced when the authors were not fully aware of `` the complex ray singer torsion . '' the formula is a more or less obvious expression of the phase of a holomorphic function in terms of its absolute value , the ray singer torsion , as positive real valued function . by theorem [ t:2 ] the invariant can be computed with combinatorial topology and by section [ s:7 ] quite explicitly in some cases . if the representations @xmath590 are unimodular then the coeuler structure is irrelevant . it is interesting to compare this invariant to the atiyah patodi singer spectral flow ; it is not the same but are related . let @xmath19 be a smooth vector field which satisfies ( h ) , ( l ) , ( ms ) , ( nct ) . suppose @xmath88 is a real valued closed one form so that @xmath89 , @xmath90 not a rest point ( lyapunov form ) . define the functions @xmath591 and @xmath592 by @xmath593 this result is proved in @xcite ; the @xmath2-acyclicity in part b ) is not necessary if ( eg ) is replaced by an apparently stronger assumption ( seg ) . prior to our work pajitnov has considered for vector fields which satisfy ( h ) , ( l ) , ( ms ) , ( nct ) an additional property , condition @xmath601 , and has verified part ( a ) of this conjecture . he has also shown that the vector fields which satisfy ( h ) , ( l ) , ( ms ) , ( nct ) and @xmath601 are actually @xmath127 dense in the space of vector fields which satisfy ( h ) , ( l ) , ( ms ) , ( nct ) . it is shown in @xcite that pajitnov vector fields satisfy ( eg ) , and in fact ( seg ) . let @xmath0 be a finitely presented group , @xmath2 a complex vector space and @xmath1 the variety of complex representations . consider triples @xmath602 where @xmath248 is a conjugacy class of @xmath0 and @xmath603 . define the rational function @xmath604 by @xmath605 where @xmath606 is a representative of @xmath607 let @xmath3 be a @xmath2-acyclic manifold and @xmath4 . note that @xmath608 $ ] identifies with the conjugacy classes of @xmath0 . suppose @xmath19 is a vector field satisfying ( l ) and ( nct ) . every closed trajectory @xmath102 gives rise to a conjugacy class @xmath109\in[s^1,m]$ ] and two signs @xmath609 . these signs are obtained from the differential of the return map in normal direction ; @xmath610 is the parity of the number of real eigenvalues larger than @xmath611 and @xmath610 is the parity of the number of real eigenvalues smaller than @xmath175 . for a simple closed trajectory , i.e. of period @xmath612 , let us consider the triple @xmath613,\epsilon_-(\hat\theta),\epsilon_+(\hat\theta)\bigr)$ ] . this gives a ( at most countable ) set of triples as in the previous paragraph . let @xmath614 be a lyapunov cohomology class for @xmath19 . recall that for every @xmath105 there are only finitely many closed trajectories @xmath102 with @xmath615)\leq r$ ] . hence , we get a rational function @xmath616 @xmath617)\leq r}\operatorname{let}_{\underline{\hat\theta}}\ ] ] where the product is over all triples @xmath618 associated to simple closed trajectories with @xmath615)\leq r$ ] . it is easy to check that formally we have @xmath619 it would be interesting to understand in what sense ( if any ) this can be made precise . we conjecture that there exists an open set with non - empty interior in each component of @xmath1 on which we have true convergence . in fact there exist vector fields @xmath19 where the sets of triples are finite in which case the conjecture is obviously true .

we describe new planetesimal accretion calculations in the kuiper belt that include fragmentation and velocity evolution . all models produce two power law cumulative size distributions , for radii 0.33 km and for radii 13 km . the power law indices are nearly independent of the initial mass in the annulus , ; the initial eccentricity of the planetesimal swarm , ; and the initial size distribution of the planetesimal swarm . the transition between the two power laws moves to larger radii as increases . the maximum size of objects depends on their intrinsic tensile strength , ; pluto formation requires 300 erg g . the timescale to produce pluto - sized objects , , is roughly proportional to and , and is less sensitive to other input parameters . our models yield 3040 myr for planetesimals with in a minimum mass solar nebula . the production of several ` plutos ' and 50 km radius kuiper belt objects leaves most of the initial mass in 0.110 km radius objects that can be collisionally depleted over the age of the solar system . these results resolve the puzzle of large kuiper belt objects in a small mass kuiper belt . # 1 # 1 # 1 = = = 1=1=0pt = 2=2=0pt 2ex 4ex scott j. kenyon harvard - smithsonian center for astrophysics 60 garden street , cambridge , ma 02138 e - mail : [email protected] 1ex and 1ex jane x. luu leiden observatory po box 9513 , 2300 ra leiden , the netherlands e - mail : [email protected] 3ex to appear in _ the astronomical journal _ july 1999

astro-ph9904115

recent discoveries of slow - moving objects beyond the orbit of neptune have radically changed our understanding of the outer solar system . these observations have revealed a large population of kuiper belt objects ( kbos ) in orbits with semi - major axes of 3945 au ( @xcite , 1995 ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . kbos with reliable orbits have a cumulative size distribution that follows @xmath14 , with @xmath15 ( @xcite ; @xcite ) . the estimated population of @xmath13 kbos with radii @xmath16 50 km indicates a total mass , @xmath17 g ] , for reasonable assumptions about the albedo and distance distribution . the small mass in kbos is a problem for current planet formation theories . in most theories , small planetesimals in the solar nebula grow by collisional accumulation ( e.g. , @xcite ; @xcite , 1984 ; @xcite , 1993 ; @xcite ) . this growth eventually produces one or more ` cores ' that accumulate most , if not all , of the solid mass in an annular ` feeding zone ' defined by balancing the gravity of the growing planetesimal with the gravity of the sun and the rest of the disk . large cores with masses of 110 @xmath18 can also accrete gas from the feeding zone ( @xcite ) . applied to the inner solar system , this model generally accounts for the masses of the terrestrial and gas giant planets ( e.g. , @xcite , @xcite ; but see @xcite ) . at 3050 au , however , the timescale to produce planet - sized , @xmath19 1000 km , objects exceeds the disk lifetime , @xmath20 100 myr , unless the mass of the outer disk is @xmath21@xmath22 ( e.g. , @xcite , 1984 ; @xcite ; @xcite , 1996 ; @xcite ) . the presence of large kbos in a small mass kuiper belt is thus a mystery . in kenyon & luu ( 1998 , hereafter kl98 ) , we began to address this issue by considering planetesimal growth in a single annulus at 35 au . we showed that calculations including accretion and velocity evolution naturally produce several `` plutos '' with radii exceeding 1000 km and numerous 50 km radius kbos on timescales of @xmath23 20200 myr for a wide range of initial conditions in plausible solar nebulae . these timescales indicate that pluto can form in the outer solar system in parallel with the condensation of the outermost large planets . in this paper , we extend kl98 by adding fragmentation to our planetesimal evolution code . the code generally matches published calculations at 1 au ( wetherill & stewart 1993 , hereafter ws93 ) and at 40 au ( davis & farinella 1997 ) . our numerical results demonstrate that fragmentation and velocity evolution damp runaway growth to provide a self - limiting mechanism for the formation of kbos and pluto in the kuiper belt . these calculations produce several plutos and at least @xmath24 50 km radius kbos on timescales of 2100 myr in annuli with modest surface densities , 2.00.14 g @xmath25 , of solid material at 35 au . our analysis sets a lower limit on the intrinsic strength of kbos , @xmath19 300 erg g@xmath8 , and indicates that the initial size distribution , the initial eccentricity of the planetesimal swarm , and the details of the fragmentation algorithm have little impact on the resulting size distribution of kbos . our results appear to resolve the mystery of large kbos in a small mass kuiper belt . planetesimal evolution at 3550 au in a minimum mass solar nebula ( @xcite ) naturally produces large kbos in numbers close to those currently observed . most of the disk mass ends up in smaller objects , 0.110 km , that can be collisionally depleted over the age of the solar system . this depletion rate depends on the intrinsic strength and eccentricity distribution of kbos ( @xcite ; @xcite ; @xcite , 1997b ) . future observations can place better constraints on these physical parameters and provide additional tests of our interpretation of kbo formation . we outline the fragmentation model and tests in sec . 2 , describe calculations for the kuiper belt in sec . 3 , and conclude with a discussion and summary in sec . the appendix contains a complete description of the fragmentation algorithms and updates of the coagulation code from kl98 . as in kl98 , we adopt safronov s ( 1969 ) particle - in - a - box method , where planetesimals are treated as a statistical ensemble of masses with a distribution of horizontal and vertical velocities about a keplerian orbit . our calculations begin with a differential mass distribution @xmath26 ) in a single accumulation zone centered at a heliocentric distance @xmath27 with an inner radius at @xmath28 and an outer radius at @xmath29 . the horizontal and vertical velocity dispersions of these bodies are @xmath30 and @xmath31 . we approximate the continuous distribution of particle masses with discrete batches having particle populations @xmath32 and total masses @xmath33 ( ws93 ) . the average mass of a batch , @xmath34 = @xmath35 , changes with time as collisions add and remove bodies from the batch . this procedure naturally conserves mass and allows a coarser grid than calculations with fixed mass bins ( @xcite ; @xcite ) . to evolve the mass and velocity distributions in time , we solve the coagulation and energy conservation equations for an ensemble of objects with masses ranging from @xmath36 g to @xmath37 g. the appendix of kl98 describes our treatment of accretion and velocity evolution , and compares numerical results with analytic solutions for standard test cases . in this paper , we add fragmentation to the problem . we adopt a simple treatment of collision outcomes , following greenberg ( 1978 ; see also @xcite , 1994 ; ws93 ) : 4ex 1 . mergers two bodies collide and merge into a single object with no debris ; \2 . cratering two bodies merge into a single object but produce debris with a mass much smaller than the mass of the merged object ; \3 . rebounds two bodies collide and produce some debris but do not merge into a single object ; and \4 . disruption two bodies collide and produce debris with a mass comparable to the mass of the two initial bodies . 4ex we consider two algorithms for treating the cratering and disruption of planetesimals . both the ws93 and the davis ( 1984 ) algorithms estimate the amount of debris @xmath38 produced from impacts with velocities exceeding a threshold velocity @xmath39 . in general , @xmath38 scales with the impact energy . ws93 assume that the debris has the same relative velocity @xmath40 of the colliding bodies ; davis ( 1984 ) assume that the debris receives a fixed fraction @xmath41 of the impact kinetic energy . in both cases , we adopt a coefficient of restitution @xmath42 to allow for rebound collisions that produce debris but no single merged body . we follow greenberg ( 1978 ) and adopt separate values of @xmath42 for collisions with ( @xmath43 ) and without ( @xmath44 ) cratering . the appendix describes these algorithms in more detail . to test our fragmentation procedures , we attempt to duplicate ws93 s calculations of planetary embryo formation at 1 au . their model begins with 8.33 @xmath45 planetesimals having radii of 8 km and a velocity dispersion of 4.7 m s@xmath8 relative to a keplerian orbit ( table 1 ; see also table 1 of @xcite ) . table 2 summarizes our results using the ws93 initial conditions with mass spacing factors of @xmath46 = 1.12 , 1.25 , 1.4 , and 2.0 between successive mass batches . we adopt the analytic cross - sections of spaute ( 1991 ) , which yield results identical to the numerical cross - sections of ws93 ( @xcite ) . figure 1 shows our reproduction of ws93 for @xmath47 . this model produces twelve 37 @xmath48 g objects with velocity dispersions of 60100 m s@xmath8 in 1.15 @xmath49 yr . in contrast , ws93 produce seven 13 @xmath50 g objects ( see figures 13 in ws93 ) . despite this factor of two difference in the total mass contained in the most massive bodies , our calculation has the same broad `` plateau '' in the cumulative number distribution @xmath51 at log @xmath52 = 2426 and a similar power law dependence , @xmath53 , at log @xmath52 = 2123 . our `` fragmentation tail '' at log @xmath52 = 718 has the standard power law dependence , @xmath54 ( @xcite ) , at large masses but flattens out more than the ws93 result at small masses . the evolution of particle velocities in our calculations generally agrees with ws93 ( figure 2 ; see figures 23 of ws93 ) . all velocities increase monotonically with time due to viscous stirring . the velocities of large bodies grow very slowly , because dynamical friction transfers their kinetic energy to small bodies . the model maintains a nearly constant ratio of vertical to horizontal velocity , @xmath55 0.53 , for all but the most massive bodies , which have @xmath56 ( figure 2 , right panel ) . this result agrees with previous calculations ( barge & pellat 1990 , 1991 ; @xcite ) . at late times , our velocities for small bodies , @xmath57 200 m s@xmath8 at @xmath58 g , are roughly a factor of two larger than those of ws93 . velocities for large bodies , @xmath57 100 m s@xmath8 at @xmath59 g , roughly equal those of ws93 . the higher velocities of our calculation lead to additional mass loss from gas drag and fragmentation . gas drag removes material from the annulus ; fragmentation produces bodies with masses less than the minimum mass , @xmath60 g , of our numerical grid . we typically lose @xmath19 25% of the initial mass to cratering and catastrophic fragmentation and another 20%25% to gas drag . ws93 lost a comparable amount of mass to fragmentation but only @xmath19 5% to gas drag . the agreement between our results and those of ws93 depends on the mass spacing factor @xmath61 . for @xmath62 30,000 yr , large @xmath61 models take longer to produce 10002000 km objects than small @xmath61 models . the ` lag ' relative to our @xmath61 = 1.12 model increases from 1%2% for @xmath61 = 1.25 up to @xmath19 10% for @xmath61 = 2 ( see also @xcite ; @xcite ; @xcite ; @xcite ; kl98 ) . the poor resolution of @xmath63 models delays the production of massive bodies that ` runaway ' from the rest of the mass distribution . in ws93 , most runaway bodies are also ` isolated ' bodies that do not interact with one another but do interact with the rest of the mass distribution ( see ws93 and the appendix ) . delays in the production of isolated , runaway bodies lead to 10%20% increases in the velocity dispersion of all bodies . larger velocities reduce gravitational focusing factors and slow the growth of the largest bodies . the mass of the largest object thus increases from 78 @xmath64 g for @xmath61 = 1.251.4 to @xmath65 g for @xmath61 = 1.12 . this trend suggests that calculations with @xmath61 = 1.011.10 , as in ws93 , would improve the agreement between our results and those of ws93 . the coagulation results also depend on the treatment of low velocity collisions . our standard expressions for long - range gravitational interactions fail when the velocity dispersion is close to the hill velocity , @xmath66 ( @xcite ; @xcite ) . we use ida s ( 1990 ) @xmath67-body calculations and adopt his simple expressions to derive the velocity evolution for collisions in the low velocity limit , @xmath68 , where @xmath69 = 25 is a constant ( see the appendix ) . this change slows down the velocity evolution , because it introduces a lower limit to the timescales for dynamical friction and the inclination component of viscous stirring ( @xcite ) . we adopt @xmath69 = 2 , following ws93 , for 1 au calculations . unlike ws93 , however , the small bodies in our calculation do not quite reach the low velocity limit and increase in velocity throughout the evolution . a modest increase in our low velocity limit , @xmath69 = 3.5 , leads to better agreement between the velocity behavior of our models and that of ws93 . to test our procedures further , we repeat the 1 au calculations using the davis ( 1984 ) fragmentation algorithm . table 3 lists our results for @xmath41 = 0.1 and @xmath61 = 1.25 , 1.4 , and 2.0 . these calculations yield solutions similar to our ws93 models . calculations with smaller @xmath61 produce larger bodies at earlier times than models with large @xmath61 ( table 3 ) . the mass of the largest object increases from @xmath70 g for @xmath61 = 2.0 to @xmath71 g for @xmath61 = 1.4 to @xmath72 g for @xmath61 = 1.25 . the velocity dispersions of the small bodies are a factor of 23 larger at @xmath73 30,000 yr because they receive a larger fraction of the fragmentation energy than in the ws93 algorithm . this trend reverses for @xmath74 50,000 yr , because the davis algorithm produces more debris for collisions between large objects than the ws93 algorithm . these large objects have small velocity dispersions , so the velocity dispersions of the small bodies increase more slowly than in the ws93 model . the mass lost from fragmentation , @xmath19 5% of the initial mass , and gas drag , @xmath19 20% of the initial mass , is also smaller . we conclude that our coagulation code with fragmentation and velocity evolution reproduces ` standard ' calculations . the results in tables 23 generally agree with published results , despite some differences in the evolution of the velocity dispersion and the radius of the largest object . these discrepancies probably result from subtle differences in the implementation of the algorithms . none seem significant , because the major results do not depend on the input parameters ( see also ws93 ) : all calculations produce several objects with @xmath75 g that contain most of the remaining mass in the annulus . as in kl98 , we rely on observations of other stellar systems and solar nebula models to choose appropriate initial conditions for our kuiper belt calculations . recent observations indicate lifetimes of @xmath19 510 myr for typical gaseous disks surrounding nearby pre main - sequence stars and for the solar nebula ( @xcite ; @xcite ; @xcite ) . in our solar system , neptune formation places another constraint on the kbo growth time , because neptune excites kbos through gravitational perturbations . recent calculations suggest neptune can form in 5100 myr ( @xcite ; @xcite ; @xcite ) . once formed , neptune inhibits kbo formation at 3040 au by increasing particle random velocities on timescales of 20100 myr ( @xcite ; @xcite ; @xcite ) . we thus adopt 100 myr as an upper limit to the kbo formation timescale at 3040 au . our starting conditions for kbo calculations are similar to kl98 . we consider an annulus centered at 35 au with a width of 6 au . this annulus can accommodate at least 10100 isolated bodies with @xmath76 g for @xmath77 . instead of the single starting radius used in kl98 , the present calculations begin with @xmath78 bodies in a size distribution with radii , @xmath79 = 180 m. this initial population has a power law cumulative size distribution , @xmath80 . we usually adopt @xmath81 ; the final size distribution appears to depend very little on @xmath82 , as discussed below . the planetesimals have a small initial eccentricity ( @xcite ) that is independent of size and an equilibrium ratio of inclination to eccentricity , @xmath83 = 0.6 ( @xcite ) . the mass density of each body is 1.5 g @xmath84 . to set the initial mass of the annulus , @xmath4 , we extend the minimum mass solar nebula to the kuiper belt and integrate the surface density distribution for solid particles , @xmath85 , across the 6 au annulus . the dust mass is then @xmath86 at 3238 au for @xmath87 = 1 au . most minimum mass solar nebula models have @xmath88 = 3060 g @xmath25 ( @xcite ; @xcite ; @xcite ) , which sets @xmath89 715 @xmath18 . we consider models with @xmath4 = 1100 @xmath18 to allow for additional uncertainty in @xmath88 . table 1 compares input parameters for kuiper belt models with initial conditions at 1 au ( see also @xcite ) . our success criteria are based on available observations of kbos . the present day kuiper belt contains ( a ) at least one object ( pluto ) with radius @xmath3 1000 km and ( b ) @xmath19 70,000 objects with radii exceeding 50 km ( @xcite ; @xcite , 1998 ) . a successful kbo calculation must satisfy both observed properties in @xmath20 100 myr . we quantify these criteria by defining @xmath90 as the radius of the largest object and @xmath91 as the radius where the cumulative number of objects is @xmath92 : a successful simulation has @xmath93 1000 km and @xmath94 50 km at @xmath62 100 myr . to provide another characteristic of these models , we define @xmath95 such that 95% of the mass is contained in objects with @xmath96 . in models with a long runaway growth phase , the largest objects contain most of the mass and @xmath97 . models with a limited runaway growth phase leave most of the mass in small objects , so @xmath98 . we end calculations with velocity evolution when @xmath90 exceeds @xmath19 1000 km . to evaluate the dependence of runaway growth on fragmentation , we extend calculations without velocity evolution to 5000 myr or to when @xmath90 exceeds @xmath19 20003000 km . to isolate how fragmentation changes the growth of kbos , we begin with constant velocity calculations of accretion . the initial size distribution has @xmath81 , @xmath61 = 1.4 , and a maximum initial radius of @xmath99 = 80 m. the initial velocities of @xmath100 = 4 m s@xmath8 and @xmath101 = 2.1 m s@xmath8 correspond to an equilibrium model with @xmath102 ( @xcite ) . we use the davis ( 1985 ) fragmentation algorithm and assume that the collision fragments receive a kinetic energy per unit mass equal to one - half of the square of the relative velocity of the colliding bodies , @xmath40 . the bodies are strong , with a tensile strength @xmath103 erg g@xmath8 . tables 1 and 4 summarize the initial conditions and results for models with @xmath4 = 1100 @xmath18 , @xmath5 = @xmath104 , and the coefficients of restitution , ( @xmath44,@xmath43 ) = ( @xmath105,@xmath105 ) and ( @xmath106,@xmath104 ) . figure 3 shows how @xmath51 evolves with time for @xmath4 = 10 @xmath18 , @xmath12 , and ( @xmath44,@xmath43 ) = ( @xmath105,@xmath105 ) . the low coefficients of restitution eliminate rebound collisions . roughly half of the most massive objects experiences at least one collision by @xmath107 16 myr , when the 18 largest bodies have @xmath108 1 km . these bodies reach @xmath109 10 km at @xmath107 135 myr and grow to 100 km sizes at 255 myr . the growth rate then increases considerably due to gravitational focusing . runaway growth ensues . the largest planetesimals reach @xmath110 200 km at @xmath107 265 myr ; @xmath90 exceeds 1000 km only 11 myr later . at @xmath107 280 myr , a single runaway body with @xmath110 2000 km begins to sweep up lower mass planetesimals and contains nearly all of the mass in the annulus a few myr later . the cumulative size distribution of small objects with @xmath111 100 m slowly approaches a power law with @xmath112 throughout the evolution . the distribution is shallow , @xmath113 , for @xmath62 20 myr as growth produces many 50500 m objects . the distribution steepens as the largest bodies grow past 1 km and settles at @xmath112 for @xmath74 150 myr . this distribution is steeper than the theoretical limit of @xmath114 ( @xcite ; @xcite ) , because growth and fragmentation never reach equilibrium . calculations with rebound collisions yield similar results . with larger coefficients of restitution , ( @xmath44,@xmath43 ) = ( @xmath106,@xmath104 ) , rebound collisions occur for @xmath111 5 m. these collisions produce fragments but no mergers into larger bodies . as a result , the size distribution is initially very steep for @xmath111 5 m and very shallow for @xmath115 5 m ( figure 4 ) . as the largest objects grow from 1 km to 10 km at @xmath116 = 16135 myr , bodies with @xmath111 5 m are swept up by the large bodies and replaced by collision fragments . this process smooths out the size distribution at @xmath109 120 m , although the slope still changes at @xmath109 5 m. the evolution of the largest bodies is unaffected by rebounds . these objects reach sizes of 100 km only slightly later than models without rebounds , 258 myr _ vs _ 255 myr , and then begin to runaway from lower mass planetesimals . several objects reach radii of 1000 km in another @xmath19 20 myr and then begin to consume all of the material left in the annulus . fragmentation is not important in any of these low eccentricity calculations . catastrophic fragmentation produces no mass loss . because the velocities remain artificially low . these models lose only 1%3% of their initial mass due to cratering . the timescale to grow into a 1000 km object is nearly identical to models without fragmentation ( see kl98 ) : pluto forms in @xmath11 276280 myr ( @xmath117)@xmath8 for @xmath99 = 80 m. calculations with larger @xmath99 should have smaller growth times ( see kl98 ) . to understand how fragmentation changes the velocity dispersions of colliding planetesimals , we consider calculations with ` limited ' velocity evolution . as in davis ( 1985 , 1994 ) , we assume that collision fragments receive a fixed fraction @xmath41 of the center - of - mass impact energy , but consider no other changes to the kinetic energy of the bodies . for most collisions , this assumption produces fragments with velocities larger than the relative collision velocity @xmath40 . to conserve kinetic energy , the merged bodies have a lower velocity dispersion after the collision . this redistribution of velocity mimics dynamical friction and should lead to more rapid growth than models with no velocity evolution . table 5 summarizes results for @xmath5 = @xmath104 , @xmath103 erg g@xmath8 , and coefficients of restitution , ( @xmath118 ) = ( @xmath119 ) , for our standard mass distribution with @xmath81 , @xmath61 = 1.4 , and @xmath99 = 80 m. figure 5 shows how @xmath51 evolves with time for @xmath4 = 10 @xmath18 . during the first 1020 myr , the size distribution has two power laws , @xmath120 for @xmath111 5 m and @xmath121 for @xmath115 5 m. this break in the slope is due to rebounds , which prevent growth of the smallest objects . as the largest objects grow to sizes of 110 km , collision debris adds to the population of 5 m objects . the slope of the size distribution is then @xmath122 for @xmath111 100 m. for large objects , @xmath123 approaches 2.5 once @xmath90 exceeds 100 km . once @xmath93 1000 km , the complete number distribution follows a power law with @xmath114 , except for a small kink at @xmath124 1 km whose amplitude decreases with time . limited velocity evolution eventually produces larger objects on shorter timescales than models without velocity evolution . in the first 2040 myr , collisions between the largest bodies , @xmath124 100500 m , add material with high velocity dispersion to batches with @xmath109 110 m. the horizontal velocity dispersions of these low mass objects increase from 4 m s@xmath8 to 1030 m s@xmath8 in 17 myr and reach a roughly constant value of @xmath19 20 m s@xmath8 after 200 myr ( figure 6 ) . this large velocity dispersion initially slows down the growth of the largest objects relative to models with no velocity evolution . it takes this model @xmath19 1 myr longer to produce objects with @xmath109 1 km , but these objects then have velocities @xmath19 1 m s@xmath8 smaller than the initial velocity of the planetesimal swarm ( see figure 6 ) . this smaller velocity dispersion enhances the growth of the largest objects once gravitational focusing becomes important . the model produces 10 km objects in 120 myr and 100 km objects in 202 myr . this rapid evolution leads to runaway growth and the production of a few 1000 + km objects after another 15 myr . figure 7 compares @xmath90 in models with no velocity evolution ( thin solid lines ) and models with limited velocity evolution ( thick solid lines ) . the annuli have initial masses @xmath4 = 1 , 10 , and 100 @xmath18 . these models are nearly indistinguishable for @xmath125 10 km . the limited velocity evolution models then produce larger objects on shorter timescales . these models do not have a mass - dependent velocity evolution such as viscous stirring or dynamical friction , so the timescale to produce 1000 km objects is still inversely proportional to the mass in the annulus , @xmath9 = 216 myr @xmath126 for @xmath5 = @xmath104 . other than the accelerated growth rate at late times , there is little difference between models with limited velocity evolution and models with no velocity evolution ( see tables 45 ) . both sets produce comparable numbers of large kbos , @xmath115 100 km , that contain most of the mass in the annulus when the first pluto forms . these models also lose @xmath19 1%3% of their initial mass due to cratering . the mass loss depends solely on the total kinetic energy of all the bodies , which is a constant in these models . catastrophic fragmentation still produces no mass loss , because we assume relatively strong objects with @xmath103 erg g@xmath8 . we describe models with weaker bodies in the next section . we now consider a complete coagulation calculation with fragmentation and velocity evolution . in addition to velocity redistribution from fragmentation , these models include ( i ) gas drag , ( ii ) dynamical friction and viscous stirring from long - range ( elastic ) collisions , and ( iii ) dynamical friction and viscous stirring from short - range ( inelastic ) collisions ( see the appendix of kl98 ) . to approximate the disappearance of the solar nebula on reasonable timescales , we assume that the gas density decays exponentially with time , @xmath127 where @xmath128 g @xmath84 ( ws93 and references therein ) . the influence of gas drag on the planetesimals thus decreases with an @xmath129-folding time of @xmath130 , which we set at @xmath130 = 10 myr ( @xcite ) . we adopt @xmath61 = 1.4 , @xmath81 , and @xmath99 = 80 m. the initial velocities are @xmath100 = 4.0 ( @xmath131 ) m s@xmath8 and @xmath101 = 2.1 ( @xmath131 ) m s@xmath8 . tables 67 summarize the initial conditions and results for models with @xmath4 = 1100 @xmath18 and @xmath5 = @xmath132 to @xmath106 . figure 8 shows how @xmath51 and @xmath100 evolve for @xmath4 = 10 @xmath18 , @xmath12 , and @xmath103 erg g@xmath8 . inelastic collisions act rapidly to circularize the orbits of objects with @xmath115 5 m , which decrease in velocity from @xmath100 = 4 m s@xmath8 to @xmath100 = 12 m s@xmath8 in 5 myr . larger bodies have less frequent collisions ; their orbits circularize on longer timescales . small bodies with @xmath111 5 m collide frequently , but rebounds prevent these bodies from circularizing their orbits . instead , collision fragments increase the velocity dispersion of the lowest mass bodies . these processes produce an inverted velocity distribution , where bodies with @xmath124 5 m have smaller velocities than objects with @xmath111 5 m and @xmath115 10 m. the steady damping of particle velocities enhances growth of the larger bodies relative to calculations with limited velocity evolution . the largest objects grow slowly for 8.5 myr , when there are @xmath19 500 objects with @xmath79 = 1 km . the size distribution then has three main features : ( i ) a pronounced fragmentation tail with a modest velocity dispersion , @xmath57 13 m s@xmath8 ; ( ii ) a transition region with pronounced kink in the size distribution ; and ( iii ) a group of rapidly growing bodies with low velocities , @xmath100 = 0.02 m s@xmath8 ( figure 8 ; left panel ) . these features remain prominent as @xmath90 increases from 1 km to 100 km in only 3.5 myr . during runaway growth , dynamical friction and viscous stirring begin to increase particle velocities ( figure 8 ; right panel ) . this phase ends when @xmath90 reaches @xmath19 300 km at 15 myr . the velocity dispersions of the small to intermediate mass bodies are then large enough , @xmath19 10 m s@xmath8 , to reduce gravitational focusing factors considerably . the largest objects enter a steady growth phase , where their radii grow slowly with time . there are 2 charon - sized objects with @xmath115 500 km at 19 myr ; 50 `` charons '' at 25 myr , and @xmath19 150 charons at 36.5 myr when the first `` pluto '' appears . the size distribution then closely follows a power law with @xmath114 for @xmath111 30 m and a steeper power law with @xmath133 for @xmath79 = 11000 km . the modest velocity dispersion of the largest objects , @xmath57 0.5 m s@xmath8 , maintains steady growth for another 60 myr : 4 objects have @xmath79 = 1450 km at 50 myr ; 8 objects have @xmath79 = 2000 km at 100 myr . figure 9 illustrates how the largest objects grow as a function of @xmath4 . all models begin with slow growth , where inelastic collisions reduce the velocity dispersions of the intermediate mass objects . slow growth lasts longer as @xmath4 decreases , because the growth rate depends on the collision rate . viscous stirring prolongs the slow growth phase of small @xmath4 models by counteracting collisional damping earlier in the evolution . slow growth thus produces more 12 km objects in models with smaller @xmath4 . runaway growth eventually turns 1 km bodies into 100 km objects . this phase ends when bodies with @xmath111 1 km have velocity dispersions exceeding 1020 m s@xmath8 . the large gravitational focusing factors that began runaway growth are then smaller by several orders of magnitude . the growth then returns to a steady phase where the largest bodies gradually approach radii of 1000 km . in these models , cratering also acts to damp runaway growth . as noted in sect . 3.3 , collision debris increases the velocities of the smallest objects . viscous stirring and dynamical friction enhance this evolution as the largest objects grow beyond 1 km . the velocity dispersions of small objects increase from @xmath19 20 m s@xmath8 at the end of runaway growth to @xmath19 100 m s@xmath8 at 100 myr . the largest objects then have @xmath124 2000 km . collision debris produces @xmath19 30% of this increase ; dynamical friction and viscous stirring are responsible for the rest . although all models with @xmath5 = @xmath104 make at least one pluto , only annuli with @xmath134 10 @xmath18 meet both success criteria . in each calculation , @xmath91 steadily increases with time during the slow growth phase ( figure 10 ) . the number of 50 km radius kbos increases dramatically during runaway growth , when gravitational focusing factors are large . at the end of runaway growth , @xmath91 begins to decrease with time as the largest bodies try to separate themselves from the rest of the mass distribution . this evolution is modest and short - lived , because the runaway growth phase ends quickly . the number of 50 km radius kbos then approaches a roughly constant value which grows with increasing @xmath4 . based on figure 10 , annuli with @xmath135 3 @xmath18 produce too few 50 km radius kbos compared with our success criterion ; annuli with @xmath136 30 @xmath18 probably produce too many . annuli with 3 @xmath18 @xmath137 30 @xmath18 can produce 1 or more plutos and roughly @xmath24 kbos in @xmath1 100 myr . this ` standard model ' has several important input parameters , including the initial eccentricity , the initial size distribution , and the intrinsic strength of the bodies . to understand how the evolution depends on these parameters , we now consider variations on the standard model . we begin with a discussion of @xmath5 and then describe models with different input strengths and size distributions . the velocity dispersion of the planetesimal swarm affects growth primarily through gravitational focusing . large bodies grow faster when velocity dispersions are small and gravitational focusing factors are large . despite small @xmath5 , all of our calculations begin in the `` high velocity regime , '' where gravitational focusing factors are near unity . collisional damping and dynamical friction reduce velocity dispersions to the low velocity regime where growth is rapid . planetesimals with small @xmath5 can reach this regime first , so we expect that smaller @xmath5 leads to more rapid growth . calculations with @xmath138 and @xmath106 confirm these general considerations . in models with @xmath139 , the larger initial velocities reduce collisional damping compared to models with @xmath12 . gravitational focusing factors thus increase slowly . additional cratering debris also counteracts collisional damping and combines with dynamical friction and viscous stirring to keep velocity dispersions large . as a result , models with @xmath139 experience a prolonged linear growth phase . this phase is @xmath19 10 times longer than for @xmath12 models . slow growth ends when the largest objects reach the sizes needed to produce modest gravitational focusing factors . a short runaway growth phase eventually produces several plutos . the final size distribution has a larger fraction of mass in more massive objects , as indicated by a large value for @xmath95 , but has roughly comparable numbers of 50 km radius kbos ( table 6 ) . these models also lose more mass to dust , @xmath19 50% , compared to the 1%3% lost in models with @xmath12 . growth is rapid in models with @xmath138 , because collisional damping quickly reduces the velocity dispersion . this change does not have a dramatic effect on the growth of kbos , because damping quickly reduces particle velocities to the low velocity limit . the damping time is then independent of velocity . viscous stirring is also less effective . in our calculations , the growth times for @xmath138 models are a factor of @xmath19 210 smaller than @xmath140 models ( table 6 ) . models with @xmath141 a few @xmath142 probably have similar growth times , but we have not investigated this possibility in detail . figure 11 compares the size distribution near the end of our calculations for three different values of @xmath5 . all models produce two power law size distributions , with @xmath114 for small objects and @xmath133 for large objects . these power laws are connected by a transition region which moves to larger radii with increasing @xmath5 . the fragmentation population thus extends to larger radii with _ larger _ @xmath5 . in contrast , the merger population ( the steep power law at larger radii ) extends to smaller radii with _ smaller _ @xmath5 . this feature of the calculations can be tested directly with observations ( @xcite ) the intrinsic strength @xmath6 of a planetesimal changes the growth rate by setting the impact energy @xmath143 needed to disrupt colliding bodies . in our disruption model , @xmath143 depends on the sum of @xmath6 and the gravitational binding energy of a colliding pair of planetesimals , @xmath144 ( eq . [ a11 ] and [ a13 ] ) . the gravitational term is small compared to @xmath6 when the combined radius of the merged object is @xmath145 km . the maximum radius of our initial size distribution , @xmath99 = 80 m , falls below this limit for @xmath146 25 erg g@xmath8 . for collisions between equal mass bodies , we can then derive a simple expression relating the minimum strength necessary for growth and the velocity dispersion , @xmath147 at the start of our calculations , this result yields @xmath148 bodies with @xmath141 0.1 initially grow for the standard case with @xmath6 = @xmath149 erg g@xmath8 . as these objects grow to 1 km sizes , viscous stirring and dynamical friction increase the velocities of small objects . once the velocities reach the threshold set by equation ( 1 ) , catastrophic disruption produces debris that is lost from the calculation . this process should limit the growth of the largest objects by reducing the reservoir of small bodies available for accretion . we thus expect the maximum size of kbos to depend on @xmath6 . to test these ideas in detail , we consider models with various @xmath6 . calculations with @xmath6 = 10 to @xmath149 erg g@xmath8 for @xmath150 and @xmath6 = @xmath151 to @xmath149 erg g@xmath8 for @xmath102 allow velocity evolution to increase particle velocities up to the disruption threshold . models with @xmath152 erg g@xmath8 and @xmath153 lose too much mass in the early stages to be of much practical value . models with @xmath154 erg g@xmath8 are identical to models with @xmath103 erg g@xmath8 . table 7 summarizes our results at 50 myr and 100 myr for @xmath155 and several different values of @xmath61 . for both values of @xmath5 , stronger objects can grow to larger sizes at 100 myr ( see figure 12 ) . in each model , accretion and collisional damping lead to a short runaway growth phase that produces 100 + km objects with low velocities ( @xmath156 0.1 m s@xmath8 ) , but leaves most of the initial mass in 0.11 km objects with much larger velocities ( @xmath157 310 m s@xmath8 ) . dynamical friction and viscous stirring then increase the velocities of these small objects to the disruption threshold . the timescale to reach this threshold increases with @xmath6 ; @xmath90 also increases with @xmath6 as indicated in table 7 and figure 12 . when all small bodies have been disrupted , the maximum radius is nearly independent of @xmath61 : @xmath158 for 10 erg g@xmath8 @xmath159 erg g@xmath8 and @xmath138 and @xmath104 . we did not run models with larger @xmath6 to the disruption threshold . future calculations will allow us to see whether @xmath90 reaches a threshold value at large @xmath6 or continues to increase as indicated by equation ( 3 ) . although @xmath90 depends on @xmath6 , both @xmath91 and the slope of the final size distribution at large radius are independent of @xmath6 . the radius limit for @xmath24 kbos has a small range , @xmath160 4560 km , for any combination of @xmath61 and @xmath6 ( table 7 ) . figure 13 shows the evolution of the size distribution for @xmath6 = 10 erg g@xmath8 . as in figure 8 and figure 11 , accretion produces a fragmentation tail with @xmath114 at small radii and a steeper power law with @xmath133 at large radii . the @xmath133 power law persists throughout the catastrophic disruption phase ; the fragmentation tail evolves into a very steep power law with @xmath161 . this behavior of the fragmentation tail occurs because larger objects initially experience more disruptive collisions than do smaller objects . catastrophic disruption adds high velocity fragments from larger objects to the smaller mass bins ; kinetic energy from this debris and viscous stirring gradually push smaller and smaller objects over the disruption threshold . eventually , the smallest objects in our grid , @xmath124 1 m , reach the disruption threshold and are slowly removed from the calculation . the initial size distribution is one of the most uncertain input parameters of our coagulation models . the growth of 110 m or larger bodies from interstellar dust grains is poorly understood . predicted sizes for conditions in the outer solar system range from @xmath1 1 m up to several hundred km depending on details of both microscopic and macroscopic physics ( e.g. , @xcite ; @xcite ; @xcite , 1993b ; @xcite ; @xcite ; @xcite ; @xcite ) . to test the sensitivity of our results to the initial conditions , we consider models with ( a ) @xmath162 , ( b ) @xmath163 , and ( c ) @xmath164 . these models produce final size distributions that are indistinguishable from our standard model with @xmath81 . both @xmath90 and @xmath91 are also independent of the initial size distribution . the time to runaway growth and the timescale to produce one pluto increase with the amount of material initially in the smallest objects , because collisional damping is more effective when there is a large reservoir of small objects . we derive @xmath9 = 30 myr for @xmath165 , @xmath9 = 37 myr for @xmath81 , @xmath9 = 42 myr for @xmath166 , and @xmath9 = 49 myr for @xmath167 . to test the importance of the mass spacing factor , we recomputed these models for @xmath61 = 1.25 . our results for ( a ) , ( b ) , and ( c ) in the preceding paragraph are indistinguishable from results with @xmath61 = 1.4 , except that the timescale to produce pluto decreases by @xmath19 5% . the model with @xmath61 = 1.25 and our standard initial size distribution , @xmath81 , produced a single large body that ran away from the rest of the large bodies . the final radius of this single object is @xmath19 50% larger than @xmath90 for other calculations with @xmath168 . several tests indicate that forming a single large body is a stochastic process sensitive to @xmath61 : small @xmath61 models produce such an object more often than large @xmath61 models . we plan to investigate this possibility further in future studies . to conclude this section , we briefly comment on the sensitivity of our calculations to other input parameters listed in table 1 . the results described above are insensitive to factor of two variations in the particle mass density @xmath169 , the relative gas velocity @xmath170 , the minimum velocity for cratering @xmath39 , and the crushing energy @xmath171 . we suspect that factor of ten variations in @xmath170 , @xmath39 , and @xmath171 will also have no impact on the results . larger variations in @xmath169 would probably change the variation of @xmath90 with @xmath6 in table 7 , but we have not investigated this possibility . increasing the fraction of kinetic energy imparted to fragmentation debris @xmath41 decreases the time needed to produce 100 + km objects 10%20% ] . the growth time decreases by @xmath19 10% for @xmath41 = 0.1 and @xmath19 20% for @xmath41 = 0.2 . variations in the input @xmath41 do not change the final size distribution or the number of kbos as a function of @xmath4 . this parameter thus has less impact on the results than either @xmath5 or @xmath6 . we summarized the major uncertainties of our planetesimal calculations in kl98 and will repeat important points for the current models here . our choice of a single accumulation zone does not allow us to follow the evolution in semimajor axis of a planetesimal swarm ( @xcite ; @xcite ) . although multi - zone calculations are important for understanding migration and other long - term aspects of planetesimal evolution , single - annulus models are a reasonable first approximation for the early evolution of kbos . the growth of large nearby bodies , such as those that will merge to form neptune , should modify the velocity evolution of kbos once these bodies reach sizes much larger than 1000 km ( see @xcite ) . for most solar nebulae , these long - distance interactions should remain small until pluto forms beyond 30 au ( kl98 ; see also @xcite ; @xcite ; @xcite ; @xcite ) . calculations now underway will test this assertion in more detail . our relatively coarse mass grid probably overestimates the timescale to produce kbos and pluto by @xmath19 5%10% ( kl98 ) . the delay in runaway growth relative to a @xmath47 model is 3%5% for @xmath61 = 1.4 and 5%10% for @xmath61 = 2.0 . although these delays are small compared to the overall uncertainties in our algorithms , calculations with @xmath172 2 rarely produce one or more isolated bodies that grow much faster than smaller bodies . the better mass resolution of @xmath173 1.4 calculations allows more mass batches to satisfy the isolation condition , defined in the appendix , which leads to more accurate calculations of the cumulative mass distribution and the velocity evolution of the lowest mass objects . the lack of a rigorous treatment of gas dynamics probably has little impact on our results . gas drag removes @xmath1 1% of the initial mass from these models . gas accretion by large bodies is also insignificant . the minimum radius needed to capture gas from the disk is @xmath109 10002000 km for typical temperatures of 50100 k at 3050 au ( @xcite ; @xcite ) . our models reach this limit on timescales exceeding the lifetime of the gaseous disk , so we expect little gas accretion by kuiper belt bodies . our calculations probably overestimate the amount of mass lost to dust . at both 1 au and 35 au , losses from catastrophic fragmentation , cratering , and gas drag grow with increasing @xmath61 . in large @xmath61 models , small delays in the formation of runaway bodies allow dynamical friction and viscous stirring extra time to increase the velocity dispersions , and hence mass loss , of the smallest objects . these effects are probably small , @xmath19 10%20% , for most of our calculations . our choice of the initial size distribution has little impact on our results . models with @xmath166 and @xmath174 produce final size distributions very similar to those for calculations with @xmath81 . the timescale to produce 1000 km objects lengthens as @xmath82 decreases . the ` equilibrium ' size distribution with @xmath133 is similar to the observed size distribution of interstellar dust grains , @xmath175 23 ( e.g , @xcite ; @xcite and references therein ) . similar size distributions are derived from calculations of the growth of very small particles using measured sticking efficiencies ( e.g. , @xcite ) . aside from the timescale to produce 1000 km objects and the amount of mass lost to gas drag and fragmentation , the initial eccentricity distribution probably also has little impact on our conclusions . the number of 50 km radius kbos is not sensitive to @xmath5 ; @xmath95 increases with @xmath5 only for @xmath176 ( table 6 ) . the slope of the final size distribution is also insensitive to @xmath5 . we suspect that initial eccentricities outside the range considered here are unrealistic . viscous stirring and gas drag appear to set a lower limit on the eccentricity of small objects , @xmath177 . models with @xmath178 should thus closely resemble @xmath5 = @xmath132 models . for @xmath179 , collisions lead to substantial fragmentation and mass loss from the numerical grid . circularization is then less effective and growth is very slow ( see table 6 ) . these calculations probably poorly approximate reality : dynamical friction between 1100 m objects and smaller dust particles not included in our grid should reduce the velocities of 1100 m objects on very short timescales , because the mass loss is @xmath1 1% of the initial mass . ] . we expect that these calculations would then more closely resemble models with smaller @xmath5 , if dust particles can grow back into 1 m bodies . the final limitation of our model is the fragmentation algorithm . we adopt an energy - scaling fragmentation law , because other types of models have not been developed and tested for the low velocity conditions appropriate in the kuiper belt . other scaling models seem to yield better results for main belt asteroids than do energy - scaling models , but these models assume that the collision can be approximated as a point - like impact on a large body ( @xcite ; @xcite ; @xcite ) . this assumption is quite good for the high speed impacts , @xmath3 100500 m s@xmath8 , of strong , dense objects in the inner solar system . point - like impacts are probably rare for the lower velocity collisions of weaker , low density objects like kbos . our consideration of kbos with a large range of intrinsic strengths suggests that an improved fragmentation model would not change our results significantly . in kl98 and this paper , we have developed a time - dependent planetesimal evolution code to calculate the formation of kbos in a single annulus outside the orbit of neptune . the computer program includes coagulation with realistic cross - sections , energy - scaling algorithms to treat cratering and disruption , and velocity evolution using the statistical expressions of hornung ( 1985 ) . our numerical solutions match standard analytic test cases and generally reproduce the results of other accretion and collision calculations ( e.g. , ws93 ; @xcite ) . our calculations demonstrate that plausible models can satisfy current observations of the kuiper belt . several plutos and @xmath13 50 km radius kbos form in minimum mass solar nebulae with @xmath180 on timescales of 2040 myr . growth is more rapid in more massive nebulae and in planetesimal swarms with smaller initial velocities . the formation time is less sensitive to the initial size distribution , the intrinsic strength of kbos , and other input parameters listed in table 1 . each kuiper belt model yields a cumulative size distribution with two main features . objects with @xmath111 0.1 km follow @xmath14 with @xmath114 , as expected for collision fragments ( @xcite ; @xcite ) . larger objects with @xmath115 110 km follow a @xmath181 power law over several orders of magnitude in radius . these slopes do not depend on @xmath4 , @xmath5 , @xmath6 , @xmath41 , and @xmath82 , among other input parameters . kenyon & luu ( 1999 ) compare these results with observations . here , we note that the @xmath181 power law for large bodies is identical to the observed slope , @xmath182 ( @xcite ; @xcite ; but see also @xcite ) . fragmentation and velocity evolution are important components in the formation of present day kbos . fragmentation produces a large reservoir of small bodies that damp the velocity dispersions of the large objects through dynamical friction . these processes allow a short runaway growth phase where 1 km objects grow into 100 km objects . continued fragmentation and velocity evolution damp runaway growth by increasing the velocity dispersions of small objects . this evolution leaves @xmath19 1%2% of the initial mass in 50 km radius kbos . the remaining mass is in 0.110 km radius objects . fragmentation will gradually erode these smaller objects into dust grains that are removed from the kuiper belt on short timescales , @xmath60 yr ( see @xcite ; @xcite ) . thus , 50 km radius kbos comprise a small fraction of the original kuiper belt . fragmentation also limits the size of the largest object in the kuiper belt . the maximum radius ranges from @xmath110 450 km for @xmath6 = 10 erg g@xmath8 to @xmath93 3000 km for @xmath6 = @xmath149 erg g@xmath8 . pluto formation sets a lower limit on the tensile strength of kbos , @xmath183 erg g@xmath8 . these results suggest a refinement of our picture for kbo formation in the outer solar system . in kl98 , we speculated that velocity perturbations due to the growth and outward migration of neptune would limit the growth of kbos at radii @xmath1 1000 km . although this hypothesis is plausible ( see , for example , @xcite ; @xcite ) , our current models demonstrate that 501000 km radius kbos form naturally at @xmath19 35 au on timescales , @xmath184 10100 myr , comparable to the neptune formation time . although a few objects can reach 2000 + km radii on timescales of 23 @xmath9 , nearly all of the initial mass beyond 30 au remains in small , 1 km radius objects that can be depleted by collisional disruption ( @xcite ) or gravitational sculpting ( @xcite ) or both on timescales exceeding 100 myr . this evolution can account for the observation of 50 + km kbos in a currently small mass kuiper belt without intervention by neptune . finally , our new results further support the notion that kbos will form in other solar systems . the dusty circumstellar disks detected in many pre main - sequence stars suggest masses of 1100 @xmath18 at distances of 30100 au ( e.g. , @xcite ; @xcite ; see also @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . kbos with 50 + km radii can grow in this material as the central stars contract to the main - sequence if the disks are not too turbulent ( see @xcite , 1993b ) . smaller , 110 km radius kbos probably form in less massive pre main - sequence disks . our results also indicate that the growth of 1001000 km radius kbos is accompanied by substantial dust production , @xmath19 0.11 @xmath18 , in models with @xmath134 10100 @xmath18 . this dust could be responsible for the ringlike structures observed in pre main - sequence stars such as gg tau ( @xcite ) and older main sequence stars such as @xmath185 eri ( @xcite ) and hr 4796 ( @xcite ; @xcite ) . the less massive disks in @xmath186 lyr , @xmath186 psa , and @xmath187 pic may also harbor kbos if the dust masses are reasonably close to the ` maximum ' masses inferred for these systems ( @xcite ) . in all of these stars , the dynamics and mass distribution of dust may well provide useful constraints on the properties of presumed kbos . we hope to explore this possibility in future studies . 4ex we thank b. bromley for making it possible to run our code on the jpl cray t3d ` cosmos ' and the hp exemplar ` neptune ' and for a generous allotment of computer time through funding from the nasa offices of mission to planet earth , aeronautics , and space science . comments from a. cameron , f. franklin , m. geller , m. holman , s. starrfield , and j. wood greatly improved our presentation . we acknowledge g. stewart for clarifying details of the ws93 calculations . as described in kl98 , our accretion model assumes that planetesimals are a statistical ensemble of masses in a cylindrical annulus of width @xmath188 and height @xmath189 centered at a radius @xmath27 from the sun . the particles have horizontal @xmath30 and vertical @xmath31 velocity dispersions relative to an orbit with mean keplerian velocity @xmath190 ( see @xcite ) . we approximate the continuous distribution of particle masses with discrete batches having an integral number of particles @xmath32 and total masses @xmath33 ( @xcite ) . the average mass of a batch , @xmath34 = @xmath35 , evolves with time as collisions add and remove bodies from the batch . to compute the evolution of particle numbers and velocities in kl98 , we solved the coagulation equation and a set of velocity evolution equations for all mass bins @xmath191 during a time step @xmath192 . we assumed that bodies merge but do not fragment during collisions . we conserved kinetic energy in each collision and adopted a kinetic approximation to calculate velocity changes due to gas drag , dynamical friction , and viscous stirring . our explicit algorithm for solving these equations reproduced standard tests and other published calculations . where @xmath195 is the cross - section , @xmath196 for @xmath197 and @xmath198 for @xmath199 . the four terms in a1a2 represent ( i ) mergers of @xmath52 and @xmath200 into a body of mass @xmath201 , ( ii ) loss of @xmath202 through mergers with other bodies , ( iii ) addition of @xmath202 from debris produced by the collisions of other bodies , and ( iv ) loss of @xmath202 by gas drag . we consider below the mass lost to small fragments @xmath203 . the second term in a1a2 includes the possibility that a collision can produce debris but no merger ( rebounds ) . as in kl98 , we calculate the `` gravitational range '' of the largest bodies @xmath204 ( @xcite ) where @xmath205 and @xmath206^{1/3}$ ] is the mutual hill radius . as in ws93 , the isolated bodies are the @xmath207 largest bodies that satisfy the summation , @xmath208 . these isolated bodies can not collide with one another but can collide with other lower mass bodies . as in kl98 , we solve the complete set of evolution equations , a1a2 above and a7a8 from kl98 , using an explicit method that limits the time step automatically to prevent large changes in the dynamical variables . section 2 of the main text compares calculations at 1 au with results from ws93 . in the rest of this appendix , we describe fragmentation algorithms ( a.2 ) and updates to our velocity evolution algorithm ( a.3 ) . algorithms for collision outcomes rely on comparisons between the kinetic energy of the two colliding planetesimals @xmath209 and the binding energy of the merged planetesimal @xmath210 . the binding energy usually consists of an intrinsic tensile strength and the gravitational binding energy , @xmath211 . this ` energy scaling ' approach is sometimes replaced by other scaling laws to model the structure of the colliding bodies more accurately . for example , housen and collaborators ( see @xcite ; @xcite ; @xcite , 1994 ) describe a strain - rate scaling model to express the collision energy needed to disrupt a body in terms of its size and impact velocity . davis ( 1994 ) showed that both types of model can match observations of the asteroid belt in our solar system ( see also @xcite ; @xcite ; @xcite ; @xcite ) . in this paper , we adopt an energy scaling model . energy scaling has two main advantages for calculating the fragmentation of kuiper belt objects . the wide use of this approach allows us to compare results with many previous calculations . other models also seem inappropriate for the low velocity collisions anticipated in the kuiper belt . these models have been scaled for conditions in the inner solar system , where collision velocities exceed 100 m s@xmath8 ( see @xcite ) . the approximations made for these collision velocities fail when the impact velocity is smaller than 1001000 m s@xmath8 ( e.g. , @xcite , p. 239 ) . we expect velocities of @xmath1 10 m s@xmath8 for most collisions in the kuiper belt ( kl98 ) , and thus can not apply the housen ( 1990 ) and other sophisticated models in our calculations . the output of any fragmentation algorithm is the mass of the merger remnant @xmath202 and the mass distribution of the fragments . we assume @xmath201 , where @xmath203 is the mass of fragments _ ejected _ from the merged planetesimal . we consider two approaches to compute @xmath203 . in the first case , we follow ws93 and set the impact velocity as where @xmath40 is the relative velocity of the two colliding bodies ( equation a12 of kl98 ) and @xmath213 = @xmath214 is the mutual escape velocity . the center of mass fragmentation energy @xmath215 and the gravitational binding energy @xmath216 per unit mass are ws93 and most other studies assume that collisions produce debris when ( i ) the impact velocity exceeds a threshold velocity , @xmath223 , and ( ii ) the amount of ejected mass exceeds a threshold value , @xmath224 . the mass ejected in the collision is derived from simple energy considerations . the collision _ disrupts _ the merged object if the collision energy exceeds the binding energy , @xmath225 , where @xmath6 is the impact strength . in most of their calculations , ws93 further require that the mass fragmented by the impact exceed half of the mass of the merged object , @xmath226 , where the fragmented mass is ( @xcite , 84 ) : if the conditions for catastrophic disruption are not met , the collision produces a large body with @xmath230 and numerous small fragments with total mass @xmath38 ( a6 ) . ws93 assume that the amount of the fragmented mass that escapes the merged body is where @xmath236 , @xmath237 , @xmath238 , and @xmath239 ( @xcite ) . tests show that this distribution produces results similar to those of ws93 when we assume that the velocity of the escaping fragments equals the relative velocity of the colliding planetesimals , @xmath40 ( see section 2 ) . as a comparison to the ws93 fragmentation algorithm , we consider the davis ( 1985 ; @xcite , 1984 ; @xcite ) approach . we adopt their energy scaling formula and write the strength ( in erg g@xmath8 ) of a planetesimal as : where @xmath241 is a constant . we adopt @xmath241 = 1 in the absence of any useful estimates for kuiper belt objects ; davis ( 1985 , 1994 ) consider @xmath241 = 1100 for collisional evolution in the asteroid belt ( see also @xcite ) . davis assume that the ejecta receive a fixed fraction @xmath242 of the impact energy and have a power law mass - velocity distribution , where @xmath244 is a reference velocity . the impact energy needed to disrupt the planetesimal and accelerate 50% of the fragments to escape velocity follows from integration of ( a12 ) and energy conservation ( @xcite ) : davis assume that collisions produce debris when the impact velocity exceeds a threshold , @xmath223 . we also adopt the ws93 threshold for the ejected mass , @xmath224 . the collision disrupts the colliding bodies if @xmath246 . if this condition is met , the total mass lost is : if the collision does not disrupt the merged planetesimal , the mass in fragments is simply @xmath38 from ( a6 ) . to estimate the fraction of this mass that escapes , we follow the davis ( 1985 ) derivation of @xmath143 and integrate the mass distribution over velocity : with these definitions , collisions with @xmath253 produce mergers and debris ; collisions with @xmath254 produce debris but no merger . we assume that rebounds conserve energy and that the velocity difference between bodies after the collision is in addition to the tests described in section 2 , we repeat the collisional evolution of kbos described in davis & farinella ( 1997 ) . with suitable modifications to our algorithms for the collision cross - section and velocity evolution , we reproduce their results to @xmath19 10%20% . our procedure for treating the evolution of the horizontal @xmath100 and vertical @xmath101 components of the velocity dispersion follows the kinetic formalism developed by hornung ( 1985 ) and barge & pellat ( 1990 , 1991 , 1992 ) . in kl98 , we adopted the ws93 expressions for dynamical friction and viscous stirring due to long range encounters and reformulated the barge & pellat ( 1990 , 1991 ) expressions for inelastic collisions in terms of @xmath129 and @xmath256 ( see equations a19a22 of kl98 ) . we verified that our algorithm for computing the velocity evolution achieved the equilibrium value for the ratio of the vertical to the horizontal velocity @xmath257 = 0.6 ( @xcite ) and generally reproduced the velocity evolution of the ws93 earth calculation . in this paper , we make two modifications to our treatment of the velocity evolution and clarify our treatment of velocity evolution at low velocities . we assume that rebound collisions do not contribute to dynamical friction and viscous stirring for inelastic collisions ( equations a21 and a22 of kl98 ) . with this assumption , the orbits of low mass objects do not circularize when their velocity dispersion exceeds the rebound velocity ( equation a17 above ) . this behavior slows down the growth time of massive objects , because the collision cross - section remains in the high velocity limit . although the kinetic prescription of hornung ( 1985 ) accurately represents dynamical evolution at high particle velocities , it fails as the particle velocity approaches the hill velocity @xmath258 ( @xcite ; @xcite ; ws93 ) . ida ( 1990 ) used @xmath207-body calculations to derive the appropriate behavior at low particle velocities , @xmath259 25 @xmath260 . we generally reproduce ida s ( 1990 ) results if we adopt for dynamical friction ( see also @xcite , @xcite ) . in both equations , the ` hv ' subscript indicates the appropriate high velocity expression from kl98 ( equations a19 and a20 ) and @xmath263 is the value of @xmath264 at @xmath265 . these equations yield constant timescales for dynamical friction and the inclination component of viscous stirring ( @xcite ; figure 10 ) . the timescale for the eccentricity component of viscous stirring varies as @xmath266 at low velocity ( @xcite ) . we describe the sensitivity of the velocity evolution to the choice of @xmath69 in sec . 2 . any @xmath267 2 slows down velocity evolution significantly , because the most massive particles in our test calculations have @xmath268 0.30.5 @xmath260 . the velocity changes are then @xmath19 25 times smaller for viscous stirring and @xmath19 500 times smaller for dynamical friction ( see also @xcite , @xcite , @xcite ) . our results are not very sensitive to @xmath69 for @xmath269 : the velocity evolution slows down gradually as @xmath69 increases from 2 to 5 ( see sec . we adopt the middle of ida s ( 1990 ) range , @xmath69 = 3.5 , in our kuiper belt calculations . for comparison , ws93 adopt @xmath69 = 2 ; @xcite adopt @xmath69 = 0.8 for viscous stirring and @xmath69 = 2.6 for dynamical friction . in many of our kuiper belt calculations , the velocity of the largest bodies approaches zero due to collisional damping , dynamical friction , and fragmentation . to avoid divergences in the stirring rates from long - range encounters , we adopt a minimum horizontal velocity of @xmath104 m s@xmath8 and a minimum vertical velocity of @xmath270 m s@xmath8 . this ` floor ' to the velocity evolution maintains the appropriate @xmath271 for collisions with @xmath272 and allows larger time steps for the brief interval when growth is rapid and the viscous stirring rate is small . tests show that the growth rate of the largest bodies does not depend on the values for these limits . lower values for the ` floor ' result in somewhat higher velocities for the small mass objects , but these changes are small . once viscous stirring begins to dominate collisional cooling , the velocities of all bodies increase above our ` floor ' values . finally , to reduce the computation time of our velocity evolution algorithm , we approximate the @xmath273 , @xmath274 , and @xmath275 integrals in equations a19a22 of kl98 with polynomial expressions derived using _ mathematica _ 1988 - 1997 wolfram research , inc . ] our results match the integral expressions to better than 1 part in @xmath276 for @xmath271 = 01 . expressions ready for use in a computer calculation appear below . davis , d. r. , weidenschilling , s. j. , farinella , p. , paolicchi , p. , & binzel , r. p. 1989 , in _ asteroids ii , _ edited by r. p. binzel , t. gehrels , & m. s. matthews , tucson , univ . of arizona press , p. 805 lissauer , j. j. , pollack , j. b. , wetherill , g. w. , & stevenson , d. j. 1996 . `` formation of the neptune system . '' in neptune and triton , eds . d. p. cruikshank , m. s. matthews , and a. m. schumann ( u. of arizona press , tucson , pp . 37 - 108 .

we obtained self - similar solutions of relativistically expanding magnetic loops by assuming axisymmetry and a purely radial flow . the stellar rotation and the magnetic fields in the ambient plasma are neglected . we include the newtonian gravity of the central star . these solutions are extended from those in our previous work by taking into account discontinuities such as the contact discontinuity and the shock . the global plasma flow consists of three regions , the outflowing region , the post shocked region , and the ambient plasma . they are divided by two discontinuities . the solutions are characterized by the radial velocity , which plays a role of the self - similar parameter in our solutions . the shock lorentz factor gradually increases with radius . it can be approximately represented by the power of radius with the power law index of . we also carried out magnetohydrodynamic ( mhd ) simulations of the evolution of magnetic loops to study the stability and the generality of our analytical solutions . we used the analytical solutions as the initial condition and the inner boundary conditions . we confirmed that our solutions are stable over the simulation time and that numerical results nicely recover the analytical solutions . we then carried out numerical simulations to study the generality of our solutions by changing the power law index of the ambient plasma density . we alter the power law index from in the analytical solutions . the analytical solutions are used as the initial conditions inside the shock in all simulations . we observed that the shock lorentz factor increases with time when the power law index is larger than , while it decreases with time when the power law index is smaller than . the shock lorentz factor can be expressed as where is the power law index of the ambient plasma . these results are consistent with the analytical studies by . [ firstpage ] relativity - stars : magnetic fields - stars : neutron - stars : flare

1102.2076

soft gamma - ray repeaters ( sgrs ) and anomalous x - ray pulsars ( axps ) are believed to be young neutron stars with strong magnetic fields ( @xmath7 ) . they are categorized as magnetars ( see , e.g. , @xcite ; , for review ) . their rotation period @xmath8 and its time derivative @xmath9 are @xmath10 and @xmath11 , respectively . a magnetic field strength inferred by assuming the dipole emission from @xmath8 and @xmath9 is about @xmath12 . persistent x - ray emissions with the luminosity of @xmath13 are observed in sgrs and axps . sgrs are identified by the hard x - ray bursts . extraordinary energetic outbursts called giant flares are observed in three sgrs . the burst energy in the sgr 1806 - 20 giant flares on december 27 , 2004 , reached @xmath14 @xcite . since sgrs and axps are not accretion powered sources , it is believed that their activity is driven by dissipation of their magnetic fields . strong magnetic fields are created by the dynamo mechanism during the core collapse of a supernova progenitor . at this stage , the star becomes unstable against the convective mode since the entropy gradient becomes negative , ( @xmath15 ) due to the neutrino cooling @xcite . thus the infant neutron star can store a large amount of magnetic energies . the magnetic fields are also amplified after the birth of the magnetar . the interior of the star rotates differentially at its birth when the equation of state is stiff . the internal magnetic fields can be amplified up to @xmath16 due to the dynamo mechanisms @xcite . as the magnetic helicity is accumulated inside the star , the lorentz force exerted by the twisted magnetic fields balances with the rigidity of the crusts . when the critical twists are accumulated , the crustal rigidity can no longer sustain the lorentz force by the strong magnetic fields . the magnetic helicity is then injected into the magnetosphere . the resulting crustal motion induces the electric fields and it results in creating the potential difference between the foot - points of the magnetic loops . the particles injected from the interior of the star are accelerated along the magnetic field lines due to the potential difference . the accelerated particles initiate the avalanches of the pair creation @xcite . these particles carry the electric current , which twists the global magnetic fields . when the dynamical equilibrium is lost by the accumulated magnetic twists , the magnetic loops expand by the magnetic pressure gradient force . inside the magnetic loops , a current sheet similar to that of solar flares is formed . the magnetic reconnections taking place in the current sheet are responsible for the magnetic energy release and resulting flares @xcite . recent observations which indicate the topological change of the global magnetic fields before and after the giant flares support these models @xcite . motivated by the magnetar flare model , @xcite performed 2-dimensional relativistic force - free simulations of magnetar flares by injecting the magnetic twists into the magnetosphere . they showed that the initially dipole magnetic fields are twisted by the foot - point motion and the loop magnetic fields then expand due to the magnetic pressure gradient force . @xcite carried out 2-dimensional relativistic force - free simulations of expanding magnetic loops and showed that the lorentz factor defined by the drift velocity @xmath17 exceeds 10 ( see , * ? ? ? * for the definition of the drift velocity ) . these simulations indicate that the magnetic loops expand self - similarly . such self - similar solutions have been found in analytical studies . in the framework of the force - free dynamics , @xcite obtained self - similar solutions of the spherically expanding magnetic shell . @xcite found self - similar solutions of the relativistic force - free field in two dimensions . @xcite derived relativistic self - similar force - free solutions and analyzed them in detail . in the framework of the relativistic magnetohydrodynamics ( mhd ) , @xcite found self - similar solutions of the spherically expanding magnetic shells . recently , @xcite obtained self - similar solutions of magnetic loops ( not shell ) by extending the non - relativistic solutions obtained by @xcite . @xcite obtained solutions by ignoring the gravity from the central star . these authors studied outflows of the the magnetized plasma lifted up from the central star . however , they did not consider the interaction between the outflow and the interstellar matter . @xcite obtained non - relativistic self - similar solutions of the expanding magnetic loops interacting with the interstellar matter . in their models , the outflows and the ambient plasma are divided by a contact discontinuity . the forward propagating wave forms another discontinuity ( shock ) . this solution is useful to understand the coronal mass ejections in solar flares . @xcite employed this solution as a test problem to check the validity and accuracy of axisymmetric mhd codes . in this paper , we extend the analytic solutions given by @xcite by including the contact discontinuity and the shock by extending the non - relativistic model by @xcite to the relativistic regime . this paper is organized as follows . in [ derivation ] , we summarize the basic equations of self - similar relativistic mhd equations given by @xcite . in [ expsol ] , we show the solutions of these equations including the two discontinuities . these solutions represent the relativistic coronal mass ejection from the central star . such solutions are expected to explain the giant flares in magnetars . the physical properties of the solutions are shown in [ property ] . we also carried out the 2-dimensional relativistic mhd simulations to study the stability of the solutions . the analytical solutions shown in [ expsol ] are used as the initial and boundary conditions for simulations . these results are shown in [ numerical ] . we summarize our results in [ summary ] . in the following , we take the light speed as unity . the complete set of the relativistic ideal mhd equations is @xmath18 @xmath19\left(h\gamma\bmath v\right)= -\nabla p + \rho_e \bmath e+\bmath j \times \bmath b-\frac{g m \rho h \gamma^2 } { r^2 } \bmath e_r,\label{eq : mhdeom}\ ] ] @xmath20\left(\ln\frac{p}{\rho^\gamma}\right)=0 , \label{eq : mhdentropy}\ ] ] @xmath21 @xmath22 @xmath23 @xmath24 @xmath25 where @xmath26 are the electric field , the magnetic field , the current density , the velocity , the lorentz factor , the charge density , the mass density , the pressure and the specific heat ratio , respectively . the vector @xmath27 is a unit vector in the radial direction . newtonian gravity of the central star is included as an external force . here @xmath28 is the gravitational constant , and @xmath29 is the distance from the centre of the star . we replaced the gravitational force @xmath30 in @xcite with @xmath31 , which can treat the gravitational force of the relativistic plasma more properly @xcite . the relativistic specific enthalpy including the rest mass energy @xmath32 is written as @xmath33 where @xmath34 is the energy density of the matter including the photon energy coupled with the plasma , and @xmath35 $ ] is the non - relativistic specific thermal enthalpy . in the following , we take @xmath36 which corresponds to the relativistic radiation pressure dominant plasma . thus we can study the evolution of a fireball confined by magnetic fields . we ignore the stellar rotation and assume a purely radial flow . we also assume axisymmetry . the magnetic fields in spherical coordinates ( @xmath37 ) are then expressed in terms of two independent scalar functions @xmath38 and @xmath39 as @xmath40 in the following , we assume that the evolution of the magnetic loops can be described by a lagrangian coordinate @xmath41 : @xmath42 where @xmath43 is a scale function of time . we further assume that the flux function @xmath38 evolves with time @xmath44 and radial distance @xmath29 through the lagrangian coordinate @xmath41 , as @xmath45 the mhd equations are then written as @xmath46 @xmath47 @xmath48 @xmath49 @xmath50 @xmath51 + q \frac{\partial}{\partial \eta}\left(\frac{q}{\gamma^2 } \right)\right\ } + \frac{gmd\gamma}{\eta^2}\left(1+\frac{4\gamma p}{zd}\right)\label{eq : eomr}\ ] ] @xmath52 @xmath53+\frac{q}{\gamma^2}\frac{\partial q}{\partial\theta}=0,\label{eq : eomtheta}\ ] ] where @xmath54 is an operator : @xmath55 ( see , * ? ? ? * for derivation ) . we can readily solve equation ( [ eq : zt ] ) as @xmath56 where @xmath57 is a constant of integral . the radial velocity is obtained from equation ( [ eq : vel ] ) as @xmath58 by substituting equation ( [ sol : z ] ) into equation ( [ eq : eomphi ] ) , we obtain the function @xmath59 which has a form : @xmath60 where @xmath61 is an arbitrary function of @xmath38 . we note that the last term in equation ( [ eq : eomr ] ) includes @xmath62 , which is a function of time . thus this term does not admit the existence of the self - similar solutions because the thermal enthalpy in the gravity term explicitly depends on time . this violation of the self - similarity comes from the difference of the scaling law between the density and the pressure ( equations [ eq : p ] and [ eq : rho ] , also , see @xcite ) . however , when the contribution of the thermal enthalpy on the gravity is sufficiently small , we can obtain self - similar solutions of the relativistic expansion . the ratio of the gravity for the thermal enthalpy to the plasma inertia is written as @xmath63 where @xmath64 is the gravitational radius . thus when @xmath65 , the gravitational force for the thermal enthalpy @xmath66 is negligible when @xmath67 . in this paper , we consider the evolution of the relativistically expanding plasma at large distance where the gravity for the thermal enthalpy @xmath68 can be neglected . in such region , the plasma inertia is sustained by the pressure gradient and lorentz forces . when @xmath69 , we assume that the rest mass energy density much exceeds the thermal energy density ( i.e. , @xmath70 ) . in this regime , the gravity is expressed as @xmath71 . the lagrangian coordinate @xmath41 ( see equation [ eq : defeta ] ) then behaves as the self - similar parameter . without loss of generality , we can change independent variables of @xmath8 from @xmath72 to @xmath73 as @xmath74 by neglecting the thermal enthalpy in the gravitational force @xmath75 , equations ( [ eq : eomtheta ] ) and ( [ eq : eomr ] ) reduce to the self - similar equations as @xmath76 , \label{eq : eomtheta2}\ ] ] @xmath77 equations ( [ def : g ] ) , ( [ eq : eomtheta2 ] ) and ( [ eq : eomr2 ] ) are the set of the self - similar mhd equations . the explicit solutions can be constructed as follows . first we prescribe an arbitrary function @xmath78 and the function @xmath61 . the pressure function @xmath8 is determined from equation ( [ eq : eomtheta2 ] ) and the density function @xmath79 is obtained from equation ( [ eq : eomr2 ] ) . before presenting the solutions of the self - similar mhd equations , we have to note that the self - similar relativistic ideal mhd equations describe the free expansion of the magnetized plasma . by taking time derivative of @xmath80 , we obtain @xmath81 where we used equations ( [ eq : defeta ] ) , ( [ sol : z ] ) and ( [ sol : v ] ) . the plasma is neither accelerated nor decelerated , but it expands with the inertial speed keeping the force balance ( this can be confirmed by inserting equation ( [ sol : v ] ) into equation ( [ eq : mhdeom ] ) , or , see equation ( 36 ) in * ? ? ? [ expsol ] in the previous section , we showed the set of the self - similar relativistic mhd equations , which describes the plasma expanding with the inertial speed . in this section , we obtain solutions of these equations by imposing appropriate boundary conditions . we adopt the simple model of magnetic explosions according to low models ( * ? ? ? * see , fig . [ fig : model ] ) . equations ( [ eq : eomtheta2 ] ) and ( [ eq : eomr2 ] ) describe magnetized plasma outflowing from the central star ( outflow region ) . the outflow sweeps up an ambient plasma while it expands . a contact surface would be formed which separates the outflowing plasma with the swept - up ambient plasma . the contact surface is situated at @xmath82 . ahead of the contact surface , a forward wave propagating into an undisturbed ambient plasma forms a shock at @xmath83 . the swept - up ambient plasma is accumulated in the region in @xmath84 ( postshock region ) . ( contact discontinuity ) and @xmath83 ( shock).,width=302 ] first , we consider the outflow region where the magnetized plasma is lifted up from the central star . various models of the magnetic field configurations have been proposed @xcite . in this paper , we utilize the following flux function @xcite @xmath85\sin^2\theta\equiv a_0 f(\lambda \eta)\sin^2\theta,\label{sol : a}\ ] ] where @xmath86 here @xmath87 , @xmath88 , and @xmath89 are constants . the flux function given by equation ( [ sol : a ] ) has local maxima ( see fig . 4 in @xcite ) . the maximum corresponds to the centre of the flux ropes . this function can thus describe the coronal mass ejection from the central star . such relativistic coronal mass ejections can be applied to giant flares in magnetars @xcite . we express the toroidal magnetic field @xmath90 as a power series of @xmath38 : @xmath91 where @xmath92 are constants . note that the dependence of the toroidal magnetic field on the polar angle is represented by @xmath93 from equations ( [ def : a ] ) , ( [ eq : b ] ) , ( [ sol : a ] ) and ( [ sol : g ] ) . this indicates that the power law index @xmath94 and @xmath92 can be interpreted as the fourier modes of the toroidal magnetic fields in polar direction and their amplitudes , respectively . the modes and the amplitudes are determined by the boundary condition at which the magnetic shear is injected into the magnetosphere . the shear injection begins before the self - similar expansion starts . in this paper , we do not consider the details of the shear injection from the central star and leave them as free parameters since we consider the self - similar stage . the explicit forms of the magnetic fields are then written as @xmath95\right\ } \cos\theta,\label{sol : br}\ ] ] @xmath96\sin\theta,\label{sol : btheta}\ ] ] @xmath97 the pressure function of the outflowing plasma @xmath98 is obtained by inserting equations ( [ sol : a ] ) and ( [ sol : g ] ) into equation ( [ eq : eomtheta2 ] ) as @xmath99 where @xmath100\right\ } , \label{sol : pa}\ ] ] @xmath101 here @xmath102 arises from the integration . by substituting equations ( [ sol : psum])-([sol : pq ] ) into equation ( [ eq : eomr2 ] ) , the corresponding density function @xmath103 is expressed as @xmath104 @xmath105 @xmath106 } { 4\pi gm\eta ( 1-\xi\eta^2)^\frac{3}{2}}\left\{\begin{array}{ll } { \displaystyle \sum_{m+n\ne 1}\frac{m\alpha_m\alpha_n a_0\tilde a^{m+n-1}(\eta , \theta)}{m+n-1}},\\ \hspace{4 cm } { \rm for \ \ } m+n\ne 1,\\ { \displaystyle \sum_{m+n=1}m\alpha_m\alpha_na_0 \ln \tilde{a}(\eta,\theta ) } , \\ \hspace{4 cm } { \rm for\ \ } m+n=1 , \end{array}\right.\label{sol : dq}\ ] ] where the function @xmath107 is defined as @xmath108\sin(\lambda \eta ) -\lambda \eta(1-\xi \eta^2)(3 + 5\xi \eta^2)\cos(\lambda \eta)\right\}.\label{def : xi}\ ] ] the function @xmath109 describes the isotropic distribution of the plasma which is related with @xmath110 through equation ( [ eq : eomr2 ] ) . to determine functions @xmath102 and @xmath111 , we need one more relation between them . we assume that there is no energy / mass injection from the central star at this stage . since the contact surface separates the outflow region from the postshock region , the outflow plasma expands adiabatically . we then obtain another relation between @xmath110 and @xmath109 from the entropy conservation equation as @xmath112 where @xmath113 , @xmath114 and @xmath115 is a constant . substituting equations ( [ eq : p ] ) , ( [ eq : rho ] ) , and ( [ sol : v ] ) into equation ( [ eq : adiabatic ] ) , we obtain @xmath116 substituting equation ( [ eq : adiabatic2 ] ) into ( [ eq : eomr2 ] ) , we obtain the solutions , @xmath117 @xmath118 where @xmath119 is a constant of integral . these solutions describe the isotropic outflowing plasma in the outflow region . next , we consider the post shock region ( @xmath120 ) . the shocked plasma also moves in radial direction . we assume that the shocked plasma evolves self - similarly and obeys the same set of self - similar equations in the outflow region . then the contact surface moves in radial direction with a constant speed and the radius of the contact surface @xmath121 is expressed as @xmath122 where @xmath123 is a constant . we assumed that the magnetic fields in the ambient plasma can be neglected . from these assumptions , the shocked plasma obeys equation ( [ eq : eomr2 ] ) . we need another relation between the shocked gas pressure @xmath124 and the shocked gas density @xmath125 . note that we can not use the adiabatic relation because the ambient plasma flows into the postshock region from the shock surface at @xmath126 . the forward shock compresses and heats up the plasma , resulting in an increase in the entropy . thus the entropy in the shocked plasma should be determined by the shock condition at @xmath83 . rather than evaluating the entropy variation by the shock , we consider the jump conditions of the plasma density , the pressure , and the velocity . the entropy variation is determined after imposing the rankin - hugoniot relations between the undisturbed and shocked plasma @xcite . by assuming the strong shocks , the relativistic rankin - hugoniot relations are written as @xmath127 @xmath128 @xmath129 with an accuracy of @xmath130 @xcite . @xmath5 and @xmath131 are the shock lorentz factor and the plasma mass density of the undisturbed ambient plasma , respectively . here we ignore the thermal pressure in the undisturbed plasma by assuming the strong shock . from equations ( [ sol : v ] ) and ( [ eq : rhg ] ) , we obtain the time evolution of the shock radius as @xmath132 where @xmath133 and @xmath134 is a constant of integral . combining equations ( [ eq : rhp ] ) and ( [ eq : rhr ] ) , we obtain the relation between the @xmath124 and @xmath125 : @xmath135 here we use equations ( [ eq : p ] ) , ( [ eq : rho ] ) , ( [ sol : rs ] ) . substituting equation ( [ eq : pds ] ) into equation ( [ eq : eomr2 ] ) , @xmath124 and @xmath125 are expressed as @xmath136 @xmath137 where @xmath138 is a constant of integral . the function @xmath139 is expressed as @xmath140 . \label{def : k}\ ] ] the ambient plasma density is obtained by substituting equations ( [ sol : rs ] ) and ( [ sol : ds ] ) into ( [ eq : rhr ] ) as @xmath141 here @xmath142 should be determined from the following equation , @xmath143 note that the radial profile of the ambient plasma density is not arbitrary but determined by equation ( [ sol : rho0 ] ) . some authors derive the self - similar solutions by prescribing the density profile of the ambient plasma as @xmath2 , where @xmath1 is a constant @xcite . in our approach , we first prescribe the self - similar variables given in equation ( [ eq : defeta ] ) . then the outflow velocity is obtained by equation ( [ eq : vel ] ) . the ambient plasma density is determined by applying the rankin - hugoniot relations at the shock . thus the ambient plasma density can not have an arbitrary form , but it is uniquely determined . finally , we apply the boundary conditions at @xmath144 . since we assumed unmagnetized ambient plasma , the magnetic fields should vanish at @xmath144 . this condition can determine the parameters @xmath88 and @xmath89 in equation ( [ sol : a ] ) . the conditions that @xmath145 and @xmath146 are expressed as @xmath147 from equations ( [ sol : br ] ) and ( [ sol : bphi ] ) . another condition is that @xmath148 . this condition is written as @xmath149 equation ( [ eq : lambdarelation ] ) determines @xmath89 and then the parameter @xmath88 is obtained from equation ( [ eq : h0relation ] ) . note that the equation ( [ eq : lambdarelation ] ) has an infinite number of roots ( see fig . 4 in @xcite ) . the first root for @xmath150 arises at @xmath151 and the second root does at @xmath152 . the first root corresponds to the position of the centre of the flux ropes . we take the second root as the contact surface , i.e. , @xmath153 ( @xmath154 in fig . 4 of @xcite ) throughout this paper . the parameter @xmath88 is then determined from equation ( [ eq : h0relation ] ) as @xmath155 . another constraint on the parameter comes from pressure balance across the contact discontinuity . the pressure @xmath98 consists of three component of the pressure , @xmath156 , @xmath157 and @xmath102 . @xmath156 and @xmath157 are exactly zero at @xmath144 since latexmath:[$\tilde{a}|_{\eta=\eta_c}=\partial \tilde{a}/\partial \eta with @xmath124 at the contact surface . from this condition , the parameter @xmath138 is expressed as @xmath159 here we used equations ( [ sol : pi ] ) and ( [ sol : ps ] ) . the remaining parameters are @xmath57 , @xmath123 , @xmath134 , @xmath92 , @xmath94 , @xmath115 , and @xmath119 , where @xmath57 denotes the scaling of time and radius and @xmath123 describes the velocity of the contact surface . equation ( [ sol : rs ] ) determines @xmath134 by prescribing the shock radius when self - similar expansion starts . the twist injection at the central star determines the amplitude @xmath92 and the fourier mode number @xmath94 of the toroidal magnetic fields . a constant @xmath115 which appears in equation ( [ eq : adiabatic ] ) denotes the entropy of the isotropic plasma in the outflow region @xmath160 . the density at the contact surface when self - similar expansion starts determines the constant @xmath119 which appears in equation ( [ sol : di ] ) . ccc ( left ) and the toroidal magnetic fields . the contact surface is situated at @xmath161 . the parameters are taken as @xmath162 , @xmath163 and @xmath164 . , width=181 ] ( left ) and the toroidal magnetic fields . the contact surface is situated at @xmath161 . the parameters are taken as @xmath162 , @xmath163 and @xmath164 . , width=181 ] first , we concentrate on the structure of the magnetic loops in @xmath165 . [ fig : mag ] shows contours of the magnetic flux @xmath38 ( left ) and the toroidal magnetic fields ( right ) in the @xmath166 plane . the parameters are @xmath162 and @xmath163 . the poloidal mode number of the toroidal magnetic fields is @xmath164 . we can see the flux rope structures emerging inside the expanding magnetic loops . the centre of the flux ropes is situated at @xmath167 ( @xmath168 ) . since the magnetic fields have both poloidal and toroidal components , they describe the twisted flux ropes . the flux ropes rise in the @xmath169 direction with time . in the limit of @xmath170 , the magnetic fields are represented as @xmath171 from equations ( [ sol : br ] ) - ( [ sol : bphi ] ) . the toroidal magnetic fluxes are diluted by the expansion according to the flux conservation equation . the configuration of the magnetic fields approaches that of the split monopole . [ fig : pol ] shows contours of the gas pressure ( left ) and the density ( right ) in the @xmath172 plane . contributions from the isotropic plasma , @xmath102 and @xmath111 , are subtracted , so that the pressure and the density can be negative in these panels . the set of the parameters is the same as those in fig . [ fig : mag ] . accompanying the flux ropes , low density voids are generated . the toroidal magnetic fields can create such voids . as mentioned at the end of [ derivation ] , since the force balance is attained in our self - similar solutions , the lorentz force by the poloidal magnetic fields should balance with the gas pressure gradient force and the toroidal magnetic pressure gradient force . this indicates that the gas pressure decreases as the toroidal magnetic pressure increases . such voids exert the buoyancy force on the plasma in radial direction . [ fig : pol ] shows that the pressure gradient force ahead of the voids balances the buoyancy force . the parameters are taken as @xmath162 , @xmath163 , @xmath164 , and @xmath173 . , width=181 ] . the parameters are taken as @xmath162 , @xmath163 , @xmath164 , and @xmath173 . , width=181 ] next we consider the isotropic part of the outflowing plasma expressed in equations ( [ sol : pi ] ) and ( [ sol : di ] ) . when the constant @xmath119 is exactly zero , the plasma distribution reduces to that in hydrostatic states . the scaling comes from the assumption of adiabatic expansions with the polytropic index of @xmath36 . when @xmath119 is not zero , the plasma distribution differs from that in the hydrostatic states . when the flow speed is non - relativistic , i.e. , @xmath174 , the solutions reduce to those in non - relativistic mhd obtained by @xcite . since the enthalpy contributes to the plasma inertia , the correction term @xmath175 arises in the relativistic mhd . the plasma tends to be hydrostatic since @xmath176 when @xmath177 . in front of the outflow region , the ambient plasma is compressed by the shocks at @xmath126 and is accumulated in the post shock region ( @xmath178 ) . the contact surface at @xmath179 divides the outflowing plasma from the shocked plasma . since we assumed that the postshock gas evolves self - similarly according to the same basic equations for the outflow region , the contact surface has the constant velocity @xmath180 . the expansion speed of the shock radius @xmath181 is , however , not constant but it increases with time . the time evolution of the shock radius is expressed by equation ( [ sol : rs ] ) . [ fig : shockvel ] shows the radial profile of the outflow lorentz factor . dashed , dotted , and dot - dashed curves denote the lorentz factor at @xmath182 , respectively . thick curve denotes the time evolution of the shock lorentz factor @xmath5 , while thin solid curve does the time profile of the outflow lorentz factor at the shocks ( @xmath183 ) . the value of the parameter @xmath134 is @xmath184 , which corresponds to @xmath185 . the shock lorentz factor is larger than the outflow lorentz factor at @xmath126 by factor @xmath186 , as expected from relativistic strong gas dynamical shocks ( see equation [ eq : rhg ] ) . the undisturbed plasma is abruptly heated up by the shocks . plasma velocity suddenly becomes zero ahead of the shocks @xmath187 where the undisturbed plasma exists . as the shock propagates in the undisturbed plasma , the shock surface is accelerated . the shock lorentz factor and the shock radius can be expressed as @xmath188^{1/4 } \left(\frac{t}{r_0}\right)^{1/4}\simeq 0.87 \left(\frac{t}{r_0}\right)^{1/4 } , \label{sol : applorentz}\ ] ] @xmath189 respectively . here we approximate @xmath190 to obtain the first equality in equation ( [ sol : applorentz ] ) . while the shock surface moves with almost constant speed @xmath191 , the shock lorentz factor increases with time with the power law index of @xmath0 . this result comes from the density profile of the undisturbed plasma . as shown later , the undisturbed plasma density decreases with radius as @xmath192 ( see equation [ sol : rho0-apx ] ) . @xcite showed that when @xmath193 , the flow is accelerated when @xmath194 . although we take into account the gravity from the central star , which is not included in @xcite , the gravitational force is smaller than the pressure gradient force in this region . thus their relation can be adopted in our analysis . the outflow is gradually accelerated when @xmath3 . next we consider the radial profile of the plasma pressure and the density . [ fig : hydro - r ] shows the radial profile of the pressure @xmath195 ( thick solid curve ) , the density @xmath196 ( thin solid curve ) and the outflow lorentz factor ( dashed curve ) on the equatorial plane . the horizontal axis denotes the radius normalized by the shock radius @xmath197 . the shock lorentz factor is taken as @xmath198 . we take @xmath173 for simplicity . the other parameters are @xmath163 , @xmath199 , @xmath200 , and @xmath201 . the contact surface is situated at @xmath202 . behind the contact surface , the gas pressure and the density decrease with radius as @xmath203 and @xmath204 , respectively . the contact discontinuity is situated at @xmath205 . we take @xmath198 and @xmath173.,width=302 ] . the contact discontinuity is situated at @xmath205 . we take @xmath198 and @xmath173.,width=302 ] the density jump appears at @xmath206 , while the pressure is continuous ( contact surface ) . the swept - up ambient plasma is accumulated in region @xmath207 . the plasma density in the postshock region ( @xmath208 ) is larger than that in the outflow region . the strong pressure gradient force in this postshock region pushes the plasma in @xmath169 direction . the pressure gradient force balances with the inertia of the accumulated plasma . thus , the plasma flows toward @xmath169 direction with the inertial velocity . ahead the contact discontinuity , a strong shock appears at @xmath126 . we assumed the strong shock and neglected the ambient plasma pressure . the plasma is abruptly heated up by the shock . the ambient plasma density jump also appears at the shocks . the density of the shocked gas is larger than that of the ambient plasma by factor @xmath209 for @xmath198 as expected from the relativistic rankin - hugoniot relations ( see equations [ eq : rhr ] and [ eq : rhg ] ) . the density of the undisturbed plasma is described by equation ( [ sol : rho0 ] ) . from this equation , the density can be approximately represented by the power law of @xmath29 as @xmath210 here we approximate @xmath211 for relativistic flows . the density decreases with radius with the power law index of @xmath212 . the ambient plasma density decreases slightly faster than that inside the shocks ( @xmath204 ) . the shock lorentz factor thus increases with radius @xcite . next we consider the total energy contained within the spherical surface @xmath197 . let @xmath213 be the total energy in @xmath214 . as shown in @xcite , the virial theorem can be applied for the relativistic inertial flow : @xmath215-(\bmath e^2+\bmath b^2)(\bmath r\cdot d \bmath{\mathcal{a}})\right\ } , \label{eq : virial}\ ] ] where @xmath216 @xmath217 and @xmath218 shows the poynting flux . here @xmath219 denotes expanding spherical surface at @xmath83 . the third , fourth , and fifth terms of the right hand side of equation ( [ eq : virial ] ) represent the poynting flux , the work done by the gas pressure , the work done by the maxwell stress , respectively . let us evaluate the non - kinetic part of the energy , @xmath220 . the second term in the right hand side of equation ( [ eq : virial ] ) is zero because the polytropic gas index is @xmath36 . the fifth term of ( [ eq : virial ] ) is zero since the electromagnetic fields vanish at @xmath126 . the third term also becomes zero after the straightforward calculations . thus the non - kinetic part of the energy can be evaluated as @xmath221 here we used equations ( [ eq : rhp ] ) and ( [ eq : rhg ] ) . the energy @xmath222 does not depend on the amplitudes of the magnetic fields . as shown in [ expsol ] , the plasma density ( pressure ) consists of three parts , @xmath223 , @xmath224 and @xmath111 ( @xmath156 , @xmath157 and @xmath102 ) . while the components @xmath223 and @xmath224 depend on the magnetic field strength , @xmath111 is independent of them . @xmath223 and @xmath224 do not contribute to the non - kinetic part of the energy from equation ( [ eq : virial ] ) since they are exactly zero at @xmath144 . this indicates that the plasma interacting with the magnetic fields is in marginally stable state . @xcite showed that the inertial flow with the polytropic index @xmath36 represents the marginally stable state in non - relativistic mhd . this situation is also valid for the relativistic mhd . total energy contained inside the shocks is thus independent of the strength of the magnetic fields . by substituting equations ( [ sol : applorentz ] ) and ( [ sol : rho0-apx ] ) into equation ( [ sol : none ] ) , the time evolution of the energy @xmath222 is written as @xmath225 where we used equations ( [ sol : applorentz ] ) , ( [ sol : apprs ] ) , and ( [ sol : rho0-apx ] ) . note that the non - kinetic part of the energy @xmath222 is positive . this means that plasma speed exceeds the escape velocity determined by the gravitational potential . the kinetic energy @xmath226 given in equation ( [ eq : kinetic ] ) is expressed as @xmath227 where we used equations ( [ eq : rho ] ) and ( [ sol : v ] ) . note that equation ( [ eq : kinetic2 ] ) depend on time through @xmath41 . when we integrate inside the sphere with radius @xmath126 , the integration is carried out in @xmath228 $ ] in the self - similar space . since the flow is relativistic , i.e. , @xmath229 , @xmath230 is constant with time . thus both non - kinetic and kinetic energies are constant with time . strictly speaking , the total energy should increase with time since the shock surface sweeps up the ambient plasma . the rest mass energy of the swept - up plasma contributes to the increase in the total energy . this energy is , however , negligible because we assume the strong shocks . from equation ( [ eq : rhp ] ) , the rest mass energy density of the undisturbed ambient plasma is smaller than the kinetic energy density of the shocked gas by factor @xmath231 and negligibly small ( note that the relations [ eq : rhp ] - [ eq : rhg ] are correct with the accuracy of @xmath232 ) . for the same reason , @xmath222 is also independent of time with the accuracy of @xmath233 . using these facts , the shock lorentz factor is expressed as @xmath234 from equation ( [ sol : none ] ) . this result is equivalent with equation ( 16 ) in @xcite . although our solutions include the magnetic fields and the gravity from the central star , they do not contribute to the total energy because the system is in a marginally stable state . thus only the hydrodynamical ( isotropic ) part contributes to the the total energy . by inserting equations ( [ sol : apprs ] ) and ( [ sol : rho0-apx ] ) into this equation , we obtain @xmath235 again ( see equation [ sol : applorentz ] ) . the shock is thus accelerated when it propagates in the ambient plasma . in this section , we show results of relativistic mhd simulations to study the stability of our solutions . for this purpose , we use the analytical solutions as the initial conditions of the numerical calculations . the relativistic mhd equations are solved in two dimensions using polar coordinates ( @xmath29 , @xmath236 ) . we assume axisymmetry ( @xmath237 ) . the number of grid points is ( @xmath238 , @xmath239)=(3600 , 360 ) on the domain of @xmath240 in normalized unit and @xmath241 . the grid sizes are @xmath242 and @xmath243 . we use the hll method @xcite to calculate numerical fluxes . we utilize the modified ctu method @xcite to achieve the second order accuracy in space . in our analytical solutions , the strong shock is expected . such a strong shock can induce the numerical oscillations . to avoid the problem , we utilize the harmonic mean for smoother prescription @xcite . we use the constraint - transport method to satisfy the no monopole condition . we impose the outflow boundary conditions in outer radial boundary at @xmath244 and the symmetric condition at the axes at @xmath245 . the inner boundary conditions are imposed at @xmath246 by applying the time - dependent analytical solutions . we solve the whole region covering @xmath247 $ ] , although our analytical solution are symmetric at @xmath248 . so we can check whether our code maintains this symmetry . the parameters are the initial time @xmath249 , @xmath250 , @xmath251 , @xmath252 , @xmath163 , @xmath253 , @xmath254 , @xmath164 , @xmath173 , and @xmath255 . the initial maximum lorentz factor ( @xmath256 ) is @xmath257 . radius and time are normalized by the inner radius @xmath258 and its light crossing time , respectively . to obtain the analytical solutions , the ambient plasma pressure @xmath259 is not specified because we adopted ranking - hugoniot relations for a strong shock , so that the ambient pressure is negligible . however , in order for the ambient plasma to be in hydrostatic equilibrium , the gas pressure gradient force should balance with the gravitational force of the central star . since it is a hard task to reconstruct the self - similar solutions taking into account the ambient plasma pressure , we use the analytical solutions obtained by using approximate rankin - hugoniot relations given in equations ( [ eq : rhp])-([eq : rhg ] ) as the initial conditions and the ambient pressure is taken initially to be constant , which means that the ambient plasma is not in hydrostatic equilibrium . the ambient plasma slowly falls toward the central star due to the gravity of the central star . the parameters we used in numerical simulations are taken so that the free fall time @xmath260 is much larger than the dynamical time @xmath261 ( typically , @xmath262 ) . thus the free fall motion of the ambient plasma does not affect the dynamics of the expanding magnetic loops . @xcite adopted a different method for this problem such that the gravitational force is artificially subtracted in the ambient plasma to numerically recover the solutions of the non - relativistic coronal mass ejection obtained by @xcite . we confirmed that the results obtained by the method proposed by them are consistent qualitatively and quantitatively with those including the gravity in the ambient plasma . the ambient plasma pressure , which is not specified in the analytical solutions , is taken so small ( @xmath263 ) that it does not affect the dynamics of the outflows . [ fig : pres ] shows the pressure profile ( colour ) and the magnetic field lines ( curves ) at @xmath264 . the left figure shows analytical solutions and the right one does numerical results . [ fig : dens ] shows the density profile ( colour ) . the magnetic field lines in fig . [ fig : pres ] and fig . [ fig : dens ] are depicted as the isocontours of the flux function . the levels of the isocontours are identical in both figures . as time goes on , the magnetic loops containing the flux ropes expand in radial direction . the flux ropes carry the toroidal magnetic fields . the ambient matter inflowing through the shock is compressed and accumulated between the contact discontinuity and the shock . the numerical results excellently recover the analytical solutions . thick curves denote the numerical results , while thin curves do analytical solutions.,width=302 ] . thick curves denote the numerical results , while thin curves do analytical solutions.,width=302 ] ccc . horizontal axis shows the time and vertical axis shows the lorentz factor on the equatorial plane . solid curve shows the analytical solutions . grey contours show the lorentz factors at the grid points closest to the reference radius @xmath265 . , width=226 ] . horizontal axis shows the time and vertical axis shows the lorentz factor on the equatorial plane . solid curve shows the analytical solutions . grey contours show the lorentz factors at the grid points closest to the reference radius @xmath265 . , width=302 ] fig . [ fig : rogm ] shows the shock structure on the equatorial plane at @xmath264 . left panel shows the lorentz factor and the right one does the density and the pressure . thin solid curves denote the analytical solutions , while thick solid curves do the numerical results . at this time , the maximum lorentz factor is @xmath266 for the analytical solution . the peak lorentz factor in the numerical simulation is , however , @xmath267 . the difference comes from the shock flattening in the simulations . note that we adopt the harmonic mean to evaluate primitive variables on the cell surface . this method is more diffusive than other interpolation methods , such as the muscl type interpolations . although we utilize this method to avoid numerical oscillations at the strong shocks , it decreases the peak lorentz factor . also the density and the pressure profiles are diffused ( right panel of fig . [ fig : rogm ] ) . another reason comes from the assumption of strong shocks adopted to derive the approximate rankin - hugoniot relations ( [ eq : rhp ] ) - ( [ eq : rhg ] ) . these relations are correct with an accuracy of @xmath268 . since we take the initial lorentz factor at the shock as @xmath269 , a few percent error arises from the approximations . such a numerical diffusion produces the sound waves from the shocks . [ fig : gamma - evol ] shows the radial profile of the lorentz factor on the equatorial plane from the initial state at @xmath270 to @xmath271 with the interval @xmath272 . thick and thin solid curves show numerical results and analytical solutions , respectively , while the dashed line denotes a reference radius @xmath265 ( see fig . [ fig : gmevol ] ) . after the simulation goes on , the shock is flattened due to the numerical diffusion , generating the sound waves propagating inward and outward from the shock . the compressional waves extract a part of the fluid kinetic energy from the shock , resulting in the increase in the numerical velocity behind / ahead of the shock from the analytical solutions . as the wave propagates away from the shock front , the radial profiles approach those of analytical solutions . note that the amplitudes of the inward propagating wave decreases with time because the wave conserves the wave energy density @xmath273 . since the density increases inward , the velocity deviation @xmath274 decreases as the wave propagates . on models of the gravity , ( i ) @xmath71 ( dashed ) ( ii ) @xmath275 ( dash - dotted ) and ( iii ) without gravity . solid curve denotes the analytical self - similar solutions . an inset shows solutions in the different range @xmath276.,width=283 ] ccc on the density parameter of the ambient gas @xmath1 . dashed curves denote the numerical results for initial condition given by the analytical solutions ( @xmath3 ) . the other curves depict those for @xmath277 . , width=302 ] on the density parameter of the ambient gas @xmath1 . dashed curves denote the numerical results for initial condition given by the analytical solutions ( @xmath3 ) . the other curves depict those for @xmath277 . , width=302 ] fig . [ fig : gmevol ] shows the time evolution of the lorentz factor . solid curve shows the analytical solution . we measure the lorentz factor at a reference radius @xmath278 to avoid the effects of the shock flattening . since the simulation is carried out at discrete grid points , we plot the range of the lorentz factors of the mesh points closest to the reference radius @xmath265 ( grey contours ) . the numerical results deviate from the analytical solution when @xmath279 because the flattening of the shock front and the emission of the sound waves temporarily increase the velocity behind the shock . the numerical results are , however , consistent with the analytical solutions after the sound waves propagate away ( @xmath280 ) . then the lorentz factor increases with time as @xmath281 . next we numerically verify the approximation of neglecting the thermal enthalpy term @xmath66 in the gravitational force @xmath282 . when we derive analytical solutions of the self - similar expansion , we neglect @xmath66 in the gravity . although the specific thermal enthalpy @xmath66 becomes larger than unity behind the strong shock ( @xmath283 , see equations [ eq : rhp]-[eq : rhg ] ) , the gravitational force itself becomes small compared to the other forces in the self - similar stage when @xmath284 . to evaluate the contribution of the gravity , we carried out numerical simulations for three models of the gravity , i.e. , ( i ) @xmath71 , ( ii ) @xmath31 ( iii ) without gravity . figure [ fig : gcomp ] shows the radial profiles of the lorentz factor on the equatorial plane at @xmath264 with different models of the gravity . solid curve shows the self - similar solutions , while dashed , dash - dotted , and dotted curves show the numerical results for models ( i)-(iii ) , respectively . an inset shows solutions in the different range of @xmath29 , @xmath285 . behind the shock , numerical results are almost independent of the models of the gravitational force , indicating that the gravitational force is much smaller than the other forces . as we mentioned in 2 , the ratio of the gravity for the thermal enthalpy to the plasma inertia decreases with radius ( see equation [ eq : fratio ] ) . the ratio @xmath286 is 0.01 at the initial state @xmath287 and @xmath288 at the final state @xmath264 in our simulations . thus the gravity for the thermal enthalpy is negligible . this explains why the numerical results are independent of the gravity models . the specific thermal enthalpy @xmath66 is larger than unity just behind the shock , but the gravitational force is much smaller than the other forces when @xmath289 . although the gravity becomes important in the region where @xmath290 , the thermal enthalpy is negligible in this region ( i.e. , @xmath291 ) . thus , we can neglect the contribution of @xmath292 in the gravity ( see the inset of fig . [ fig : gcomp ] ) . we note that the numerical results for model ( iii ) deviate from the analytical solutions in this region . the plasma is accelerated in radial direction by the pressure gradient force , leading to the formation of shocks ( @xmath293 ) . since we use analytical solutions for inner boundary conditions at @xmath294 , the plasma is supplied from the inner boundary . when the plasma is not confined by gravity , the outflowing plasma forms second shocks at @xmath295 . when @xmath197 is close to @xmath296 , the contribution of @xmath66 in the gravity is not negligible , but we have to take into account the general relativistic effects in such region , so that the characteristic length @xmath296 enters into the formulations . in such regime , no self - similar solutions can be obtained . it is out of the scope of this paper to obtain solutions in this regime . next we carried out simulations with different density profiles of the ambient plasma to study the generality of the analytical solutions and the effects of the ambient density distribution on the loop dynamics . we substitute the density profile given in equation ( [ sol : rho0 ] ) with the power law profile as , @xmath297 where @xmath298 . the analytical solutions correspond to @xmath299 . we study the several case ( @xmath300 ) . the initial condition is given by the analytical solutions inside the shock . the dependence of the equatorial lorentz factor ( left ) and the density profile ( right ) at @xmath264 on the density parameter of the ambient gas @xmath1 are plotted in fig . [ fig : rogm - delta ] . dashed curve denotes the numerical results for initial conditions given by the analytical solutions ( @xmath299 ) . other curves depict those for @xmath301 . the peak lorentz factor decreases as @xmath1 decreases . this is because the shell becomes massive for a smaller @xmath1 by sweeping up the larger ambient plasma . the swept up plasma is accumulated behind the shock surface . the inertia from the excess plasma accumulated behind the shock decelerates the outflows and creates another discontinuity behind the shock . the discontinuity can be considered as the reverse shock . the compression ratio of the reverse shock is larger for the denser ambient plasma . . circles depict the numerical results for the initial conditions given by the analytical solution for @xmath3 . other symbols show the numerical results for @xmath302 . the lorentz factor is evaluated at @xmath303 , where @xmath304 is the radius where the lorentz factor is its maximum.,width=283 ] time evolution of the equatorial lorentz factor at the reference radius @xmath305 are shown in fig . [ fig : gmpeak ] . circles depict the numerical results for the initial conditions given by the analytical solutions for @xmath299 . other symbols show the numerical results for @xmath301 . the lorentz factor is evaluated at @xmath303 , where @xmath304 is the radius where the lorentz factor is its maximum . the peak lorentz factor increases with time when @xmath279 . this increase comes from the emission of the sound waves propagating from the shock front . when @xmath306 , the peak lorentz factor increases with time when @xmath307 . the critical value of @xmath1 whether the outflow is accelerated or not can be evaluated from the mass conservation . the plasma density of the outflow decreases with radius by @xmath308 according to the mass conservations ( see equation [ eq : rho ] ) . the rest mass energy of the ambient plasma accumulated in the shell thus increases with time when @xmath309 . on the other hand , when @xmath194 , the outflow is accelerated since the plasma inertia of the outflow decreases with time . naively , we can understand these processes from equation ( [ eq : gs ] ) . according to this equation , the flow is accelerated when the ambient plasma density decreases faster than @xmath308 . when @xmath310 , the shock is accelerated and its lorentz factor is proportional to the radius ( @xmath311 time ) . it indicates that the flow expands freely . the influence of the ambient plasma is almost negligible . @xcite derived the self - similar solution of the free expansion . the shock lorentz factor then increases with radius linearly . the shock profile for @xmath312 or a larger @xmath1 is consistent with the solution obtained by @xcite . inside the shock surface , our solutions are , however , not identical with their analytical solutions since our solutions include the magnetic fields and are intrinsically non - spherical . we derived axisymmetric relativistic self - similar solutions of the magnetic flux rope expansion by assuming the purely radial flow and ignoring the stellar rotation . by taking the self - similar variable as @xmath313 , the arbitrary function @xmath43 has a unique form given in equation ( [ sol : z ] ) . the mhd equations are then solved analytically . the solutions obtained in this paper are the extension of our previous work @xcite by considering the two discontinuities , the contact discontinuity and the shock . the contact discontinuity separates the outflowing plasma and the ambient plasma . the outgoing waves propagating in the ambient plasma form shocks . such a self - similar solutions including two discontinuities are derived by @xcite in non - relativistic mhd . our solutions are the extension of their solutions to the relativistic mhd . for the non - relativistic case , the compression ratio at the shock is determined by the specific heat ratio . the specific heat ratio is taken as @xmath36 in non - relativistic mhd equations . the system is marginally stable for the inertial flow . the compression ratio is then up to 7 for the strong shocks . in relativistic plasma , the sound speed is limited to @xmath314 for the ideal gas . the differences between the upstream flow velocity and the downstream wave velocity are larger for the larger flow velocity . this fact results in forming the strong shocks and the compression ratio can be larger than 7 . the ambient plasma is abruptly heated by the strong shocks . the hot plasma is accumulated in the shell between the shocks and the contact discontinuity . inside the contact discontinuity , the magnetic loops anchored to the central star are assumed to follow the flux rope solutions obtained by @xcite . the flux ropes are contained inside the global magnetic loops . such magnetic field configuration can be expected for the sgr flares after the magnetic energy is dissipated @xcite . we also carried out numerical simulations of two dimensional relativistic mhd by using the self - similar solutions as the initial and inner boundary conditions . since analytical solutions are obtained in this paper , we can apply them to check the accuracy of multi - dimensional relativistic mhd codes . many previous authors reported that the relativistic mhd code is verified by using the one dimensional shock tube problems . there are only a few standard multi - dimensional problems , such as the blast wave problem or the rotor problem . however , no exact solutions are known for these problems . we have shown that the self - similar solutions can be applied to check the accuracy of the relativistic mhd codes . numerical calculations show that the shock velocity strongly depends on the ambient plasma density . when the density profile is steeper than @xmath204 , the shock lorentz factor increases with radius . on the other hand , it decreases for profiles shallower than @xmath308 . such a behavior is expected from the consideration of the mass conservation @xcite . the density inside the shocks approximately decreases with radius as @xmath204 . when the decrease in the ambient plasma density is steeper than the shocked plasma density , the outflows can be accelerated . we can understand this from the energy conservation given in equation ( [ eq : gs ] ) . the time dependence of the shock lorentz factor is related to the ambient plasma density . when @xmath315 and the ambient plasma density is represented by the power law on @xmath29 ( @xmath193 ) , the shock lorentz factor is expressed as @xmath316 . numerical results agree with this relation . especially when @xmath317 , the shock lorentz factor linearly increases with radius . finally let us apply our results to the magnetar flares . in the discussion below , we concentrate on the extraordinarily energetic outbursts ( giant flares ) observed in sgrs . although axps as well as sgrs would be magnetars , the giant flares have not been detected from the axps . the reason would be that the giant flares are very rare events ( 1 per a few decades in the whole sky ) . the shock lorentz factor @xmath5 is estimated from the mass ejected by the flare @xmath318 , and equation ( [ sol : none ] ) as @xmath319 the expansion speed of the magnetic loop is relativistic when @xmath320 ( see , also * ? ? ? * ) . the mass density behind the forward shock is estimated from equations ( [ eq : rhr ] ) , ( [ eq : rhg ] ) and ( [ eq : gsn ] ) as @xmath321 the mass density in the ambient plasma @xmath322 estimated from equations ( [ eq : rhr ] ) , ( [ eq : rhg ] ) , ( [ eq : gsn ] ) and ( [ eq : rosn ] ) , is much larger than the goldreich - julian density . such dense coronal pair plasmas would be created by the pair production when the magnetically trapped fireball is formed on the surface of the magnetar @xcite . the temperature of the shocked gas is evaluated by using equation ( [ eq : rhp ] ) as @xmath323 here we assume the pair plasma . although the radiation spectrum of the initial spike in the giant flare is not well determined because of its short duration , the typical temperature indicated by the spectrum is a few @xmath324 @xcite , which is lower than @xmath325 estimated from our analytical solutions . we have to note that the temperature at the photosphere @xmath326 should be smaller than @xmath325 because the temperature decreases with decreasing the radius behind the shock ( see , fig . [ fig : hydro - r ] and fig . [ fig : rogm ] ) . the radius of the photosphere @xmath327 can be determined by the condition that the optical depth of the expanding magnetic loops @xmath328 satisfies @xmath329 where @xmath236 is the angle between the velocity vector and the direction of the photon propagation @xcite , and @xmath330 is the effective opacity . here @xmath331 and @xmath332 are the opacity for the electron scattering and the free - free absorption , respectively . we assume that the ambient plasma ( @xmath333 ) is optically thin . the electron scattering is the dominant source for the opacity inside the shocks . this is because the plasma temperature is increased by the shock heating , so that the free - free opacity ( @xmath334 ) is much smaller than that of the electron scattering . we numerically integrate equation ( [ eq : tau ] ) assuming that @xmath335 , the expansion energy @xmath336 , and the mass of the central star @xmath337 , where @xmath338 is the solar mass . the temperature at the photosphere @xmath326 calculated at @xmath339 from the analytical solutions ( equations [ eq : zt ] [ eq : p ] , [ eq : rho ] , [ sol : ps ] and [ sol : ds ] ) can roughly be fitted by @xmath340 when @xmath341 and @xmath342 . the result is almost independent of the radius of the contact discontinuity ( i.e. , @xmath343 ) . the temperature in the observer s frame is @xmath344 when @xmath345 . here we assume @xmath346 since the photosphere is very close to the shock surface . this result is consistent with the observational results when @xmath345 and @xmath347 . such hot , relativistically expanding magnetic loops are expected to be formed with the help of the magnetic reconnections . according to the scenario by @xcite , the magnetic reconnections inside the magnetic loops are responsible for the initial spike of the flares ( see , also * ? ? ? subsequently , the plasma is heated up by the shocks produced by the magnetic reconnection . by applying our models to the giant flares , the time evolution of the luminosity from expanding plasma is expressed as @xmath348 from @xmath349 and equation ( [ eq : tpsn ] ) . here we assumed @xmath350 . since @xmath351 from equations ( [ eq : gs ] ) and ( [ eq : ropower ] ) , we obtain @xmath352 because @xmath353 from equation ( [ sol : apprs ] ) . @xcite reported that the observed photon counts decreased exponentially with time after the initial spike . however , the photon counts at the earlier stage , whose decay time is very short ( @xmath354 ) , is not inconsistent with the power - law decay . in addition to the initial spikes , a hump is observed a few hundred seconds after the initial spike @xcite . it is considered that the energy is re - injected from the central star . we now consider another possibility for this hump . before the magnetic energy release , the toroidal magnetic energy can be comparable to that of the poloidal magnetic fields inside the magnetic loops . when the magnetic reconnection takes place inside the magnetic loops , the magnetic energy of the poloidal magnetic fields is converted to the plasma energies , generating the twisted flux ropes . the toroidal magnetic field energy of the flux ropes does not dissipate in this process . when the twisted flux ropes cross the alfvn radius ( @xmath311 light cylinder ) , the rest magnetic energy will be dissipated by interaction with the ambient global magnetic fields . the time scale that a flux rope crosses the light cylinder and releases the magnetic energy is about @xmath355 , which is consistent with the observations . the energy dissipation at the light cylinder can be responsible for the humps . we need further study to verify these processes . we are grateful to an anonymous referee for improving our manuscript . we thank hiroaki isobe , jin matsumoto , kazunari shibata , and youhei masada for useful discussions . numerical computations were carried out on cray xt4 at center for computational astrophysics , cfca , of national astronomical observatory of japan and on fujitsu fx-1 at jaxa supercomputer system ( jss ) of japan aerospace exploration agency ( jaxa ) . this work was supported by the grants - in - aid for scientific research of ministry of education , culture , sports , science , and technology ( rm:20340040 ) . k. , boggs s. e. , smith d. m. , duncan r. c. , lin r. , zoglauer a. , krucker s. , hurford g. , hudson h. , wigger c. , hajdas w. , thompson c. , mitrofanov i. , sanin a. , boynton w. , fellows c. , von kienlin a. , lichti g. , rau a. , cline t. , 2005 , , 434 , 1098 t. , tanaka y. t. , takei y. , kawai n. , yoshida a. , nomoto k. , yoshikawa i. , saito y. , kasaba y. , takashima t. , mukai t. , noda h. , murakami t. , watanabe k. , muraki y. , yokoyama t. , hoshino m. , 2005 , , 434 , 1110

to investigate the link between weak tidal interactions in disk galaxies and the boosting of their recent star formation , we obtain images and spatially integrated spectra ( ) for 40 late - type spiral galaxies ( sab - sbc ) with varying degrees of lopsidedness ( a dynamical indicator of weak interactions ) . we quantify lopsidedness as the amplitude , of the fourier component of the azimuthal surface brightness distribution , averaged over a range of radii . the median spectrum of the most lopsided galaxies shows strong evidence for a more prominent young stellar population ( i.e. strong balmer absorption , strong nebular emission , a weak break and a blue continuum ) when compared to the median spectrum of the most symmetric galaxies . we compare the young stellar content , quantified by and the strength of the break ( ) , with lopsidedness and find a correlation between the two . we also find a correlation between and lopsidedness . using the evolutionary population synthesis code of bruzual & charlot we model the spectra as an `` underlying population '' and a superimposed `` boost population '' with the aim of constraining the fractional boost in the sfr averaged over the past gyr ( the characteristic lifetime of lopsidedness ) . from the difference in both and the strength of the break ( ) between the most and least symmetric thirds of our sample , we infer that of stars are formed over the duration of a lopsided event in addition to the `` underlying '' sfh ( assuming a final galactic stellar mass of ) . this corresponds to a factor of increase in the sfr over the past years . for the nuclear spectra , all of the above correlations except vs. are weaker than for the disk , indicating that in lopsided galaxies , the sf boost is not dominated by the nucleus .

astro-ph0003109

galaxies do not live isolated lives , but exist in the tidal fields of their environment . arp ( 1966 ) , in his atlas of peculiar galaxies , lay the observational groundwork for the modern study of interacting galaxy systems by identifying many `` peculiar '' systems , later interpreted as various stages of major galaxy mergers . strong galaxy - galaxy interactions may dramatically alter the stellar populations ( e.g. larson & tinsley 1978 ; kennicutt _ et al . _ 1987 ; turner 1998 ; kennicutt 1998 ) , morphology ( e.g. toomre & toomre 1972 ; hernquist , heyl & spergel 1993 ) and kinematics of galaxies ( e.g. toomre & toomre 1972 ; barnes & hernquist 1992 ) driving evolution along the hubble sequence . massive mergers are also capable of funneling gas into the center of galaxies causing nuclear starbursts ( barnes & hernquist 1991 ; mihos , richstone & bothun 1992 ; barnes & hernquist 1996 ) and qso activity ( e.g. sanders _ 1988 ) . at the present epoch , however , major mergers are fairly rare events ( e.g. kennicutt _ et al . _ 1987 ) and their broad evolutionary importance is unclear . minor mergers and , in general , weak tidal interactions between galaxies occur with much higher frequency than major ones ( e.g. lacey & cole 1993 ) . by weak interactions we mean those which do not destroy the disk of the `` target '' spiral . hierarchical structure formation models ( e.g. cold dark matter ) predict that the merging histories for high mass objects today contained multiple low mass accretion events in their past ( e.g. lacey & cole 1993 ) . the specific roles which weak interactions play in the evolution of galaxies , however , is uncertain . weak interactions may cause disk heating ( e.g. toth & ostriker 1992 ; quinn , hernquist & fullagar 1993 ) and satellite remnants may build up the stellar halo ( e.g. searle & zinn 1978 ; johnston , hernquist & bolte 1996 ) . ( 1987 ) studied the relation between interaction strength and star formation by making a comparison between isolated galaxies , close pairs , and galaxies from the arp atlas . they found that close pairs have larger values of @xmath15 , i.e. higher star formation rates ( sfr ) than isolated galaxies . while pair spacing is weakly correlated with the sfr , they could not determine the specific role of interaction strength on the sfr . ( 1998 ) and allam _ ( 1999 ) both studied the hubble type specific effects of environment on the sfr in galaxies . they found that the sfr/_mass of existing stars _ was inversely proportional to the local galaxy density . they postulate that the anti - correlation is due partly to gas stripping and due partly to the anti - correlation of the merger cross - section with the galaxy - galaxy velocity dispersion . there is also evidence that interactions excite nuclear activity . in their close pair and strongly interacting sample kennicutt _ ( 1987 ) found a strong correlation between @xmath16 emission in the disk and that in the nucleus . such a correlation between disk and nuclear emission is supported by theoretical work ; mihos & hernquist ( 1994 ) and hernquist & mihos ( 1995 ) demonstrated that minor interactions form bar instabilities in the disk which in turn funnel large amounts of gas into the nucleus . the effectiveness of this process is suppressed by the presence of a dense bulge , which prevents bar formation . due to the numerical expense in computing high resolution n - body / sph ( collisionless particle / smoothed particle hydrodynamics ) models , the exact interaction parameters which result in such activity are uncertain . weak interactions may also manifest themselves as kinematic or structural irregularities . roughly @xmath17 of all spiral galaxies have asymmetric hi profiles and rotation curves ( baldwin , lynden - bell & sancisi 1980 ; richter & sancisi 1994 ; haynes _ et al . ( 1980 ) postulated that these asymmetries are caused by weak interactions in the galaxy s past or by lopsided orbits . ( 1999 ) examined the optical rotation curves of a set of observed and simulated interacting disk galaxies . they showed that interactions can cause large scale , time dependent asymmetries in the rotation curves of their sample galaxies . ( 1999 ) studied the kinematic asymmetries present in two galaxies lopsided in their optical and hi distributions . they qualitatively reproduced the kinematic asymmetries by placing closed orbits in mildly lopsided potential . a dynamical indicator of weak interactions may be `` lopsidedness . '' in the context of this paper ( following rudnick & rix 1998 ; hereafter rr98 ) , lopsidedness is defined as a bulk asymmetry in the _ mass _ distribution of a galactic disk . surveys for lopsidedness in the stellar light of galaxies were first carried out by rix & zaritsky ( 1995 ; hereafter rz95 ) and zaritsky & rix ( 1997 ; hereafter zr97 ) . using near - ir surface photometry of face - on spiral galaxies ( spanning all hubble types ) they examined the magnitude of the @xmath2 azimuthal fourier component of the i and k - band surface brightness , thus characterizing the global asymmetry of the stellar light . rz95 and zr97 found that a quarter of the galaxies in their sample were significantly lopsided . using a larger , magnitude limited sample restricted to early type disks ( s0 to sab ) and imaged in the r - band , rr98 found that the fraction of significantly lopsided early type disks is identical to that for late - type disks . rr98 convincingly demonstrated that lopsidedness is not an effect of dust , but is in fact the asymmetric distribution of the light from old stars and hence from the _ stellar mass _ in the disk . some theoretical work has been done in examining long lived @xmath2 modes ( syer & tremaine 1996 ; zang & hohl 1978 ; sellwood & merritt 1994 ) , little convincing evidence however has been put forth to show that isolated galaxies will form stable @xmath2 modes without external perturbations or significant counter - rotating populations . without invoking the special cases above , long lived lopsidedness is possible if the disk resides in a lopsided potential . the question remains however : how is a lopsided potential created / maintained ? numerical simulations of hyperbolic encounters between disk galaxies fail to produce @xmath2 modes of amplitude @xmath18 without destroying the pre - existing stellar disk ( naab , t. ; private communication ) . minor mergers and possibly some weak interactions therefore remain as the most probable cause of lopsidedness ( rr98 ) . recent work has shown that perturbations in the outer halo of a galaxy may be amplified and even transmitted down into the disk ( weinberg 1994 ) . work by walker , mihos & hernquist ( 1996 ) and zr97 showed that the type and magnitude of lopsidedness seen in rz95 , zr97 and rr98 is comparable to the result of the accretion of a small satellite , if the mass ratio with the main galaxy is @xmath19 . in a preliminary study ( i.e. a rigid halo with no dynamical friction ) levine and sparke ( 1998 ) showed that lopsided galaxies may be formed by disks orbiting off center and retrograde in a flat - cored , dark matter dominated halo . they postulated that a galaxy may be pushed off center by a satellite accretion . using phase mixing arguments ( baldwin _ et al . _ 1980 ; rz95 ) and analysis of n - body simulations ( walker _ et al . _ 1996 ; zr97 ) the lifetime of lopsided features has been estimated at @xmath20 gyr . that lopsidedness is transient ( @xmath21 ) yet common , requires that it must be recurring and therefore lopsidedness may have significant evolutionary consequences . the current paper focuses on the impact that minor mergers ( observed as lopsidedness ) may have on boosting the sfr and the recent star formation history ( sfh ) of disk galaxies . for the purpose of this discussion , we will assume that lopsidedness is caused by minor mergers . regardless of what causes lopsidedness however , the perturbation in the gravitational potential manifestly exists and therefore may affect the gas in the galaxy to such a degree as to boost the sfr . indeed , zr97 find that lopsidedness is correlated ( at @xmath22 confidence ) with the `` excess '' of blue luminosity ( over what is predicted by the tully - fisher relation ) . modeling the integrated spectral evolution of starbursts using evolutionary population synthesis ( eps ) codes has been been well studied ( e.g. couch and sharples 1987 ; barger _ _ 1996 ; turner 1998 ) and despite its limitations , is a useful tool in determining the relative sfh over the past @xmath23 gyr . the same techniques used to probe the sfh in massive starbursts should also work to probe the recent sfh in the putative mini - bursts which we seek to study . by comparing measured indicators of recent sf ( e.g. @xmath4 , @xmath3 break strength , a star content ) , to the same indicators derived from the eps models , we will place limits on the mini - burst mass and duration . we have obtained spatially integrated spectra of a sample of 40 late type spiral galaxies ( sab - sbc ) of varying degrees of lopsidedness with the intent of using their relative stellar populations ( as determined from stellar template fitting and eps models ) to determine their recent sf histories . unlike the mass - normalized blue light excess , @xmath24 used in zr97 , our method operates independently of assumptions about a galaxy s mass , inclination or luminosity . in addition to probing the recent ( @xmath25 gyr ) sfh with studies of the stellar continuum we probe the current sfr by measuring the integrated balmer line emission strengths ( e.g. kennicutt _ et al . _ 1994 ) . the layout of the paper is as follows . in 2 we discuss the sample selection , observations , data reduction and determination of galaxy lopsidedness ; in 3 we examine our methods for determining the current sfr and recent sfh via the measurement of emission and stellar continuum properties as a function of lopsidedness . the discussion of the significance of these results , including the correlation of the boost parameters with other galaxy characteristics and the impact of our results on previous works ( i.e. rz95,zr97 & rr98 ) is contained in 4 . in 5 we present a summary and possible directions for future work . to build a sample of galaxies with varying degrees of lopsidedness , we imaged a large number of galaxies ( @xmath26 ) taken from the rc3 catalog ( de vaucouleurs et al . 1991 ) , selected according to the following criteria : apparent blue magnitude m@xmath2714 , redshift cz@xmath2810,000 km / s , axis ratio @xmath290.64 ( @xmath30 ) , de vaucouleurs type @xmath31 , and a maximum diameter of 4@xmath32 . the median diameter of the galaxies in our sample was 1.8@xmath32 . the magnitude and redshift limits were chosen to minimize the required exposure times . the axis ratio of the galaxies was constrained because it is hard to measure an azimuthal @xmath2 component in a highly inclined galaxy . once imaging was obtained and lopsidedness determined for each galaxy ( see _ 2.2.1 _ ) , we constructed a sample for spectroscopy consisting of 40 of our imaged galaxies ( see table 1 ) . these were selected to give the sample equal numbers of lopsided and symmetric targets . our sample is partially selected according to hubble type , and we must explore the effects which morphological evolution induced by lopsidedness may have on our conclusions . walker _ ( 1996 ) suggested that minor mergers increase bulge size , heat the galactic disk vertically and consume a large fraction of the galaxy s gas , eventually resulting in a low post - merger sfr . these two effects may drive galaxies towards earlier hubble type after they experience minor mergers . evolutionary processes such as these however become dominant either after the expected lifetime of lopsidedness ( @xmath33 gyr ) , or after repeated merger events ( walker _ et al . _ 1996 ) . during the interaction itself the irregularity in structure made manifest by lopsidedness as well as the creation of spiral arms via tidal perturbations will temporarily move a galaxy later in hubble type . this will effectively push lopsided early type disk galaxies into our sample while pushing those of later type out of it . due to their lower gas masses ( roberts & haynes 1994 ) , early type spirals have less potential for a large absolute increase in their sfr than late types . small boosts , however , may be easily noticeable against the typically older stellar population of early type disks . the exact interplay of these two effects may bias our measurement of the relation between sfr and lopsidedness . our imaging data were obtained during runs at steward observatory s 2.3-m bok reflector on kitt peak ( 1997 november 6 - 7 and 1998 february 1 - 2 ) and at its 61-inch ( 1.5-m ) reflector on mt . bigelow ( 1998 may 15 - 18 ) . the ccd pixel scales at the bok reflector and bigelow reflector were @xmath34 with fields of view @xmath35 and @xmath36 respectively . the median seeing at the 2.3-m for the november and february runs were 13 and 15 , respectively , while the median seeing at the 1.5-m in may was 13 . a nearly - mould @xmath37band filter ( @xmath38 ) was used at the 2.3-m and a kron - cousins r - band filter with @xmath38 was used at the 1.5-m . the effects of wavelength on observed lopsidedness are discussed in 2.1 of rr98 , and found not to be critical . to determine the lopsidedness of a galaxy we perform an azimuthal fourier decomposition of the r - band surface brightness , as in rr98 : @xmath39 } \},\ ] ] where for each radius , @xmath40 is the average intensity and @xmath41 describes the lopsidedness . we define the luminosity normalized quantities @xmath42 . instead of @xmath43 , we use @xmath44 ( the error corrected value which accounts for the positive definite nature of our measurements and the presence of errors ; see rr98 for details ) as our measure of asymmetry . we calculate the mean asymmetry of each galaxy , @xmath1 ( see table 2 ) , from 1.5 to 2.5 disk scale lengths using the weighted average described in rr98 . spectra were obtained with the bollers & chivens spectrograph at the 2.3-m bok reflector during the nights , 1998 march 22 - 25 , 1998 may 25 - 28 , and 1998 june 29 . we used a @xmath45 in @xmath46 order , blazed at @xmath47 , and a @xmath48 slit , resulting in a resolution of @xmath49 and a wavelength range of @xmath50 . this range includes the entire balmer series redward to @xmath51 , the @xmath3 break , ca h+k doublet , [ ] @xmath52 and [ ] @xmath53 . to reduce read noise , the ccd was binned in the spatial direction ; on 22 march we binned by 2 for a resultant pixel scale of @xmath54 while on all other nights we binned by 4 for a pixel scale of @xmath55 . a cuso@xmath56 filter was used to block @xmath57 order light . aside from using standard stars to calibrate the relative spectral response of the instrument , no absolute flux calibration was attempted . this was done partly because of the non - photometric conditions of some of our nights and partly due to the independence of our analysis methods on absolute flux levels . following kennicutt ( 1992 ) , we obtained spatially integrated spectra of the galaxies by repeatedly scanning the slit across the galaxy between @xmath58 . for each galaxy we obtained 2 exposures of 25 minutes each . to isolate the disk contributions to the integrated spectra , we also obtained a 5 minute exposure of the nucleus for each galaxy for later subtraction . the basic image reduction was carried out with standard iraf routines . the total dark current was @xmath59 and so it was ignored . the images were flat - fielded with a combination of twilight and smoothed night sky flats . high @xmath60 twilight flats ( @xmath61 ) were used to take out small - scale variations and the lower @xmath60 smoothed night sky flats ( @xmath62 ) were used to take out large - scale sensitivity fluctuations . the large - scale flat - field quality was determined by measuring variance in the median sky level at the four corners of the images . in all images where the galaxy was small in the field of view , the images were found to be flattened to better than @xmath63 . in some cases , the large size of the galaxies in comparison to the field of view at the 1.5-m telescope precluded such an estimate . point sources in the images were selected using daofind , and surrounding pixels were excised in the subsequent analysis to a radius where the stellar point spread function declined to the level of the sky . the basic spectral reduction was carried out with iraf routines . the measured dark current from multiple 25 minute dark exposures was @xmath64 and so was not accounted for . small - scale variations were removed with a combined series of 2 min . dome flats . twilight flats were fit in the spatial direction with a @xmath65 order cubic spline at several different positions in the spectral direction to construct a map of the slit response as a function of wavelength . non - linear pixels were interpolated over using a bad pixel mask generated from the ratio between a combined series of 2 minute and 10 second dome flats . removing cosmic rays from spectra without accidently removing emission lines is best automated by using the shape of the spectrum itself in the cleaning process : for a given column ( spatial direction ) , we medianed together the target column with the eight columns on either side to construct a slit profile , @xmath66 . using @xmath67 minimization , we fit the target column with the the following function : @xmath68 where the @xmath69 term accounts for the sky background . pixels deviating by more than 7 @xmath70 from the best fit are replaced with the value of the best fit model at that point . flagged segments larger than the spectral resolution of the instrument were not removed so as to avoid the accidental cleaning of emission lines . this algorithm is very efficient and , once a suitable set of parameters ( i.e. @xmath71 , threshold ) has been chosen , can remove almost all of the cosmic rays on the chip . after cosmic ray removal , the spectra were rectified using he - ar lamps taken at various times during the night . background contributions were then subtracted . to extract the spectra , we summed the number of center rows which corresponded to @xmath72 for the disk spectra and extracted the center row only for the nuclear spectra . using standard stars taken during the night , we then removed the spectral response of the system from the 1d , extracted spectra . finally , the wavelength scale of each spectra was shifted to the rest frame of the galaxy . the guided nuclear exposures were scaled by the effective exposure times on the nucleus during the drift scanning . these scaled exposures were then subtracted from the drift spectra . in this way , we separated the disk and nuclear contributions . finally , all spectra were scaled to their median flux level . the error calculation for @xmath14 is described in rr98 ; it accounts for both photon statistics and systematic flat - fielding uncertainties . using poisson statistics to describe the error in a raw spectrum we created an `` error spectrum '' and performed on it all of the standard reduction steps discussed in 2.3.2 except for the cosmic ray cleaning . after normalizing the `` science spectra , '' the `` error spectra '' were scaled by the median of the `` science spectra '' to form a detailed representation of the error at each pixel . the `` error spectra '' were used in all of the following analysis steps . by examining spatially integrated spectral characteristics as a function of @xmath14 , we can determine whether lopsidedness affects the sf histories of galaxies . it is difficult , even in major mergers , to invert optical spectrophotometry into a sfh estimate ( e.g. turner 1998 ) . however , the relative strength of the current and recent sfr in different galaxies may be studied with moderate @xmath60 , non flux - calibrated spectra ( e.g. couch & sharples 1987 ; barger _ the equivalent width of @xmath16 in emission ( @xmath15 ) integrated over the whole disk is a robust measure of the current global sfr in terms of the previously formed stars ( kennicutt et al . much of the balmer emission inn a galactic disk comes from the hii regions seen in sfr regions . in this regime , the balmer decrement relates the emission flux in @xmath51 to that in @xmath16 . if the continuum level at these two wavelengths is similar , then the decrement will also relate the @xmath73 of the two lines . therefore , the absorption corrected @xmath74 ( see 3.2 ) serves as an indirect indicator of the current sfr in the disk . the likely lifetime of lopsidedness has been estimated as @xmath20 gyr ( baldwin _ et al . _ 1980 ; rz95 ; zr97 ) . to examine sf on these timescales , we need a sf tracer with a comparable lifetime . main sequence a - stars have lifetimes of @xmath75 gyr ( clayton 1983 ) , have strong spectral signatures ( e.g. strong balmer lines , a blue continuum , and a weak @xmath3 break ) , and so serve as appropriate probes of the recent sfh . fitting population synthesis models ( e.g. bruzual & charlot 1993 ) to our spectra can give us a measure of the recent sfh . however for four reasons we choose to simply fit with empirical stellar templates ( jacoby _ et al . _ 1984 ) : 1 ) the spectral resolution of available population synthesis models ( @xmath76 ) is significantly less than that of our spectra ( @xmath77 ) . 2 ) the empirical templates of jacoby _ et al . _ ( 1984 ) have a spectral resolution ( @xmath78 ) slightly less than that of our spectra . 3 ) an adequate fit to the spectra can be obtained with a small number of high s / n stellar templates . to estimate the recent , relative sfh , we therefore need not deal with the complexities of the imf and metallicity of the stellar populations . as a qualitative measure of the relative contributions to our spectra from young and old stars , we synthesize the global absorption spectra of the galaxies in our sample with a linear combination of two stellar templates , an a0v and a g0iii spectra from the jacoby _ ( 1984 ) stellar library . we also individually fit and subtract a @xmath79 order polynomial of zero mean from the spectra and from each stellar template to correct for color terms ( e.g. from calibration errors , from approximating the spectra with only two stellar templates ) . the spectra ( with the polynomial subtracted ) are fit by @xmath67 minimization with the following model:@xmath80 where @xmath81 and the weights , c were determined iteratively . @xmath82 and @xmath83 are the weights for the normalized stellar template spectra , @xmath84 and @xmath85 . @xmath86 is the sum of the weighted template polynomial components , @xmath81 represents the doppler broadening of the stars which is convolved with the stellar templates , and @xmath87 is the convolution operator . for more details on the fitting procedure ( 1995 ) and turner ( 1998 ) . independent of continuum slope , there is a unique set of line shapes and strengths for stars of each spectral type . the polynomial fit minimizes the effects of a global continuum slope on our best fit solution so that we are instead performing a global fit to the spectral features ( @xmath3 break , ca h@xmath88k , balmer series , etc . ) we find that with only two , polynomial subtracted , stellar templates , we are able to consistently achieve fits with @xmath89 . our template fit to the galaxy spectra should reflect only the stellar populations in the galaxy , not interstellar emission . therefore , we first fit all portions of spectra , omitting expected emission regions , and use the residual of the best model fit to construct an emission spectrum . we remove the large scale variations in the residual by fitting it with a high order polynomial ( @xmath90 ) at all locations where no emission is expected . we then fit each emission feature with a gaussian . to isolate the stellar continuum , we subtract these gaussian components from the original spectra . we then re - fit our @xmath91 to this cleaned , pure absorption spectrum to determine @xmath92 . this process is illustrated in figure 1 . we measure @xmath74 ( see table 2 ) from the complementary absorption corrected emission line spectra for all of our galaxies . a linear continuum was fit on either side of @xmath51 ( @xmath93 ) . we then measured the equivalent width of the emission line integrating from @xmath94 . the boundaries of our integration were chosen by visual inspection to minimize noise contributions from the continuum while maximizing the amount of line flux . we can quantify the relative a star abundances by simply using the value of @xmath82 ( see table 2 ) in @xmath92 . because the basis templates are normalized to their median fluxes ( as are the data , ) the scaling factor of the individual templates gives a measure of how much a - stars contribute to the integrated spectra . balmer absorption lines are strongest in a - stars , the @xmath3 break is weak , and so we may also use these two features to measure the recent sfh ( couch & sharples 1987 ; barger _ we use the equivalent width of @xmath95 in absorption ( @xmath4 ) as our indicator of balmer line strength in order to minimize emission contamination and sample a relatively isolated region of the spectrum . a linear continuum was fit on either side of @xmath95 ( @xmath96 ) . we measure the @xmath97 itself from @xmath98 . instead of measuring @xmath4 directly from the spectrum , we decide to measure @xmath99 ( see table 2 ) of the best matching template spectrum , @xmath92 . these template spectra are in general a good match to the data and provide an essentially noiseless estimate of @xmath4 with a value derived from the best fit to a broad wavelength range . when measuring @xmath4 directly from the spectra , the relation between @xmath4 and @xmath100 ( see 4.3 ) clearly remains , but the scatter in @xmath4 at a given @xmath100 is slightly larger . to further parameterize the recent sfh , we use the strength of the @xmath3 break ( @xmath5 ) ( see table 2 ) . our measure of the break strength is defined as : @xmath101 the suppression of the continuum blueward of @xmath102 manifested in the break is caused by the combined absorption from the ca h@xmath88k and @xmath103 absorption line . as young , massive stars ( with blue continua and weak metal lines ) contribute more to the overall spectrum , @xmath5 decreases . to examine qualitatively the link between lopsidedness and sfr , we constructed composite disk spectra from the median of our nine most symmetric and nine most lopsided galaxies , and these we show in figure 2 . the lopsided composite spectrum has a much bluer continuum , stronger balmer lines , stronger emission and a shallower @xmath3 break than its symmetric counterpart regardless of exactly how many galaxies we include in the median spectrum . this result indicates that lopsidedness is in fact correlated with the recent sfh . however , a quantitative treatment of the relation is still needed . figure 3 shows the relation between @xmath74 and @xmath104 for the entire spectroscopic sample . using a spearman - rank test , we find that the distribution of @xmath74 vs. @xmath100 deviates from a random one at the @xmath105 level . the median @xmath74 of our nine most symmetric galaxies is @xmath106 while the median for our nine most lopsided galaxies is @xmath107 . the correlation of @xmath74 , and hence the current sfr , with @xmath14 is clearly shown in fig . this implies that an elevation in the instantaneous sfr occurs over the same timescales as lopsidedness , lasting maybe as long as @xmath23 gyr after the onset of lopsidedness . figure 4 shows the relation of a - star content , @xmath99 , and @xmath5 with @xmath108 . they deviate from random distributions at the @xmath109 , @xmath110 , and @xmath111 levels respectively . our sample is not large enough to study these correlations separately for different hubble types ; instead we analyze the change in spectral properties of the sample as a whole . figure 5 shows that the data are differentiated by lopsidedness in the @xmath99 vs. @xmath5 plane . the large scatter is not surprising , as even within a single hubble type there is a considerable variation in the specific sfhs ( kennicutt _ et al . _ 1994 ) . between the most lopsided and symmetric @xmath112 of our galaxies the median values of @xmath99 and @xmath5 differ by , @xmath113 and @xmath114 , respectively . this defines a boost vector ( fig . 5 ) whose direction and magnitude characterize the difference in spectral indices between a symmetric and lopsided galaxy in our sample . a number of efforts have shown that the detailed sfh , even over the last @xmath115 years , can not be determined unambiguously from integrated spectra ( e.g. turner 1998 and references therein ; leonardi & rose 1996 ) . instead , we restrict ourselves to a set of model sfhs consisting of a `` normal '' or `` underlying '' spiral galaxy sfh , which in lopsided galaxies has been boosted in the past by a factor @xmath116 . we then attempt to characterize the boost strength , @xmath116 , and an associated timescale . using the eps code of bruzual & charlot , ( gissel 1995 ; in preparation ) we construct seds representing the light from the underlying stellar populations of galaxies in our sample . we specify the sfh by the birthrate parameter @xmath117 ( scalo 1986 ) , where @xmath118 assuming an exponential sfh , i.e. @xmath119 $ ] , each @xmath117 has an associated timescale , @xmath120 defined through @xmath121 -1 = \frac{t_{gal}}{\tau(b ) } \frac{1}{b},\ ] ] with @xmath122 as the age of the galaxy . we adopt an age of @xmath123 gyr and a metallicity of @xmath124 . we also use the salpeter ( 1955 ) imf with a mass range of @xmath125 . this imf provides a better fit than does the scalo imf ( 1986 ) to the global photoionization rates and to the colors of galactic disks ( kennicutt et al . 1994 ) . applying the scalo imf results in slightly lower fractional boost masses than the salpeter imf , but the imf choice does not qualitatively affect our conclusions . the model sed is then given by the integral : @xmath126 where @xmath127,\ ] ] where @xmath128 is the hubble type dependent initial sfr of the galaxy used to normalize the present day masses to the same value , and where @xmath129 is the sed for a population @xmath130 years after a delta function burst ( bruzual & charlot s simple starburst model ) . we account for the change in the spectral properties by _ adding _ an exponentially declining increase , or `` boost '' in sf on top of the underlying galactic sfh . we assume that the boost originated about @xmath131 years ago , comparable to the presumed dynamical age of the boost and the a - stars seen in figure 2 . the model sed for a galaxy with a boost population is given by : @xmath132 dt ' + \int^{t_{boost}}_{0 } s(t',\lambda ) c_{boost } \exp \left [ -\frac{t'}{\tau_{boost } } \right ] dt'\ ] ] . for each hubble type , i.e. for each @xmath120 we construct model tracks by varying @xmath116 and @xmath133 in equation ( 9 ) , effectively adding differing boosts to the unperturbed galaxy sfh . as shown in fig . 6 , moving the underlying population later in hubble type , or to a larger @xmath120 , shifts the tracks to greater values of @xmath99 and smaller values of @xmath5 . for a given hubble type , varying the boost timescale changes the slope of the track of increasing boost mass . it is interesting to note that our empirical boost vector can not be matched directly with any of the models in the @xmath99 vs. @xmath5 plane , suggesting that the eps models may have systematic errors . assuming , however , that the correct models will differ from these by a simple shift in the @xmath99 vs. @xmath5 plane , we choose the underlying population model which minimizes the distance ( in the @xmath99 vs. @xmath5 plane ) to the beginning of our observed boost vector . we then match the slope of the boost vector by varying the boost timescale . the best match we find was for @xmath134 ( hubble type sb ; kennicutt et al . 1994 ) and @xmath135 myr . for reference , we also show the tracks for underlying galaxy spectra with @xmath117 parameters corresponding to hubble types of sab and sbc ( @xmath136 and @xmath137 respectively ; kennicutt et al . for each of these hubble types , we plot tracks corresponding to boosts with @xmath138 myr and @xmath135 myr . we also plot the @xmath138 myr boost for the @xmath134 model . measuring the projected length of the boost vector on the best model , and assuming the galaxy disk has a final stellar mass of @xmath139 , we find that @xmath10 of stars have been formed over the the past @xmath9 gyr in addition to the `` underlying '' sfh . this corresponds to an factor of @xmath140 increase in the sfr over the past @xmath141 years . extinction will have two primary effects on our data . it will redden the continuum from the galaxy . by allowing for a polynomial element to our template fits , we minimize the effect of the continuum slope on our results . extinction may also affect regions of current sf ; obscuring both emission lines and the continuum from massive young stars . we do not make a further account of reddening but acknowledge that our boost masses may increase if reddening is included . it is important to verify that our result is not an artifact of correlations between the underlying sfh and lopsidedness . to determine whether mass , gas content or hubble type effects mimic our relation between lopsidedness and sfr , we examined the distributions of @xmath74 , @xmath99 , @xmath5 , a star content and @xmath14 vs. @xmath142 , @xmath143 and t - type ( de vaucouleurs et al . a hidden correlation exists if any of the latter three parameters are correlated both with a sf indicator and with @xmath14 . none of the above distributions , except for @xmath74 and @xmath5 vs. t - type and @xmath143 , have more than a @xmath144 probability of being correlated , with a @xmath145 significance for the four remaining correlations . given the known correspondence between sfr and hubble type , mean stellar population and hubble type , and sfr and gas surface density ( kennicutt et al . 1994 ) , these correlations with @xmath74 and @xmath5 are expected . however , none of our canonical galaxy parameters show a correlation with lopsidedness at above the @xmath146 level and we can therefore be confident that we are seeing a true correlation between our sf indicators and @xmath147 . as an initial diagnostic of the nuclear contributions , we examine the fraction of galactic light originating in the nucleus . even without absolute spectrophotometry , we can still examine what percentage of the total galaxy light is made up of light from the nucleus as long as the spectra are calibrated with respect to each other . the fractional luminosity of the nucleus ranged from @xmath148 with a median value of @xmath149 . the fractional luminosity of the nucleus is uncorrelated with @xmath150 . a spearman - rank test showed that there was a @xmath151 probability that the data was drawn from a random sample . figure 7 shows that while the nuclear @xmath74 vs. @xmath150 , and @xmath99 vs. @xmath150 are correlated , the relations show more scatter than in the disk . the spearman - rank test shows that these two relations are less significantly correlated in the nucleus than in the disk by @xmath152 and @xmath153 , respectively . only for @xmath5 vs. @xmath150 was the correlation greater than in the disk . our data indicate that the enhanced sf in lopsided galaxies occurs in the disk with equal or greater strength than in the nucleus . this result appears to run contrary to what numerical simulations suggest about the reactions of gas in the disk to tidal perturbations from external sources . mihos & hernquist ( 1994 ) and hernquist & mihos ( 1995 ) show that even high mass ratio mergers ( @xmath154 ) can form a bar which torques gas in the disk and funnels much of it into the nucleus . however the presence of a dense bulge easily inhibits bar formation and hence gas is not funneled as effectively . another explanation is that star formation is occurring in large amounts at the nucleus , but is partly masked by extinction . high extinction is often found to be coincident with high star formation regions in mergers and is easily capable of hiding the presence of young populations ( turner 1998 ) . to estimate the volume density of lopsided galaxies from a magnitude limited sample requires the understanding of any systematic differences in luminosity between lopsided and symmetric galaxies ( zr97 ) . our best fit boost brightens the galaxies by 1.3 , 1.0 and 0.8 magnitudes ( in b , v , and r , respectively ) over the brightness of the underlying population . lopsided galaxies will hence be overrepresented in our b - band selected sample , and their volume density will be @xmath155 of that derived from any magnitude limited survey . to compare the frequency of lopsidedness for different hubble types ( e.g. rr98 ) , we need to understand how this brightening depends on the `` underlying '' sfh . unfortunately , we have too few galaxies in our sample to discuss hubble type dependent effects on the boost strength and luminosity evolution . at present , the main constraint on the frequency of lopsidedness is from its frequency in imaging samples ( @xmath156 ; rr98 ) . coupled with our brightening estimates , we find that @xmath157 of the galaxies in a volume limited sample will be lopsided . lopsidedness is likely caused by minor mergers , the frequency of which increases with look back time . given that lopsidedness lasts for @xmath158 gyr , we can estimate that the average galaxy has been lopsided at least once in its lifetime . to quantify the correlation between the recent sf histories of present - day spiral galaxies and their global asymmetry , we compare the integrated spectral properties of late - type spirals of varying lopsidedness . we find that the recent ( @xmath159 gyr ) sfh and current ( @xmath160 years ) sfr are both strongly correlated with @xmath150 although there is appreciable scatter in the individual galaxy - to - galaxy properties . for @xmath74 , reflecting the current sfr , we find a @xmath7 spearman - rank correlation with @xmath150 . we fit a combination of a0v and g0iii stellar spectra to our galaxy spectra to quantify the relative abundance of a - stars in the disk ( which traces the sfr within 0.5 gyrs ) . from these best fit model spectra , @xmath92 , we measure a number of spectral indices , and find that @xmath99 , @xmath5 , and @xmath82 are correlated with @xmath150 at the @xmath161 , @xmath162 , and @xmath163 levels , respectively . we measure the same spectral indices in the nucleus , and find them less correlated with @xmath150 ( except @xmath5 ) . unless a nuclear starburst is obscured , the disk and not the nucleus is the primary site of the sf increase we see in lopsided galaxies . this is in contrast to numerical simulations where minor mergers funnel gas into the nucleus of galaxies , causing intense starbursts ( mihos & hernquist 1994 ; hernquist & mihos 1995 ) . only by the presence of a dense bulge can the formation of a bar , and the subsequent funneling of gas , be prevented . to quantify the mass of additional stars formed in lopsided galaxies , we defined a boost vector in @xmath99 vs. @xmath5 space , by comparing the median values of these properties for the most symmetric third and the most lopsided third of our sample . we find @xmath164 and @xmath165 . we fit this vector with an `` underlying population @xmath88 boost '' eps model corresponding to a progenitor galaxy with @xmath134 , @xmath135 myr , and boost age of @xmath9 gyr . using this best fit eps model , we find that @xmath10 is formed in the boost in addition to the `` underlying '' sfh ( assuming a final stellar mass of @xmath166 ) . this is a considerable fraction ( @xmath167 ) of the final stellar mass of the galaxy and corresponds to a factor of 8 increase in the sfr over the past @xmath13 years . given the increasing merger rates and increasing gas fractions towards higher redshifts , minor merger induced sf boosts of short duration played an important role in assembling the present day stellar content of galaxies . finally , we address by how much the frequency of lopsidedness from a magnitude limited sample is increased by the corresponding luminosity boost . our best fit eps boost model corresponds to a @xmath158 magnitude brightening when galaxies becomes lopsided , increasing their presence four - fold in magnitude limited samples . we lack the statistics however , to examine any hubble type dependent differences in the luminosity boost . it is obvious that more work needs to be done to fully understand the cause of lopsidedness as well as the sfh of lopsided galaxies . to quantify the hubble type specific boost in the recent sfh , a large sample should be obtained with significant numbers of galaxies in each hubble type bin . since imaging and spectroscopy will be needed for this project , a volume limited sample may be constructed which bypasses many of the problems encountered when selecting galaxies according to an apparent magnitude limit . companion searches to sufficiently faint magnitudes will help to study the possible link between environment and lopsidedness ( as caused by weak tidal interactions ) . with the recent commissioning of large area imaging and spectroscopy surveys such as sloan digital sky survey , constructing such a sample will become relatively straightforward . numerical simulations have shown to be a useful tool in studying the evolution of the stellar and gas distributions in minor mergers . high resolution simulations with a live halo are crucial for studying the detailed response of the disk to the merger ( walker _ et al . a thorough exploration of interaction parameter space is needed to quantify the structural and kinematic response in the stellar and gas components . high resolution n - body studies are also needed to explore the global stability of isolated galactic disks . this work was completed with partial support from nsf grants ast9870151 , ast9421145 and ast9900789 . greg rudnick and h .- w . rix would like to thank nelson caldwell for many valuable discussions on measuring sfhs from integrated spectra . greg rudnick would like to thank craig kulesa and christopher gottbrath for many useful discussions in the early hours of the morning . greg rudnick would also like to thank megan sosey and chris gottbrath for assisting with our observing program . finally , we would like to thank the steward observatory 2.3-m telescope operators for their assistance in the completion of this project . allam , s. s. , tucker , d. l. , lin , h. , & hashimoto , y. 1999 , , 522 , l89 baldwin , j. e. , lynden - bell , d. , & sancisi , r. 1980 , , 193 , 313 barger , a. j. , aragon - salamanca , a. , ellis , r. s. , couch , w. j. , smail , i. , & sharples , r. m. 1996 , , 279 , 1 barnes , j. , & hernquist , l. 1991 , , 370 , l65 barnes , j. , & hernquist , l. 1992 , , 30 , 705 barnes , j. , & hernquist , l. 1996 , , 471 , 115 bruzual , a. g. , & charlot , s. 1993 , , 405 , 538 barton , e. j. , bromley b. c. , & geller , m. j. 1999 , , 511 , l25 clayton , d. 1983 , _ principles of stellar evolution and nucleosynthesis . _ chicago : univ . of chicago press couch w. j. , & sharples , r. m. 1987 , , 229 , 423 de vaucouleurs , g. , de vaucouleurs , a. , corwin , h. g. , jr . , buta , r. j. , paturel , g. , & fouque , p. 1991 , _ third reference catalog of bright galaxies . _ new york : springer - verlag hashimoto , y. , oemler , a. , lin , h. , & tucker , d. l. 1998 , , 499 , 589 haynes , m. p. , van zee , l. , hogg , d. e. , roberts , m. s. , & maddalena , r. j. 1998 , , 115 , 62 hernquist , l. , heyl , j. , & spergel , d. 1993 , , 416 , l9 hernquist , l. , & mihos , c. j. 1995 , , 448 , 41 jacoby , g. h. , hunter , d. a. , & christian , c. a. 1984 , , 56 , 257 johnston , k. v. , hernquist , l. , & bolte , m. 1996 , , 465 , 278 kennicutt , r. c. , jr . , roettiger , k. a. , keel , w. c. , van der hulst , j. m. , & hummel , e. 1987 , , 93 , 1011 kennicutt , r. c. , jr . 1992 , , 388 , 310 kennicutt , r. c. , jr . , tamblyn , p. , & congdon c. w. 1994 , , 435 , 22 kennicutt , r. c. , jr . 1998 , , 498 , 541 lacey , c. , & cole , s. 1993 , , 262 , 627 larson , r. b. , & tinsley , b. m. 1978 , , 219 , 46 leonardi , a. j. , rose , j. a. 1996 , , 111 , 182l levine , s. e. , & sparke , l. s. 1998 , , 496 , 13 mihos , c. j. , richstone , d. o. , & bothun , g. d. 1992 , , 400 , 153 mihos , c. j. , & hernquist , l. 1994 , , 425 , l13 quinn , p. j. , hernquist , l. , & fullagar , d. p. 1993 , , 403 , 74 richter , o .- g . , & sancisi , r. 1994 a & a , 290 , l9 rix , h .- w . , & zaritsky , d. 1995 , , 447 , 82 rix , h .- w . , kennicutt , r. c. , jr . , braun , r. , & walteros , r. a. m. 1995 , , 438 , 155 roberts , m. s. , & haynes , m. p. 1994 , , 32 , 115 rudnick , g. , & rix , h .- w . 1998 , , 116 , 1163 salpeter , e. e. 1955 , , 121 , 161 sanders , d. b. , soifer , b. t. , elias , j. h. , madore , b. f. , matthews , k. , neugebauer , g. , & scoville , n. z. 1988 , , 325 , 74 scalo , j. m. 1986 , _ fund . _ , 11 , 1 searle , l. , & zinn , r. 1978 , , 225 , 357 sellwood , j. a. , & merritt , d. 1994 , , 425 , 530 swaters , r. a. , schoenmakers , r. h. m. , sancisi , r. , & van albada , t. s. 1999 , , 304 , 330 syer , d. , & tremaine , s. 1996 , , 281 , 925 toomre , a. , & toomre , j. 1972 , , 178 , 623 toth , g. , & ostriker , j. p. 1992 , , 389 , 5 turner , a. 1998 , phd . thesis walker , i. r. , mihos , c. j. , & hernquist , l. 1996 , , 460 , 121 weinberg , m. d. 1994 , apj , 421 , 481 zang , t. a. , & hohl , f. 1978 , , 226 , 521 zaritsky , d. , & rix , h .- w . 1997 , , 477 , 118 llllll a 1219 + 41 & 3.0 & 1.19 & 6927 & 0.85 & [email protected] ic 0520 & 2.0 & 1.95 & 3528 & 0.79 & [email protected] ic 0900 & 4.0 & 1.62 & 7067 & 0.65 & [email protected] ic 1269 & 4.0 & 1.70 & 6116 & 0.74 & [email protected] ngc 2347 & 3.0 & 1.78 & 4422 & 0.71 & [email protected] ngc 2582 & 2.0 & 1.23 & 4439 & 1.00 & [email protected] ngc 2744 & 2.0 & 1.66 & 3428 & 0.65 & [email protected] ngc 2916 & 3.0 & 2.45 & 3730 & 0.68 & [email protected] ngc 3066 & 4.0 & 1.10 & 2049 & 0.89 & [email protected] ngc 3162 & 4.0 & 3.02 & 1298 & 0.83 & [email protected] ngc 3177 & 3.0 & 1.44 & 1302 & 0.81 & [email protected] ngc 3310 & 4.0 & 3.09 & 0980 & 0.78 & [email protected] ngc 3353 & 3.0 & 1.35 & 0944 & 0.71 & [email protected] ngc 3681 & 4.0 & 2.51 & 1239 & 0.79 & [email protected] ngc 3684 & 4.0 & 3.09 & 1163 & 0.69 & [email protected] ngc 3897 & 4.0 & 1.95 & 6411 & 1.00 & [email protected] ngc 3928 & 3.0 & 1.51 & 0982 & 1.00 & [email protected] ngc 3963 & 4.0 & 2.75 & 3186 & 0.91 & [email protected] ngc 4017 & 4.0 & 1.78 & 3454 & 0.78 & [email protected] ngc 4041 & 4.0 & 2.69 & 1234 & 0.93 & [email protected] ngc 4351 & 2.0 & 2.00 & 2310 & 0.68 & [email protected] ngc 4412 & 3.0 & 1.41 & 2294 & 0.89 & [email protected] ngc 4430 & 3.0 & 2.29 & 1443 & 0.89 & [email protected] ngc 4595 & 3.0 & 1.74 & 0633 & 0.65 & [email protected] ngc 4639 & 4.0 & 2.75 & 1010 & 0.68 & [email protected] ngc 4814 & 3.0 & 3.09 & 2513 & 0.74 & [email protected] ngc 4911 & 4.0 & 1.45 & 7970 & 0.91 & [email protected] ngc 5218 & 3.0 & 1.82 & 2807 & 0.69 & [email protected] ngc 5614 & 2.0 & 2.45 & 3892 & 0.83 & [email protected] ngc 5653 & 3.0 & 1.74 & 3567 & 0.74 & [email protected] ngc 5713 & 4.0 & 2.75 & 1883 & 0.89 & [email protected] ngc 5915 & 4.0 & 1.82 & 5570 & 1.00 & [email protected] ngc 5923 & 2.0 & 1.74 & 2291 & 0.72 & [email protected] ngc 5936 & 3.0 & 1.45 & 4004 & 0.89 & [email protected] ngc 6012 & 2.0 & 2.09 & 1854 & 0.72 & [email protected] ngc 6195 & 3.0 & 1.55 & 9000 & 0.69 & [email protected] ngc 6574 & 4.0 & 1.41 & 2282 & 0.78 & [email protected] ngc 6711 & 4.0 & 1.35 & 4671 & 0.93 & [email protected] ngc 6824 & 3.0 & 1.70 & 3337 & 0.69 & [email protected] ngc 7177 & 3.0 & 3.10 & 1150 & 0.64 & [email protected] llllllll a 1219 + 41 & 19.7 & 3.15 & [email protected] & [email protected] & [email protected] & 0.34 & [email protected] ic 0520 & 19.8 & 3.82 & [email protected] & [email protected] & [email protected] & 0.16 & [email protected] ic 0900 & 11.6 & 6.54 & [email protected] & [email protected] & [email protected] & 0.38 & [email protected] ic 1269 & 13.0 & 2.93 & [email protected] & [email protected] & [email protected] & 0.32 & [email protected] ngc 2347 & 10.6 & 5.34 & [email protected] & [email protected] & [email protected] & 0.32 & [email protected] ngc 2582 & 20.0 & 2.82 & [email protected] & [email protected] & [email protected] & 0.24 & [email protected] ngc 2744 & 10.8 & 4.28 & [email protected] & [email protected] & [email protected] & 0.60 & [email protected] ngc 2916 & 23.0 & 3.29 & [email protected] & [email protected] & [email protected] & 0.36 & [email protected] ngc 3066 & 13.7 & 3.40 & [email protected] & [email protected] & [email protected] & 0.37 & [email protected] ngc 3162 & 26.0 & 2.91 & [email protected] & [email protected] & [email protected] & 0.52 & [email protected] ngc 3177 & 8.6 & 6.58 & [email protected] & [email protected] & [email protected] & 0.37 & [email protected] ngc 3310 & 9.92 & 7.63 & [email protected] & [email protected] & [email protected] & 0.45 & [email protected] ngc 3353 & 12.1 & 4.23 & [email protected] & [email protected] & [email protected] & 0.38 & [email protected] ngc 3681 & 24.9 & 3.03 & [email protected] & [email protected] & [email protected] & 0.12 & [email protected] ngc 3684 & 16.8 & 4.51 & [email protected] & [email protected] & [email protected] & 0.34 & [email protected] ngc 3897 & 12.6 & 4.49 & [email protected] & [email protected] & [email protected] & 0.25 & [email protected] ngc 3928 & 18.7 & 4.03 & [email protected] & [email protected] & [email protected] & 0.35 & [email protected] ngc 3963 & 25.4 & 2.98 & [email protected] & [email protected] & [email protected] & 0.34 & [email protected] ngc 4017 & 16.0 & 3.89 & [email protected] & [email protected] & [email protected] & 0.48 & [email protected] ngc 4041 & 10.8 & 6.97 & [email protected] & [email protected] & [email protected] & 0.35 & [email protected] ngc 4351 & 18.1 & 4.18 & [email protected] & [email protected] & [email protected] & 0.42 & [email protected] ngc 4412 & 23.7 & 2.90 & [email protected] & [email protected] & [email protected] & 0.40 & [email protected] ngc 4430 & 28.3 & 2.67 & [email protected] & [email protected] & [email protected] & 0.35 & [email protected] ngc 4595 & 16.3 & 3.82 & [email protected] & [email protected] & [email protected] & 0.46 & [email protected] ngc 4639 & 23.4 & 3.24 & [email protected] & [email protected] & [email protected] & 0.15 & [email protected] ngc 4814 & 13.5 & 5.58 & [email protected] & [email protected] & [email protected] & 0.17 & [email protected] ngc 4911 & 14.5 & 3.54 & [email protected] & [email protected] & [email protected] & 0.27 & [email protected] ngc 5218 & 8.7 & 4.86 & [email protected] & [email protected] & [email protected] & 0.29 & [email protected] ngc 5614 & 11.9 & 4.76 & [email protected] & [email protected] & [email protected] & 0.10 & [email protected] ngc 5653 & 6.8 & 5.63 & [email protected] & [email protected] & [email protected] & 0.36 & [email protected] ngc 5713 & 14.6 & 4.27 & [email protected] & [email protected] & [email protected] & 0.48 & [email protected] ngc 5915 & 5.0 & 3.48 & [email protected] & [email protected] & [email protected] & 0.51 & [email protected] ngc 5923 & 15.2 & 2.52 & [email protected] & [email protected] & [email protected] & 0.23 & [email protected] ngc 5936 & 11.7 & 3.61 & [email protected] & [email protected] & [email protected] & 0.44 & [email protected] ngc 6012 & 12.2 & 3.46 & [email protected] & [email protected] & [email protected] & 0.15 & [email protected] ngc 6195 & 11.2 & 4.17 & [email protected] & [email protected] & [email protected] & 0.25 & [email protected] ngc 6574 & 8.6 & 4.46 & [email protected] & [email protected] & [email protected] & 0.31 & [email protected] ngc 6711 & 9.2 & 3.13 & [email protected] & [email protected] & [email protected] & 0.38 & [email protected] ngc 6824 & 14.5 & 4.64 & [email protected] & [email protected] & [email protected] & 0.24 & [email protected] ngc 7177 & 10.9 & 5.66 & [email protected] & [email protected] & [email protected] & 0.12 & [email protected] [ tbl : table2.7.23.99.in ]

at finite temperature , the hydrodynamic collective modes of superfluid trapped fermi gases are coupled to the motion of the normal component , which in the bcs limit behaves like a collisionless normal fermi gas . the coupling between the superfluid and the normal components is treated in the framework of a semiclassical transport theory for the quasiparticle - distribution function , combined with a hydrodynamic equation for the collective motion of the superfluid component . we develop a numerical test - particle method for solving these equations in the linear response regime . as a first application we study the temperature dependence of the collective quadrupole mode of a fermi gas in a spherical trap . the coupling between the superfluid collective motion and the quasiparticles leads to a rather strong damping of the hydrodynamic mode already at very low temperatures . at higher temperatures the spectrum has a two - peak structure , the second peak corresponding to the quadrupole mode in the normal phase

cond-mat0612361

most of the current experiments involving trapped atomic fermi gases focus on the bec - bcs crossover . by changing the magnetic field around a feshbach resonance , the scattering length @xmath0 of the atoms can be varied from small positive values through very large values near the resonance to small negative values . for @xmath1 , @xmath2 ( where @xmath3 denotes the fermi momentum ) the system can be considered as a bose - einstein condensate ( bec ) of diatomic molecules . the crossover region , @xmath4 , is not yet very well understood from a theoretical point of view . finally , on the other side of the resonance , when @xmath5 , @xmath6 , the system should be in the bcs phase if the temperature is sufficiently low . however , the bcs critical temperature @xmath7 is extremely low , and very soon the magnetic field reaches the point where @xmath7 becomes smaller than the actual temperature @xmath8 , and the system undergoes the phase transition to the normal ( non - superfluid ) phase . one possibility to study the crossover experimentally is to measure the properties of certain collective oscillations . for example , the radial and axial breathing modes of a cigar - shaped trapped fermi gas have been measured over the whole crossover region @xcite . in these experiments one can observe how the frequencies and damping rates of the modes change from what one expects for a bec to what one expects for a collisionless normal fermi gas . assuming that , except in the collisionless normal phase , hydrodynamics is valid , the measured frequencies can give some information on the equation of state in the crossover region . however , this schematic picture is not completely accurate . since the system is in a trap , there is no sharp transition from the superfluid to the normal phase . this can be seen as follows : the bcs critical temperature @xmath7 depends on the atom density @xmath9 , and the density depends on the position @xmath10 . in the center of the trap , the density @xmath11 and hence the local critical temperature @xmath12 are higher than in the outer part of the trap . as a consequence , for a given temperature , the outer part gets already normal at a magnetic field where the inner part is still superfluid . to be more precise , a system in the bcs phase at finite temperature behaves effectively like a mixture of superfluid and normal components with densities @xmath13 and @xmath14 , respectively , which become @xmath15 , @xmath16 in the limit @xmath17 and @xmath18 , @xmath19 in the limit @xmath20 . as a consequence , if @xmath21 , the superfluid inner part of the trap behaves like a mixture of normal and superfluid components , while only the outer part with @xmath22 is completely normal @xcite . if the collision rate was high enough , also the normal component of the gas would behave hydrodynamically . such a system could be described by landau s two - fluid hydrodynamics which has been applied to collective modes in trapped superfluid gases at finite temperature @xcite . however , although in the recent experiments the transition to the normal phase seemed to occur at a value of @xmath23 @xcite ( i.e. , the bcs phase has not really been reached ) , the system behaved already like a collisionless normal fermi gas . hence it seems to be clear that the normal component can not be treated in terms of hydrodynamics , but a description in terms of a vlasov equation is required . we note that there are other approaches to the description of the collective modes at finite temperature . in particular , let us mention the quasiparticle random phase approximation ( qrpa ) @xcite , which can be seen as the linearized form of the time - dependent bogoliubov - de gennes ( bdg ) equations . however , for practical reasons this method is limited to systems with spherical symmetry and numbers of particles up to a few times @xmath24 . another disadvantage of this method is that it does not allow to include a collision term . for the case of clean superconductors , a semiclassical transport theory taking the coupling between normal and superconducting components into account has been developed by betbeder - matibet and nozires @xcite . transport theories of this type have also been used for describing the dynamics of superfluid @xmath25he @xcite . in a preceding paper @xcite , we derived the semiclassical transport equations for the case of trapped atomic fermi gases and applied them to the quadrupole mode of a gas in a spherical trap . we found that the presence of the normal component leads to a strong damping of the hydrodynamic collective mode . the same mechanism might explain the strong damping observed experimentally near the transition to the collisionless behavior @xcite . however , in ref . @xcite we had to replace the gap @xmath26 by a constant in order to find an analytical solution of the transport equations . due to this simplification , which can not really be justified , the damping of the hydrodynamic mode at a given temperature was much weaker than that obtained in qrpa calculations @xcite . in the present paper we will work out a numerical method which allows us to treat the realistic @xmath10-dependence of the gap . in addition , the method is very versatile and allows to treat much more general cases than can be solved analytically in the constant - gap approximation . the basic idea is to replace the continuous phase - space distribution function of the quasiparticles by a sum of a finite number of delta functions in phase space , called `` test particles . '' in the normal phase , the test - particle method is routinely used for solving the vlasov equation , e.g. for simulating heavy - ion collisions in nuclear physics @xcite . it has also been applied to the simulation of the dynamics of normal trapped atomic fermi gases with collision term @xcite and of bose - fermi mixtures @xcite . however , to our knowledge , the test - particle method has not yet been used in the context of superfluid systems , and in fact the numerical difficulties are quite different from those encountered in the usual applications . the article is organized as follows . in sec . [ transportbcs ] , we give a brief summary of the transport equations for the bcs phase and their linearization in the case of small deviations from equilibrium . we also give arguments why some terms which appear in the equations can be neglected . [ testparticlemethod ] we introduce the test - particle method for the case of small oscillations around equilibrium . we describe in detail a number of tricky points we encountered during the implementation of the method , in particular the calculation of the test - particle trajectories , the generation of the test - particle distribution in phase space , and the initialization after a delta - like perturbation . in sec . [ firstresults ] we present the first results obtained with the help of this method , again for the quadrupole mode in a spherical system . finally , in sec . [ conclusions ] , we summarize and draw our conclusions . in this subsection we will give a brief summary of the kinetic equation approach developed by betbeder - matibet and nozires @xcite for the case of clean superconductors and adapted to the case of trapped atomic fermi gases in ref . @xcite . in this paper we will only give the final equations . for details of the derivations , see ref . @xcite and the footnote . should be replaced by @xmath27 . in eq . ( 34b ) , @xmath28 should be replaced by @xmath29 . in eq . ( 61 ) , @xmath30 should be replaced by @xmath31 . in addition , we point out that @xmath32 if @xmath33 , which is not evident from eq . ( 61 ) but can be derived from eq . ( 54 ) . ] we consider a dilute gas of fermionic atoms of mass @xmath34 in two equally populated hyperfine states @xmath35 and @xmath36 , trapped by an external potential @xmath37 and interacting via an attractive short - range interaction which leads to a scattering length @xmath38 . the corresponding classical mean - field hamiltonian ( minus the chemical potential @xmath39 ) reads @xmath40 where @xmath41 denotes the sum of the external and the hartree potential , @xmath42 in the latter equation , @xmath43 denotes the coupling constant and @xmath9 is the density per spin state . the vlasov equation ( without collision term ) for the phase - space distribution function @xmath44 in the normal phase can be written in the compact form @xmath45 where @xmath46 denotes the poisson bracket . one way to derive this equation is to perform a wigner - kirkwood expansion up to order @xmath47 of the time - dependent hartree - fock equation @xcite . in the superfluid phase the derivation of an analogous transport equation is much more complicated due to the presence of the complex order parameter ( gap ) @xmath26 whose phase describes the collective motion of the cooper pairs . in addition to the density matrix @xmath48 , there exists now an anomalous density matrix ( pairing tensor ) @xmath49 . the gap @xmath50 and the anomalous density are related by the gap equation @xmath51 the time - dependence of @xmath48 and @xmath49 is governed by the time - dependent hartree - fock - bogoliubov or bdg equations . as in the normal phase , the semiclassical transport theory can be derived from these equations by performing a wigner - kirkwood expansion up to order @xmath47 . however , it turns out that it is necessary to introduce a gauge transformation with a phase @xmath52 that makes the order parameter @xmath50 real . this corresponds to a transformation into the local rest frame of the cooper pairs , which are moving with the collective velocity @xmath53 . the effect of this transformation is to change the gap @xmath50 , the single - particle hamiltonian @xmath54 , the normal and anomalous density matrices @xmath48 and @xmath49 according to @xmath55-\hbar{\dot{\phi}}({\bm{\mathrm{r}}})\,,\label{htil}\\ { \tilde{\varrho}}({\bm{\mathrm{r}}},{\bm{\mathrm{p } } } ) & = \varrho[{\bm{\mathrm{r}}},{\bm{\mathrm{p}}}-\hbar{\bm{\nabla}}\phi({\bm{\mathrm{r}}})]\,,\\ { \tilde{\kappa}}({\bm{\mathrm{r}}},{\bm{\mathrm{p } } } ) & = \kappa({\bm{\mathrm{r}}},{\bm{\mathrm{p}}})e^{2i\phi({\bm{\mathrm{r}}})}\,.\end{aligned}\ ] ] roughly speaking , the phase @xmath56 determines the dynamics of the superfluid component of the system , while the dynamics of the normal component , consisting of thermally excited quasiparticles , has to be described separately . the distribution of these quasiparticles , denoted by @xmath57 , obeys the following equation of motion : @xmath58 this equation looks formally very similar to the vlasov equation ( [ vlasov ] ) , except that the hamiltonian @xmath54 is replaced by the quasiparticle energie @xmath59 , which is defined as @xmath60 throughout this article , the indices `` @xmath61 '' and `` @xmath62 '' denote the time - even and time - odd parts of a phase - space function , i.e. , the parts which are even and odd in @xmath63 , respectively . the quasiparticle - distribution function @xmath64 is related to the normal and anomalous density matrices in the new gauge , @xmath65 and @xmath66 , by @xmath67 the vlasov - like equation ( [ qpvlasov ] ) has to be complemented with an equation of motion for the phase @xmath56 . it turns out that @xmath56 has to be determined from the continuity equation @xmath68 where the density @xmath9 and the current @xmath69 are given by @xmath70 let us now assume that the external potential @xmath37 can be written as @xmath71 where @xmath72 is time - independent and @xmath73 is a small perturbation . the equilibrium quantities ( corresponding to the potential @xmath72 ) will be marked by an index `` 0 '' . in particular , since in equilibrium the gap can be chosen to be real , we have @xmath74 the quasiparticle distribution function in equilibrium is given by @xmath75\,,\quad\ ] ] where @xmath76 denotes the fermi function , @xmath77 our aim is to calculate the small deviations from equilibrium induced by the perturbation @xmath73 , which will be marked by an index `` 1 '' . to that end we linearize the transport equation ( [ qpvlasov ] ) for the quasiparticle - distribution function , @xmath78 where @xmath79 . the deviation of the quasiparticle energy , @xmath80 , which appears on the r.h.s . , depends itself on @xmath81 through the deviation of the hartree field , @xmath82 , and the deviation of the gap , @xmath83 . expressing @xmath84 and @xmath83 in terms of @xmath81 , we can write eq . ( [ qpvlasovlin1 ] ) as @xmath85 where @xmath86 and @xmath87 are the quasiparticle contributions to @xmath84 and @xmath83 , @xmath88 while @xmath89 is a function which depends only on equilibrium quantities . the explicit expression for the function @xmath89 reads @xmath90\,,\ ] ] where the local fermi momentum @xmath91 is defined as usual by @xmath92 , and the temperature dependence of @xmath89 is governed by the function @xmath93 with @xmath94 . in the two limiting cases @xmath17 and @xmath95 , the function @xmath96 takes the values 0 and 1 , respectively . as a consequence , @xmath97 if @xmath98 . in order to determine the phase @xmath99 , we also linearize the continuity equation ( [ conti ] ) : @xmath100 again , we express all quantities in terms of equilibrium quantities and the unknown quantities @xmath81 and @xmath99 . according to eq . ( [ defcurrent ] ) , the current @xmath101 can be decomposed into quasiparticle and superfluid contributions , @xmath102 where the quasiparticle contribution is given by @xmath103 ( note that only the time - odd part of @xmath81 contributes to the integral , and @xmath104 ) . the time derivative @xmath105 which appears when one writes down the explicit expression for @xmath106 can be eliminated with the help of eq . ( [ qpvlasovlin ] ) . as a result , the l.h.s . of the continuity equation becomes @xmath107\ , . \label{contilin}\end{gathered}\ ] ] as noted in ref . @xcite , the continuity equation is trivially satisfied in the normal phase ( @xmath108 ) . this becomes evident if its l.h.s . is written in the form ( [ contilin ] ) , since in the normal phase we have @xmath33 and @xmath109 . ( [ qpvlasovlin ] ) is still very complicated . in order to simplify the problem , let us look more closely at the different terms in order to see if some of them are less important than others . the basic assumption being that @xmath50 , @xmath110 , and @xmath111 are much smaller than @xmath112 , the distribution function is sharply peaked near the fermi surface . under this condition it is useful to express the distribution function @xmath64 in terms of the variables @xmath10 , @xmath113 , and @xmath114 instead of @xmath10 and @xmath63 , where @xmath115\,,\quad { \bm{\mathrm{\hat{p}}}}= \frac{{\bm{\mathrm{p}}}}{|{\bm{\mathrm{p}}}|}\,,\ ] ] and @xmath116 . in terms of these variables , @xmath64 is sharply peaked near @xmath117 , and the relevant values of @xmath113 are of the same order of magnitude as @xmath50 , @xmath110 , and @xmath111 . if @xmath81 is written as a function of the new variables , the poisson bracket on the l.h.s . of eq . ( [ qpvlasovlin ] ) becomes @xmath118 with the short - hand notation @xmath119 where @xmath120 and @xmath121 denote the angles characterizing the unit vector @xmath122 . in addition to the assumption @xmath123 , our semiclassical theory requires that all quantities vary slowly in space , i.e. , on a length scale @xmath124 which should be larger than the coherence length @xmath125 @xcite . then , using @xmath126 , @xmath127 , @xmath128 , and @xmath129 , all terms in eq . ( [ poissonxi ] ) can be estimated to be of the order of magnitude @xmath130 , except the last one , which is of the order @xmath131 . hence , the last term of eq . ( [ poissonxi ] ) is negligible . let us now distinguish different kinds of contributions to @xmath81 , depending on whether they are even or odd functions in @xmath113 and @xmath114 : * the part of @xmath81 which is odd in @xmath113 and even in @xmath114 describes , roughly speaking , a change of the fermi momentum , i.e. , fluctuations of the density , and contributes to @xmath86 , @xmath132 with @xmath133 , while its contribution to @xmath87 is suppressed by one power of @xmath134 and can be neglected . * the part of @xmath81 which is even in @xmath113 and odd in @xmath114 describes a shift of the fermi sphere and therefore contributes to the current @xmath135 , @xmath136 and also to the other integral in the continuity equation ( [ contilin ] ) . * the part of @xmath81 which is even in @xmath113 and in @xmath114 describes , roughly speaking , a local temperature fluctuation and leads to a non - vanishing value of @xmath83 ( via @xmath87 ) , @xmath137 while its contribution to @xmath86 is suppressed by one power of @xmath134 and can be neglected . * the part of @xmath81 which is odd in @xmath113 and odd in @xmath114 gives only a negligible contribution to the current @xmath135 ( suppressed by one power of @xmath134 ) . if one neglects the last term in eq . ( [ poissonxi ] ) , the poisson bracket in eq . ( [ qpvlasovlin ] ) leads only to a coupling between @xmath138 and @xmath139 and between @xmath140 and @xmath141 . to be more specific , @xmath141 and @xmath140 do not contribute to the dynamics of @xmath139 and @xmath138 . since we are interested in density oscillations and currents , which are determined by @xmath139 and @xmath138 , we might wonder if we could disregard completely @xmath141 and @xmath140 . to that end we have to check that also on the r.h.s . of eq . ( [ qpvlasovlin ] ) there is no term which couples the undesired quantities @xmath141 and @xmath140 to @xmath139 or @xmath138 . actually , on the r.h.s . ( [ qpvlasovlin ] ) there is no term containing @xmath140 and only one term containing @xmath141 , namely the third one , @xmath142 this term clearly contributes to @xmath143 , but at least to leading order in @xmath134 it does not contribute to @xmath144 or @xmath145 . in the continuity equation ( [ contilin ] ) , @xmath141 and @xmath140 do not appear , i.e. , the undesired quantities @xmath141 and @xmath140 do not contribute to the dynamics of @xmath99 , either . we are therefore allowed to disregard them . now , since we are not interested any more in @xmath141 and @xmath140 , we can remove all the terms on the r.h.s . of eq . ( [ qpvlasovlin ] ) which contribute only to the dynamics of these uninteresting quantities . as mentioned above , this is the case for the third term , eq . ( [ thirdterm ] ) , which contributes only to @xmath143 . the last term on the r.h.s . of eq . ( [ qpvlasovlin ] ) , @xmath146 can be omitted , too , since it contributes only to @xmath147 . in conclusion , we are left with a simplified version of eq . ( [ qpvlasovlin ] ) , which reads @xmath148 the aim of the present work is to solve the vlasov - like equation ( [ qpvlasov ] ) for the quasiparticle - distribution function @xmath64 together with the continuity equation ( [ conti ] ) for the phase of the order parameter with the help of the test - particle method , in analogy to the test - particle method which is used to solve the usual boltzmann equation . the basic idea of this method is to replace the continuous distribution function @xmath57 by a sum of delta functions in phase space , @xmath149\delta[{\bm{\mathrm{p}}}-{\bm{\mathrm{p}}}_i(t)]\ , , \label{tpstandard1}\ ] ] corresponding to a finite number of test particles , each of which follows the classical equation of motion @xmath150 as can be seen by inserting eq . ( [ tpstandard1 ] ) into eq . ( [ qpvlasov ] ) . note that , contrary to the usual test - particle method , our test particles here can not be identified with real particles but rather with bogoliubov quasiparticles . in its general form , the test - particle method can be applied to situations far from equilibrium . however , here we are only interested in the linear - response regime , i.e. , in the limit of small deviations from equilibrium . in this case it is possible to formulate the method in such a way that only the classical trajectories corresponding to the unperturbed system appear . to that end , we make the following ansatz for the deviation of the distribution function from equilibrium : @xmath151\ , . \label{nu1ansatz}\ ] ] inserting this into the linearized transport equation ( [ qpvlasovsimp ] ) , we obtain the following equation of motion for the function @xmath152 : @xmath153 where @xmath154 denoting by @xmath155 and @xmath156 the classical trajectories satisfying the equations of motion @xmath157 with the initial conditions @xmath158 one can easily show that @xmath159 = f[{\bm{\mathrm{r}}}({\bm{\mathrm{r}}},{\bm{\mathrm{p}}};t),{\bm{\mathrm{p}}}({\bm{\mathrm{r}}},{\bm{\mathrm{p}}};t);t]\ , . \label{eomy1}\ ] ] let us now replace the quasiparticle - distribution function by @xmath160 delta functions in phase space . since the order of magnitude of @xmath81 is dominated by @xmath161 , it is clear that these delta functions should be distributed near the fermi surface . to be more specific , we choose @xmath160 points @xmath162 , @xmath163 in phase space which are distributed according to a probability density which is proportional to @xmath161 , in such a way that for arbitrary but sufficiently smooth phase - space functions @xmath164 the integral of @xmath164 times the function @xmath165 $ ] can be approximated by @xmath166 \approx -c \sum_{i=1}^{n_\nu } \chi({\bm{\mathrm{r}}}_i,{\bm{\mathrm{p}}}_i)\ , . \label{montecarlo}\ ] ] note that , if @xmath162 , @xmath163 are distributed in such a way , the same is true for @xmath167 , @xmath168 , since the quasiparticle energy @xmath169 $ ] is a constant of the motion . in particular , defining @xmath170 $ ] and using eq . ( [ montecarlo ] ) , we can approximate the integral of an arbitrary function @xmath171 times the distribution function @xmath81 as @xmath172\ , . \label{nuintegral}\ ] ] in other words , we have replaced @xmath81 by @xmath173 \delta[{\bm{\mathrm{p}}}-{\bm{\mathrm{p}}}_i(t)]\,.\ ] ] according to eq . ( [ eomy1 ] ) , the equation of motion of the coefficients @xmath174 is reduced to @xmath175\ , . \label{eomy}\ ] ] above we assumed the function @xmath164 to be sufficiently smooth . of course , this causes some trouble if we want to calculate local quantities like the density or the current . for instance , we obtain @xmath176\ , , \label{rho1nuy}\end{gathered}\ ] ] where @xmath177 $ ] . this result makes sense only after the delta functions have been averaged over a volume containing a sufficiently large number of test particles in order to have a reasonable statistics . supposing that this can be done , and supposing that @xmath178 and the phase @xmath179 are known , we can use the result for @xmath86 in the explicit expression for @xmath180 in order to obtain a system of @xmath160 coupled first - order differential equations of the form ( [ eomy ] ) for the coefficients @xmath174 . this represents a tremendous simplification with respect to the original partial differential equation ( [ qpvlasovsimp ] ) in seven dimensions ( @xmath10 , @xmath63 , and @xmath181 ) . however , the phase @xmath182 is not known , but it has to be determined from the continuity equation ( [ contilin1 ] ) . this is , again , very difficult . hence , instead of solving the continuity equation exactly , we make an ansatz for @xmath99 and determine the parameters by minimizing the violation of the continuity equation , @xmath183 the explicit expression for @xmath184 being given by eq . ( [ contilin ] ) . the idea is to expand @xmath99 on an appropriately chosen set of orthogonal functions @xmath185 , @xmath186 inserting this ansatz into eq . ( [ contilin ] ) , we see that the integral in eq . ( [ contimin ] ) depends on @xmath187 and @xmath188 . at a given time , we regard @xmath187 and @xmath189 as given ( e.g. , at the moment when the perturbation is switched on , we know that @xmath190 ) . hence , in order to have a minimal violation of the continuity equation , we have to minimize eq . ( [ contimin ] ) with respect to @xmath188 by demanding @xmath191 at this stage it turns out to be convenient to choose the basis functions @xmath185 such that they satisfy the orthogonality relation @xmath192 then we obtain the following differential equation for the coefficients @xmath187 : @xmath193 where @xmath0 is a time - independent matrix , @xmath194 using eq . ( [ nuintegral ] ) and integrating by parts , we can rewrite eq . ( [ eomxmatrix1 ] ) in a more convenient form as @xmath195 where @xmath196 denotes the matrix @xmath197 and the vector @xmath198 is defined by @xmath199 mainly for formal purposes , we note that also the equation ( [ eomy ] ) for the coefficients @xmath174 can be rewritten in matrix notation as @xmath200 + f_i(t)\\ + \sum_{j=1}^{n_\nu } g_{ij}(t ) y_j(t)\ , , \label{eomymatrix}\end{gathered}\ ] ] where @xmath201 and @xmath202}{1+ga}\\ -\frac{\delta_0}{e_i^2}{\bm{\nabla}}\frac{\delta_0 { \tilde{\delta}}[{\bm{\mathrm{r}}}-{\bm{\mathrm{r}}}_j(t)]}{1+ga}\big)_{{\bm{\mathrm{r}}}_i(t)}\ , . \label{gijsmeared}\end{gathered}\ ] ] in the latter equation , @xmath203 denotes a kind of `` smeared '' delta function which accounts for the averaging mentioned below eq . ( [ rho1nuy ] ) . however , as mentioned above , eqs . ( [ eomymatrix ] ) and ( [ gijsmeared ] ) will be used for formal purposes only . in practice , it is much faster to calculate @xmath204 after each time step on a discrete mesh , and to interpolate the stored values when performing the next time step for the coefficients @xmath174 . for the calculation of @xmath204 , we replace the delta function in eq . ( [ rho1nuy ] ) by a gaussian having a width @xmath205 . in summary , the coupled system of partial differential equations , namely the transport equation for the distribution function @xmath81 and the continuity equation for the phase @xmath99 [ eqs . ( [ qpvlasovsimp ] ) and ( [ contilin1 ] ) ] , has been replaced by a coupled system of ordinary linear differential equations for the coefficients @xmath174 and @xmath187 [ eqs . ( [ eomymatrix ] ) and ( [ eomxmatrix ] ) ] , which can formally be written as @xmath206 in practice , the solution of the classical equations of motion for the test particles , eqs . ( [ trajectory ] ) , faces us with some unusual features which are not present with the usual newtonian equations of motion . note that we are not dealing with ordinary particles but with bogoliubov quasiparticles , which have some surprising properties . for instance , @xmath207 being a constant of the motion and @xmath208 , it is evident that the energy @xmath209 can not be conserved if the gap @xmath210 depends on @xmath10 . in particular , when a test particle with quasiparticle energy @xmath207 reaches the surface where @xmath211 , it is reflected ( andreev reflection ) . during this reflection , the momentum @xmath212 stays almost constant , but the energy @xmath209 changes its sign ( i.e. , a particle is transformed into a hole or vice versa ) , such that the velocity @xmath213 is reversed . as a consequence , the quasiparticle is reflected into the direction where it came from , which is very surprising if the incident angle is different from @xmath214 . in order to find the test - particle trajectories numerically , it does not seem very efficient to start directly from eqs . ( [ trajectory ] ) , since a small numerical error in the momentum of the order of @xmath215 would immediately lead to a completely wrong behavior . it is therefore advantageous to make use of the variable @xmath209 , whose equation of motion reads @xmath216 solving this equation together with the equations for @xmath217 and @xmath212 , we can correct @xmath212 after each time step according to @xmath218 in practice , the variable @xmath209 also allows us to introduce a very reliable method for determining the step size . let us denote by @xmath219 the result we obtain after one time step of size @xmath220 , and by @xmath221 the result we obtain after two time steps of size @xmath222 each . then the quantity @xmath223 is a measure for the numerical error and can be used for adapting the step size @xmath220 to the situation . it turns out that the step size has to become very small only during andreev reflection . now let us give some examples for typical test - particle trajectories . for that purpose , let us restrict ourselves to the most simple case which is a spherical harmonic trap , @xmath224 this potential defines the so - called trap units , i.e. , energies are measured in units of @xmath225 , temperatures in units of @xmath226 , lengths in units of @xmath227 , etc . in this example , due to spherical symmetry , not only the quasiparticle energy @xmath59 , but also the angular momentum @xmath228 of a test particle is a constant of the motion . within the local - density approximation ( lda ) @xcite , the density @xmath229 has its maximum at the center of the trap and vanishes approximately ( except for very small temperature effects ) at the thomas - fermi radius @xmath230 . the gap @xmath231 has its maximum at the center of the trap , too , and goes to zero at some critical radius @xmath232 which is temperature dependent and determined by the equation @xmath233 . in order to avoid numerical problems arising from the infinite derivative of @xmath231 at @xmath234 , we convolute the lda result for @xmath231 with a gaussian of width @xmath235 . in fact , this is more realistic than the lda result since the exact solution of the bdg equations also leads to a gap @xmath231 which has an exponential tail @xcite . as parameters we choose @xmath236 , @xmath237 , and @xmath238 . the corresponding number of atoms in the trap is approximately @xmath239 . for these parameters quantum mechanical ( bdg , qrpa ) results are available for comparison . the width @xmath235 is chosen such as to optimize the agreement with the bdg gap , which for the present parameters is achieved with @xmath240 . in fig . [ figgroundstate1 ] we show the corresponding gap @xmath231 as a function of the distance @xmath241 from the center of the trap . from this figure it is evident that due to the condition @xmath242 , the relevant quasiparticles ( having @xmath243 ) are excluded from the region @xmath244 . in addition to the gap , we display the potential @xmath245 , since the motion of a quasiparticle with given energy @xmath59 and angular momentum @xmath246 is also limited by the condition @xmath247 . it has been shown that also within the fully quantum - mechanical bdg theory the lowest - lying quasiparticle states are localized in this region @xcite . in our example , the motion of the relevant quasiparticles is restricted to the region @xmath248 . most of these quasiparticles will undergo andreev reflection . their trajectories are approximately described by an ellipse which is cut at the points where @xmath249 . if @xmath250 , the quasiparticle will move hence and forth on the same partial ellipse . such trajectories with @xmath251 and @xmath252 are shown in the left panel of fig . [ figtraj ] . however , if the quasiparticle energy is higher , the change in energy from @xmath253 to @xmath254 ( or vice versa ) during the andreev reflection results in a change of momentum which is no more negligible . then , due to angular momentum conservation , the angle of reflection is slightly different from the angle of incidence , and the whole trajectory is precessing . an example for such a trajectory with @xmath255 is also shown in the left panel of fig . [ figtraj ] . a completely different picture arises if the initial conditions are such that the quasiparticle does never reach the point where @xmath256 . then the trajectory is just a precessing , slightly deformed ellipse , as shown in the right panel of fig . [ figtraj ] for the case of a trajectory with @xmath257 . there is a striking analogy between these trajectories and the `` glancing '' orbits discussed , e.g. , in ref . @xcite in the context of a superconducting cylinder which is coated by a normal - metal layer . ( solid line ) and potential @xmath245 ( dashed line ) for the case of a spherical trap with frequency @xmath258 , chemical potential @xmath236 , coupling constant @xmath237 , and temperature @xmath259 . @xmath210 and @xmath260 are in units of @xmath225 , @xmath241 is in units of the oscillator length @xmath261 . roughly speaking , these two curves determine the classically allowed region for a quasiparticle with given energy @xmath59.,width=264 ] . the three trajectories shown in the left panel belong to quasiparticles with @xmath262 , @xmath252 , and @xmath263 , respectively . the trajectory displayed in the right panel belongs to a quasiparticle with @xmath264.,title="fig:",height=177 ] . the three trajectories shown in the left panel belong to quasiparticles with @xmath262 , @xmath252 , and @xmath263 , respectively . the trajectory displayed in the right panel belongs to a quasiparticle with @xmath264.,title="fig:",height=177 ] in sec . [ descriptionmethod ] we supposed that one can generate a distribution of points @xmath162 , @xmath163 in phase space such that eq . ( [ montecarlo ] ) is approximately satisfied for sufficiently smooth functions @xmath164 . in practice , this distribution is obtained in two steps . first we generate the coordinates @xmath162 , and in a second step the momenta @xmath163 . the mean density of test particles at a certain point @xmath10 is given by @xmath265 where @xmath203 denotes a smeared delta function in order to account for the averaging . using eq . ( [ montecarlo ] ) , we conclude @xmath266\equiv \frac{w({\bm{\mathrm{r}}})}{c}\ , . \label{ntestparticle}\ ] ] the algorithm for the generation of the coordinates @xmath162 is now very simple . first we look for the maximum @xmath267 of the function @xmath268 . defining @xmath269 , we obtain a function whose values lie between 0 and 1 . then we generate uniformly distributed random points @xmath270 in a volume which contains the whole system , and retain each point with the probability @xmath271 , until the desired number of points , @xmath160 , is reached . the formula ( [ ntestparticle ] ) for the test - particle density @xmath272 can also be used for the determination of the normalization constant @xmath273 . integrating @xmath272 over space , we must recover the total number of test particles . this implies @xmath274 now we turn to the distribution of the momenta @xmath163 . it is evident that the angular distribution of the momenta is isotropic , i.e. , the interesting part of the problem is the distribution of the absolute values , @xmath275 , which is , of course , directly related to the distribution of the energies @xmath209 . let us define the mean number of test particles per energy and volume @xmath276 again , with the help of eq . ( [ montecarlo ] ) , this becomes @xmath277 with @xmath278 , i.e. , for given spatial coordinates @xmath10 , the probability density for finding a particle at energy @xmath113 is proportional to @xmath279 . such a distribution can be generated in the following way . starting from random numbers @xmath280 which are uniformly distributed in the interval @xmath281 , it is straight - forward to show that the energies @xmath282 are distributed according to the probability density @xmath283 . it is evident that negative energies with @xmath284 have to be removed . furthermore , it is preferable to cut the distribution at energies which lie too far away from the fermi surface , e.g. , @xmath285 ( the probability that this happens is less than @xmath286 ) . the momenta @xmath287 are thus limited by @xmath288 , and the function defined by @xmath289 can not become greater than 1 and can serve as a probability . if we retain each energy @xmath290 generated according to eq . ( [ xidistrib ] ) with the probability @xmath291 , the remaining energies are distributed according to the desired distribution . in order to give an illustration for the resulting distribution of test particles , we show in fig . [ figtestparticles ] the radial distribution of @xmath292 test particles in a trap with the same parameters as in fig . [ figgroundstate1 ] . in agreement with what we discussed in the preceding subsection , we see that the test particles are mainly located in the region @xmath293 , corresponding to the region where the system is mainly normal fluid . due to the angular average the statistical fluctuations around the ideal distribution , eq . ( [ ntestparticle ] ) , which is represented by the dotted line , are very small . we verified that , apart from the statistical fluctuations , our test - particle distribution stays constant , which is a good numerical test of both the initial test - particle distribution and of the test - particle trajectories . test particles in a trap with parameters given below fig . [ figgroundstate1 ] , counted in bins of size @xmath294 . for comparison , the dotted curve represents the ideal distribution according to eq . ( [ ntestparticle]).,width=264 ] in the linear response regime , as the name implies , the response to a time - dependent perturbation of the form @xmath295 , with an arbitrary time dependence @xmath296 , can be obtained as convolution of @xmath296 with the response to a delta function in time . it is therefore sufficient to study perturbations of the form @xmath297 we thus set the inhomogeneous terms in eq . ( [ eommatrix ] ) to @xmath298 and @xmath299 , respectively , @xmath300 and @xmath301 being defined analogously to eqs . ( [ vvector ] ) and ( [ fvector ] ) but with @xmath73 replaced by @xmath302 . assuming that the system was in equilibrium before this perturbation , we may ask the question : what are the values of the coefficients @xmath174 and @xmath187 immediately after the perturbation , i.e. , at infinitesimally small @xmath303 ? this question can be answered exactly , since during the infinitesimal period where the perturbation is active , the matrix in eq . ( [ eommatrix ] ) can be regarded as time - independent . integrating eq . ( [ eommatrix ] ) over time from @xmath304 to @xmath305 , we obtain in the limit @xmath306 @xmath307 let us now assume that the function @xmath302 lies in the space spanned by the functions @xmath185 . then it is evident that the corresponding linear combination is given by the coefficients @xmath308 , i.e. , @xmath309 note that the functions @xmath185 do not necessarily have to have this property . for example , we could define a basis of functions satisfying the orthogonality relation ( [ orthogonality ] ) and maybe even a suitably defined completeness relation if @xmath310 , but which all vanish identically outside the superfluid region , i.e. , in the region where @xmath33 ( and @xmath109 ) . ( [ spanv1 ] ) would then be satisfied inside the superfluid region , but not outside . hence , it is an additional requirement for the choice of the functions @xmath185 . combining eqs . ( [ initial ] ) and ( [ spanv1 ] ) , we find @xmath311 eq . ( [ spanv1 ] ) also leads to a simplification of the initial value of the coefficients @xmath174 and the quasiparticle - distribution function . using the explicit expressions for the matrix @xmath312 and the vector @xmath301 [ eqs . ( [ cmatrix ] ) and ( [ fvector ] ) with @xmath73 replaced by @xmath302 ] , we obtain from the third line of eq . ( [ initial ] ) @xmath314 as a consequence , if eq . ( [ spanv1 ] ) is satisfied , the initial values of the coefficients @xmath174 vanish , which implies @xmath315 in fact , the simple result of this subsection , which is summarized in eqs . ( [ initialphi ] ) and ( [ initialnu ] ) , could have been anticipated without any calculation . the effect of a perturbation of the form ( [ perturbationdelta ] ) is to give a particle at position @xmath10 a kick @xmath316 since this kick does not depend on the momentum of the particle , the local fermi sphere is shifted as a whole , there is no change in density and no fermi surface deformation . within the present theoretical framework , cooper pairs are not broken either , they just acquire a center of mass momentum . thus , the distribution function in the local rest frame stays unchanged ( @xmath317 ) , and the collective velocity is given by @xmath318 . note , however , that in reality a perturbation which has the form of a short pulse would lead to much more complicated effects ( e.g. , pair breaking ) . since our semiclassical description requires that the time dependence of the perturbation is slow , our formal result for a delta - like excitation becomes physically meaningful only after it has been convoluted with a function @xmath296 which varies slowly in time . in other words , we can only calculate the low - frequency part of the response function . in this section we will discuss first numerical results which have been obtained using the test - particle method . our intention here is to see whether this method is in principle capable to describe the most important features of collective excitations in superfluid trapped fermi gases . to that end , we will study the quadrupole excitation of a spherical system , which is excited by @xmath319 ( the factor @xmath320 has been introduced in order to make the coefficient @xmath321 dimensionless ) . for practical purposes , we will make an additional approximation : we will restrict our ansatz for the phase , eq . ( [ phiansatz ] ) , to only one or two functions @xmath185 . it is clear from rotational symmetry that in the case of a quadrupole excitation of the form ( [ vperturbation ] ) the most general form the phase can have is @xmath322\,,\ ] ] such that that the functions @xmath185 can be written as @xmath323\,.\ ] ] it is known from superfluid hydrodynamics that at zero temperature the velocity field is essentially linear in the coordinates , i.e. , the function @xmath324 is almost constant . as a first guess we will assume that this is still true at non - zero temperature , and hence we will take only one single function ( @xmath325 ) in the ansatz ( [ phiansatz ] ) for the phase , @xmath326 . the proportionality constant will be determined from the normalization condition ( [ orthogonality ] ) . such a restricted ansatz means of course that the continuity equation will not be exactly satisfied in the superfluid region ( remember that outside the superfluid region the phase has no effect whatsoever ) . we will therefore improve this initial ansatz by including a second function ( @xmath327 ) which allows to modulate @xmath324 in the superfluid region . the first idea one might have is to use for @xmath328 polynomials in @xmath329 and to orthogonalize the resulting functions @xmath185 . however , it turns out that this leads to numerical instabilities due to the fast growing of the resulting polynomials outside the superfluid region . let us explain this effect in some more detail . as seen from the transport equation for the quasiparticle - distribution function , the phase @xmath99 outside the superfluid region enters directly the dynamics of @xmath81 . although the net effect of the phase and of the quasiparticle motion should be independent of the choice of @xmath99 outside the superfluid region , each of these contributions depends on this choice . if @xmath99 changes too rapidly , the numerical solution of the equation of motion for the coefficients @xmath174 becomes less accurate and the cancellation of the two effects does not work any more . we therefore have to look for functions @xmath330 which are linearly independent inside the superfluid region , but which do not grow outside . here we will choose the functions @xmath331 and @xmath332 ^ 2 $ ] , where @xmath333 is the function defined in eq . ( [ defphi ] ) . the function @xmath334 has its maximum in the center of the trap and goes smoothly to zero at the boundary of the superfluid region . from @xmath335 and @xmath336 the functions @xmath337 and @xmath338 are determined according to the orthogonality condition ( [ orthogonality ] ) with the help of the gram - schmidt orthogonalization method . as we will see , the results obtained with @xmath339 and @xmath327 are very similar and we therefore claim that they would not change a lot if we included additional functions . let us now present the results . as in the examples shown in the preceding section , we consider a spherical harmonic trap with @xmath340 , containing approximately @xmath239 atoms . the resulting density profile @xmath229 is shown in fig . [ figgroundstate2 ] as the dashed line . the critical temperature within lda is @xmath341 . as before , the lda result for the gap @xmath231 is convoluted with a gaussian having a width @xmath342 . we will study the quadrupole mode for three different temperatures , @xmath343 , @xmath344 , and @xmath345 . the equilibrium gap @xmath231 for these three temperatures is also displayed in fig . [ figgroundstate2 ] . ( dashed line ) and gap @xmath231 ( solid lines ) in a spherical harmonic trap containing @xmath346 atoms ( @xmath236 , interaction strength @xmath237 ) . the gap is displayed for three different temperatures , @xmath343 , @xmath344 , and @xmath345 , while the density profile is practically independent of @xmath8.,width=264 ] after the system is excited , its shape will oscillate . a measure for this quadrupole deformation is the ratio @xmath347 where @xmath348 denotes the mean square radius in equilibrium , which in the present case has the value @xmath349 . in the linear response , the quadrupole deformation is of course proportional to the strength of the perturbation , and we therefore divide our results by this strength [ denoted @xmath321 in eq . ( [ vperturbation ] ) ] . in our simulation we use @xmath292 test particles , the width of the gaussians used for smearing @xmath86 ( see sec . [ descriptionmethod ] ) is set to @xmath350 . in fig . [ figq ] we display the time dependence of the quadrupole deformation after the perturbation for the three temperatures mentioned before . the corresponding spectra , obtained by fourier transformation , are shown in fig . [ figqfft ] . the results for the two cases @xmath339 and @xmath327 are displayed as dashed and solid curves , respectively . in all cases the two curves are in reasonable agreement , such that we can say that the use of @xmath351 independent functions in the ansatz for the phase is sufficient . we see that the temperature dependence of the spectrum is highly non - trivial . at low temperatures ( upper panel of fig . [ figqfft ] ) , we see essentially the hydrodynamic quadrupole mode , which at zero temperature lies at @xmath352 @xcite and which is now damped as a consequence of its coupling to the normal component . at higher temperatures ( middle of fig . [ figqfft ] ) , a second peak builds up in the spectrum , corresponding to the quadrupole mode in the normal phase , which lies slightly above @xmath353 @xcite ( for the present set of parameters , its frequency is @xmath354 @xcite ) . as the temperature approaches @xmath7 ( lower panel of fig . [ figqfft ] ) , the strength contained in this second peak increases , while the hydrodynamic mode , whose frequency is slightly shifted downwards , disappears . these findings are in good agreement with quantum mechanical qrpa calculations @xcite . . the parameters are the same as in fig . [ figgroundstate2 ] , the three panels correspond , from top to bottom , to @xmath343 , @xmath344 , and @xmath345 . the dashed and solid lines correspond to @xmath339 and @xmath327 , respectively.,title="fig:",width=264 ] + . the parameters are the same as in fig . [ figgroundstate2 ] , the three panels correspond , from top to bottom , to @xmath343 , @xmath344 , and @xmath345 . the dashed and solid lines correspond to @xmath339 and @xmath327 , respectively.,title="fig:",width=264 ] + . the parameters are the same as in fig . [ figgroundstate2 ] , the three panels correspond , from top to bottom , to @xmath343 , @xmath344 , and @xmath345 . the dashed and solid lines correspond to @xmath339 and @xmath327 , respectively.,title="fig:",width=264 ] + .,title="fig:",width=264 ] + .,title="fig:",width=264 ] + .,title="fig:",width=264 ] + we note that the damping width of the hydrodynamic mode at low temperature is now comparable with that found within the qrpa and much stronger than that found in our previous work @xcite , where we replaced the gap @xmath231 by a constant . the reason is in fact very simple : with a constant gap , the fraction @xmath355 of the normal component is independent of @xmath241 , whereas in the case of an @xmath241-dependent gap the normal component in the outer part of the system is already important at very low temperatures @xcite . as the temperature @xmath8 approaches @xmath7 , the quadrupole mode of the normal phase ( that at @xmath356 ) becomes undamped , as it is the case within the qrpa . however , even though collisions are strongly suppressed at these low temperatures , it should be kept in mind that the collision term , which is neglected in the present work , is non - zero and its inclusion would lead to a finite lifetime of this oscillation , too . finally , let us compare our semiclassical results more quantitatively to quantum mechanical qrpa results . in fig . [ figqrpa ] we show the qrpa , @xmath357 , @xmath259 . the semiclassical result ( solid line ) was obtained with @xmath327 basis functions for the phase . the qrpa result ( dotted line ) was taken from ref . @xcite.,width=264 ] result of ref . @xcite for the quadrupole excitation spectrum ( dotted line ) together with the semiclassical result we obtain with the same parameters ( solid line ) . as one can see , the total normalization and the relative weights of the two peaks are in reasonable agreement . also the widths of the qrpa peaks are well reproduced by the semiclassical calculation . the main differences are that within the semiclassical calculation the two peaks lie a bit too high and that they are not as well separated as within the qrpa . a comparison of the two curves for @xmath339 and @xmath327 in the @xmath358 case shown in the middle of fig . [ figqfft ] , whose parameters are quite close to those of fig . [ figqrpa ] , suggests that the latter effect might be partly due to the restricted ansatz for the phase . however , as one can deduce from the irregular structure of the qrpa spectrum , even in a system with 32000 atoms shell effects , i.e. , effects which depend on the discrete single - particle spectrum , are still quite pronounced . it is clear that such effects can not be reproduced within a semiclassical calculation . in this sense the agreement between the two spectra is very satisfactory , in particular since one can assume that the shell effects decrease with increasing number of particles . in this paper , we developed a numerical test - particle method for solving the semiclassical transport equations for an ultracold trapped fermi gas in the bcs phase in the collisionless limit . these transport equations take into account the coupling between the dynamics of the cooper pairs ( superfluid component ) and the thermally excited bogoliubov quasiparticles ( normal component ) . we developed the method for the case of small deviations from equilibrium , so that the test - particle trajectories can be calculated in the equilibrium state . since the test - particles describe bogoliubov quasiparticles rather than real particles , the trajectories have very unusual properties compared with the trajectories one has to deal with when applying the test - particle method to the normal vlasov equation . our test particles can have the character of particles as well as holes , depending on whether their energy @xmath113 is positive or negative , and they can also be transformed from the one into the other if they hit the region where the gap @xmath50 becomes larger than their quasiparticle energy @xmath59 ( andreev reflection ) . another complication as compared with the normal vlasov equation is that the dynamics of the quasiparticles is coupled to the collective motion of the superfluid component , which is described by the phase @xmath56 of the order parameter . this phase has to be determined simultaneously with the evolution of the quasiparticle - distribution function by solving the continuity equation . in the present work , we make an ansatz for @xmath56 with time - dependent coefficients , leading to an approximate solution of the continuity equation . as a first application , we calculated the response of a gas trapped in a spherical trap to a delta - like perturbation of quadrupole form . after this perturbation , the shape of the gas shows a damped oscillation . at low temperatures , this oscillation is just the hydrodynamic quadrupole mode which is damped by its coupling to the normal component . with increasing temperatures , the extension of the normal component increases , and , as a consequence , the normal component can perform its own quadrupole oscillation . since the frequency of the quadrupole mode in the normal collisionless fermi gas is higher than that of the hydrodynamic mode , this leads to a two - peak structure in the response function . as the temperature approaches @xmath7 , the strength of the hydrodynamic mode disappears and only the normal mode survives . the next step will be to apply the method presented here to more realistic cases , namely to the axial and radial breathing modes of a gas in a cigar - shaped trap containing a larger number of particles . in fact , the deformation and the large particle number do not pose a big problem , which is one of the main advantages of the present method as compared with quantum mechanical qrpa calculations . another possible application of the method is to study the dynamics of a vortex , where already the equilibrium situation is characterized by a non - vanishing phase of the order parameter . however , there are still a number of unsolved problems and possible improvements of the method . first of all , the collision term @xcite should be included , which is an additional source of damping of the collective oscillations . as mentioned in the introduction , the possibility to include collisions is an important advantage of the present method as compared with the qrpa , where collision effects can not be taken into account since this would necessitate to include four - quasiparticle excitations . second , from a fundamental point of view , the fact that the continuity equation is only approximately fulfilled is of course unsatisfactory and one should think about another numerical method for solving the continuity equation . finally , one might ask the question how the present theory can be extended to the strongly interacting regime . unfortunately , this question is up to now completely open , since in this regime thermal fluctuations of the order parameter , which are not contained in the bdg equations , play a crucial role ( see , e.g. , ref . @xcite ) . the author wishes to thank p. schuck for numerous fruitful discussions and the critical reading of the manuscript . * m. bartenstein , a. altmeyer , s. riedl , s. jochim , c. chin , j. hecker denschlag , and r. grimm , phys . lett . * 92 * , 203201 ( 2004 ) ; a. altmeyer , s. riedl , c. kohstall , m. wright , r. geursen , m. bartenstein , c. chin , j. hecker denschlag , and r. grimm , cond - mat/0609390 ( 2006 ) ; a. altmeyer , s. riedl , c. kohstall , m.j . wright , j. hecker denschlag , and r. grimm , e - print cond - mat/0611285 ( 2006 ) . j. kinast , s.l . hemmer , m.e . gehm , a. turlapov , and j.e . thomas , phys . lett . * 92 * , 150402 ( 2004 ) ; j. kinast , a. turlapov , and j.e . thomas , phys . a * 70 * , 051401(r ) ( 2004 ) . m. urban , phys . rev . a * 71 * , 033611 ( 2005 ) . e. taylor and a. griffin , phys . a * 72 * , 053630 ( 2005 ) . bruun and b.r . mottelson , phys . lett . * 87 * , 270403 ( 2001 ) ; g.m . bruun , phys . . lett . * 89 * , 263002 ( 2002 ) . m. grasso , e. khan , and m. urban , phys . a * 72 * , 043617 ( 2005 ) . o. betbeder - matibet and p. nozires , ann . * 51 * , 392 ( 1969 ) . serene and d. rainer , phys . rep . * 101 * , 221 ( 1983 ) . d. vollhardt and p. wlfle , _ the superfluid phases of helium 3 _ ( taylor & francis , london , 1990 ) m. urban and p. schuck , phys . a * 73 * , 013621 ( 2006 ) . bertsch and s. das gupta , phys . rep . * 160 * , 190 ( 1988 ) . f. toschi , p. vignolo , s. succi , and m.p . tosi , phys . rev . a * 67 * , 041605(r ) ( 2003 ) ; f. toschi , p. capuzzi , s. succi , p. vignolo , and m.p . tosi , j. phys . b * 37 * , s91 ( 2004 ) . t. maruyama , h. yabu , t. suzuki , phys . a * 72 * , 013609 ( 2005 ) . p. ring and p. schuck , _ the nuclear many - body problem _ ( springer - verlag , berlin 1980 ) , a.l . fetter and j.d . walecka , _ quantum theory of many - particle systems _ , ( mcgraw - hill , new york 1971 ) . m. houbiers , r. ferwerda , h.t.c . stoof , w.i . mcalexander , c.a . sackett , and r.g . hulet , phys . a * 56 * , 4864 ( 1997 ) . m. grasso and m. urban , phys . a * 68 * , 033610 ( 2003 ) . m.a . baranov and d.s . petrov , phys . a * 58 * , r801 ( 1998 ) . g. bruun , y. castin , r. dum , and k. burnett , eur . j. d * 7 * , 433 ( 1999 ) . bruun and h. heiselberg , phys . a * 65 * , 053407 ( 2002 ) . c. bruder and y. imry , phys . lett . * 80 * , 5782 ( 1998 ) . bruun and c.w . clark , phys . 83 * , 5415 ( 1999 ) . m.a . baranov and d.s . petrov , phys . a * 62 * , 041601(r ) ( 2000 ) . m. cozzini and s. stringari , phys . lett . * 91 * , 070401 ( 2003 ) . l. vichi and s. stringari , phys . a * 60 * , 4734 ( 1999 ) ; c. menotti , p. pedri , and s. stringari , phys . lett . * 89 * , 250402 ( 2002 ) . a. perali , p. pieri , l. pisani , and g.c . strinati , phys . 92 * , 220404 ( 2004 ) .

inelastic light ( raman ) scattering has been used to study electronic excitations and phonon anomalies in detwinned , slightly overdoped ybacuo and moderately overdoped ycabacuo single crystals . in both samples modifications of the electronic pair - breaking peaks when interchanging the- and-axis were observed . the lineshapes of several phonon modes involving plane and apical oxygen vibrations exhibit pronounced anisotropies with respect to the incident and scattered light field configurations . based on a theoretical model that takes both electronic and phononic contributions to the raman spectra into account , we attribute the anisotropy of the superconductivity - induced changes in the phonon lineshapes to a smallwave admixture to the pair wave function . our theory allows us to disentangle the electronic raman signal from the phononic part and to identify corresponding interference terms . we argue that the raman spectra are consistent with anwave admixture with an upper limit of 20 percent .

0907.4587

the symmetry of the pair wavefunction provides important clues to the mechanism of high temperature superconductivity . phase sensitive experiments have shown that the dominant component of the order parameter exhibits @xmath8 symmetry . @xcite in cuprates of the yba@xmath0cu@xmath1o@xmath9-type with orthorhombic crystal symmetry , an @xmath7wave admixture resulting in an anisotropy of the superconducting ( sc ) energy gap , @xmath10 , between the @xmath5 and @xmath6axes is expected . a more general trend of increasing @xmath7-wave admixture with increasing doping level has also been reported . @xcite in yba@xmath0cu@xmath1o@xmath11 ( ybco@xmath11 ) , such an admixture has been confirmed by angle - resolved photoemission spectroscopy ( arpes ) @xcite and flux measurements on josephson junctions . @xcite however , the interpretation of the arpes data is complicated because the surface properties of ybco@xmath11 may differ from those of the bulk . @xcite the phase - sensitive measurements , on the other hand , have yielded only lower bounds ( 9% ) to the @xmath7-wave admixture . in addition , phenomenological calculations based on the two - dimensional hubbard model @xcite have demonstrated that the in - plane anisotropy of the spin fluctuation spectrum determined by inelastic neutron scattering on detwinned ybco @xcite is consistent with a small isotropic @xmath7-wave admixture to the @xmath8-pairing symmetry . however , the same calculations show that the dispersion and spectral weight of the spin fluctuations are also influenced by anisotropic hopping parameters that are difficult to determine independently . moreover , different interpretations of the neutron data have also been proposed @xcite . further experiments are therefore required to conclusively establish the magnitude of the @xmath7-wave contribution to the sc gap in ybco@xmath11 . inelastic light ( raman ) scattering is a powerful probe of electronic and lattice vibrational excitations in high - t@xmath12 superconductors . @xcite both electronic and phononic raman scattering have been applied to study the superconducting gap anisotropy in ybco@xmath11 . the anisotropy can be inferred from the energies of gap features of the electronic continuum in various polarization geometries , @xcite but the accuracy of this method is limited by overlap with raman - active phonons . possible manifestations of the gap anisotropy and an anisotropic electron - phonon coupling in the lineshape of a raman - active out - of - plane vibration of the in - plane oxygen atoms of @xmath13 symmetry have also been investigated . @xcite however , the analysis of these data has been contested , @xcite and a quantitative estimate of the gap anisotropy has not been extracted from them . the present work was in part motivated by a recent theoretical study that yielded quantitative predictions for the electronic raman continua in a @xmath14-wave superconductor . @xcite in order to enable a detailed comparison with these predictions , we have performed raman scattering measurements of the temperature dependence of the electronic continua and phonon modes in twin - free , slightly overdoped ybco@xmath2 and moderately overdoped y@xmath3ca@xmath4ba@xmath0cu@xmath1o@xmath2 ( henceforth ybco@xmath2:ca ) single crystals . we found that interference between the electronic and phononic scattering channels imposed severe limitations on our capability to extract information about the gap anisotropy from the electronic continuum alone , in accordance with previous work . @xcite we have therefore employed a phenomenological model that treats the electronic and phononic contributions to the raman signal on equal footing . this formalism yields predictions for the energy , intrinsic linewidth , and fano parameters of phonons coupled to the electronic continuum as a function of temperature . from a comparison to the experimentally determined lineshapes of two phonon modes , the @xmath13 vibration of the in - plane oxygen at 340 @xmath15 as well as an apical - oxygen vibration at 501 @xmath15 , we extract an upper bound of 20% on the @xmath7-wave admixture to the sc energy gap . this is consistent with a 10% to 15% admixture of @xmath7-wave to the @xmath8 pair - breaking peak observed in the electronic continuum in the superconducting state , and it coincides with the lower bound extracted from the phase - sensitive measurements @xcite . similar effects have been recently reported for an organic superconductor with the same point group as yba@xmath0cu@xmath1o@xmath11 ( @xmath16 ) . in this case , however , the analysis led to the conclusion that only an isotropic @xmath7-wave state is present . @xcite this paper is organized as follows . in sec . [ spectra ] we describe experimental details and discuss the raw raman spectra . an introduction to our theoretical model is given in sec . [ model ] . in particular , we focus on a simultaneous description of both , the electronic raman response and the phonon anomalies above and below @xmath17 . based on this description we are able to analyze the superconductivity - induced changes in the phonon lineshape in sec . [ phonons ] . a summary of our results and a comparison to prior work is contained in sec . [ conclusions ] . high quality single crystals of ybco@xmath2 and ybco@xmath2:ca were grown by the top - seeded solution growth method . @xcite the orientations of the crystallographic axes were determined by laue x ray diffraction and polarized light microscopy . the samples were then cut into rectangular shapes of typical size 3@xmath183@xmath181 mm@xmath19 , annealed at 520@xmath20c in a flow of oxygen gas for 150 hours , and quenched in liquid nitrogen in order to avoid further oxygen diffusion . the magnetization curves of the ybco@xmath2 ( ybco@xmath2:ca ) crystal show an onset of @xmath21 k ( 75 k ) . a transition width @xmath22 of less than 3 k ( 6 k ) indicates good homogeneity of the samples . employing tallon s phenomenological expression , @xcite the hole doping level of our samples is estimated as @xmath23 ( 0.21 ) per planar cu ion , i.e. , they are slightly ( moderately ) overdoped . the crystals were detwinned individually using the procedure described in refs . and . raman spectra measured at different spots of the _ ab_-surface of each sample look basically the same , thus confirming the homogeneity of our single crystals . the raman scattering experiments were performed in backscattering geometry using a dilor xy - triple grating raman spectrometer with a ccd camera as detector . for excitation , the 514.5 nm line of an ar@xmath24/kr@xmath24 mixed gas laser was used . the resolution of our spectrometers for this experiment was about 3 @xmath15 . in order to avoid heating of the sample , the power of the incident laser was kept below 10 mw at the sample surface with a laser spot of 100 @xmath25 m in diameter . the direction of the incident laser light was always parallel to the crystallographic @xmath26axis . for the selection rule analysis , a polarizer and an analyzer were placed into the light path before and after the sample . ybco@xmath2 and ybco@xmath2:ca are orthorhombic with the space group @xmath27 ( @xmath16 symmetry ) . in polarized raman scattering experiments , excitations of @xmath28 , @xmath13 , @xmath28+@xmath13 , and @xmath28 symmetries are accessible by using @xmath29 ( or @xmath30 ) , @xmath31 ( or @xmath32 ) , @xmath33 ( or @xmath34 ) , and @xmath35 ( or @xmath36 ) polarizations of the incident and scattered light fields , respectively . @xcite in this notation @xmath37 and @xmath38 correspond to the direction of the electric field of light along the @xmath5 and @xmath6axes , whereas @xmath39 and @xmath40 correspond to the diagonal directions , i.e. @xmath41 and @xmath42 , respectively . as most previous work on ybco has been performed on twinned specimens , we follow previous publications and use the tetragonal notation of the polarization symmetries throughout this paper . in this notation , the structure is described in terms of the closely related tetragonal point group @xmath43 , and the excitations of @xmath28+@xmath13 symmetry are accessible for either @xmath29 or @xmath30 polarization , @xmath44 for either @xmath31 or @xmath32 polarization , @xmath28+@xmath44 for either @xmath33 or @xmath34 polarization , and @xmath13 for either @xmath35 or @xmath36 polarization of the incident and scattered electric field vectors . figure [ opt_spec ] show raman spectra of detwinned ybco@xmath2 crystals measured at various temperatures . as in other high-@xmath17 superconductors , @xcite the normal - state spectra exhibit a flat electronic continuum with superimposed phonons . below @xmath17 , a significant fraction of the electronic spectral weight is transferred to higher energies , resulting in a broad pair - breaking peak in the electronic continuum . in addition , the phonon lineshapes reveal characteristic changes . in b@xmath45 polarization this spectral - weight redistribution is most pronounced because electronic raman scattering in this geometry is sensitive to the maximum of the sc gap along the anti - nodal direction of the two - dimensional sc gap . we therefore focus in this work on this geometry , along with the @xmath29 and @xmath30 polarization channels that provide direct information about the in - plane anisotropy of the electronic response . raman spectra of a detwinned slightly overdoped yba@xmath0cu@xmath1o@xmath2 single crystal in @xmath29 and @xmath30 polarization taken with an ar@xmath46 laser line ( @xmath47=514.5 nm ) . the mode assignment corresponds to refs . . the spectra were shifted by a constant offset with respect to each other . the intensity scales in left and right panels are the same ] . figure [ opt_spec ] gives an overview of the phonon modes of ybco@xmath2 in the @xmath29 and @xmath30 polarization geometries . as expected based on a group - theoretical analysis , @xcite five phonons appear in both polarization channels . the lowest - frequency phonon at 113 @xmath15 ( @xmath48 ) originates predominantly from vibrations of the barium atoms . the next lowest frequency phonon at 148 @xmath15 ( @xmath48 ) corresponds mainly to the vibrations of copper ( cu2 ions ) . the vibrations of the apical oxygen ions ( o4 ) appears at 501 @xmath15 ( @xmath48 ) . the two modes at 340 @xmath15 ( @xmath13 ) and 446 @xmath15 ( @xmath48 ) originate from out - of - phase and in - phase vibrations of the planar oxygen ions ( o2 and o3 ) , respectively . @xcite two additional defect - induced modes appear in the @xmath30 symmetry at 232 @xmath15 and at 579 @xmath15 . they originate from vibrations of the copper ( cu1 ) and the oxygen ( o1 ) ions , respectively , in the cu o chains , which are aligned along the crystallographic @xmath6axis . @xcite in ybco crystals with fully oxygenated chains , these two modes are raman forbidden but infrared allowed ( b@xmath49 symmetry ) . @xcite they become raman - active due to breaking of the translational symmetry by defects , _ i.e. _ , unoccupied oxygen positions in the cu1-o1 chains . @xcite the absence of the strong mode at 579 @xmath15 in the data with @xmath29 polarization confirms the high detwinning ratio of our crystal ( @xmath50% ) . the measured raman intensity @xmath51 in a given polarization channel @xmath52 is related to the imaginary part of the response function @xmath53 via @xmath54 \mathrm{im } \ , \chi_{\sigma } ( \omega ) $ ] , where @xmath55 denotes the bose distribution and @xmath56 is a coupling constant . the raman response in cuprate superconductors in the optimally and overdoped regimes consists of electronic ( intraband ) and phononic excitations . in order to disentangle these two contributions it is necessary to employ a proper fitting procedure . a phonon interacting with intraband excitations acquires a renormalized self - energy and exhibits an asymmetric , fano - type raman lineshape . to describe such an electron - phonon coupled raman spectrum the following formula can be employed @xcite @xmath57 } \\ & & \qquad \times \left\ { s^2(\omega ) - 2 \epsilon ( \omega ) s(\omega ) \rho_{\sigma } ( \omega ) - \rho^2_{\sigma } ( \omega ) \right\ } , \nonumber % \end{aligned}\ ] ] where @xmath58 and @xmath59 . the renormalized phonon frequency and the renormalized phonon line width are given by @xmath60 , and @xmath61 , respectively , with the intrinsic phonon frequency @xmath62 and the bare phonon line width @xmath63 . the parameter @xmath64 can be expressed in terms of the electron - phonon coupling @xmath65 , the raman phonon matrix element @xmath66 , and the raman electronic matrix element @xmath67 , that is , @xmath68 . in our model , @xmath65 , @xmath66 , and @xmath67 are assumed to be real and therefore s@xmath69 is real . the kramers - kronig related functions @xmath70 and @xmath71 denote real and imaginary parts of the electronic response function , respectively , @xmath72 . while the first term in eq . , @xmath73 , describes the `` bare '' electronic raman response , the second term represents the phononic contribution and its coupling to the electronic background . we will use expression to describe the coupling of the electronic background to the phonons whose lineshapes exhibit the clearest manifestations of the electron - phonon interaction ( e.g. , the in - plane @xmath13 phonon ) . the electronic response function @xmath74 can either be computed from a microscopic model @xcite or be determined from a fit to the raman data using a phenomenological model function . @xcite these two approaches will be described in the following two subsections . it is instructive to note that formula can be brought into the form of the widely used `` standard '' fano profile @xcite @xmath75 with the asymmetry parameter @xmath76\end{aligned}\ ] ] and @xmath77 as in eq . . to extract renormalized phonon parameters the raman spectra are often fitted using the above fano profile , eq . , with the fano parameter @xmath78 , the intrinsic phonon frequency @xmath62 , and the renormalized phonon line width @xmath79 ( half - width at half - maximum , hwhm ) kept frequency independent . while such an approach can give valuable insights into the temperature dependence of the phonon lineshapes , the _ intrinsic _ electron - phonon coupling constants and the shape of the electronic continuum @xmath80 in the sc state can not be determined . we will compare the simplified fano formula with our generalized theory [ eq . ( [ eq : fano raman intensity ] ) ] in sec . [ phonons ] . raman spectra of ybco@xmath2 ( @xmath81 nm ) in the normal state at @xmath82 k ( a ) and in the superconducting state at @xmath83 k ( b ) . open circles show the experimental data , the solid curve the fitting result with formula and four lorentzians for the remaining @xmath48 phonons . the insets show the corresponding data after phonon subtraction . panels ( c ) and ( d ) show the phonon - subtracted spectra for the @xmath29 and @xmath30 channels , respectively . the vertical lines indicate the maxima of the electronic peak intensity at 450 @xmath15 and 480@xmath15 , respectively.,scaledwidth=42.0% ] to apply the phenomenological model function [ eq . ] we need to either assume an expression that describes the real and imaginary parts of the electronic response function @xmath80 , or compute the electronic raman response from a microscopic model . let us first describe a phenomenological model function for @xmath84 . following along the lines of refs . we express the imaginary part of the raman efficiency with three terms @xmath85 \nonumber\\ & & \qquad - \left [ \frac{c_{2 } } { 1 + \epsilon^2_{2 } ( \omega ) } - ( \omega \to - \omega ) \right ] , \end{aligned}\ ] ] where @xmath86 and @xmath87 . the first term in eq . models an incoherent electronic background , which dominates the response in the normal state of cuprate superconductors . it is linear in @xmath88 at small frequencies and becomes constant for large raman shifts @xcite . the second and third terms are lorentzians describing the pair - breaking peak located below @xmath89 , and the suppression of spectral weight at low frequencies , respectively . the latter is due to the opening of the sc gap . to reduce the number of free fitting parameters , we set @xmath90 . the last two terms in eq . decrease in intensity and shift spectral weight to lower frequency as the superconducting transition temperature @xmath17 is approached from below . ( @xmath91 ) terms , similar to the c@xmath92 and c@xmath93 terms but with @xmath88 replaced by @xmath94 , are essential to achieve the symmetry requirements for the raman response . in the normal state , the electronic response function is entirely described by the incoherent contribution [ first term in eq . ] . the real part of the electronic response function , @xmath95 , is obtained from the kramers - kronig transform of @xmath96 . since @xmath97 renormalizes the phonon frequency @xmath62 and the parameter @xmath64 , formula together with eq . and its kramers - kronig transform yield a self - consistent analysis of the raman spectra . to compute the kramers - kronig transform of the incoherent part in eq . , a cut - off frequency @xmath98 has to be introduced , which results in a constant offset in the real part of the electronic response @xmath95 . provided @xmath98 is chosen large enough , however , this error only leads to negligibly small corrections . we have employed a non - linear fit procedure with ten ( six ) independent fit parameters to the raman spectra of ybco@xmath2 and ybco@xmath2:ca in the superconducting ( normal ) state . @xmath99 and @xmath100 describe the intensity and position of the maximum of the normal - state electronic background given by the square root of a rational function . in the superconducting state @xmath101 , @xmath102 , and @xmath103 describe amplitude , position and width of a lorentzian function reflecting the region of the pair - breaking peak . finally , @xmath63 , @xmath62 , and @xmath64 effectively characterize amplitude , width , position , and asymmetry of a generalized fano function ( describing the @xmath13 phonon ) . note that for not too strong frequency dependences of @xmath96 and @xmath95 , the renormalization of the fano formula due to the electronic background can be simply viewed as an offset of the parameters entering in the fano formula [ cf . ( 1 ) , ( 2 ) , and ( 3 ) ] . .[tab : param]extracted parameter values of the @xmath13 oxygen vibration in ybco@xmath2 measured at 20 k and 100 k , respectively . left part : asymmetry parameter @xmath64 , intrinsic phonon linewidth ( fwhm ) @xmath104 , and intrinsic phonon frequency @xmath62 using the generalized fano approach of eq.([eq : fano raman intensity ] ) . right part : @xmath62 extracted with eq.([eq : fano profile ] ) for different phenomenological electronic backgrounds [ sqrt = first term in eq.([rho model function ] ) ; lin . = linear background with offset at @xmath105 . for comparison we display in the middle part @xmath62 obtained from inelastic neutron scattering ( ins ) experiments by reznik _ @xcite [ cols="<,^,^,^,^,^,^",options="header " , ] the results of this fit procedure are shown in fig . [ b1g_spec ] and fig . [ ca_b1gxxyy ] together with the experimental data . table [ tab : param ] lists the corresponding fit parameters . note that on general grounds the extracted @xmath62 and @xmath63 parameter values need to be identical at a given temperature for all measurements of a given phonon . in table [ tab : param ] we show , however , the parameter values that correspond to the best ( non - linear least squares ) fit to the data . the differences reflect the error bars of our procedure . it is important to emphasize that the fano profile of the @xmath13-mode shown in figs . [ b1g_spec ] , [ ca_b1gxxyy ] and [ sc_b1g_spec](a ) results from the interaction with the electronic raman signal ; _ both _ electronic and phononic contributions and their interdependence can be described by eqs . ( [ eq : fano raman intensity ] ) and ( [ rho model function ] ) . the final result agrees well with the measured data . thus , our model allows us to some extent to disentangle the electronic and phononic parts of the raman response , and to identify the shape of the electronic background . however , strictly speaking , only a combination of parameters such as @xmath106 and @xmath107 can be extracted . in the next subsection this procedure will be improved for the sc state by employing a microscopic description of the pair - breaking excitations . but for the overdoped sample ybco@xmath2:ca . the vertical lines indicate the maxima of the electronic peak intensity at 290 @xmath15 ( xx ) and 355@xmath15 ( yy ) , respectively.,scaledwidth=42.0% ] to describe the polarization - dependent electronic response function @xmath108 in the superconducting state , we employ a microscopic model using a realistic tight - binding band structure with anisotropic hopping parameters and a superconducting gap with a mixture of @xmath8- and @xmath7-wave symmetries . such a microscopic approach in conjunction with the analysis of the raman spectra based on eq . allows us to obtain precise information about the wavevector dependence of the superconducting order parameter . in particular , we are interested in estimating the magnitude of a possible @xmath7-wave admixture to the @xmath8 pair wave function . the starting point of our calculation is the microscopic model introduced in ref . . the electronic raman response function in the sc state for a given symmetry channel @xmath52 is described by @xcite @xmath109 @xmath110 being the tsuneto function . the angular brackets denote the average over the brillouin zone , that is , @xmath111 the quasiparticle dispersion @xmath112 contains an effective one - band description of a single copper - oxygen plane @xmath113 \nonumber\\ & & \qquad - 4t'\cos k^{\ } _ x\cos k^{\ } _ y - \mu , \label{eq : dispersion}\end{aligned}\ ] ] and the superconducting gap @xmath114 for the computation of the electronic raman response in the superconducting state , eq . , we have assumed the same band structure parameters as in ref . together with @xmath115 mev and @xmath116 mev . the raman vertices @xmath117 in eq . can be classified according to the irreducible representations of the symmetry group of the crystal . since we shall consider a model with small distortions , @xmath118 , @xmath119 , we use the notation corresponding to tetragonal symmetry , as in sec . [ spectra ] above . we can thus express , for example , the @xmath120 raman vertex as @xmath121 \quad .\ ] ] raman spectra of ybco@xmath2 in the superconducting state at @xmath83 k ( @xmath47=514.5 nm ) . open circles show the experimental data . the short dashed , short dotted , and short dash - dotted curves are the theoretical result obtained with eq . . the @xmath13 phonon was also taken into account . we show results for various @xmath7-wave contributions @xmath122 . the dashed , dotted , and dash - dotted curves show the calculated imaginary part of the electronic response , @xmath123 . ( b ) subtracted spectra [ 20k ( sc - state)-100k ( n - state ) ] for the b@xmath45 polarization channel of ybco@xmath2 . before subtraction the spectra were divided by the bose factor . the solid curve depicts the theoretical result for 10% @xmath7-wave contribution.,scaledwidth=45.0% ] to describe the combined electronic and phononic raman response in the sc state , we use eq . together with the real and imaginary parts of the electronic raman response in the sc state , eq . . furthermore , we assume that the bare fit parameters do not change as we go from the normal state to the sc state . in figure [ sc_b1g_spec](a ) we show numerical results obtained from our theoretical model and their comparison with the data on ybco@xmath2 obtained in @xmath13-polarization . the calculations have been performed at @xmath83 k and for various @xmath7-wave contributions . the best description for both the @xmath13-mode and the electronic response is found if @xmath124 is assumed to be 10 percent of the maximum of the @xmath8-wave gap ( short dotted line ) ; this also accounts for the slight shift of the pair breaking peaks in @xmath29 and @xmath30 polarizations ( fig . [ b1g_spec ] c and d ) . furthermore , our numerical results solely for the electronic response , i.e. the 2@xmath125-pair - breaking peak , are also displayed in fig . [ sc_b1g_spec ] ( dashed , dotted and dash - dotted lines ) . they are obtained by setting all phononic parts and the corresponding interference terms in eq . ( [ eq : fano raman intensity ] ) to zero . interestingly , we find that the pair - breaking peak shifts to lower energies with increasing @xmath124 , a fact which has been discussed in a previous paper by schnyder _ in addition , the cubic low - energy response , i.e. its @xmath126-behavior found for @xmath127 changes if @xmath128 : we obtain linear correction terms which are , however , proportional to @xmath129 , and thus barely observable . @xcite raman spectra of ybco@xmath2:ca in the superconducting state at @xmath130 k ( @xmath47=514.5 nm ) . open circles show the experimental data . the short dashed , short dotted , and short dash - dotted curves are the theoretical result obtained with eq . . the @xmath13 phonon was also taken into account . we show results for various @xmath7-wave contributions @xmath131 . the dashed , dotted , and dash - dotted curves show the calculated imaginary part of the electronic response , @xmath132 . ( b ) subtracted spectra [ 12k ( sc - state)-80k ( n - state ) ] for the b@xmath45 polarization channel of ybco@xmath2:ca . before subtraction the spectra were divided by the bose factor . the solid curve depicts the theoretical result for 15% @xmath7-wave contribution.,scaledwidth=45.0% ] figure [ sc_b1g_spec](b ) shows the differences ( @xmath133 ) , which were obtained by subtracting the spectra at 20 and 100 k after dividing by the bose factor . the solid line is obtained after subtracting the model function for the normal state [ see fig . [ b1g_spec](a ) ] from the results for the sc state when @xmath134 [ dotted line in fig . [ sc_b1g_spec](a ) ] . we find that the position of the pair - breaking peak at @xmath135 @xmath15 ( and partly its shape ) is well described by our theory . this confirms that 10% @xmath7-wave contribution [ short dotted line in fig . [ sc_b1g_spec](a ) ] yields an excellent description of the electronic raman response . furthermore , the temperature - dependence of the related @xmath13 phonon is also reproduced ( see below ) . note that the difference between experimental and calculated data at small energies is likely to be due to elastic impurity scattering which is not taken into account in our theoretical model . @xcite finally , we return to the moderately overdoped sample . raman data on ybco@xmath2:ca above and below @xmath17 in several polarization geometries are displayed in fig . [ ca_b1gxxyy ] . in the absence of specific information about the electronic band dispersions of this material , we modeled these data by scaling the magnitude of the energy gap @xmath136 by the ratio of transition temperatures , keeping all other model parameters identical to those used for ybco@xmath2 . the best fit was obtained for @xmath137 ( however the estimated error bars are about @xmath138 ) , slightly larger than the corresponding quantity in ybco@xmath2 . note that the quality of the fit is comparable to the one for ybco@xmath2 , although separate plane and chain subsystems were not introduced in the analysis . @xcite the previously observed difference between spectra in @xmath29 and @xmath30 geometry was confirmed ( fig . [ ca_b1gxxyy ] ) , but our model calculations suggest that this is a consequence of the @xmath7-wave admixture to the gap , @xcite obviating the need to introduce quantum interference between scattering from chains and planes . ( color online ) fano - analysis [ eq.([eq : fano profile ] ) ] of three different phonons measured at a temperature of 20 k ( @xmath47=514.5 nm ) . ( a , b ) the fit of the 340 @xmath15 mode , which corresponds to the out - of - phase vibrations of the planar oxygen , along @xmath30 and @xmath29 polarizations , respectively.(c , d ) same as ( a , b ) but for the in - phase planar oxygen mode around 440 @xmath15 and apical oxygen mode at 501 @xmath15 . open circles show the experimental data . the dash - dotted lines are the results of the overall fit to the experimental data . the solid lines are the results of fits to fano profiles [ eq.([eq : fano profile ] ) ] as described in the text . the dotted lines correspond to a linear background . the intensity units are arbitrary but the same in the four vignettes . ] several phonons in ybco display a pronounced asymmetric lineshape suggesting a strong interaction with the electronic continuum . as shown in fig . [ fit_phonons ] the asymmetry is most pronounced for the 340 @xmath15 mode . interestingly , the phonon peak reflecting the vibration of the apical oxygen at 501@xmath15 exhibits a strong asymmetry for a polarization of incident and scattered light along the @xmath5axis , whereas along the @xmath6axis the phonon appears to be almost symmetric ( see fig . [ fit_phonons ] ) . two additional modes are present in this spectral range . the mode at 440 @xmath15 originates from an in - phase vibration of the oxygen atoms o2 and o3 . additional modes are present at about 472 @xmath15 and 480 @xmath15 for polarizations parallel to the @xmath5- and @xmath6-axis , respectively . as mentioned above , these modes are raman forbidden , but correspond to ir allowed vibrations involving the cu1-o1 chains . @xcite due to defects ( oxygen vacancies ) in the cu1-o1 chains they become raman active . similar to previous temperature dependent raman experiments on ybco@xmath11 @xcite and hgba@xmath0ca@xmath1cu@xmath139o@xmath140 , @xcite we have fitted the phonons of ybco@xmath2 by using simple fano profiles . the solid lines in fig . [ fit_phonons ] are the results of fits to the experimental data using eq . , the dash dotted lines correspond to the resulting fitted lineshape of the entire spectrum . the calculated profiles agree well with the measured spectra . in order to obtain estimates of the intrinsic phonon positions and linewidths , we have corrected the peak position and linewidth by the procedure used in ref . . we find for the @xmath29-symmetry , i.e. , polarizations along the @xmath5-axis : @xmath141 @xmath142 denotes the intrinsic full width at half maximum ( fwhm)@xcite . for the @xmath30-polarizations one obtains an analogous equation . fits to the simplified fano profile of eq . ( [ eq : fano profile ] ) are much less complicated than the fitting procedure to the full spectrum discussed above . this is a key advantage especially in situations in which several closely spaced phonons partially overlap as is the case , for instance , for the apical - oxygen vibration in fig . [ fit_phonons ] . the multi - parameter , global fit [ eq . ( [ eq : fano raman intensity ] ) ] yields unstable results for these phonons . it is important to note , however , that the quantities extracted from simple fano fits are renormalized by the electronic response function , and therefore deviate slightly from the intrinsic phonon frequency @xmath62 and the bare linewidth @xmath63 of table [ tab : param ] . we will compare the results of both procedures in detail at the end of the next subsection . figure [ sc_shifts ] shows the temperature dependence of the energies and linewidths of two particular phonons , the 340 @xmath15 and 501 @xmath15 modes , measured with light polarization along the crystallographic @xmath5 and @xmath6axes . the spectra were taken at temperatures ranging from 20 to 300 k. the temperature dependence of the phonon energy and linewidth in the normal state arises from anharmonic phonon - phonon interactions , i. e. , the decay of a high - energy optical phonon into two phonons of lower energy with opposite momenta @xcite . for simplicity , assuming the resulting phonons to have the same energy @xcite and using bose - einstein statistics , this decay process leads to @xmath143 . this process , implying decay through real transitions , is strictly valid for the line width , but has been also used for the frequency shift although , in this case , virtual transitions also play a role . in fig . [ sc_shifts ] , both linewidth and peak position were fitted simultaneously for the temperature range above @xmath17 ( solid lines ) . both of these quantities show abrupt changes at the sc transition temperature due to the opening of the superconducting gap , as previously observed in ybco ( see for example refs . ) and also other superconducting compounds like hgba@xmath0ca@xmath1cu@xmath139o@xmath144 @xcite and bi@xmath0sr@xmath0cacu@xmath0o@xmath145 . @xcite ( color online ) frequency and line width ( fwhm ) obtained using eqs . ( [ eq : fano profile ] ) and ( [ eq : q_xx ] ) versus temperature of the 340 @xmath15 mode and the 501 @xmath15 mode of ybco@xmath2 for the @xmath29- and @xmath30-polarization , respectively ( @xmath47=514.5 nm ) . below @xmath17 a comparison between calculations for @xmath146 ( dashed line ) , @xmath147 ( dash - dotted line ) , and @xmath148 ( dotted line ) is displayed . the solid line corresponds to the temperature dependence of pure phonon phonon interaction ( interpreted as klemens decay into two phonons of equal frequencies ) . ] within the error bars of @xmath149 @xmath15 in ( a)-(c ) and 2.0 @xmath15 in ( d ) , the phonon peak position and linewidth of the 340 @xmath15 mode is the same for the @xmath29 and @xmath30 polarizations ( see fig . [ sc_shifts]a and b ) . taking the phonon positions and linewidths in the sc state as @xmath150 and @xmath151 , the maximum change in the phonon position and linewidth is obtained from @xmath152 @xmath153 and @xmath154 @xmath155 , @xcite where @xmath156 denotes the temperature at which the maximum of the sc - induced changes occurs . the additional softening below @xmath17 due to the electron phonon interaction is about -6 @xmath15 . the change in linewidth ( fwhm ) is + 3 @xmath15 , reflecting a broadening . for the vibration of the apical oxygen , i.e. the 501 @xmath15 mode , the corresponding shift and broadening are + 3 @xmath15 and + 5 @xmath15 , respectively . the sc - induced changes for the 340 @xmath15 and 501 @xmath15 modes are in good agreement with previous data obtained on twinned ybco . @xcite the sc - induced changes in the phonon linewidth and peak position ( for a phonon labeled as @xmath25 ) can be related to changes in the phonon self energy @xmath157 ( with @xmath158 ) , resulting from the interaction between the phonons and the electronic system below @xmath17 . @xcite the induced frequency shifts @xmath159 are related to the real part of the phonon polarization @xmath160 by @xcite @xmath161 while the induced changes in the linewidth can be calculated via @xmath162 here , @xmath163 denotes the electronic density of states at the fermi level and @xmath164 is the dimensionless electron - phonon coupling constant . @xmath165 denotes the spectral function for phonon @xmath25 under consideration . @xcite taking into account screening effects , i.e. the long - range coulomb force , the self - energy is given by : @xcite @xmath166 where the angular brackets are defined by eq . . the symmetry of the optical phonons is reflected in the matrix element @xmath167 . the electron - phonon coupling of phonons of @xmath168 and @xmath169 symmetry are in a first approximation given by @xmath170 with the electron - phonon coupling strength @xmath171 and @xmath172 ( note that there is more than one phonon of @xmath48 symmetry ) . in general , the phonons below the energy of the superconducting gap @xmath173 should soften below t@xmath174 ( i.e. they should shift to lower energies ) , whereas phonons above @xmath175 should harden . this is confirmed in our experiments ( not all data are shown here ) and in previous work ( see refs . , , and ) . the energy of the apical oxygen phonon ( 501 @xmath15 = 62.5 mev ) is right at the gap energy and is therefore sensitive to small changes in the energy of the sc gap . the 340 @xmath15 mode is well below the sc energy gap for @xmath176 . with increasing temperature the energy of the @xmath89 gap shifts to lower energies and moves through the energy of the 340 @xmath15 mode . this explains the maximum of the linewidth at 75 k. a similar behavior was observed by limonov @xmath177 @xmath178 @xcite . finally , we discuss the role of the @xmath7-wave contribution to the superconducting gap which has been introduced in eq . ( [ eq : s - wavegap ] ) . our numerical results obtained with eqs . ( [ eq : chi_micro])([eq : s - wavegap ] ) and eqs . ( [ eq : phonon self energy_re])([eq : g_constants ] ) using the fits reported in sec . [ model ] are also displayed in fig . [ sc_shifts ] . we compare @xmath127 ( dashed line ) with @xmath179 mev ( dash - dotted line ) and @xmath180 mev ( dotted line ) . we find that the results obtained with 20 percent @xmath7-wave contribution can not describe our data . this is clearly visible in figs . [ sc_shifts](a ) and ( c ) which show the sc - induced changes @xmath159 in the position of the corresponding phonon . in particular , for the 501 @xmath15-mode which is known to be very sensitive to the superconducting gap , @xcite one predicts a softening for @xmath180 mev , while instead a hardening is observed . therefore , our data imply an upper limit of @xmath181 . the best agreement is obtained for @xmath182 s - wave admixture . at the end of this subsection , we are contrasting our generalized fano theory [ see eq . ( [ eq : fano raman intensity ] ) ] with the standard fano approach described by eqs . ( [ eq : fano profile ] ) and ( [ eq : q_xx ] ) . the main difference between both approaches is the theoretical description of the electronic raman response . while in the standard fano approach the background is assumed either to be linear ( see fig . [ fit_phonons ] ) or to follow a square - root behavior @xcite [ first term in eq . ( [ rho model function ] ) ] in both the normal and sc - state , our generalized fano theory is able to take the rearrangement of spectral weight due to the opening of the superconducting gap into account . within our microscopic description [ eqs . ( [ eq : chi_micro])([eq : s - wavegap ] ) ] it is then possible to determine the ratio @xmath183 . another difference between both fano theories concerns the asymmetry parameter @xmath78 : in our generalized theory it becomes @xmath88-dependent which allows a self - consistent description of the @xmath13-phonon and the electronic raman response from 80 to about 1000 @xmath15 . thus , in other words , no specific frequency interval close to the phonon peak position [ see fano lineshape analysis in figs . [ fit_phonons](a ) and ( b ) ] needs to be selected . we have summarized our comparison in table i in which the intrinsic frequency of the @xmath13-phonon extracted from both fano theories is displayed : @xmath62 obtained from eq . ( [ eq : fano raman intensity ] ) [ left part , third column ] , @xmath62 obtained from eqs . ( [ eq : fano profile ] ) and ( [ eq : q_xx ] ) either for a square - root [ @xmath184 or linear background [ @xmath185 ( right part ) . these values have to be compared with @xmath62 obtained from inelastic neutron scattering experiments [ @xmath186 ( middle column ) . one clearly sees that the values for @xmath62 obtained from our generalized fano theory are similar to those measured in ins experiments . on the other hand , the values extracted from the standard fano approach differ by 1.1 [ b@xmath45(100k ) , @xmath62(sqrt ) ] to 5.9 [ xx(20k ) , @xmath62(lin . ) ] wavenumbers . note , however , that these values can be substantially modified if a specific frequency interval around the peak position of the @xmath13-phonon is defined and the corresponding lineshape analysis is then restricted to this interval ( see fig . [ fit_phonons ] . ) these improved values are shown in fig . [ sc_shifts ] . in the case of the apical oxygen vibration analyzed in figs . [ fit_phonons](c ) and ( d ) , a fit with eq . ( [ eq : fano raman intensity ] ) is hardly possible due to the multiple peak structure between 400 @xmath15 and 540 @xmath15 . therefore , we have used eqs . ( [ eq : fano profile ] ) , ( [ rho model function ] ) , and ( [ eq : q_xx ] ) . on the whole , we find characteristic differences between our self - consistent , generalized fano theory and the standard fano approach , but our conclusions about the ratio @xmath129 are independent of the theory used provided that eqs . ( [ eq : fano profile ] ) , ( [ rho model function ] ) , and ( [ eq : q_xx ] ) are restricted ( or strongly weighted ) to an interval around the peak position of the phonon being investigated . nm ) of both polarization channels for the 340 @xmath15 and 503 @xmath15 mode of ybco@xmath2 , respectively . results of the theoretical calculations for different s - wave admixtures ( 0 , 0.1 , and 0.2 ) are also shown . , scaledwidth=45.0% ] figure [ asymmetry ] shows the temperature dependence of the asymmetry parameters @xmath187 and @xmath188 in eq . ( [ eq : fano profile ] ) as obtained from the fits of the fano profile to the experimental spectra . we discuss these data in the framework of the approximate expression @xmath189 already applied in raman work on other high @xmath190 superconductors . @xcite we first focus on temperatures above @xmath17 and note that @xmath191 for the 340 @xmath15 mode , whereas a pronounced difference in the asymmetry parameters along in @xmath29 and @xmath30 geometries is observed for the 501 @xmath15 mode ( @xmath192 , @xmath193 ) . as @xmath194 is a materials parameter independent of the light - field configuration and @xmath195 is weakly @xmath88-dependent , the strong variation of asymmetry parameters must be attributed to differences in @xmath196 . our results are in fair agreement with the phononic and electronic raman efficiencies ( proportional to @xmath197 and @xmath198 , respectively ) calculated in refs . using density functional methods . for instance , the calculated @xmath199 ( @xmath200 ) in the normal state is about 1.2 and 0.65 for the 340 @xmath15 and 501 @xmath15 modes , respectively . the changes of the asymmetry parameters with temperature in the normal state may reflect a temperature dependence of @xmath67 , _ i.e. _ parameters in the electronic band structure as well as that of the anharmonic linewidth , but their detailed origin can not be disentangled moving to the superconducting state , we note that the only parameter in eq . ( [ eq : new_q ] ) expected to change significantly across @xmath17 is the response function @xmath195 . the temperature dependence of @xmath201 and @xmath202 below @xmath17 therefore also reflects the _ ab_-anisotropy in the 2@xmath125-gap . thus we have computed the temperature evolution of the asymmetry parameters in the superconducting state , using the same model parameters described in sec . [ model ] . the reasonable agreement with the data ( fig . [ asymmetry ] ) demonstrates the self - consistency of our approach . however , since the superconductivity - induced modification of @xmath201 and @xmath202 is relatively subtle ( compared to that observed in the normal state ) , this analysis does not provide constraints on the ratio @xmath129 beyond those already discussed above . it is interesting to note that other @xmath28 modes also reveal a temperature - dependent asymmetry . in particular , we have observed that the ba mode near 115 @xmath15 is symmetric ( _ i.e. _ @xmath203 ) for @xmath204 and its asymmetry increases monotonically to @xmath205 ( not shown ) . its intensity , however , is rather weak compared with the @xmath13 mode ( and this applies even more so to the cu mode near 150 @xmath15 ) , so we have not used these modes further in our discussion . their asymmetry has been analyzed in more detail in ref . , however without considering a possible @xmath7-component to the @xmath8-wave gap . in this work we have revisited the superconductivity - induced effects on phononic and electronic raman scattering in detwinned slightly overdoped yba@xmath0cu@xmath1o@xmath2 and moderately overdoped y@xmath3ca@xmath4ba@xmath0cu@xmath1o@xmath2 single crystals , both from an experimental and a theoretical point of view . in particular , we have performed a detailed study of the in - plane anisotropies in the electronic continuum and in the phonon lineshapes and assessed their implications for an @xmath7-wave admixture to the @xmath206-wave superconducting gap . to this end , we developed and applied a formalism that treats the frequency dependence of both electronic and phononic raman scattering on equal footing . thus we can disentangle both parts and clarify the role of interference terms . due to this procedure we are able to extract the intrinsic frequency @xmath62 of the phonon position reflecting the intrinsic electron - phonon interaction . a comparison shows that @xmath62 agrees well with ins data at the @xmath79-point . the best agreement with the raman data was obtained by model calculations based on admixtures of 10% and 15% @xmath207 5% @xmath7-wave contribution for yba@xmath0cu@xmath1o@xmath2 and y@xmath3ca@xmath4ba@xmath0cu@xmath1o@xmath2 , respectively . this agrees with values obtained by other experimental methods @xcite and confirms the previously observed trend @xcite of an increase in the @xmath7-wave contribution with increasing doping level . our data do not show evidence of the previously reported @xcite unusual quantum interference between electronic raman scattering from planes and chains in overdoped y@xmath208ca@xmath209ba@xmath0cu@xmath1o@xmath11 . we would like to thank r. zeyher and j. unterhinninghofen for helpful discussions and k. syassen for a critical reading of the manuscript . thanks v. hinkov , r.merkle , and b. baum for their help in sample preparation and a. schulz and h. uhlig for technical support . this work was partially supported by the international max planck research school for advanced materials ( imprs ) . a. s. thanks the swiss nsf for its financial support and the max planck institute for hospitality . wollman , d.j . van harlingen , w.c . lee , d.m . ginsberg , and a.j . leggett , phys . lett . * 71 * , 2134 ( 1993 ) . tsuei , j.r . kirtley , c.c . chi , lock see yu - jahnes , a. gupta , t. shaw , j.z . sun , and m.b . ketchen , phys . lett . * 73 * , 593 ( 1994 ) ; c. c. tsuei , j. r. kirtley , z. f. ren , j. h. wang , h. raffy , z. z. li , nature * 387 * , 481 ( 1997 ) . t. strohm and m. cardona , solid state comm . * 104 * , 233 ( 1997 ) ; t. strohm , ph.d . thesis , university of stuttgart ( 1999 ) . t. masui , m. limonov , h. uchiyama , s. lee , s. tajima , and a. yamanaka , phys . rev . b * 68 * , 060506(r ) ( 2003 ) . t. hiramachi , t. masui , s. tajima , physica c , * 463 * , 89 ( 2007 ) . r. nemetschek , r. hackl , m. opel , r. philipp , m.t . bal - monod , j.b . bieri , k. maki , a. erb , and e.walker , eur . j. b * 5 * , 495 ( 1998 ) . lu , d.l . feng , n.p . armitage , k.m . shen , a. damascelli , c. kim , f. ronning , z .- x . shen , d.a . bonn , r. liang , w.n . hardy , a.i . rykov , and s. tajima , phys . * 86 * , 4370 ( 2001 ) ; h. uchiyama , t. masui and s. tajima , j. low temp . phys . * 131 * , 287 ( 2003 ) . see j.r . kirtley , c.c . tsuei , a. ariando , c.j.m . verwijs , s. harkema , and h. hilgenkamp , nat . phys . * 2 * , 190 ( 2006 ) , and references therein . zabolotnyy , s.v . borisenko , a.a . kordyuk , j. geck , d.s . inosov , a. koitzsch , j. fink , m. knupfer , b. bchner , s .- l . drechsler , h. berger , a. erb , m. lambacher , l. patthey , v. hinkov , and b. keimer , phys . b * 76 * , 064519 ( 2007 ) . i. eremin and d. manske , phys . * 94 * , 067006 ( 2005 ) . a. p. schnyder , d. manske , c. mudry , and m. sigrist , phys . b * 73 * , 224523 ( 2006 ) . h.a . mook , p. dai , f. dogan , and r.d . hunt , nature * 404 * , 729 ( 2000 ) . v. hinkov , s. pailhs , p. bourges , y. sidis , a. ivanov , a. kulakov , c.t . lin , d.p . chen , c. bernhard , and b. keimer , nature * 430 * , 650 ( 2004 ) . g. s. uhrig , k. p. schmidt , and m. grninger , phys . lett . * 93 * , 267003 ( 2004 ) . m. vojta , th . vojta , and r. k. kaul , phys . * 97 * , 097001 ( 2006 ) . for a review , see t.p . devereaux and r. hackl , rev . * 79 * , 175 ( 2007 ) . c. thomsen , in _ light scattering in solids vi _ , edited by m. cardona and g. gntherodt , springer verlag berlin ( 1991 ) , p.285 . m. f. limonov , a. i. rykov , s. tajima , and a. yamanaka , phys . b * 61 * , 12412 ( 2000 ) . limonov , a.i . rykov , s. tajima , and a. yamanaka , phys . lett . * 80 * , 825 ( 1998 ) . t. strohm , v.i . belitsky , v.g . hadjiev , and m. cardona , phys . lett . * 81 * , 2180 ( 1998 ) . schnyder , c. mudry , and d. manske , phys . rev . b * 75 * , 174525 ( 2007 ) . e. faulques , v. g. ivanov , c. mzire , and p. batail , phys . rev . b * 62 * , r9291 ( 2000 ) . c. t. lin , b. liang , h. c. chen , j. cryst . growth * 237 - 239 * , 778 ( 2002 ) . j. l. tallon , c. bernhard , h. shaked , r. l. hitterman , and j. d. jorgensen , phys . b * 51 * , 12911 ( 1995 ) . voronkova and th . wolf , physica c * 218 * , 175 ( 1993 ) . c. t. lin and a. kulakov , physica c * 408 - 410 * , 27 ( 2004 ) . f. slakey , s. l. cooper , m. v. klein , j. p. rice , and d. m. ginsberg , phys . b * 39 * , 2781 ( 1989 ) . k. c. hewitt and j. c. irwin , phys . rev . b * 66 * , 054516 ( 2002 ) . x. k. chen , j. c. irwin , h. j. trodahl , m. okuya , t. kimura , k. kishio , physica c * 295 * , 80 ( 1998 ) . x. k. chen , j. c. irwin , r. liang , w. n. hardy , physica c * 227 * , 113 ( 1994 ) . m. le tacon , a. sacuto , a. georges , g. kotliar , y. gallais , d. colson , a. forget , nat . * 2 * , 537 ( 2006 ) . m. cardona , physica c * 317 - 318 * , 30 ( 1999 ) . c. thomsen , m. cardona , b. gegenheimer , r. liu , and a. simon , phys . b * 37 * , 9860 ( 1988 ) ; b. friedl , c. thomsen , and m. cardona , phys . rev . lett . * 65 * , 915 ( 1990 ) . c. thomsen and m. cardona in _ physical properties of high temperature superconductors i _ , ed . by d.m . ginsberg ( world scientific , singapore , 1988 ) , p.411 . d.r . wake , f. slakey , m.v . klein , j.p . rice , and d.m . ginsberg , phys . . lett . * 67 * , 3728 ( 1991 ) . c. thomsen and g. kaczmarczyk in _ handbook of vibrational spectroscopy _ , ed . by j. m. chalmers & p. r. griffiths ( wiley , chichester , 2002 ) , pp . 2651 - 2669 . m. n. iliev , v. g. hadjiev , s. jandl , d. le boeuf , v. n. popov , d. bonn , r. liang , w. n. hardy , phys . b * 77 * , 174302 ( 2008 ) . a. g. panfilov , m. f. limonov , a. i. rykov , s. tajima , and a. yamanaka , phys . b * 57 * , r5634 ( 1998 ) . s. bahrs , s. reich , a. zwick , a. r. goi , w. bacsa , g. nieva , and c. thomsen , phys . b * 241 * , r63 ( 2004 ) . a. bock , s. ostertun , r. das sharma , m. rbhausen , k .- o . subke , and c. t. rieck , phys . b * 60 * , 3532 ( 1999 ) . a. bock , ann . ( leipzig ) * 8 * , 441 ( 1999 ) . x. k. chen , e. altendorf , j. c. irwin , r. liang , and w. n. hardy , phys . rev . b * 48 * , 10530 ( 1993 ) . t. p. devereaux , a. virosztek , a. zawadowski , phys . b * 51 * , 505 ( 1995 ) . m. le tacon , a. sacuto , y. gallais , d. colson , and a. forget , phys . b * 76 * , 144505 ( 2007 ) . m. limonov , d. shantsev , s. tajima , and a. yamanaka , phys . b * 65 * , 024515 ( 2001 ) . m. v. klein , in _ light scattering in solids i _ , edited by m. cardona , springer verlag berlin ( 1983 ) , p. 169 . m. limonov , d. shantsev , s. tajima , and a. yamanaka , physica c * 357 - 360 * , 265 ( 2001 ) . devereaux , phys . b * 50 * , 10287 ( 1994 ) . we emphasize that the low - energy power laws in figs . [ b1g_spec ] and fig . [ sc_b1g_spec](a ) are changed in the presence of an @xmath7-wave contribution while there exists no power law for the substracted data in fig . [ sc_b1g_spec](b ) . t. masui , m. limonov , h. uchiyama , s. tajima , and a. yamanaka , phys . lett . * 95 * , 207001 ( 2005 ) ; m. limonov , t. masui , h. uchiyama , s. lee , s. tajima , and a. yamanaka , physica c * 392 - 396 * , 53 ( 2003 ) . t. masui , t. hiramachi , k. nagasao , and s. tajima , phys . rev . * 79 * , 014511 ( 2009 ) . c. bernhard , t. holden , j. humlicek , d. munzar , a. golnik , m. klser , th . wolf , l. carr , c. homes , b. keimer , and m. cardona , solid state commun . * 121 * , 93 ( 2002 ) . e. altendorf , x.k . chen , j.c . irwin , r. liang , and w.n . hardy , phys . rev . b * 47 * , 8140 ( 1993 ) ; e. altendorf , j. chrzanowski , j. c. irwin , a. oreilly , and w. n. hardy , physica c * 175 * , 47 ( 1991 ) . v.g . hadjiev , xingjiang zhou , t. strohm , m. cardona , q.m . lin , and c.w . chu , phys . b * 58 * , 1043 ( 1998 ) . note that the derivation of eq . ( [ eq : q_xx ] ) assumes that the raman intensity @xmath210 can be described by eq . ( [ eq : fano profile ] ) and an @xmath88-independent background . on the other hand , if @xmath211 ( with @xmath212 ) is assumed , one obtains higher correction terms for @xmath213 : @xmath214 . since , however , the slope @xmath215 is relatively small , those higher corrections are of the order of 2% for the @xmath13-phonon , for example . therefore , it is reasonable to work with eq . ( [ eq : q_xx ] ) . g. klemens , phys . rev . * 148 * , 845 ( 1966 ) . j. menndez and m. cardona , phys b * 29 * , 2051 ( 1984 ) . m. limonov , s. lee , s. tajima , and a. yamanaka , phys . b * 66 * , 054509 ( 2002 ) . martin , j.a . sanjurjo , k.c . hewitt , x.z . wang , j.c . irwin , and m.j.g . lee , phys . b * 56 * , 8426 ( 1997 ) . k. c. hewitt , x. k. chen , c. roch , j. chrzanowski , j. c. irwin , e. h. altendorf , r. liang , d. bonn , and w. n. hardy , phys . b * 69 * , 064514 ( 2004 ) . r. zeyher and g. zwicknagl , z. phys . b * 78 * , 175 ( 1990 ) . e. j. nicol , c. jiang , and j. p. carbotte , phys . b * 47 * , 8131 ( 1993 ) . g. d. mahan , _ many - particle physics _ , plenum press , new york ( 1981 ) . d. reznik , b. keimer , f. dogan , and i. a. aksay , phys . * 75 * , 2396 ( 1995 ) . e. t. heyen , m. cardona , j. karpinski , e. kaldis , and s. rusiecki , phys . b * 43 * , 12958 ( 1991 ) . e.t . heyen , ph.d . thesis , university of stuttgart , germany ( 1991 ) . m. c. krantz , i. i. mazin , d. h. leach , w. y. lee , and m. cardona , phys . b * 51 * , 5949 ( 1995 ) . this is similar to a @xmath216-behavior of the electronic background assumed in ref .. e. faulques and v. g. ivanov , phys . b * 55 * , 3974 ( 1997 ) .

we theoretically study the acoustic phonon limited mobility in-doped two - dimensional mos for temperatures k and high carrier densities using the boltzmann equation and first - principles calculations of the acoustic electron - phonon ( el - ph ) interaction . in combination with a continuum elastic model , analytic expressions and the coupling strengths for the deformation potential and piezoelectric interactions are established . we furthermore show that the deformation potential interaction has contributions from both normal and umklapp processes and that the latter contribution is only weakly affected by carrier screening . consequently , the calculated mobilities show a transition from a high - temperature behavior to a stronger behavior in the low - temperature bloch - grneisen regime characteristic of unscreened deformation potential scattering . intrinsic mobilities in excess of v s are predicted at k and high carrier densities ( ) . at k , the mobility does not exceed v s . our findings provide new and important understanding of the acoustic el - ph interaction and its screening by free carriers , and is of high relevance for the understanding of acoustic phonon limited mobilities in general .

1206.2003

two - dimensional ( 2d ) atomic crystals @xcite such as graphene @xcite are promising candidates for future electronic applications . monolayers of semiconducting transition metal dichalcogenides ( mx@xmath1 ) constitute a new family of 2d materials @xcite which have interesting electronic and optical properties @xcite . in conjunction with the excellent gate control inherent to atomically thin materials their finite gap makes them desirable materials for various electronic applications . however , in spite of the recent progress in sample fabrication and transport measurements on gated single to few - layer samples @xcite , little is so far known about the intrinsic carrier properties such as factors limiting the achievable mobilities . experimentally , monolayer mos@xmath1 has been demonstrated to be a direct - gap semiconductor with a band gap of @xmath131.8 ev @xcite and a room - temperature mobility in @xmath0-type samples ranging from @xmath131 to @xmath13200 @xmath6 v@xmath7 s@xmath7 depending on the device structure @xcite . the highest values have been obtained in top - gated samples with a high-@xmath14 gate dielectric @xcite , indicating that impurity scattering can be strongly suppressed by dielectric engineering @xcite , and mobilities close to our theoretically predicted intrinsic phonon limited mobility of @xmath15 @xmath6 v@xmath7 s@xmath7 can be achieved @xcite . other theoretical studies have addressed different issues related to the performance of monolayer mos@xmath1 transistors @xcite . at low temperatures where optical phonon scattering is suppressed , scattering by acoustic phonons can be expected to become an important limiting factor for the mobility of the two - dimensional electron gas ( 2deg ) confined to the atomic layer of an extrinsic 2d semiconductor . this is the case in conventional heterostructure - based 2degs where impurity and acoustic phonon scattering are dominating scattering mechanisms at low temperatures @xcite . in contrast to impurity scattering which can be suppressed by e.g. dielectric engineering , scattering by acoustic phonons is intrinsic to the semiconductor and can not be eliminated . the intrinsic mobility determined by acoustic phonon scattering alone therefore provides an important upper limit for the achievable mobilities . in the low - temperature regime , acoustic phonon - dominated transport manifests itself in a strong change in the temperature dependence of the carrier mobility once the temperature is lowered below the bloch - grneisen ( bg ) temperature @xmath16 . it is given by @xmath17 , where @xmath18 is the fermi wave vector , @xmath19 is the acoustic sound velocity , and @xmath20 is the boltzmann constant , and marks the temperature below which full backscattering at the fermi surface by acoustic phonons is frozen out ( see fig . [ fig : mos2 ] ) . for heterostructure - based 2degs the bg regime is well established @xcite , and recently , transport in the bg regime has been studied in graphene both experimentally @xcite and theoretically @xcite . in a 2deg the fermi wave vector @xmath18 scales with the carrier density as @xmath21 and the bloch - grneisen temperature acquires a similar density dependence @xmath22 with @xmath23 and @xmath24 denoting the spin and valley degeneracy , respectively @xcite . for monolayer mos@xmath1 ( @xmath25 ) this results in bg temperatures @xmath26 for the transverse ( ta ) and longitudinal ( la ) acoustic phonon , respectively , with the carrier density @xmath27 @xmath10 measured in units of @xmath28 @xmath10 . these numbers are on the same order of magnitude as those for graphene @xcite , and transport in the high - mobility bg regime should be achievable in monolayer mos@xmath1 ( and other 2d transition metal dichalcogenides ) . however , the above considerations also emphasize the importance of high extrinsic carrier densities @xmath29 @xmath10 in order for the bg transition to occur at sufficiently high temperatures where acoustic phonon scattering is significant @xcite . such large carrier densities can be achieved with e.g. advanced electrolytic gating where densities on the order of @xmath30 @xmath10 have been reached in 2d samples of graphene and mos@xmath1 @xcite . valleys of the brillouin zone showing the phase space available for scattering below and above the bloch - grneisen temperature @xmath16 . the part of the hexagonal brillouin zone marked by the gray - shaded box indicates the plotting range for the contour plots in fig . [ fig : m ] . bottom : lattice and primitive unit cell of 2d hexagonal mos@xmath1 . ] valleys of the brillouin zone showing the phase space available for scattering below and above the bloch - grneisen temperature @xmath16 . the part of the hexagonal brillouin zone marked by the gray - shaded box indicates the plotting range for the contour plots in fig . [ fig : m ] . bottom : lattice and primitive unit cell of 2d hexagonal mos@xmath1 . ] in the present work , we study the acoustic phonon limited mobility of @xmath0-type 2d mos@xmath1 at low temperatures ( @xmath2 k ) taking into account both deformation potential ( dp ) and piezoelectric ( pe ) scattering . the flexural phonon couples weakly to charge carriers and is here neglected . in our previous work considering scattering of both acoustic and optical phonons @xcite , only the deformation potential interaction was taken into account in the coupling to the acoustic phonons . there we found that the mobility at higher temperatures ( @xmath31 k ) was dominated by optical phonon scattering . with piezoelectric interaction included this is still the case , however , with a slightly lower room - temperature mobility of 320 @xmath6 v@xmath7 s@xmath7 at @xmath32 @xmath10 . otherwise the conclusions of ref . remain unaffected . for the temperatures considered in this work , scattering by intervalley acoustic phonons and optical phonons is strongly suppressed and can be neglected @xcite . using a first - principles approach , we calculate the deformation potential and piezoelectric interactions in 2d mos@xmath1 . supported by continuum model calculations of the acoustic el - ph interaction in 2d hexagonal lattices , this allows us to establish analytic expressions and the individual coupling strengths for the two scattering mechanisms . the calculated intrinsic low - temperature mobility provides a platform for comparison with future measurements of the carrier mobility in monolayer mos@xmath1 which can ( i ) lead to an experimental verification of the theoretical deformation potentials and piezoelectric constant reported here @xcite , ( ii ) reveal to what extent the mobility is affected by extrinsic surface acoustic / optical phonons @xcite which have turned out to be important in substrate - supported graphene samples @xcite , and ( iii ) address the importance of the interplay between scattering of acoustic phonons and impurities which results in a complex temperature and density dependence of the mobility @xcite . in this context previous studies have emphasized the importance of including both the ta and la phonon in order to obtain good agreement with experiment @xcite . with the present work we uncover new important aspects of the acoustic el - ph interaction and how it is affected by carrier screening . these are issues of high relevance for the understanding of acoustic phonon limited 2deg mobilities in semiconductors and , in particular , monolayers of transition metal dichalcogenides . the paper is organized as follows . section [ sec : ii ] briefly summarizes the boltzmann transport theory for acoustic phonon scattering . in sec . [ sec : iii ] the first - principles results for the acoustic electron - phonon ( el - ph ) interaction are presented and the deformation potential and piezoelectric interactions are discussed in closer detail along with a microscopic description of carrier screening . finally , the results for temperature and density dependence of the acoustic phonon limited mobility are presented in sec . [ sec : iv ] . two - dimensional mos@xmath1 has a hexagonal lattice structure like graphene with the bottom of the conduction band residing in the @xmath33 points at the corners of the brillouin zone @xcite . the two @xmath33 valleys are perfectly isotropic with an effective electron mass of @xmath34 @xcite . the satellite valleys located at the @xmath35-@xmath36 path inside the brillouin zone are well separated in energy from the @xmath33-valleys and therefore not important for the low - field transport properties @xcite . the conduction band spin splitting of a few mev due to the intrinsic spin - orbit interaction in 2d transition metal dichalcogenides @xcite can be safely neglected here . at the same time we note that a rashba - type spin - orbit interaction can affect the phonon limited 2deg mobility @xcite . in boltzmann theory , the drift mobility @xmath37 , where @xmath38 is the conductivity , is in the presence of ( quasi ) elastic scattering given by the drude - like expression @xcite @xmath39 where @xmath40 is the energy - dependent relaxation time and the energy - weighted average @xmath41 is defined by @xmath42 here , @xmath0 is the two - dimensional carrier density , @xmath43 is the density of states in 2d , @xmath44 and @xmath45 are the spin and valley degeneracy , respectively , @xmath46 is the carrier energy , @xmath47 } \right\}^{-1}$ ] is the equilibrium fermi - dirac distribution function , and @xmath48 is the chemical potential . for a degenerate electron gas , only scattering within a shell of width @xmath49 around the fermi level is relevant and @xmath50 applies . in valley - degenerate semiconductors scattering in inequivalent valleys is not necessarily identical . in such cases the boltzmann equation must be solved explicitly in all inequivalent valleys . in the absence of intervalley scattering this amounts to replacing the relaxation time in with a valley - averaged relaxation time : @xmath51 , where @xmath52 denotes the valley index , @xmath53 is the number of inequivalent valleys , and @xmath54 denotes the individual valley relaxation times . for acoustic phonon scattering , which to a good approximation can be treated as a quasielastic scattering process , the relaxation time for the individual acoustic phonons is given by @xcite @xmath55 where @xmath56 denotes the branch index ( @xmath57ta , la ) , @xmath58 is the scattering angle , and @xmath59 is understood . the transition matrix element is given by @xmath60 , \end{aligned}\ ] ] where @xmath61 , @xmath62 is the el - ph coupling , @xmath63 is the wave vector and temperature dependent static dielectric function of the 2deg , and @xmath64 is the acoustic phonon energy . the phonons are assumed to be in equilibrium and populated according to the bose - einstein distribution function @xmath65 . screening of the el - ph interaction by the carriers themselves is accounted for by the dielectric function @xmath66 . as we here show , the presence of both normal and umklapp processes in the acoustic deformation potential interaction requires a microscopic description of carrier screening . the consequence of this is a central result of this work , and will be discussed in further detail in sec . [ sec : screening ] . in the present work the expression for the relaxation time in eq . is evaluated numerically assuming quasielastic scattering ; i.e. , the phonon energies are omitted in the @xmath67 functions of eq . ( implying @xmath68 ) but included in the fermi function @xmath69 of eq . . this is particularly important in the bg regime where the phonon energy becomes comparable to the thermal smearing at the fermi level ; i.e. , @xmath70 . in the following , we present first - principles calculations of the el - ph interaction in 2d mos@xmath1 obtained with the density - functional based method outlined in ref . and implemented in the gpaw electronic structure package @xcite . as a complement to our first - principles calculations , we calculate in app . [ app : continuum ] the acoustic el - ph interaction in 2d materials using an elastic continuum model . the interaction with the acoustic phonons can be written in the general form @xmath71 where @xmath72 is the area of the sample , @xmath73 is the mass density , @xmath74 is the matrix element between the bloch states with wave vectors @xmath75 and @xmath76 , and @xmath77 is the change in the crystal potential due to a phonon with wave vector @xmath78 and branch index @xmath56 . the couplings in the @xmath33 valleys are related through time - reversal symmetry as @xmath79 . as the hexagonal lattice of two - dimensional mos@xmath1 lacks a center of symmetry , charge carriers in monolayer mos@xmath1 interact with acoustic phonons through both the deformation potential and the piezoelectric interaction . the coupling matrix element therefore has contributions from both coupling mechanisms , i.e. @xmath80 the two coupling mechanisms are often assumed to be out of phase , i.e. one is real and the other imaginary @xcite ( see also app . [ app : continuum ] ) . this implies that piezoelectric and deformation potential interactions do not interfere in lowest - order perturbation theory , i.e. @xmath81 , and can therefore be treated as separate scattering mechanisms . -valley of the conduction band in monolayer mos@xmath1 . the contour plots show the absolute value of the calculated coupling matrix elements @xmath82 at @xmath83 for the ta ( left ) and la ( right ) phonons as a function of the two - dimensional phonon wave vector @xmath78 . the absolute value of the relative phase @xmath84 between the two interactions ( see eq . ) is shown in the bottom plots.,title="fig : " ] -valley of the conduction band in monolayer mos@xmath1 . the contour plots show the absolute value of the calculated coupling matrix elements @xmath82 at @xmath83 for the ta ( left ) and la ( right ) phonons as a function of the two - dimensional phonon wave vector @xmath78 . the absolute value of the relative phase @xmath84 between the two interactions ( see eq . ) is shown in the bottom plots.,title="fig : " ] -valley of the conduction band in monolayer mos@xmath1 . the contour plots show the absolute value of the calculated coupling matrix elements @xmath82 at @xmath83 for the ta ( left ) and la ( right ) phonons as a function of the two - dimensional phonon wave vector @xmath78 . the absolute value of the relative phase @xmath84 between the two interactions ( see eq . ) is shown in the bottom plots.,title="fig : " ] -valley of the conduction band in monolayer mos@xmath1 . the contour plots show the absolute value of the calculated coupling matrix elements @xmath82 at @xmath83 for the ta ( left ) and la ( right ) phonons as a function of the two - dimensional phonon wave vector @xmath78 . the absolute value of the relative phase @xmath84 between the two interactions ( see eq . ) is shown in the bottom plots.,title="fig : " ] -valley of the conduction band in monolayer mos@xmath1 . the contour plots show the absolute value of the calculated coupling matrix elements @xmath82 at @xmath83 for the ta ( left ) and la ( right ) phonons as a function of the two - dimensional phonon wave vector @xmath78 . the absolute value of the relative phase @xmath84 between the two interactions ( see eq . ) is shown in the bottom plots.,title="fig : " ] -valley of the conduction band in monolayer mos@xmath1 . the contour plots show the absolute value of the calculated coupling matrix elements @xmath82 at @xmath83 for the ta ( left ) and la ( right ) phonons as a function of the two - dimensional phonon wave vector @xmath78 . the absolute value of the relative phase @xmath84 between the two interactions ( see eq . ) is shown in the bottom plots.,title="fig : " ] in fig . [ fig : m ] we show our first - principles results for the deformation potential and piezoelectric interactions with the ta and la phonons in 2d mos@xmath1 for @xmath85 @xcite . the two coupling mechanisms have been obtained from the total coupling matrix element in eq . using the real - space partitioning scheme outlined in app . [ app : partition ] . the scheme is based on the observation that the deformation potential is short range while the piezoelectric interaction is long range , and can therefore be separated in real space . while the deformation potential couplings have the three fold rotational symmetry of the conduction band in the vicinity of the @xmath33 points , the six - fold rotational symmetry of the piezoelectric couplings stems from the hexagonal crystal lattice . the interference between the deformation potential and piezoelectric interaction can be inferred from the relative phase @xmath84 between their complex - valued matrix elements , @xmath86 .\ ] ] the absolute value of the relative phase is shown in the bottom row of fig . [ fig : m ] . at long wavelengths , we find that the above - mentioned out - of - phase property ( @xmath87 ) holds for the la phonon only . for the ta phonon the two coupling mechanisms interfere ( @xmath88 ) and must hence be considered together . deviations from this behavior occur at short wavelengths where the strictly transverse and longitudinal character of the ta and la phonons vanishes . however , for long - wavelength acoustic phonon scattering , the interaction with the ta and la phonons is given by fully interfering and noninterfering couplings , respectively . in order to gain further understanding of the deformation potential interaction , we here quantify the contributions from normal and umklapp processes . formally , the two are associated with terms in the fourier expansion of the short - range phonon - induced change in the crystal potential , @xmath89 with @xmath90 and @xmath91 for normal and umklapp processes , respectively ( @xmath92 is a reciprocal lattice vector , see also app . [ app : normalumklapp ] ) . since the coupling to the ta phonon vanishes when umklapp processes are neglected altogether @xcite , they are essential for a correct description of the acoustic el - ph interaction . and their contributions from normal and umklapp processes given by eq . ( full lines ) . the dashed lines show the analytic deformation potential interaction in eq . .,title="fig : " ] and their contributions from normal and umklapp processes given by eq . ( full lines ) . the dashed lines show the analytic deformation potential interaction in eq . .,title="fig : " ] in fig . [ fig : m_df_normalvsumklapp ] we show the normal and umklapp contributions to the acoustic deformation potential interactions obtained with the fourier filtering method outlined in app . [ app : normalumklapp ] . the plots show the absolute value of the @xmath78-direction average of the matrix element @xmath93 where @xmath94 denotes the phonon - induced potential with normal ( @xmath95 ) and umklapp ( @xmath96 ) processes included , respectively . due to the complex - valued matrix elements , the absolute values of the normal and umklapp contributions do not add up to the total matrix element . in accordance with the statement below eq . , we find that in the long - wavelength limit the deformation potential interactions for the ta and la phonons are completely dominated by umklapp and normal processes , respectively . at shorter wavelengths both processes contribute . the separation of the deformation potential interaction into contributions from normal and umklapp processes is not only of technical character . as we show below in sec . [ sec : screening ] , it has important consequences for the screening of the deformation potential interaction . in the following , the analytic expressions for the deformation potential and piezoelectric couplings obtained in app . [ app : continuum ] are introduced and the coupling strengths are determined from the first - principles el - ph couplings . the deformation potential originates from the local changes of the crystal potential caused by the atomic displacements due to an acoustic phonon . the determination of the interaction strength thus requires a microscopic calculation , such as the first - principles approach used in this work . the interaction with acoustic phonons via the deformation potential interaction is most commonly assumed to be isotropic and linear in the phonon wave vector , i.e. @xmath97 where @xmath98 is the acoustic deformation potential @xcite . since the _ true _ deformation potential couplings in fig . [ fig : m ] are anisotropic and show a more complex @xmath99-dependence at shorter wavelengths , the deformation potential of eq . must be regarded as an _ effective _ coupling parameter . in valley - degenerate semiconductors where the couplings in the different valleys are related through time - reversal symmetry , the _ effective _ deformation potential furthermore accounts for the variation in the angular dependence of the coupling between inequivalent valleys . we have here recalculated the acoustic deformation potentials from ref . in order to avoid undesired contributions from the piezoelectric interaction that potentially were included there . the new deformation potentials are given in tab . [ tab : parameters ] and the couplings are shown in fig . [ fig : m_df_normalvsumklapp ] ( dashed lines ) together with the angular average of the first - principles couplings . for deformation potential scattering above the bg temperature where the equipartition approximation @xmath100 applies and with screening neglected , the relaxation time in eq . becomes independent of the carrier energy and is given by @xcite @xmath101 this results in a @xmath3 temperature dependence of the 2deg mobility characteristic of acoustic deformation potential scattering in the high - temperature regime . piezoelectric coupling to acoustic phonons occurs in crystals lacking an inversion center and originates from the macroscopic polarization that accompanies an applied strain @xmath102 . the strength of the interaction is given by the piezoelectric tensor @xmath103 here given in voigt notation . .material parameters for monolayer mos@xmath1 used in the present work . apart from the acoustic deformation potentials and the piezoelectric constant all parameters have been adopted from ref . . [ cols="<,^,^",options="header " , ] [ tab : parameters ] in app . [ app : piezo ] we obtain the piezoelectric interaction in a 2d hexagonal lattice using continuum theory . we find that the piezoelectric interaction is given by @xmath104 where @xmath105 ( units of c / m ) is the only independent component of the piezoelectric tensor of the 2d hexagonal lattice @xcite , @xmath106 is the vacuum permeability , erfc is the complementary error function , @xmath107 is an effective width of the electronic wave functions , and @xmath108 is an anisotropy factor that accounts for the angular dependence of the piezoelectric interaction . it is given by @xmath109 for the ta and la phonon , respectively , and results in a highly anisotropic piezoelectric interaction . the first - principles results for the piezoelectric interaction in fig . [ fig : m ] are in overall good agreement with the analytic expression in eq . ( see also figs . [ fig : m_analytic ] and [ fig : m_fit ] ) @xcite . from a fit to the first - principles results , the piezoelectric constant of 2d mos@xmath1 is estimated to be @xmath110 c / m ( @xmath130.01 @xmath111/bohr ) . this is an order of magnitude smaller than a recently reported value ( @xmath112 ) obtained with a berry s phase approach @xcite . we are at present , however , not able to clarify the origin of this disagreement . the @xmath99-dependence of the 2d piezoelectric interaction in eq . is qualitatively different from the one in 3d bulk systems where @xmath113 @xcite . in the long - wavelength limit where @xmath114 , the 2d piezoelectric interaction acquires a linear @xmath99-dependence @xmath115 . hence , the deformation potential and piezoelectric interaction in a 2d lattice behave qualitatively the same in the long - wavelength limit . assuming that the linear long - wavelength behavior holds , the high - temperature relaxation time for piezoelectric scattering is given by eq . with the replacement @xmath116 where the factor @xmath117 stems from the angular mean of the piezoelectric interaction @xcite . this corresponds to an effective isotropic piezoelectric coupling with @xmath118 in eq . . the relative strength of the deformation potential and piezoelectric interactions is thus governed by the ratio @xmath119 of the prefactors in eqs . and . since this is of the order of unity with the parameters for 2d mos@xmath1 listed in table [ tab : parameters ] , both coupling mechanisms must be taken into account . the first - principles el - ph interactions presented above have been obtained for the neutral material and therefore do not take into account screening by the 2deg in extrinsic 2d mos@xmath1 . in the following we apply a microscopic theory for carrier screening and show that the normal and umklapp contributions to the deformation potential interaction are screened differently . formally , the screened el - ph interaction can be obtained by replacing the phonon - induced potential @xmath77 in the matrix element @xmath120 of eq . with its screened counterpart @xmath121 where @xmath122 is the ( static ) microscopic dielectric function of the 2deg . inserting , this can be recast in fourier space in terms of the @xmath78-dependent dielectric matrix @xmath123 @xcite as @xmath124 where @xmath92 and @xmath125 are reciprocal lattice vectors and the second equality holds in the diagonal approximation @xmath126 . to a good approximation , the screened phonon - induced potential thus follows by dividing the individual fourier components in eq . by the components @xmath127 of the diagonal dielectric matrix . the latter is related to the 2deg polarizability @xmath128 through the expression @xcite @xmath129 which is similar to the standard long - wavelength expression for the dielectric function in eq . , however , with the important difference that the denominator in the second term of eq . contains a factor @xmath130 instead of a factor @xmath99 . for intravalley scattering where @xmath131 , this implies that @xmath132 behaves differently at long ( @xmath90 ) and short ( @xmath133 ) wavelengths ; while it diverges as @xmath134 in the @xmath135 limit in the former case , it approaches a finite value in the latter . as an immediate consequence , normal and umklapp components of the el - ph interaction are renormalized differently with a significantly stronger screening of the former . while a correct description of the screened el - ph interaction can only be obtained from eq . , the calculation of the microscopic dielectric function from first principles is , however , beyond the scope of the present study . instead , we adopt the following _ ad hoc _ approach to carrier screening . in order to account for the qualitative difference between the screening of normal and umklapp processes , the dielectric function of the 2deg in eq . is approximated as follows . for the long - wavelength component ( @xmath90 ) of the dielectric function , well - established approximations exist in the literature @xcite . we here apply the finite - temperature rpa theory due to maldague @xcite , @xmath136 where @xmath48 is the chemical potential and the static polarizability at finite temperatures is obtained as @xmath137 here , @xmath138 is the zero - temperature rpa polarizability given by the density of states @xmath139 for @xmath140 ( see also app . [ app : screening ] ) @xcite . the integral in eq . is evaluated numerically using the approach of ref . . we have here neglected the form factor in the polarizability arising from the finite thickness of the electronic bloch functions . first - principles calculations of the rpa dielectric function ( see e.g. refs . ) could be helpful in clarifying to which extent this leads to an overestimation of the screening strength . for the short - wavelength part ( @xmath133 ) of the dielectric function , we introduce an _ effective _ dielectric constant @xmath141 which acts as a simple scaling parameter for the @xmath133 components of the potential in eq . . from the relations @xmath142 , where @xmath143 is the thomas - fermi screening wave vector ( see app . [ app : screening ] ) and the latter inequality follows from the expression for the microscopic polarizability @xcite , we observe that the short - wavelength screening efficiency is relatively weak ; i.e. @xmath144 . we can hence to a good approximation set @xmath145 thus leaving the umklapp contribution to the deformation potential interaction unscreened . with our findings above , the screened matrix element for the acoustic el - ph interaction can be written as @xmath146 here , the normal contribution to the deformation potential interaction and the long - range piezoelectric interaction are screened by the long - wavelength dielectric function in eq . , while the umklapp contribution to the deformation potential interaction is screened by @xmath141 . it should be noted that this differs from conventional descriptions of the screened acoustic el - ph interaction where the deformation potential interaction is either screened with a long - wavelength dielectric function or left unscreened ( see e.g. refs . ) . as the deformation potential interactions for the ta and la phonons are largely dominated by umklapp and normal processes , they can to a good approximation be screened by @xmath141 and @xmath147 in eq . , respectively , thus leaving the deformation potential interaction for the ta phonon unscreened . taking into account the interference between the deformation potential and piezoelectric interaction , we can hence approximate the coupling matrix elements for the ta and la phonon as @xmath148 and @xmath149 respectively , where @xmath82 are given by eqs . and . in the following we provide a qualitative estimate of the efficiency of long - wavelength carrier screening given by the dielectric function in eq . . the screening strength is in this case governed by the dimensionless parameter @xmath150 where @xmath151 is the finite - temperature thomas - fermi wave vector and @xmath152 denotes a typical scattering wave vector . in the case of acoustic phonon scattering @xmath152 is given by @xmath153 in the degenerate ( @xmath154 ) and nondegenerate ( @xmath155 ) regime , respectively , and where @xmath156 is the fermi temperature . here @xmath157 is a typical scattering wave vector in the bg regime where the accessible phase space is restricted by the availability of thermally excited phonons . above the bg temperature where scattering on the full fermi surface is possible , @xmath18 becomes a typical scattering wave vector . in the case of a nondegenerate 2deg where the average carrier energy is @xmath158 , @xmath159 is a typical wave vector . in eq . for acoustic phonon ( full lines ) and charged impurity ( dashed lines ) scattering in 2d mos@xmath1 . ] in the low - temperature limit , the screening parameter for acoustic phonon scattering is given by @xmath160 which is independent of the carrier density . the @xmath161 divergence of the low - temperature screening parameter implies that scattering of acoustic phonons via normal process deformation potential and piezoelectric interaction is strongly suppressed for a degenerate 2deg in the bg regime . it is interesting to compare this with the screening parameter for charged impurity scattering . in this case , scattering on the full fermi surface is possible at low temperatures . hence , @xmath162 in the degenerate / nondegenerate regime and the low - temperature limit of the screening parameter becomes @xmath163 which is independent of temperature and decreases with the carrier density . this results in a less efficient screening of impurity scattering compared to acoustic phonon scattering at high densities and low temperatures . the full temperature dependence of the screening parameter for scattering of acoustic phonons ( full lines ) and impurities ( dashed lines ) is shown in fig . [ fig : q_s ] for different carrier densities and @xmath164 m / s representative of the acoustic sound velocities in 2d mos@xmath1 . in the low temperature regime , the limits in eqs . and are approached . due to the density and temperature dependence of the debye - hckel wave vector , @xmath165 , the screening strength becomes independent of the scattering mechanism and increases ( decreases ) with the carrier density ( temperature ) in the nondegenerate high - temperature regime . the overall large values of the screening parameter ( @xmath166 ) in fig . [ fig : q_s ] follow from a large effective mass and valley degeneracy . carrier screening in monolayer mos@xmath1 and other 2d transition metal dichalcogenides is therefore inherently strong and the screening strength exceeds that of e.g. si and gaas based 2degs @xcite . for scattering of acoustic phonons this has the important consequence that normal process deformation potential and piezoelectric interaction is strongly reduced already at relative low carrier densities @xmath167 @xmath10 . in the following we use the boltzmann equation approach outlined in section [ sec : ii ] to study the temperature and density dependence of the acoustic phonon limited mobility in 2d mos@xmath1 for temperatures @xmath168 k and high carrier densities @xmath169 to @xmath170 @xmath10 . the mobility limited by scattering of acoustic phonons follows a generic temperature dependence @xmath171 where the exponent @xmath172 depends on temperature , carrier density , and the dominating scattering mechanism . the same holds for the resistivity @xmath173 with a change in the sign of the exponent . in order to establish the relative strength of deformation potential and piezoelectric scattering in 2d mos@xmath1 , we start by considering the scattering rate given by the expression for the inverse relaxation time in eq . with the replacement @xmath174 . figure [ fig : tau_vs_e ] shows energy dependence of the individual scattering rates due to deformation potential ( dashed lines ) and piezoelectric ( full lines ) scattering for carrier densities @xmath32 @xmath10 ( left ) and @xmath175 @xmath10 ( right ) and temperatures @xmath176 k and @xmath177 k. with the fermi temperature given by @xmath178 k ( @xmath179 mev ) , the two plots correspond to a nondegenerate and degenerate carrier distribution , respectively . the bg temperatures for the ta and la phonons are in the two plots : ( left ) @xmath180 k , and ( right ) @xmath1336 k and @xmath1357 k , respectively . at temperatures @xmath176 k and @xmath177 k and carrier densities @xmath32 @xmath10 ( left ) and @xmath175 @xmath10 ( right ) . this corresponds to fermi energies of @xmath181 mev and @xmath182 mev , respectively . in the right plot , the bg temperature is @xmath183 @xmath184 k for the ta ( la ) phonon . at @xmath176 k , the dip in the scattering rate that appears at the fermi energy is a clear fingerprint of bloch - grneisen physics.,title="fig : " ] at temperatures @xmath176 k and @xmath177 k and carrier densities @xmath32 @xmath10 ( left ) and @xmath175 @xmath10 ( right ) . this corresponds to fermi energies of @xmath181 mev and @xmath182 mev , respectively . in the right plot , the bg temperature is @xmath183 @xmath184 k for the ta ( la ) phonon . at @xmath176 k , the dip in the scattering rate that appears at the fermi energy is a clear fingerprint of bloch - grneisen physics.,title="fig : " ] in the nondegenerate regime shown in the left plot of fig . [ fig : tau_vs_e ] , carrier screening is weak implying that deformation potential and piezoelectric scattering are of the same order of magnitude . at low energies , however , deformation potential scattering of the la phonon and piezoelectric scattering are strongly screened and unscreened deformation potential scattering of the ta phonon dominates the scattering rate . the saturation of the piezoelectric scattering rate at high energies is a consequence of the nonmonotonic @xmath99-dependence of the matrix element in eq . ( see also fig . [ fig : m_fit ] ) . in the degenerate high - density regime shown in the right plot of fig . [ fig : tau_vs_e ] , carrier screening is so strong that the piezoelectric scattering rate is diminished by almost three orders of magnitude relative to the low - density scattering rate ( note the scaling factor in the legend of the right plot in fig . [ fig : tau_vs_e ] ) . in this regime , unscreened deformation potential scattering of the ta phonon therefore completely dominates . the dip in the scattering rate that develops at the fermi level with decreasing temperature is a signature of transport in the bg regime . in this temperature regime the freezing out of short - wavelength acoustic phonons and the sharpening of the fermi surface strongly limit the phase space available for acoustic phonon scattering resulting in a strong suppression of scattering at the fermi level . next , we consider the temperature and density dependence of the mobility . due to the strong anisotropy of the piezoelectric interaction , the mobility in 2d mos@xmath1 is slightly anisotropic . along the different high - symmetry directions of the hexagonal lattice we find that the variation in the mobility is less than @xmath1310% . in the following we shall focus on a single direction of the applied electric field @xcite . the temperature dependence of the mobility is shown in fig . [ fig : mobility_vs_t ] for carrier densities @xmath169 to @xmath185 @xmath10 corresponding to bg temperatures up to @xmath1362 k ( @xmath1399 k ) for the ta ( la ) phonon . both the mobility ( upper ) and the exponent @xmath186 ( lower ) of its power - law dependence @xmath187 are shown . at the lowest densities the characteristic temperatures @xmath16 and @xmath156 are comparable while @xmath188 for @xmath189 @xmath10 . as a consequence , the crossover to the high - mobility bg regime at @xmath190 , marked by the dots in the lower plot , appears clearly for all carrier densities . in the high - temperature regime @xmath191 , the mobility shows an approximate linear temperature dependence with @xmath192 . this is in good agreement with the individual high - temperature limits for unscreened deformation potential and piezoelectric scattering which we find to be @xmath193 and @xmath194 , respectively . at the lowest carrier densities , the larger value of @xmath195 appearing at @xmath196 originates from the temperature dependence of the dielectric function for a nondegenerate carrier distribution . in @xmath197 for the same set of carrier densities . the dots mark the bg temperatures of the ta ( @xmath198 ) and la ( @xmath199 ) phonons , respectively.,title="fig : " ] in @xmath197 for the same set of carrier densities . the dots mark the bg temperatures of the ta ( @xmath198 ) and la ( @xmath199 ) phonons , respectively.,title="fig : " ] in the low - temperature bg regime @xmath200 , a stronger temperature dependence with @xmath201 appears and the mobility approaches a @xmath202 limiting behavior at @xmath203 . numerically , we find that the @xmath204 limiting behavior is characteristic of unscreened deformation potential scattering . screened deformation potential and piezoelectric scattering share the same @xmath205 limit due the identical long - wavelength limits of their respective couplings in eqs . and . the mobility at @xmath203 is thus completely dominated by unscreened deformation potential scattering of the ta phonon . our findings for the low - temperature limits of the mobility due to scattering of 2d phonons differ from the usual low - temperature limits of the 2deg mobility with scattering of bulk 3d phonons in heterostructures . in this case , the limits are given by @xmath206 for unscreened deformation potential scattering and @xmath207 and @xmath206 for screened deformation potential and piezoelectric scattering , respectively @xcite . as the low - temperature limits of the mobility are only realized deep inside the bg regime @xmath203 , they may , however , be difficult to observe experimentally . , 20 , 50 k. the dots mark the quantum - classical crossover from a nondegenerate to a degenerate carrier distribution at @xmath208 . ] in fig . [ fig : mobility_vs_n ] we show the calculated density dependence of the mobility for temperatures @xmath209 , 20 , 50 k. in the nondegenerate low - density regime , the density dependence of debye - hckel screening results in a mobility that increases with the density . for the lowest temperature where carrier screening becomes strong , unscreened deformation potential scattering of the ta phonon dominates the mobility resulting in a weaker density dependence . the same holds for densities in the vicinity of the quantum - classical crossover at @xmath210 ( marked by dots in fig . [ fig : mobility_vs_n ] ) where the mobility shows almost no density dependence . in the degenerate high - density regime , the mobility approaches a @xmath211 behavior at low temperatures . the strong density dependence of the mobility in this regime can be ascribed to two factors : ( i ) a fermi velocity that increases with the carrier density as @xmath212 , and ( ii ) transport in the bloch - grneisen regime where the scattering rate for fixed temperature @xmath213 decreases with the density . the latter is a consequence of a reduction in the fraction of the fermi surface that is probed by acoustic phonon scattering when the density ( fermi wave vector ) increases . we end by briefly commenting on the applicability of the expression @xmath214 often used to fit experimental mobilities @xcite in the linear high - temperature regime . here , @xmath215 is the residual mobility due to e.g. impurity scattering and @xmath216 is the density - dependent slope of the linear temperature dependence . in fig . [ fig : mobility_vs_t ] it is seen to apply at @xmath217 except for the lowest densities where the inverse mobility becomes slightly nonlinear . from the density dependence of the mobility in fig . [ fig : mobility_vs_n ] we conclude that the temperature coefficient @xmath216 is a monotonically decreasing function of the carrier density . in this work we have combined analytic and first - principles calculations of the el - ph interaction with an semianalytic solution of the boltzmann equation to study the temperature and density dependence of the acoustic phonon limited mobility in 2d @xmath0-type mos@xmath1 . the acoustic deformation potentials for the ta and la phonons and the piezoelectric constant in 2d mos@xmath1 were extracted from the first - principles el - ph interaction and from a microscopic description of carrier screening it was shown that the umklapp contribution to the deformation potential interaction is not affected by screening . due to strong screening of deformation potential scattering of the la phonon and piezoelectric scattering of both the ta and la phonon , the mobility was found to be dominated by unscreened deformation potential scattering of the ta phonon at high densities . at low carrier densities @xmath169@xmath218 @xmath10 deformation potential and piezoelectric scattering were found to be comparable . for @xmath8 k and moderate to high carrier densities @xmath219 @xmath10 , intrinsic mobilities in excess of @xmath220 @xmath6 v@xmath7 s@xmath7 were predicted . at temperatures @xmath221 k , the acoustic phonon limited mobility does not exceed @xmath222 @xmath6 v@xmath7 s@xmath7 . in the low - temperature bg regime @xmath223 , the mobility acquires a @xmath202 dependence characteristic of unscreened deformation potential scattering of 2d phonons . the mobility was furthermore found to increase monotonically with the carrier density . similar conclusions can be expected to hold for monolayers of other transition metal dichalcogenides which have similar atomic and electronic structures . apart from our findings for the mobility , we here list a few other key results of our work : ( i ) in the long - wavelength limit the acoustic deformation potential interactions for the ta and la phonons were found to be completely dominated by umklapp and normal processes , respectively ; ( ii ) from a microscopic treatment of carrier screening we showed that normal and umklapp processes in the el - ph interaction are screened differently , and that the deformation potential interaction with the ta phonon to a good approximation can be left unscreened ; ( iii ) our further developments for first - principles calculations of the el - ph interaction included in app . [ app : firstprinciples ] . as a final remark , we note that our conclusion regarding the screening of the acoustic el - ph interaction can be verified experimentally . for a purely long - wavelength treatment of carrier screening , we find that the theoretically predicted mobility is substantially higher and shows a much richer temperature dependence with higher values of @xmath172 and no linear temperature dependence ( see also ref . ) . experimental low - temperature mobility data on high - mobility monolayer mos@xmath1 samples matching the predictions of this work will hence provide strong support for our findings . as our findings for the acoustic el - ph interaction must be expected to be relevant in other semiconductors , we believe that the present study is of high importance for an improved understanding of the acoustic el - ph interaction and phonon limited mobilities in semiconductor based 2degs . we thank o. hod and a. konar for fruitful discussions . the center for nanostructured graphene ( cng ) is sponsored by the danish national research foundation , project no . dnrf58 . this work was supported by the villum kann rasmussen foundation . in the thomas - fermi ( tf ) approach to screening , the finite - temperature dielectric function of a 2deg is in the long - wavelength limit given by @xcite @xmath224 where @xmath225\ ] ] is the temperature - dependent screening wave vector , @xmath226 the zero - temperature tf wave vector , @xmath227 the constant density of states in 2d , and @xmath228 the fermi level . this reproduces the rpa and debye - hckel results @xmath229 for the screening wave vector in the low and high - temperature limits , respectively . in a degenerate 2deg , thomas - fermi screening overestimates the screening strength at @xmath230 where 2deg screening becomes less efficient . rpa corrections to the dielectric function are required to cure this problem @xcite . however , for quasi - elastic scattering with @xmath231 , tf theory provides a good approximation to the dielectric function . in this appendix , we calculate the acoustic el - ph interaction in 2d materials using continuum theory . for this purpose , the electronic states are described by plane - wave solutions @xmath232 where @xmath72 is the area of the sample , @xmath75 is the two - dimensional electronic wave vector , @xmath233 , and @xmath234 is the normalized envelope of the electronic wave functions accounting for its confinement in the direction perpendicular to the material layer . the hamiltonian for the el - ph interaction takes the usual form @xmath235 where @xmath236 is the two - dimensional phonon wave vector , @xmath56 is the acoustic branch index and @xmath237 is the coupling constant . in the following , the matrix element @xmath120 is obtained by applying an elastic continuum model for the acoustic phonons . in a lattice without an inversion center , the acoustic el - ph is a sum of deformation potential ( dp ) and piezoelectric ( pe ) interactions . the unscreened el - ph interaction which couples to the carrier density is in real space given by @xcite @xmath238 where @xmath239 is the displacement field due to the acoustic phonons , @xmath240 is the deformation potential , and @xmath84 is the electrostatic potential from the piezoelectric polarization of the lattice . considering an isolated 2d material sheet , the in - plane acoustic phonons can be described by the quantized two - dimensional displacement field @xmath241 where @xmath236 is the two - dimensional phonon wave vector , @xmath242 is a unit vector describing the polarization of the acoustic branch @xmath56 , @xmath243 is the @xmath244-profile of the displacement field in the direction perpendicular to the material sheet , and @xmath245 is the vibrational normal coordinate . in the long - wavelength limit the polarization vectors for the ta and la phonons are perpendicular ( @xmath246 ) and parallel ( @xmath247 ) to the phonon wave vector @xmath78 , respectively , and can to a good approximation be assumed independent on @xmath99 @xcite . with the displacement field written in the form in eq . , the hamiltonian can be recast as a sum over terms from the individual phonons , @xmath248 e^{i { \mathbf{q}}\cdot { \mathbf{r}}_\parallel } , \end{aligned}\ ] ] where @xmath249 are to be determined below . the el - ph coupling constant is given by the matrix element @xmath250 in the following , the envelope function @xmath251 is assumed independent of the electronic wave vector @xmath75 . taking the divergence of the displacement field in eq . , the deformation potential interaction is found to be @xmath252 because of the dot product between the phonon wave vector and the polarization vector , the interaction with the ta phonon vanishes . the coupling matrix element for the la phonon is given by @xmath253 where the result of the @xmath244-integral in eq . has been absorbed in the deformation potential constant . the fact the ta phonon does not couple , illustrates the limitation of the often assumed form for deformation potential interaction in eq . and underlines the importance of more involved descriptions @xcite . piezoelectric interaction with acoustic phonons appears in lattices which lack a center of symmetry . in this case , the displacement field @xmath254 associated with the acoustic phonons leads to a polarization @xmath255 of the lattice given by @xcite @xmath256 where @xmath257 is the strain tensor and @xmath111 is the tensor of piezoelectric moduli having symmetry @xmath258 . in 2d materials the piezoelectric coupling has units of c / m ( c / m@xmath259 in 3d)the displacement field can be thought to have a normalized @xmath244-profile with units of m@xmath7 . for a 2d hexagonal lattice with a basis there is only one independent piezoelectric component @xmath105 ( voigt notation ) which is related to the other nonzero components as @xcite @xmath260 here , the primitive lattice vectors of the hexagonal lattice have been chosen as @xmath261 where @xmath262 is the lattice constant . expanding the strain tensor as in eq . , its ( @xmath78,@xmath56)-components follow directly from eq . as @xmath263 and the associated polarization in eq . is given by @xmath264 the potential @xmath84 resulting from the piezoelectric polarization field is given by poisson s equation @xmath265 where @xmath266 is the polarization charge . since we are considering an isolated material sheet , the only boundary condition that applies is @xmath267 for @xmath268 . fourier transforming in all three directions , we find @xmath269 where @xmath270 is the fourier variable in the direction perpendicular to the plane of the layer . the fourier components of the branch - resolved piezoelectric polarization charge are given by @xmath271 where the angular dependencies have been collected in the anisotropy factor @xmath272 . it is given by @xmath273 for the ta and la phonon , respectively . the @xmath244-dependence of the piezoelectric potential is given by the inverse fourier transform of with respect to @xmath270 which yields @xmath274 where in the last equality , a @xmath67-function @xmath244-profile with @xmath275 has been assumed . for atomically thin materials , this should be a reasonable approximation . in eq . [ eq : m_pe_continuum ] for the ta ( left ) and la ( right ) phonon as a function of the phonon wave vector @xmath78 . the parameters for mos@xmath1 in tab . [ tab : parameters ] have been used.,title="fig : " ] in eq . [ eq : m_pe_continuum ] for the ta ( left ) and la ( right ) phonon as a function of the phonon wave vector @xmath78 . the parameters for mos@xmath1 in tab . [ tab : parameters ] have been used.,title="fig : " ] the piezoelectric el - ph interaction in eq . is now given by @xmath276 , and assuming , for simplicity , a gaussian envelope in eq . , the long - wavelength limit of the piezoelectric coupling matrix element becomes @xcite @xmath277 where @xmath107 is the effective width of the electronic envelope function . the piezoelectric interaction has the general property that @xmath278 @xcite . the absolute value of the piezoelectric interaction is shown in fig . [ fig : m_analytic ] . here , the six fold rotational symmetry stems from the hexagonal crystal lattice and is accounted for by the anisotropy factors in eq . . contrary to the 3d bulk case where the piezoelectric interaction is independent of @xmath99 @xcite , the 2d interaction in eq . depends on the magnitude of the phonon wave vector and acquires a linear @xmath99-dependence for @xmath135 . the latter is a consequence of the 2d crystal lattice which does not support a piezoelectric potential from the acoustic phonons in the long - wavelength limit ( see eq . ) . in this appendix , we present the first - principles method applied in the calculation of the el - ph interaction . in order to support the new developments included in this work , we start by briefly highlighting the most important aspects of the main method presented in full detail in ref . . the coupling matrix element @xmath120 in the el - ph interaction in eq . of the main text involves the change in the crystal potential @xmath77 due to a phonon with wave vector @xmath78 and branch index @xmath56 . under the assumption that the atomic displacements are small , the phonon - induced change in the potential can be constructed as a sum over individual atomic gradients , @xmath279 here , @xmath216 is an atomic index in the primitive unit cell , @xmath280 is the lattice vector of unit cell @xmath281 ( relative to the reference unit cell at @xmath282 ) , @xmath283 is the mass - scaled phonon polarization vector , @xmath284 denotes the gradient with respect to displacements of atom @xmath285 in the @xmath286 directions , @xmath287 is the crystal potential , and @xmath95 is the number of unit cells in the lattice . the matrix element of the phonon - induced potential between the bloch states with wave vectors @xmath75 and @xmath288 are evaluated by expanding the bloch function in an lcao basis . the resulting expression for the matrix element follows by exploiting the periodicity of the crystal lattice and takes the form @xcite @xmath289 where @xmath290 is a composite atomic ( @xmath216 ) and orbital ( @xmath48 ) index , @xmath291 denotes the atomic orbital @xmath48 on atom @xmath216 in the primitive unit cell @xmath292 , @xmath293 are the lcao expansion coefficients , and @xmath294 is the gradient of the crystal potential with respect to atomic displacements in the reference unit cell . the quantity in the last line of eq . is the lcao supercell matrix of the potential gradient . the real - space structure of its matrix elements is illustrated schematically in fig . [ fig : supercell_matrix ] . measured from the position of the atomic site where the gradient of the potential is taken is used to separate out the short- and long - range part of the el - ph interaction . matrix elements involving lcao orbitals located beyond the cutoff as the one shown are defined as long range ( see also eq . ) . ] in first - principles calculations of the el - ph interaction both short range ( deformation potential ) and long range ( piezoelectric and frhlich interaction ) are included in the coupling in eq . . however , due to their different origin it may sometimes be desirable to consider them separately . in the following we outline a real - space partitioning scheme to separate the short - range and long - range contributions to the el - ph interaction . the central quantity of the partitioning scheme is the lcao supercell matrix illustrated schematically in fig . [ fig : supercell_matrix ] . the partitioning scheme consists in splitting up the summations in eq . into two contributions ; ( i ) a short - range part which neglects all matrix elements involving lcao orbitals beyond a chosen real - space cutoff @xmath295 , and ( ii ) a long - range part which includes the remaining matrix elements ; i.e. in general , the real - space cutoff @xmath295 must be chosen small enough that the short - range part does not include contributions from truly long - range effects in the relevant range of phonon wave vectors ; i.e. @xmath298 where @xmath299 is the maximum phonon wave vector of interest . at the same time , the cutoff can not be chosen too small that short - range effects are cut off . as these guidelines do not provide a unique way to choose the real - space cutoff , it should be verified in an actual calculation that the results do not change significantly for different values of the cutoff . obtained with the real - space partitioning scheme for different values of the real - space cutoff @xmath295 . the plots show the absolute value of the matrix elements @xmath82 averaged over the high - symmetry directions of the hexagonal lattice . the analytic couplings in eqs . and are shown with dashed lines for the mos@xmath1 parameters listed in tab . [ tab : parameters].,title="fig : " ] obtained with the real - space partitioning scheme for different values of the real - space cutoff @xmath295 . the plots show the absolute value of the matrix elements @xmath82 averaged over the high - symmetry directions of the hexagonal lattice . the analytic couplings in eqs . and are shown with dashed lines for the mos@xmath1 parameters listed in tab . [ tab : parameters].,title="fig : " ] obtained with the real - space partitioning scheme for different values of the real - space cutoff @xmath295 . the plots show the absolute value of the matrix elements @xmath82 averaged over the high - symmetry directions of the hexagonal lattice . the analytic couplings in eqs . and are shown with dashed lines for the mos@xmath1 parameters listed in tab . [ tab : parameters].,title="fig : " ] obtained with the real - space partitioning scheme for different values of the real - space cutoff @xmath295 . the plots show the absolute value of the matrix elements @xmath82 averaged over the high - symmetry directions of the hexagonal lattice . the analytic couplings in eqs . and are shown with dashed lines for the mos@xmath1 parameters listed in tab . [ tab : parameters].,title="fig : " ] in fig . [ fig : m_fit ] we show the deformation potential ( top row ) and piezoelectric ( bottom row ) interactions for the ta and la phonons in 2d mos@xmath1 obtained with the partitioning scheme for different values of the cutoff @xcite . the dashed lines show the analytic forms for the deformation potential and piezoelectric interactions in eqs . and with the parameters listed in tab . [ tab : parameters ] . while the variation in the deformation potential interaction is relatively insignificant , the piezoelectric interaction is more sensitive to the chosen cutoff . the best agreement between the analytic expression for the piezoelectric interaction in eq . and the first - principles results is obtained for @xmath300 . it should be emphasized that the finite value of the first - principles piezoelectric interaction in the limit @xmath135 in fig . [ fig : m_fit ] is an artifact inherent to a supercell method . the finite real - space range of the el - ph interaction in supercell methods naturally sets a lower limit for the magnitude of the phonon wave vector @xmath301 at which long - range interactions can be obtained reliably . it is given by @xmath302 where @xmath303 is the size of the supercell ( measured as the diameter of a sphere that can be contained within the supercell ) . in the calculations presented here @xmath304 implying that @xmath305 @xmath306 . in order identify the normal and umklapp processes in the el - ph interaction , we start by noticing that the gradients of the potential @xmath307 in eq . are localized functions in real space ( see fig . [ fig : supercell_matrix ] ) , where @xmath308 denotes an atom - specific function . due to the periodicity of the lattice , the gradient in unit cell @xmath281 is related to the gradient in the reference cell through a translation by the lattice vector @xmath280 . we now express the gradients of the potential in terms of their fourier series , @xmath309 where @xmath310 is a three - dimensional fourier variable . it is here important to distinguish between the projections @xmath311 of @xmath312 onto the periodic / nonperiodic directions of the lattice . the sum over the unit cell index @xmath281 in eq . now only contains two exponential factors , @xmath313 where the reciprocal lattice vector @xmath92 and the phonon wave vector @xmath78 by definition have the dimensionality of the lattice . this restricts the projection @xmath314 to values @xmath315 . inserting in eq . , we find for the phonon - induced potential @xmath316 first of all , we note that this allows us write the phonon - induced potential on the bloch form in eq . . secondly , the separation into normal ( @xmath90 ) and umklapp ( @xmath133 ) processes is now formally straight - forward . . the square lattice denotes the reciprocal lattice with the shaded cell indicating the first brillouin zone . the dashed arrows show a pair of reciprocal lattice vectors @xmath317 from the electronic bloch functions that conserve crystal momentum by bringing @xmath315 back onto @xmath78 in the first brillouin zone . ] it is important to note that in first - principles calculations of the el - ph interaction , the term `` umklapp process '' has a more general meaning than the one encountered in conventional textbook discussions of the subject @xcite . in order to illustrate this , it is instructive to evaluate the matrix element @xmath120 in eq . with the electronic bloch functions expanded as @xmath318 , where @xmath319 are the fourier components of the periodic part of the bloch functions . this yields @xmath320 where @xmath317 are the reciprocal lattice vectors from the bloch functions , @xmath321 is a reciprocal lattice vector that shifts @xmath288 to the first brillouin zone in case it falls outside , and the @xmath322 dependence has been integrated out . the @xmath67 function resulting from the integral in the last line can be regarded as a generalized statement of conservation of crystal momentum taking into account the reciprocal lattice vectors from the bloch functions . it differs from the standard textbook definition where @xmath323 and @xmath324 implying that umklapp processes only contribute to the matrix element if @xmath325 ; i.e. @xmath326 @xcite . for the generalized conservation of crystal momentum this is not the case . here , umklapp processes involving all fourier components @xmath92 of the scattering potential contribute regardless of the value of @xmath321 . this is illustrated in fig . [ fig : umklapp ] which shows an umklapp process ( @xmath133 ) for @xmath327 and a @xmath328 pair that conserves crystal momentum . for intravalley scattering where @xmath76 is _ always _ inside the first brillouin zone ( if not , the first brillouin zone can chosen such that this is the case ) and hence @xmath329 , coupling to phonons via umklapp processes takes place through the type of process shown in fig . [ fig : umklapp ] . in practice , the matrix elements of the el - ph interaction are evaluated using eq . . the expression for the phonon - induced potential change in eq . is therefore not directly applicable for the separation of the normal and umklapp processes . instead we note that the sum over unit cell indices @xmath281 in eq . provides an automatic selection of the fourier components in @xmath294 that contribute in eq . . the fourier expansion of @xmath294 can therefore be used directly in the calculation of the matrix element in eq . . writing the fourier expansion as @xmath330 the two terms with @xmath314 lying inside and outside the brillouin zone ( bz ) of the crystal lattice define the normal and umklapp contribution to the el - ph interaction , respectively . numerically , the atomic gradients @xmath294 are represented on a three - dimensional real - space grid in the supercell having length @xmath331 and @xmath332 number of grid points in the direction of the lattice vector @xmath333 . the resulting grid spacing is @xmath334 . the values of the gradients on the grid are denoted @xmath335 . the fourier expansion is obtained using the fast fourier transform ( fft ) , @xmath336 with the corresponding fourier space grid in the direction of the primitive reciprocal lattice vector @xmath337 given by @xmath338 with @xmath339 , @xmath340 and @xmath341 . as the supercell has a real - space grid spacing significantly smaller and a size significantly larger than the lattice constant @xmath262 , i.e. @xmath342 and @xmath343 , respectively , we have that @xmath344 and @xmath345 . the normal and umklapp processes can now be separated by fourier filtering @xmath346 . this is done by zeroing the fourier components at @xmath347 with @xmath314 lying outside or inside the brillouin zone . using a tilde to denote the filtered quantities , we have @xmath348 applying the inverse fft ( ifft ) @xmath349 we obtain the filtered gradients in real space which can be used in the numerical evaluation of the matrix element in eq . . since the gradients of the potential @xmath294 are real valued , the imaginary part of @xmath350 can be discarded .

in order to study the outflows from accretion disks , we solve the set of hydrodynamic equations for accretion disks in the spherical coordinates ( ) to obtain the explicit structure along the direction . using self - similar assumptions in the radial direction , we change the equations to a set of ordinary differential equations ( odes ) about the-coordinate , which are then solved with symmetrical boundary conditions in the equatorial plane , and the velocity field is obtained . the viscosity prescription is applied and an advective factor is used to simplify the energy equation.the results display thinner , quasi - keplerian disks for shakura - sunyaev disks ( ssds ) and thicker , sub - keplerian disks for advection dominated accretion flows ( adafs ) and slim disks , which are consistent with previous popular analytical models . however , an inflow region and an outflow region always exist , except when the viscosity parameter is too large , which supports the results of some recent numerical simulation works . our results indicate that the outflows should be common in various accretion disks and may be stronger in slim disks , where both advection and radiation pressure are dominant . we also present the structure dependence on the input parameters and discuss their physical meanings . the caveats of this work and possible improvements in the future are discussed .

1103.4237

the accretion disk models have been developed much in the past several decades . many disk models have been proposed and some of them are widely adopted in astrophysical studies nowadays , including but not restricted to shakura - sunyaev disk ( ssd ; see shakura & sunyaev 1973 ) , advection dominated accretion flow ( adaf ; see narayan & yi 1994 ; abramowicz et al . 1995 ) , and slim disk ( abramowicz et al . 1988 ) . however , the outflow structure of accretion disks still remains an open problem . the equations that describe the hydrodynamic processes are the navier - stokes equations , which are quite difficult to solve in the case of accretion disks which involve viscosity and radiation . therefore , in most works , some kind of simplification , such as one - zone or polytropic distribution and hydrostatic equilibrium , are usually applied in the vertical direction , and the vertical variation ( @xmath4 dependence in cylindrical coordinates ) of the velocity field is usually neglected . in this way the equations are changed to a set of ordinary differential equations ( odes ) in the radial direction , which can be solved numerically . however , by taking these assumptions , one can not get a clear picture of the vertical structure of accretion flows ; the velocity is always radially inward and no mass will cross the disk surface , displaying no outflow structure . among the exceptions is a work done by narayan & yi in 1995(hereafter ny95 ) , which used self - similar assumptions in the radial direction and solved the structure along the @xmath1 direction in spherical coordinates ( @xmath0 ) . however , in their work they assumed @xmath5 and thus can not get a proper velocity field , and their solutions compose of only pure inflow . they argued that the bernoulli parameter is positive in their solutions so that a bipolar outflow is expected to develop near the vertical axis . blandford & begelman ( 1999 , hereafter bb99 ) relaxed the mass conservation assumption and assumed that the mass inflow rate varies with radius , and obtained solutions with outflow called adiabatic inflow - outflow solutions ( adios ) . their solutions are one - dimensional self - similar solutions that are height - averaged and they also applied the bernoulli parameter to argue the presence of outflow . however , abramowicz et al.(2000 ) pointed out that the positive bernoulli function ( instead of bernoulli parameter , which is defined in convenience of self - similar assumptions ) is not sufficient for outflow ( also see the simulation works done by stone et al . 1999 and yuan & bu 2010 ) . blandford & begelman ( 2004 ) furthered their work and presented some self - similar two - dimensional solutions of radiatively inefficient accretion flows with outflow . they assumed hydrostatic equilibrium in the vertical direction and that convection dominates the heat transport , which may only be applicable in certain cases . xu & chen ( 1997 , hereafter xc97 ) relaxed @xmath5 and obtained two types of solutions with outflow : accretion and ejection solutions . however their solutions require the net accretion rate to be 0 , which is not realistic . xue & wang(2005 , hereafter xw05 ) followed ny95 and solved the disk structure along the @xmath1 direction considering @xmath6 . they arbitrarily set a disk surface , at which @xmath7 , and the sound speed on the surface for their calculation . their solutions display a field of inflow near the equatorial plane with wind blowing out of the upper boundary , however the boundary is set as an input parameter rather than being calculated , and they only investigated several cases of adafs . in 3.2 we will see that according to our calculation , their assumption does nt hold for accretion flows with large @xmath2 value . sadowski et al.(2010 ) abandoned the self - similar assumptions and solved the accretion disk structure in the radial and vertical direction simultaneously . because the navier - stokes equations for accretion disks can not be decoupled intrinsically , they adopted other assumptions , e.g. the disk is not geometrically thick , to derive the equations . as in their work they did nt consider @xmath8 while @xmath9 and @xmath10 were supposed not to vary vertically , they were not able to study outflows . in summary , in the analytical model study , the vertical or @xmath1-direction structure of accretion disks and outflows can still not be dealt with satisfactorily , and many improvements are still needed to be made . on the other hand , observationally there are more and more evidences for outflows in accretion systems , such as sgr a*(marrone et al . 2006 ; xie & yuan 2008 ) , soft x - ray transients(loeb et al . 2001 ) and quasars with blueshifted absorption lines(e.g . , pg1115 + 80 ; chartas et al . 2003 ) . many numerical simulation works have also discovered outflows in their results ( e.g. stone et al . , 1999 ; igumenshchev & abramowicz , 2000 ; okuda et al . 2005 ; ohsuga et al.2005 ; ohsuga & mineshige 2007 ; ohsuga et al . the common existence of outflows in these works inspire us to investigate the vertical structure of accretion flows explicitly , to find solutions which can deal with @xmath6 and positive @xmath9 to get a clear velocity field , and with more reasonable boundary conditions . as a first step , we followed the work done by ny95 and xw05 , using self - similar assumptions in the radial direction and solving the odes along the @xmath1 direction in spherical coordinates ( @xmath0 ) . we used the @xmath2 viscosity prescription and assumed that the @xmath11 component of the viscosity stress tensor is dominant . by neglecting other components of the viscosity stress tensor , the necessary number of boundary conditions is reduced and we will only need the boundary conditions in the equatorial plane , which is obatined by symmetry and is thus quite physical . as we did nt set other arbitrary restrictions to @xmath9 and @xmath6 other than the self - similar assumptions , we can get a velocity field containing positive @xmath9 , to discuss the flow structure with possible outflows . these assumptions are applicable to different kinds of accretion disk models , ranging through ssd , adaf and slim disk , so we also did many calculations with different sets of parameters ( which is very difficult to do in a numerical simulation work as it would be too time - consuming ) , trying to find the flow structure dependence on the parameters , in order to understand the inflow / outflow mechanism more physically . these results can also be helpful to future numerical simulation works when they set the input parameters . it should be noted that there is another branch of researches which investigate accretion flows with an outflow ( usually wind ) plus accretion disk model . in these researches , the configuration of the outflow and accretion disk is usually preset , and the calculations are focused on either the outflow or the accretion disk while the influence of the other part is simplified or parameterized ( e.g. fukue , 1989 ; takahara et al . , 1989 ; kusunose , 1991 ; bb99 ; misra & taam , 2001 ; fukue , 2004 ; xie & yuan , 2008 ) . recently there are also several works done in this way which deal with the outflow and the accretion disk simultaneously ( kawabata & mineshige , 2009 ; dotan & shaviv , 2010 ) . compared with our work , in these studies the accretion disk is usually height - integrated and the configuration of the accretion flow is assumed rather than being calculated , while in our work we solve the full hydrodynamic equations to get the configuration of the accretion flow . our work focuses on studying the general structure of disks , outflows and the physical mechanism behind them , and the results can be complementary to one another . in 2 we present the basic equations and assumptions we used in our calculations . in 3.1 we discuss our numerical methods and present solutions corresponding to typical parameters of ssd , adaf and slim disk . in 3.2 we show the disk structure dependence on different parameters and discuss their physical meanings . 4 we present our summary and discussion . we consider the hydrodynamic equations of an accretion flow in the spherical coordinates ( @xmath0 ) . the flow is assumed to be steady(@xmath12 ) and axisymmetric(@xmath13 ) . thus the continuity equation can be written as @xmath14 we assume that for the accretion flow , only the @xmath11-component of the viscous tensor , @xmath15 , is dominant . and we use the newtonian gravitational potential , @xmath16 . then the equations of motion can be written as(kato et al . 2008 , also see appendix a of xw05 ) @xmath17 @xmath18 @xmath19 in our calculation , we adopt the @xmath2 prescription of viscosity @xmath20 and @xmath21 is the total pressure . the shearing box radiation magnetohydrodynamics ( mhd ) simulations done by hirose et al . ( 2009 ) found that the vertically integrated stress is approximately proportional to the vertically averaged total thermal ( gas plus radiation ) pressure . in our work , however , we also apply this relation locally . following ny95 , we use the advective factor , @xmath22 , i.e. a fraction @xmath3 of the dissipated energy is advected as stored entropy and a fraction @xmath23 is lost due to radiation . in our calculation we assume that @xmath3 is constant in the accretion flow , so the energy equation can be written as @xmath24 where @xmath25 is the internal energy of the material and can be expressed as(kato et al . 2008 ) @xmath26 where @xmath27 is the heat capacity ratio and is considered to be a constant input parameter in our calculation . we adopt the self - similar assumptions in the radial direction , therefore we seek a solution of the form @xmath28 this set of self - similar solutions is similar to that of ny95 and is actually identical to that of xw05 . compared with those works , here we use the total pressure @xmath21 in equation ( 11 ) instead of the sound speed @xmath29 , which is proportional to @xmath30 . as stated in ny95 , the only length scale in the problem is @xmath31 and the only frequency is @xmath32 , thus all velocities must scale with radius as @xmath33 . in ny95 , narayan and yi set @xmath34 , which implies @xmath5 according to the continuity equation , thus they intrinsically set no outflow for the accretion disk . here we relax this parameter @xmath35 following bb99 and xw05 , to allow outflows from the disk . if we substitute equations ( 7)-(11 ) into equations ( 1)-(4 ) , the @xmath31 components can be eliminated , leaving only the dimensionless fuctions @xmath36 , @xmath37 , @xmath38 , @xmath39 , @xmath40 , the variable @xmath1 and some constants which are set as input parameters . the energy equation ( equation ( 5 ) ) is somewhat more complex . as the internal energy depends differently on gas pressure and radiation pressure , we discuss the energy equation more carefully here . first , one may notice that only the total pressure @xmath21 appears in equations ( 1)-(4 ) . this is not hard to understand as the dynamical processes do nt recognize whether the pressure is from gas or radiation . @xmath41 and @xmath42 do affect the energy equation in a different way , and to describe this effect , we define the gas pressure ratio @xmath43 , @xmath44 there is one particular case that when @xmath45 , no matter what the value of @xmath43 is , the solution remains the same , as in this case equation ( 6 ) becomes @xmath46 ; if the accretion flow is radiation pressure dominated ( @xmath47 ) , then equation ( 6 ) also becomes @xmath46 , and the flow structure wo nt depend on @xmath27 ; if the accretion flow is gas pressure dominated , then equation ( 6 ) becomes @xmath48 , and the result will depend on the value of @xmath27 . for a general case , equation ( 6 ) can be written as @xmath49p.\ ] ] physically @xmath43 is a value between 0 and 1 , so @xmath50 is always between @xmath51 and @xmath52 . then we can expect that a general solution should be somewhere between two extreme cases : the gas pressure dominated case and the radiation pressure dominated case . for these two extreme cases , we can calculate the exact solutions of the problem , as equation ( 6 ) will be simplified to forms without @xmath43 . these two cases also provide the typical solutions we emphasize in 3.1 . however we also want to know how the solutions change from the gas pressure dominated case to the radiation pressure dominated case . for the accretion flows in which both gas pressure and radiation pressure are important , we have to assume the gas pressure ratio @xmath43 to be constant to solve the problem . solving the problem with variant @xmath43 is beyond the capability of this self - similar treatment , and we re planning to do that in our next work . as we already mentioned , the radiation pressure dominated case has the same result as that of the gas pressure dominated case with @xmath45 , so the question eventually becomes examining how the solution changes according to @xmath27 for the gas pressure dominated case . we define an equivalent @xmath27 here @xmath53 so that equation ( 6 ) becomes @xmath54 and this @xmath55 represents the equivalent @xmath27 that the accretion flow would have if the flow is treated as gas pressure dominated ( even @xmath56 ) . we can see that if the flow is radiation pressure dominated ( @xmath57 ) , then no matter what the value of @xmath27 is , @xmath55 is always 4/3 . here we treat @xmath27 as a constant , and for any specific case , we can get a constant @xmath55 and all the situations can be treated as a gas pressure dominated flow with @xmath55 as an input parameter . usually @xmath27 is taken as a value between 7/5(the value for diatomic ideal gas ) and 5/3(the value for monatomic ideal gas ) , so that @xmath55 ranges between 4/3 and 5/3 . with equations ( 7)-(11 ) and ( 15 ) , the equations ( 1)-(5 ) can be reduced to a set of five odes : @xmath58=0,\ ] ] @xmath59=0,\ ] ] @xmath60=0,\ ] ] @xmath61=0,\ ] ] @xmath62 - 2\gamma_\mathrm{equ } v_\theta(\theta)\frac{d\rho(\theta)}{d\theta}\right\}=0,\ ] ] with 5 dimensionless fuctions @xmath36 , @xmath37 , @xmath38 , @xmath39 , @xmath40 , the variable @xmath1 and four input parameters ( @xmath2,@xmath3,@xmath55,@xmath35 ) . this set of odes can be numerically solved with proper boundary conditions . we assume the structure of the disk is symmetric to the equatorial plane , and thus we have @xmath63 in which only four conditions are independent . for the last boundary condition we set @xmath64 , which can be normalized by a scale factor if the effective accretion rate at a certain radius is set ( ny95 , xw05 , etc . ) . these boundary conditions are enough for our calculations and we did nt introduce other arbitrary boundary conditions . we obtained numerical solutions of equations ( 16)-(20 ) with different sets of input parameters ( @xmath2,@xmath3,@xmath55,@xmath35 ) . some typical solutions are shown in figures 1 - 4 . the calculation starts from the equatorial plane ( @xmath65 ) towards the vertical axis ( @xmath66 ) . generally the mass density @xmath67 and total pressure @xmath21 will decrease as @xmath1 decreases , and at certain inclination they will get very close to 0 as shown in the figures . we take this as the upper boundary of the flow structure . if we continue the calculation through this inclination , it will encounter numerical errors . we think the reason is that we ca nt describe the flow near the vertical axis with a simple self - similar solution in the radial direction . this effect is also to be expected . if we describe the upper boundary ( minimum @xmath1 ) that we reach in our calculation as @xmath68 , then the effective accretion rate @xmath69 across a sphere at radius @xmath31 within the region calculated by us is @xmath70 which is a function of @xmath31 unless @xmath34 . in equation ( 22 ) ( and subsequently in this paper ) negative values of @xmath69 represent inflow while positive values represent outflow . if we describe the accretion rate in the region between the vertical axis and the inclination @xmath68 as @xmath71 , then according to the steady nature of the flow , we have @xmath72 in which @xmath73 represents the total accretion rate across any sphere at a reasonable radius centered by the central accretor and should be a constant for a steady accretion flow . if the solution does nt end at an upper boundary , and instead can describe the entire flow structure in the whole space , then @xmath74 and @xmath75 , which should be a constant . according to equation ( 22 ) , this can only happen in the following two cases : ( 1 ) @xmath34 which enforces @xmath69 not to change with radius @xmath31 ; ( 2 ) when @xmath76 , the integration term in equation ( 22 ) must be 0 , in which case @xmath77 and is a constant along radius @xmath31 . the first case was discussed in ny95 . because @xmath34 , @xmath78 is independent of @xmath31 , and the continuity equation ( 1 ) shows that @xmath5 , resulting in a solution in which the flow is always radial ( with rotation ) . the second case was discussed in xc97 , and the fact that @xmath79 requires that the outflow rate exactly equals the inflow rate at any radius . however this is unrealistic for an accretion flow , which is also discussed in xw05 . the reason is that , when material is accreted in the form of an accretion flow , gravitational energy is released and part of it is changed to internal energy via viscous friction . the restriction that outflow rate equals inflow rate requires that the internal energy released from gravitational energy should be fully returned to gravitational energy , which violates the second law of thermodynamics . therefore , the inflow rate must be larger than the outflow rate at a certain radius , in order to compensate for the increase in entropy in the hydrodynamic process . as we have mentioned before , there are many observations of accretion systems with outflows , and they can be neither of the two cases mentioned above ( case ( 1 ) has no outflow while case ( 2 ) violates the second law of thermodynamics ) . so the self - similar solutions for an accretion disk with outflow have to be truncated at some inclination . it should be noted that this effect has already been discussed in xw05 , though they assumed that the self - similar solution is only valid for the inflow part . our solutions , however , show that at least part of the outflow ( with @xmath80 ) can be described with self - similar assumptions . from the above analysis , we can also see that @xmath35 is a very important parameter . when @xmath34 , the entire flow structure can be described by the same set of self - similar solutions , however this kind of solutions have no outflow . when @xmath81 , self - similar solutions can only describe part of the entire flow structure . according to equation ( 22 ) , when @xmath82 , the effective accretion rate @xmath69 decreases as @xmath31 decreases , implying that material is lost due to outflows . this kind of solutions contain outflows , which is also consistent with the results of many numerical simulation works ( e.g. okuda et al . 2005 ; ohsuga et al.2005 ; ohsuga & mineshige 2007 ; ohsuga et al . 2009 ) . on the other hand , when @xmath83 , the solutions have an effective accretion rate @xmath69 that increases towards the central accretor , which implies that there must be matter injection into the accretion flow from high latitudes . this kind of matter injection is not likely to happen in real cases . so we would like to investigate solutions with @xmath84 . solutions with @xmath34 have been examined in ny95 , and they are only for adafs , while solutions with @xmath82 in our work is applicable to a much wider range of accretion disk types . so in this subsection , we take @xmath85 to calculate the typical solutions . in this way we can observe how the solutions differ from those of @xmath34 when @xmath35 does nt change much . we also calculated how the solutions change with different @xmath35 , which will be discussed in detail in 3.2 . the result shows that , when other parameters are the same , for a large range of @xmath35 with @xmath82 the solutions display similar structures qualitatively . the value of the viscosity parameter @xmath2 can be inferred both from observations and from mhd simulations of magneto - rotational instability ( mri ) . as reviewed by king et al . ( 2007 ) , the value of @xmath2 for ionized disks given by numerical simulations is @xmath86 , while that obtained through observations is @xmath87 . however , the observational determinations of @xmath2 are strongly model - dependent and it still remains an open question what the value of @xmath2 should be for hot , ionized disks . in this subsection we take @xmath88 , which is a typical value used in theoretical models of accretion disks . we also discussed how the solutions would change with different @xmath2 in 3.2 . among the most popular analytical disk models are the ssd , adaf and slim disk models , so here we show the solutions with input parameter corresponding to them respectively . for accretion flows composed mostly of fully ionized hydrogen , the material can be regarded as monatomic ideal gas , of which @xmath89 , and that is the value we take here ( accretion flows composed of unionized @xmath90 would have @xmath91 . we did nt discuss this situation here , but some relative results can be found in 3.2 ) . figures 1 & 2 are both for the ssd model which has little advection ( @xmath92 ) , while figure 1 is for the gas pressure dominated region and figure 2 for the radiation pressure dominated region . the adaf model is treated as advection dominated(@xmath93 ) , and gas pressure dominated ( @xmath94 , so that @xmath95 ) , as shown in figure 3 . the slim disk model is treated as advection dominated(@xmath93 ) , and radiation pressure dominated ( @xmath57 , so that @xmath96 ) , as shown in figure 4 . the other two parameters are all set as @xmath88 and @xmath85 . in each figure , we show the function curves of @xmath37 , @xmath38 , @xmath39 , @xmath40 , @xmath36 and the mach numbers of @xmath9 and @xmath6 , in which the adiabatic sound speed is calculated as @xmath97 as shown in equations ( 7 - 11 ) , all the velocities scale with @xmath98 . the density @xmath36 is scaled to 1 in the equatorial plane and it can be determined in a real case once the accretion rate at a certain radius is given . the pressure @xmath21 is also scaled , and it satisfies the relation that @xmath99 . so once we get the actual density @xmath67 at a certain radius , we can calculate the corresponding pressure @xmath21 easily . the mach number curves for @xmath9 first decreases with @xmath1 , reaching 0 at some inclination , and then increases . that is because it changes sign along @xmath1 and we use absolute values in calculating the mach numbers . there are several common features among figures 1 - 4 . the value of @xmath6 is always non - positive . the value of @xmath9 is always negative close to the equatorial plane , and becomes positive at smaller inclinations . in each of these figures , there exists a certain inclination @xmath100 at which @xmath101 . this @xmath100 has similar meaning with the @xmath100 in xw05 , except that it is obtained through calculation rather than being preset as an input parameter in xw05 . in the region @xmath102 , @xmath37 is negative , meaning that the accretion flow is moving towards the central accretor as inflow . so this region corresponds to the normal accretion disk part in basic models and we call it the inflow region in this paper . in the region @xmath103 , @xmath37 is positive and the accretion flow is moving away from the central accretor , and we call this region the outflow region . the region between the upper boundary and the vertical axis contains the outflow which blows out of the upper boundary in the form of wind as shown in figure 5 , and we call this region the wind region , to distinguish it from the outflow region . it should be noted that our solutions actually can not describe the structure of the wind region . however , it is natural to suppose that the flow structure in the wind region with @xmath104 is also outflow , and the outflow region ( @xmath103 ) can either be regarded as a transition region between inflow and outflow , or as part of the outflow which can still be described with the self - similar assumptions . the physical properties shown in figures 1 & 2 agree quite well with the ssd model . the rotation is very close to being keplerian ; the radial velocity is much smaller than the azimuthal velocity ; and the disk is geometrically thin . however , as mentioned above , our calculation shows that even in the ssd case , there exists outflow in the accretion disk . as shown in figures 5(a ) & 5(b ) , this outflow is in the form of wind blowing out from the disk surface . both @xmath9 and @xmath6 are subsonic in the ssd cases . the advection dominated solutions , as shown in figures 3 & 4 , are geometrically much thicker than the solutions with low advection , and are able to describe the structure of the majority of the space , except only a small region near the vertical axis . the rotation is sub - keplerian and the radial velocity can be comparable with the azimuthal velocity . both the adaf and the slim disk case show strong outflows compared with the ssd cases , which can be seen more clearly in figure 5 . @xmath6 keeps to be subsonic for both cases . however , for the slim disk case ( figure 4 ) , @xmath9 becomes supersonic near the vertical axis , displaying a stronger outflow than the other three typical solutions . outflow is the result of the competition among the pressure gradient , the centrifugal force and the gravitational force . as we have mentioned before , the key feature of the outflow in our work is positive @xmath9 , so we would like to investigate the properties of these forces along the radial direction . the radial components of the pressure gradient , the centrifugal force and the gravitational force can be written respectively as @xmath105 in which @xmath106 we plot the values of @xmath107 , @xmath108 and @xmath109 for the four typical solutions in figure 6 . the dotted lines correspond to the gravitational force which is scaled as 1 ; the dashed lines correspond to the radial component of the centrifugal force ; the dot - dashed lines correspond to the radial component of the pressure gradient ; and the solid lines correspond to the sum of the radial components of the centrifugal force and the pressure gradient , which drives the outflow . figure 6(a ) shows the properties of gas pressure dominated ssds , while figure 6(b ) shows the properties of radiation pressure dominated ssds . it can be seen that for both of the ssd cases , the influence of pressure gradient is very small , and almost in the entire range of the solution , the gravitational force is balanced with the sum of the centrifugal force and the pressure gradient in the radial direction . this is consistent with the ssd model in which both the advection and the pressure gradient are small enough to be neglected . therefore these two solutions have similar accretion flow structure and only some quantitative differences exist between each other . on the other hand , for the advection dominated cases , such as the adaf case ( figure 6(c ) ) and the slim disk case ( figure 6(d ) ) , the pressure gradient plays a more important role than the centrifugal force and drives a significant outflow as @xmath1 decreases . the slim disk case has qualitative differences from the other three cases . in both ssd cases and the adaf case , the radial component of pressure gradient starts decreasing at smaller inclination angle , and to balance this effect , the disk rotates faster as @xmath1 decreases to increase the centrifugal force , as shown in figures 1 - 3 . however in the slim disk case , the radial component of the pressure gradient keeps increasing as @xmath1 decreases . it not only drives a much stronger outflow than the other three cases , but also reduces the required centripetal force to keep the disk rotating . thus the disk rotates slower as @xmath1 decreases , as shown in figure 4 . as a comparison with ny95 , we also present here a solution with parameters @xmath88 , @xmath34 , @xmath110 and @xmath93 in figure 7 . this set of parameters corresponds to @xmath111 in ny95 . in this case , the accretion flow does nt contain outflow and the three components of velocity and the sound speed remain constant in the structure . the curves of @xmath36 and @xmath37 do nt agree well with ny95 , however this is because ny95 also sets the boundary conditions on the vertical axis and the values of @xmath36 and @xmath37 are limited by these boundary conditions . our solution only uses the boundary conditions in the equatorial plane , and can still reach the vertical axis in this case and has a reasonable distribution of physical properties , which further proves the applicability of our method . in this section we show how the solutions change according to different input parameters respectively , and discuss their physical meanings . as shown in 3.1 , the whole space is generally divided into three regions : an inflow region , an outflow region and a wind region . as the accretion flow is symmetrical both axially and under reflection in the equatorial plane , we only need to investigate the flow structure in a quadrant of the @xmath112 plane . in this case , the inflow region starts from the equatorial plane ( @xmath65 ) to the inclination @xmath113 where @xmath101 ; the outflow region starts from the inclination @xmath113 to the upper boundary of the flow structure at @xmath114 , and beyond that is the wind region ( @xmath115 ) in which the motion of the matter does nt satisfy the self - similar assumptions . here we use @xmath100 and @xmath68 as indicators of the disk size in @xmath1-direction , which directly reflects the size of the inflow region and the whole inflow / outflow region , respectively . in the figures showing how the properties of solutions change according to input parameters , each line represents a set of solutions , and we assign a unique label for each solution set for clarity . table 1 lists all the solution set labels , their input parameters and corresponding lines at the end of 3.2 . ` 1 ) solution dependence on \alpha ` the dependence of @xmath100 and @xmath68 on @xmath2 is shown in figure 8 . first we can clearly see that over a large range of @xmath2 , the advection dominated solutions have both larger inflow region and larger inflow / outflow region . that s because in these solutions most of the viscous heating is stored in the accretion flow as internal energy , resulting in larger pressure at given density and thus larger size of the whole inflow / outflow region . for solutions with the same advective factor @xmath3 , gas pressure dominated solutions generally have larger inflow region . the inflow region lasts from the equatorial plane to the inclination @xmath100 where @xmath9 reaches 0 , so its size depends on the behavior of @xmath9 . in the equatorial plane , the reflection symmetry gives equation ( 21 ) , and when it is substituted into equations ( 17 ) , ( 19 ) & ( 20 ) , we can get @xmath116 equation ( 30 ) describes the hydrodynamical balance in the radial direction among the pressure gradient , the centrifugal force and the gravitational force , which determines the acceleration of @xmath9 ( the term @xmath117 represents the acceleration of @xmath9 under the self - similar assumptions ) . equation ( 31 ) describes how the viscosity affects the angular momentum transfer rate in the equatorial plane . equation ( 32 ) describes the energy mechanism of the accretion flow that a fraction @xmath3 of the dissipated energy is advected as stored entropy . it should be noted that @xmath9 is generally negative in the equatorial plane , so the absolute value of @xmath9 can be written as ( according to equation ( 32 ) ) @xmath118.\ ] ] if @xmath119 remains the same , then in the range of parameters we study in this paper , @xmath120 increases with @xmath55 . however @xmath119 also changes with input parameters , and the exact value of @xmath121 should be calculated via equations ( 30)-(32 ) . figure 10 shows the profile of @xmath121 vs @xmath55 when other parameters are taken as @xmath85 and @xmath88 . it can be seen that for both the advection dominated solutions and the solutions with little advection , @xmath120 increases with @xmath55 . with a larger value of @xmath9 in the equatorial plane , the accretion flow is inclined to undergo a larger range of @xmath1 to change @xmath9 to 0 , resulting in a larger inflow region . for the two ssd cases , as their flow structures are quite similar , larger values of @xmath9 in the equatorial plane give the gas pressure dominated solution set @xmath122 larger inflow regions than @xmath123 . for the advection - dominated cases , not only the equatorial value of @xmath9 is smaller in the slim disk case than that in the adaf case , but its radial direction pressure gradient also increases as @xmath1 decreases , giving a larger outward acceleration to @xmath9 , thus the gas pressure dominated solutions @xmath124 have larger inflow regions than the radiation pressure dominated solutions @xmath125 . the solution set @xmath125 , which corresponds to the slim disk model , has a large outflow size over a broad range of @xmath2 , displaying a large outflow area . the solutions corresponding to the ssd model , @xmath122 and @xmath123 , have very small outflow regions , and their outflows are mainly in the form of wind blowing out of the disk surface instead of escaping in the outflow region . the solution set @xmath124 , which corresponds to the adaf model , has a moderate outflow region with small @xmath2 . however as @xmath2 increases , this outflow region quickly disappears and the outflow takes the form of wind blowing out of the upper surface . the velocity field displayed in figure 5 is an example for @xmath88 . @xmath2 is connected with both the viscous heating and the angular momentum transport of the accretion flow . larger @xmath2 will increase the viscous heating for a fixed surface density . but at the same time it also increases the angular momentum transfer rate , thus increasing @xmath9 and decreasing surface density , which decreases the total viscous heating rate . therefore the dependence of the viscous heating rate on @xmath2 , is somewhat complex . viscous heating represents how much gravitational energy is converted to internal energy . but some of it is lost due to radiation and eventually there is only an @xmath3 fraction of the viscous heating which is converted to internal energy . the increase in the internal energy will raise the temperature and subsequently the pressure , thus increasing the size of the accretion flow . however , for the ssd cases shown in @xmath122 & @xmath123 , @xmath3 is very small , thus the energy converted to internal energy is negligible and the disk structure is dominated dynamically by the gravity and the rotation of the material , rather than the pressure gradient . as a result , the sizes of accretion flows change very little with different @xmath2 for @xmath122 & @xmath123 . on the other hand , for the advection dominated solutions @xmath124 & @xmath125 , a large fraction of the gravitational energy is eventually stored as internal energy , so that the pressure gradient plays a more important role in determining the flow structure as mentioned in 3.1 . therefore the sizes of accretion flows are closely connected with @xmath2 . in figure 8 we can see that for @xmath124 & @xmath125 , when @xmath2 is very small , the viscous process is not quite effective , therefore both the inflow region size and the outflow region size are small ( it s easy to understand if we consider the situation @xmath126 , in which case no accretion flow will be formed ) . from this small value , along with the increasing of @xmath2 , the viscous process starts to take effect and the whole inflow / outflow region size starts to increase , until it reaches a maximum size at @xmath127 for line @xmath128 and @xmath129 for line @xmath130(the exact value of @xmath2 for maximum size also depends on other parameters ) . from then on , the larger values of @xmath9 starts to dominate the total effect and the corresponding accretion flow size starts to decrease . this property can also be seen in the top left panel of figure 17 , which displays the ratio of the outflow rate to the inflow rate . it should be noted that , in figure 8 , the size change based on @xmath2 is different between the whole inflow / outflow region and the pure inflow region . the size of the whole inflow / outflow region decreases after a peak at a certain @xmath2 while the size of the pure inflow region keeps increasing . for solution sets @xmath122 , @xmath124 and @xmath125 , this difference is enough to generate a critical @xmath131 respectively , at which the outflow region totally disappears . so for @xmath132 , the accretion flow that we can resolve with self - similar assumptions is composed of pure inflow and we did nt find a surface at which @xmath133 . that s why we say that the assumption used in xw05 may not be true for large @xmath2 . kluzniak and kita ( 2000 ) also found this critical @xmath2 which was calculated to be @xmath134 by them . in our work , however , @xmath131 depends on other parameters and can change significantly , as shown in figure 8 ( @xmath135 for @xmath122 , @xmath136 for @xmath124 and @xmath137 for @xmath125 ) . for @xmath123 , this @xmath131 is above 1 and it s not displayed . figures 9 displays the @xmath9 distribution along the @xmath1-direction for solutions with different @xmath2 . figures ( a ) , ( b ) , ( c ) and ( d ) correspond to the four sets of solutions in figure 8 respectively , and the solid , dashed and dotted lines correspond to different @xmath2 as indicated in the legends . the range of the axes are adjusted to show more details in comparison . we can see that in the solutions corresponding to the ssd model , i.e. figures 9(a ) & 9(b ) , @xmath9 is almost proportional to @xmath2 . that is because when alpha increases , so does the angular momentum transfer rate . for the ssd case , the accretion flow structure is mostly determined by gravity and rotation as discussed in 3.1 , and the rotation is quasi - keplerian , so the increase in angular momentum transfer rate increases @xmath9 almost proportionally . a special case in the equatorial plane is shown by equation ( 31 ) , in which when @xmath10 is keplerian , the value of @xmath9 is proportional to @xmath2 . for the adaf and the slim disk cases , because pressure gradient plays an important role in the accretion flow structure , @xmath9 is generally not proportional to @xmath2 , but still positively correlated with @xmath2 as shown in figure 9(c ) & 9(d ) . we can also see that some lines in figure 9 do nt contain positive @xmath9 . that s because the @xmath2 values for these solutions are greater than the critical value @xmath131 mentioned above . ` 2 ) solution dependence on n ` the parameter @xmath35 describes how the density changes along the radius in our self - similar assumptions . as discussed in 3.1 , when @xmath34 , the entire flow structure can be described by the same set of self - similar solutions , but does nt contain outflow ; when @xmath82 , the solution contains outflow but the self - similar assumptions can not be applied to the region near the vertical axis ; solutions with @xmath83 are unlikely to happen in real cases . the dependence of @xmath100 and @xmath68 on @xmath35 is shown in figure 11 . figures 12 & 13 display the @xmath9 and @xmath6 distributions along the @xmath1-direction for solutions with different @xmath35 . we can see that solutions all achieve a maximum size at @xmath34 for both the whole inflow / outflow region and the inflow only region . this is because when @xmath34 the entire space can be described by the same set of self - similar equations as discussed in 3.1 . in this case , @xmath5 and mass is conserved on the streamlines pointing exactly at the central accretor , thus no outflow is required . before reaching the maximum value at @xmath34 , the flow size is positively correlated with @xmath35 as a general trend . as @xmath35 decreases , according to equation ( 22 ) @xmath69 decreases faster inwards , which means a larger mass loss rate into the area beyond @xmath68 , resulting in the increase in the size of the wind region and the decrease in the size of the whole inflow / outflow region in figure 11 , and the increase in @xmath6 in figure 13 . in figure 11 line @xmath138 has 2 peaks . according to equation ( 25 ) , the radial pressure gradient is proportional to @xmath139 , so when @xmath140 does nt change much , the radial pressure gradient increases with @xmath35 . in advection - dominated solutions @xmath141 & @xmath142 , the radial pressure gradient is more important than centrifugal force in accelerating @xmath9 , so for larger @xmath35 which is inclined to have larger radial pressure gradient , one would expect to see faster increase in @xmath9 and stronger outflow in the outflow region . these features of @xmath9 can be seen in figure 12 . the radial pressure gradient is dominant in the slim disk case as mentioned in 3.1 , so the acceleration of @xmath9 increases significantly with increasing @xmath35 in the slim disk case , which can be seen in figure 12(d ) with a much steeper profile of @xmath9 when @xmath85 . this increased outflow velocity in the outflow region can cause a steeper density drop along @xmath1 direction towards the upper boundary and therefore make the outflow region become smaller , creating a valley in line @xmath138 around @xmath143 . the top right panel in figure 17 shows how the ratio of the outflow rate to the inflow rate changes depending on @xmath35 . we can see that when @xmath35 becomes small enough , the ratio changes very little with @xmath35 . in this range of @xmath35 , the size of different regions also does nt change much in figure 10 , and the flow structure is similar to each other . for @xmath95 , we can not get a reasonable set of equatorial values for physical properties when @xmath35 is larger than 3/2 . for @xmath96 , i.e. solutions @xmath144 and @xmath142 , we can still get reasonable values of equatorial physical properties when @xmath35 is larger than 3/2 , so that we can calculate solutions with @xmath83 in this case . solutions with @xmath83 have an effective accretion rate @xmath69 that increases towards the central accretor , which implies that there must be matter injection into the accretion flow from high latitudes . this kind of matter injection is not likely to happen in real cases , so we wo nt pay more attention to the solutions with @xmath83 here . solutions with @xmath34 and @xmath82 both exist and have quite different structure . however , as in ny95 , solutions with @xmath34 are only for adafs , while solutions with @xmath82 in our work is applicable to a much wider range of accretion disks . in this sense , we believe that outflows should be common in various accretion disks . ` 3 ) solution dependence on f ` the dependence of @xmath100 and @xmath68 on @xmath3 is shown in figure 14 . it has already been discussed that as the advective factor @xmath3 increases , the size of the whole inflow / outflow region also increases , which is shown more clearly here in figure 14 . the gas pressure dominated solutions have larger inflow region in agreement with figures 8 & 11 , and the reason has been discussed in 3.2.1 . as @xmath3 increases , the size of the outflow region increases faster for radiation pressure dominated flows , displaying a larger outflow size for solutions corresponding to the slim disk model . the reason is that , as @xmath3 increases , a larger fraction of viscous heating is converted to internal energy instead of being lost via radiation , raising the temperature of the accretion flow . the gas pressure is proportional to @xmath145 while the radiation pressure is proportional to @xmath146 , so in the radiation pressure dominated solutions , @xmath21 increases faster as @xmath3 increases . the increased pressure inflates the outflow region and causes a stronger outflow . on the other hand , as @xmath3 gets close to 1 , eventually the outflow velocity at small inclination becomes so large , that it will instead reduce the size of the outflow region because the material lost in the outflow region is limited . as shown in the bottom left panel of figure 17 , at a certain @xmath147 , the fraction of material lost in the outflow region reaches an upper limit , and with larger @xmath3 , this fraction almost remains unchanged . in this range of @xmath3 , the larger outflow velocity will instead shrink the size of the outflow region . this causes line @xmath148 in figure 14 to reach a peak of @xmath149 at @xmath147 ( corresponding to figure 17(c ) ) , and to start decreasing as @xmath3 becomes larger . ` 4 ) solution dependence on \gamma_\mathrm{equ } ( and consequently \beta ) ` the dependence of @xmath100 and @xmath68 on @xmath55 is shown in figure 15 . the top axis displays the corresponding values of @xmath43 when @xmath27 takes the value of 5/3 . if the intrinsic heat capacity ratio is not 5/3 ( e.g. when dealing with proto - stellar accretion disks with unionized gas ) , the corresponding @xmath43 in figure 15 may not be correct , but one can still calculate the correct @xmath43 with @xmath55 and @xmath27 according to equation ( 14 ) . figure 16 displays the @xmath9 and @xmath6 distributions along the @xmath1-direction for solutions with different @xmath55 . the bottom right panel of figure 17 shows how the ratio of the outflow rate to the inflow rate changes depending on @xmath55 ( and consequently @xmath43 ) . in figure 15 , the size of the pure inflow region generally increases with @xmath55 ( and consequently @xmath43 ) . as shown in figure 10 & 16 and discussed in 3.2.1 , solutions with larger @xmath55 ( and consequently larger gas pressure ratio @xmath43 ) have larger values of @xmath9 in the equatorial plane . this effect combined with the acceleration properties of @xmath9 , causes solutions with larger @xmath55 ( and consequently larger @xmath43 ) to have larger inflow region . for solutions with low advection , as the outflow region is very small , the size of the whole inflow / outflow region also displays similar profile with increasing @xmath55 ( and consequently @xmath43 ) . that s why lines @xmath150 , @xmath151 and @xmath152 all have positive slope in figure 15 . in figure 17(d ) , there exists a transition point for the advection dominated solutions , and for solutions with @xmath153 ( corresponding to @xmath154 ) , the ratio of the outflow rate to the inflow rate keeps being large . as discussed before , for the advection dominated cases , radiation pressure dominated solutions have larger pressure gradient , shown as steeper velocity profiles in figure 16 , which drives stronger outflow . as shown in figure 17(d ) , the outflow to inflow ratio increases as @xmath55 decreases from 5/3 ( which indicates the decreasing of gas pressure ratio @xmath43 from 1 ) . however , the outflow rate can not exceed the inflow rate and eventually the increase will cease , reaching a transition point at @xmath155 ( corresponding to @xmath156 ) . for @xmath153 ( corresponding to @xmath157 ) , the outflow to inflow ratio remains almost unchanged . further decrease in @xmath55 will continue to increase the outflow velocity , while the fraction of the inflow lost in the outflow region is almost unchanged , so the size of the outflow region will shrink . this feature can be also seen in figure 15 , which shows that the size of the outflow region for advection dominated solutions first increases with decreasing @xmath55 , reaching an upper limit at @xmath155 ( corresponding to @xmath156 ) , and decreases with decreasing @xmath55 after that . it should be noted that , this transition value of @xmath55 depends on other input parameters , and this value of @xmath158 corresponds to @xmath88 , @xmath85 and @xmath93 . for the solutions with low advection , the ratio of the outflow rate to the inflow rate do nt change much and are always very small as shown in figure 17(d ) . ` 5 ) bends in solutions ` we have shown the solution dependence on different input parameters . the solutions sometimes show bends which we would like to discuss more specifically here . the bend about @xmath2 is closely connected with the energy mechanism . as discussed in 3.2.1 , the dependence of viscous heating on @xmath2 is somewhat complex . it has a peak at a certain value of @xmath2 ( which depends on other input parameters ) , and decreases on both sides . for the advection dominated solutions @xmath124 & @xmath125 , most of the viscous heating is stored as internal energy in the accretion flow , which raises the temperature and subsequently the pressure , thus increasing the size of the accretion flow . on the other hand , the increase in @xmath2 also increases the absolute value of @xmath9 of the inflow near the equatorial plane , in which case more work needs to be done to drive an outflow , decreasing the outflow region as @xmath2 increases . so naturally the size of the whole inflow / outflow region also has a peak at some value of @xmath2 . as @xmath2 increases from a very small value , at first the increase in viscous heating dominates the total effect , and the size of the accretion flow increases , until it reaches a maximum value ; from then on , the increase in @xmath9 starts to dominate the total effect , which makes the outflow more difficult to produce and also reduces viscous heating when @xmath2 becomes large enough , thus shrinking the size of the whole accretion flow . to show the energy profile for solutions with different @xmath2 , here we calculate the @xmath1-averaged bernoulli function @xmath159 . the bernoulli function represents the specific total energy in the accretion flow . a positive value of the bernoulli function is only a necessary , not a sufficient , condition for outflow formation ( abramowicz et al . 2000 ) . however , as our solutions have already showed outflow structure in the velocity fields , a larger value of the bernoulli function indicates that the accretion flow is more likely to contain stronger outflow . locally the bernoulli function can be written as @xmath160,\ ] ] in which @xmath161 is the specific enthalpy , @xmath162 is the velocity ( all three components included ) and @xmath163 is the gravitational potential . so the @xmath1-averaged bernoulli function can be written as @xmath164 \sin\theta d\theta } { \int_{\theta_\mathrm{b}}^{90\textordmasculine } \rho(\theta)\cdot \sin\theta d\theta}.\end{aligned}\ ] ] figure 18 displays how the @xmath1-averaged bernoulli function @xmath159 changes with variant @xmath2 for the advection dominated cases . it can be seen that the positions of the peaks agree quite well with the bends of line @xmath128 and line @xmath130 in figure 8 . figure 19 displays the velocity fields of solutions around the bend for variant @xmath2 . the values of @xmath2 are chosen so that the solutions on both sides of the peaks have similar upper boundaries ( @xmath165 ) . it can be seen that the velocities in the outflow region are more aligned to the @xmath1 direction in the right panel of each row than those in the left panel . that s because larger @xmath2 increases the angular momentum transfer rate , thus increases the absolute value of @xmath9 near the equatorial plane where most of the material is accreted . the value of @xmath9 is negative in the equatorial plane and changes gradually to positive values in the outflow region under the effect of pressure gradient and centrifugal force . the absolute value of @xmath9 in the equatorial plane in the right panel of each row is significantly larger than that in the left panel , so when pressure gradient does nt differ much , the positive values of @xmath9 in the outflow region will be smaller for larger @xmath2 and the velocities will be more aligned to the @xmath1 direction . the solutions corresponding to slim disks ( the second row ) show velocities more aligned to the radial direction in the outflow region than those corresponding to adafs ( the first row ) , due to larger pressure gradient and thus larger values of @xmath9 in the outflow region . the bends in solution dependence on @xmath35 , @xmath3 and @xmath55 have different physical mechanism from that on @xmath2 . it has been mentioned that , as @xmath35 increases , @xmath3 increases or @xmath55 decreases , the pressure gradient increases and drives stronger outflow . however , the outflow rate can not exceed the inflow rate . at seen in figure 17 , the ratio of the outflow rate to the inflow rate for lines @xmath142 , @xmath166 and @xmath152 will eventually reach an upper limit at certain values of corresponding parameters . for solution sets @xmath166 and @xmath152 , further increase in @xmath3 or further decrease in @xmath55 will continue to increase the radial outflow velocity , while the outflow rate in the outflow region is almost unchanged , so the size of the outflow region will shrink . for the solution dependence on @xmath35 , the bend near @xmath167 has similar physical reason as the bends for @xmath3 and @xmath55 . however , the size of the accretion flow has another peak at @xmath34 , where the flow is always radial ( with rotation ) and no outflow is needed . so line @xmath142 in figure 17 starts to drop as @xmath35 gets close to 3/2 and line @xmath138 in figure 11 gets to another peak at @xmath34 . figure 20 displays the velocity fields of solutions around the bend for variant @xmath35 , @xmath3 and @xmath55 . it can be seen that , as @xmath35 increases ( the first row ) , @xmath3 increases ( the second row ) or @xmath55 decreases ( the third row ) , the flow patterns display similar evolvement for these parameters : at first both the accretion flow size and the radial outflow velocity increase ; after the corresponding parameter gets past the respective value at the transition point ( the middle panels in all three rows of figure 20 ) , the radial outflow velocity continues to increase while the size of the accretion flow decreases . the velocities in the outflow regions become more aligned to the radial direction from left to right in all three rows , due to increased radial outflow velocities . the labels for solution sets and lines which appear in the text and figures are summarized in table 1 along with their corresponding input parameters . 0.2 cm [ cols="<,^,^,^,^,^,<",options="header " , ] with the self - similar assumptions along the radial direction and boundary conditions obtained from the reflection symmetry in the equatorial plane , we are able to solve the navier - stokes equations along the @xmath1 direction explicitly , and the velocity field is obtained . the result shows that outflows are common in all kinds of accretion disk models , which is consistent with some numerical simulation works ( e.g. ohsuga et al.2005 ; ohsuga & mineshige 2007 ; ohsuga et al . generally the accretion flow consists of three different regions : an inflow region near the equatorial plane which contains the largest portion of mass , an outflow region above the inflow region in which the matter starts escaping the central accretor in the @xmath31-direction , and a wind region which contains the material blowing out from the boundary of the outflow region . the structure of the inflow and the outflow region can be resolved in our solutions , while the wind region does nt obey the self - similar assumptions and can not be resolved by our calculation . the inflow region and the wind region are essential to form a steady accretion flow structure . the outflow region is missing in the cases with large viscosity parameter @xmath132 . the critical value @xmath131 depends on other parameters and is @xmath136 for adafs , @xmath137 for slim disks and @xmath135 for gas pressure dominated ssds when @xmath85 . to compare with the popular analytical models , we calculated solutions with parameters corresponding to the ssd , the adaf and the slim disk models . the solutions corresponding to the ssd model have relatively thin , quasi - keplerian disks with very small outflow regions , and the outflows mostly take the form of wind from the disk surface . the solutions corresponding to the adaf and slim disk models have thick , sub - keplerian disks . the solution corresponding to the adaf model has a moderate outflow region which shrinks quickly with the increase of @xmath2 , and the outflow mainly takes the form of wind . however as the upper boundary of the outflow region is close to the vertical axis , eventually the outflow may take the form of bipolar jets , though we can not actually resolve the explicit structure around the vertical axis . the solutions corresponding to the slim disk model has a large outflow region in which most of the outflow takes place , and near the upper boundary of the outflow region , the material escapes with supersonic velocity . our calculation displays very small outflow regions for input parameters corresponding to the ssd model , and for these solutions the inflow region has similar structure with the one - dimensional ssd model . however the accretion rate inside the disk may not be conserved along the radius due to mass lost via wind . for the adaf and the slim disk cases , our result displays an outflow region too large to be neglected , and we think it s necessary to consider the explicit structure in the @xmath1 or @xmath4 direction to get more realistic results than the traditional analytical models . it was proposed in several works ( gu & lu , 2007 ; jiao et al . 2009 ) that the slim disk model has an upper limit of accretion rates , above which outflows seem to be inevitable . later in another paper ( jiao & lu 2009 ) , one - dimensional steady transonic global solutions for slim disks with presumed amount of outflows were calculated , and in this case the proposed upper limit of accretion rates for slim disks can be exceeded . another paper written by takeuchi et al . ( 2009 ) used the outflow data from two - dimensional radiation - hydrodynamic ( rhd ) simulations to construct an one - dimensional steady solution , and the result shows that the emergent spectra do not sensitively depend on the amount of mass outflow . however , these models are still one - dimensional models and they only incorporated the mass loss effect of outflow into their models , so the results still remain debatable . we think that two - dimensional solutions considering the hydrodynamical processes and radiative transfer along both the radial and the @xmath1 direction are required to examine the topic , which we are planning to do in our future work . we also investigated the dependence of the accretion flow structure on different input parameters respectively . @xmath3 and @xmath55 ( consequently @xmath43 ) both have some transition values as discussed in 3.2.3 and 3.2.4 , at which the outflow region reaches a maximum size . generally when @xmath168 and @xmath169 ( for intrinsic @xmath89 ) , the accretion flow will have a large outflow region and strong outflow , as the aforementioned solutions corresponding to the slim disk model . however , in what conditions the transition happens depends on other input parameters and it wo nt happen in case of large @xmath2 due to the existence of the critical value @xmath131 . solutions with @xmath34 do nt contain outflow and they correspond to the solutions in ny95 , which is only applicable to adafs . solutions with @xmath82 have the aforementioned three - region ( or two - region for @xmath132 ) structure . solutions with @xmath83 are not likely to happen in real cases . finally , we would like to mention some caveats in our work . the solutions obtained in this paper are steady solutions , and thus it is not guaranteed that they are stable solutions . these solutions do not cover the whole space , so for the unresolved region near the axis , we know nothing about the detailed structure other than that it contains the material blowing out of the upper boundary in the form of wind . our work is based on the simple @xmath170 prescription of viscosity , which is not bad according to recent mhd simulation works(hirose et al . 2009 ; ohsuga et al . 2009 ) , especially when the flow is steady . however , it may be necessary to consider other components of the viscous stress tensor for solutions with large @xmath9 . we also used the advective factor @xmath3 to simplify the energy equation , and assumed @xmath3 to be a constant . realistically @xmath3 should vary with both @xmath1 and @xmath31 , which could be improved by considering the details of radiative transfer . we used newtonian gravitational potential in our work , and it needs to be corrected when applied to study the structure close to the innermost stable circular orbit . in our calculation we did nt consider the effects of convection ( e.g. stone et al . 1999 ; igumenshchev & abramowicz 2000 ; yuan & bu 2010 ) or magnetic field ( e.g. blandford & payne 1982 ; naso & miller 2010 ) , which may be important in studying the accretion flow and the generation of outflows . finally , it is necessary to abandon the self - similar assumptions if we want to investigate the disk and outflow structure more quantitatively and precisely . these caveats will be improved in our future work . we thank fu - guo xie for discussions and suggestions . we are also grateful to marek abramowicz , feng yuan , wei - min gu and li xue for helpful comments . we also thank the referee for his / her constructive suggestions . this work was supported by the nsfc grant ( no . 11033001 ) and the 973 program ( no . 2007cb815405 ) . abramowicz , m. a. , chen , x. , kato , s. , lasota , j .- regev , o. 1995 , , 438 , l37 abramowicz , m. a. , czerny , b. , lasota , j. p. , & szuszkiewicz , e. 1988 , , 332 , 646 abramowicz , m. a. , lasota , j .- p . , & igumenshchev , i. v. 2000 , , 314 , 775 blandford , r. d. , & begelman , m. c. 1999 , , 303 , l1 blandford , r. d. , & begelman , m. c. 2004 , , 349 , 68 blandford , r. d. , & payne , d. g. 1982 , , 199 , 883 chartas , g. , brandt , w. n. , & gallagher , s. c. 2003 , , 595 , 85 igumenshchev , i. v. , & abramowicz , m. a. 2000 , , 130 , 463 jiao , c .- , & lu , j .- f . 2009 , chinese physics letters , 26 , 049701 jiao , c .- l . , xue , l. , gu , w .- m . , & lu , j .- f . 2009 , , 693 , 670 kato , s. , fukue , j. , & mineshige , s. 2008 , black - hole accretion disks : towards a new paradigm ( kyoto : kyoto univ . press ) kawabata , r. , & mineshige , s. 2009 , , 61 , 1135 king , a. r. , pringle , j. e. , & livio , m. 2007 , , 376 , 1740 kluzniak , w. , & kita , d. 2000 , arxiv : astro - ph/0006266 naso , l. , & miller , j. c.2010 , , 521 , a31 okuda , t. , teresi , v. , toscano , e. , & molteni , d. 2005 , , 357 , 295 ohsuga , k. , mori , m. , nakamoto , t. , & mineshige , s. 2005 , , 628 , 368 ohsuga , k. , & mineshige , s. 2007 , , 670 , 1283 ohsuga , k. , mineshige , s. , mori , m. , & kato , y. 2009 , , 61 , l7 stone , j. m. , pringle , j. e. , & begelman , m. c. 1999 , , 310 , 1002 takahara , f. , rosner , r. , & kusunose , m. 1989 , , 346 , 122 takeuchi , s. , mineshige , s. , & ohsuga , k. 2009 , , 61 , 783 xie , f .- , & yuan , f. 2008 , , 681 , 499

according to a theorem of courcelle monadic second - order logic and guarded second - order logic ( where one can also quantify over sets of edges ) have the same expressive power over the class of all countable-sparse hypergraphs . in the first part of the present paper we extend this result to hypergraphs of arbitrary cardinality . in the second part , we present a generalisation dealing with methods to encode sets of vertices by single vertices .

0910.3085

guarded second - order logic ( @xmath1 ) is the variant of monadic second - order logic ( @xmath2 ) where one can not only quantify over sets of vertices but also over sets of edges . this modification results in a large increase of expressive power . statements that can be expressed in guarded second - order logic , but not in monadic second - order logic , include the existence of certain minors in a graph and the existence of hamiltonian paths . the high expressive power of guarded second - order logic means that most @xmath1-theories are quite complicated . in @xcite seese has shown that every class of graphs with infinite tree width has an undecidable @xmath1-theory . this result immediately generalises to hypergraphs . it follows that all classes of hypergraphs with a decidable @xmath1-theory are _ @xmath0-sparse , _ for some @xmath0 , which roughly means that their members have few edges . for classes of countable @xmath0-sparse hypergraphs , courcelle @xcite has shown that every @xmath1-formula is equivalent to an @xmath2-formula over such a class . it follows that over every class of countable hypergraphs with a decidable @xmath1-theory guarded second - order logic and monadic second - order logic have the same expressive power . unfortunately , the proof of theorem 1.4 in @xcite contains an error . in the first part of the present article we give a new proof of this theorem . in addition , we extend the result from countable hypergraphs to hypergraphs of arbitrary cardinality . when we look at the results of the first part we see that most of them concern the coding of sets of vertices by single vertices . in the abstract , this problem can be stated as follows.08em : given a set @xmath3 of finite sets of vertices , find a definable function @xmath4 that is injective . in our concrete case , @xmath5 is the set of edges . in the second part of the paper we consider more general instances of this problem where @xmath6 can be arbitrary . this generalisation is inspired by a result of colcombet and lding @xcite on set interpretations . their main technical result is a method to transform a definable finite - to - one function @xmath7 into an injective one . colcombet and ldung consider as background structure only the infinite binary tree . below we show that using guarded second - order parameters , i.e. , sets of edges , we can extend some of their results to arbitrary graphs . the overview of the article is a follows . we start in section [ sect : sparse orientations ] with basic definitions and a survey of results on definable orientations of sparse hypergraphs . in section [ sect : spanning tree ] we prove the general version of the one technical result of @xcite whose proof does not extend to arbitrary cardinalities . in section [ sect : gso and mso ] we summarise the consequences for the expressive power of guarded second - order logic on sparse hypergraphs . section [ sect : flow ] contains the second part of the article . we study network flow problems and we show how to use flows to transform definable finite - to - one maps into injective ones . let us fix our terminology regarding graphs and hypergraphs . when we say ` graph ' we will mean an undirected one . undirected graphs will always be simple and loop free , whereas directed graphs will be simple , but they may contain loops . when dealing with hypergraphs we will sometimes allow multiple edges . such a hypergraph is a two - sorted structure @xmath8 where @xmath9 is the set of vertices , @xmath10 the set of edges , and @xmath11 the incidence relation . using sloppy notation we will tacitly identify an edge @xmath12 of such a hypergraph with the set @xmath13 of its vertices and we write @xmath14 instead of @xmath15 . similarly , if @xmath16 is a set of edges then the union @xmath17 consists of all vertices incident with at least one edge of @xmath6 . we will use this notation even if there are multiple edges . _ monadic second - order logic _ ( @xmath2 ) extends first - order logic by variables and quantifiers that range over sets of vertices . similarly , _ guarded second - order logic _ ( @xmath1 ) extends first - order logic by variables and quantifiers ranging over sets of vertices or sets of edges ( for detailed definitions see @xcite ) . we will also consider _ weak monadic second - order logic _ ( @xmath18 ) where quantification is restricted to _ finite _ sets of vertices . let @xmath19 be a hypergraph . a. we say that @xmath20 has _ rank _ @xmath21 if every edge of @xmath20 has at most @xmath21 vertices . b. a _ subhypergraph _ of @xmath20 is a hypergraph @xmath22 with @xmath23 and @xmath24 . c. let @xmath25 . the subhypergraph _ induced by @xmath26 _ is @xmath27 in order to translate @xmath1-formulae into @xmath2-formulae , we have to encode sets of edges by sets of vertices . a simple way to do so consists in choosing an _ orientation _ of the hypergraph , i.e. , a function assigning to each edge one of its vertices . let @xmath19 be a hypergraph . a. an _ _ orientation _ _ of @xmath20 is a function @xmath28 with @xmath29 , for all @xmath12 . we say that a formula @xmath30 _ defines _ an orientation @xmath31 of @xmath20 if we have @xmath32 b. an orientation @xmath31 is _ bounded by @xmath0 _ if @xmath33 c. we call @xmath20 _ @xmath2-orientable _ if there exist an @xmath2-formula @xmath34 with parameters @xmath35 defining an orientation of @xmath20 . similarly , we say that @xmath20 is _ @xmath1-orientable _ if there exist a @xmath1-formula @xmath36 with parameters @xmath35 and @xmath37 defining an orientation of @xmath20 . in this paper we are mainly interested in _ sparse _ hypergraphs , i.e. , hypergraphs with few edges . a hypergraph @xmath19 is _ _ @xmath0-sparse__-sparse . _ courcelle also introduces a notion of a _ @xmath0-sparse _ graph . since uniform sparsity is the more robust notion , and the only one we will use in this paper , we have changed terminology for brevity . a related notion is the _ arboricity _ of a graph ( see , e.g. , section 2.4 of @xcite ) . ] if @xmath38 let @xmath39 be a graph . if the degree of @xmath40 is at most @xmath41 , then @xmath40 is @xmath0-sparse . if @xmath40 is planar , then it is @xmath42-sparse . if @xmath43 is a finite induced subgraph of @xmath40 then @xmath44 this follows from the fact that every planar graph with @xmath45 vertices has at most @xmath46 edges ( see , e.g. , corollary 4.2.10 of @xcite ) . in the next section we will prove that every hypergraph of bounded rank is @xmath1-orientable . in the remainder of this section we show that @xmath0-sparse hypergraphs are even @xmath2-orientable . for countable hypergraphs these results are all due to courcelle @xcite . the only thing new in the present section are two applications of the compactness theorem for first - order logic to extend the results to uncountable hypergraphs . the proofs in section [ sect : spanning tree ] , on the other hand , are mostly new . [ lem : k - sparse hypergraphs and orientations ] a hypergraph @xmath19 ( possibly with multiple edges ) of finite rank is @xmath0-sparse if and only if there exists an orientation of @xmath20 that is bounded by @xmath0 . for @xmath47 , let @xmath48 be finite . then @xmath49 @xmath50 first , let us consider the case where @xmath20 is finite . if @xmath31 is an arbitrary orientation of @xmath20 then @xmath51 hence , if there is some element @xmath52 with @xmath53 then there must be some other element @xmath54 with @xmath55 . let us define the _ weight _ of an orientation by @xmath56 we have to construct an orientation of weight @xmath57 . to do so we transform an orientation @xmath31 with @xmath58 into one with smaller weight . given @xmath31 , fix an element @xmath52 with @xmath53 . let @xmath16 be the smallest subset of @xmath10 such that @xmath59 belongs to the set @xmath60 and we have @xmath61 , for every element @xmath62 . the subhypergraph @xmath63 induced by @xmath64 is @xmath0-sparse . hence , there exists some element @xmath65 with @xmath55 . by choice of @xmath6 we can find a sequence of edges @xmath66 with @xmath67 we define a new orientation @xmath68 by setting @xmath69 it follows that @xmath70 hence , @xmath71 . repeating this construction we obtain an orientation @xmath31 with @xmath72 . the general case where @xmath20 may be infinite can be proved using the compactness theorem for first - order logic . let @xmath73 be the elementary diagram of @xmath20 ( i.e. , the set of all first - order formulae with parameters that hold in @xmath20.08em ; see @xcite for details ) where we consider @xmath20 as a two - sorted structure @xmath8 with a binary incidence relation @xmath74 . we can write down a formula @xmath75 stating that @xmath28 is a function such that 1 . @xmath76 , for all @xmath77 2 . @xmath78 , for all @xmath79 by assumption and the first part of the proof , every finite subset of @xmath80 is satisfiable . therefore , according to the compactness theorem , there exists a model @xmath81 of @xmath80 . by the diagram lemma ( see , e.g. , @xcite ) , we can find an elementary embedding @xmath82 ( i.e. , an embedding preserving every first - order formula ) . since every edge of @xmath20 has only finitely many vertices it follows that @xmath83 hence , we can define the desired orientation of @xmath20 by @xmath84 . it turns out that the orientation obtained via the preceding lemma is @xmath2-definable . the following sequence of lemmas shows how we can encode such an orientation by a finite set of unary predicates . let @xmath19 be a directed graph and @xmath40 an undirected one . every orientation @xmath31 of @xmath40 induces an directed graph @xmath85 by orienting every edge @xmath86 of @xmath40 such that it points to the vertex @xmath87 . an _ @xmath20-orientation _ of @xmath40 consists of a pair @xmath88 where @xmath31 is an orientation of @xmath40 and @xmath89 is a homomorphism @xmath90 . we say that an @xmath20-orientation @xmath88 is _ bounded by @xmath0 _ if @xmath31 is bounded by @xmath0 . we say that a family @xmath91 of unary predicates _ encodes _ an @xmath20-orientation @xmath88 of @xmath40 if @xmath92 , for all @xmath93 . [ lem : encoding of h - orientations definable ] for every finite graph @xmath20 , there exists a first - order formula @xmath94 such that @xmath95 let @xmath96 be an enumeration of the vertices of @xmath20 . all @xmath94 has to say is that the @xmath97 form a partition of the vertices ( some @xmath97 may be empty ) and that there is no edge @xmath98 of @xmath40 such that @xmath99 , @xmath100 and @xmath101 is _ not _ an edge of @xmath20 . for every @xmath102 , there exists a finite loop - free directed graph @xmath103 with antisymmetric edge relation that has the following property . for every finite directed graph @xmath40 , with irreflexive and antisymetric edge relation and indegree at most @xmath0 , there exists a homomorphism @xmath104 . [ cor : existence of bounded orientations ] every @xmath0-sparse undirected graph has a @xmath103-orientation which is bounded by @xmath0 . in lemma [ lem : k - sparse hypergraphs and orientations ] , we have shown that such a graph @xmath39 has an orientation @xmath28 that is bounded by @xmath0 . it follows that @xmath85 has indegree at most @xmath0 . by the theorem , there exists a homomorphism @xmath105 . thus , @xmath88 is the desired @xmath103-orientation . [ lem : t - orientations definable ] for every @xmath102 , there exists a first - order formula @xmath106 such that @xmath107 note that the homomorphism @xmath89 of a @xmath103-orientation @xmath88 uniquely determines the orientation @xmath31 since the edge relation of @xmath103 is antisymmetric . in particular , the parameters @xmath108 encoding @xmath88 tell us whether @xmath31 is bounded by @xmath0 . hence , we can obtain @xmath106 by adding a check for boundedness to the formula @xmath109 of lemma [ lem : encoding of h - orientations definable ] . the class of all @xmath0-sparse undirected graphs is finitely @xmath2-axiomatisable . by lemma [ lem : k - sparse hypergraphs and orientations ] and corollary [ cor : existence of bounded orientations ] it follows that a graph @xmath40 is @xmath0-sparse if and only if it has a @xmath103-orientation that is bounded by @xmath0 . hence , we can use the formula @xmath110 where @xmath111 is the formula from lemma [ lem : t - orientations definable ] . in order to apply these results to hypergraphs we use the following construction associating a graph with every hypergraph . let @xmath19 be a hypergraph with orientation @xmath31 . we define a directed graph @xmath112 with edge relation @xmath113 let @xmath19 be a @xmath0-sparse hypergraph of rank @xmath21 where @xmath114 and @xmath115 . then @xmath20 has an orientation @xmath31 that is bounded by @xmath116 such that the edge relation of @xmath117 is antisymmetric . first , we consider the case that @xmath20 is finite . we call an element @xmath52 _ bad _ for an orientation @xmath31 of @xmath20 if there is some element @xmath54 such that @xmath117 contains both edges @xmath118 and @xmath119 . note that this implies that the vertex @xmath120 is also bad . we construct a sequence of orientations @xmath121 such that @xmath122 and the number of bad elements decreases at every step . we start with an arbitrary orientation @xmath123 bounded by @xmath0 . given an orientation @xmath124 with the above properties we construct a new orientation @xmath125 with fewer bad elements as follows . let @xmath59 be a bad element , set @xmath126 , and let @xmath127 since @xmath59 is bad we have @xmath128 note that every element of the form @xmath129 with @xmath130 is also bad since , by definition of @xmath131 , there is an edge @xmath132 with @xmath133 consequently , @xmath134 contains the edges @xmath119 ( since @xmath135 ) and @xmath118 ( since @xmath136 ) . it follows that @xmath137 we define the new orientation @xmath125 by @xmath138 then we have @xmath139 in particular , @xmath125 is bounded by @xmath116 . by construction , the element @xmath59 is not bad for @xmath125 . furthermore , if @xmath140 is an edge in @xmath141 with @xmath142 then this edge is induced by an edge @xmath86 in @xmath20 with @xmath143 . hence , @xmath140 is also an edge of @xmath134 . therefore , every element that is bad for @xmath125 is also bad for @xmath124 . it remains to prove the claim for infinite hypergraphs @xmath20 . let @xmath144 be the union of the elementary diagram of @xmath20 and formulae stating that @xmath31 is an orientation of @xmath20 that is bounded by @xmath116 and that @xmath117 has an antisymmetric edge relation . if @xmath145 is a model of @xmath144 then there exists an embedding @xmath146 and the desired orientation of @xmath20 can be obtained via @xmath89 from that of @xmath145 . hence , it is sufficient to show that @xmath144 is satisfiable . note that every finite subset @xmath147 is satisfiable since every finite substructure of @xmath20 has an orientation of the desired form . by the compactness theorem it follows that @xmath144 is satisfiable . while @xmath0-sparse hypergraphs are @xmath2-orientable there are hypergraphs without an @xmath2-definable orientation . for instance , the countably infinite clique is such a graph . in this section we will show that every hypergraph of bounded rank is at least @xmath1-orientable . a basic tool the proof below is based on is the notion of a spanning tree of a hypergraph . before presenting the rather involved definition for hypergraphs let us start with considering the simpler case of graphs . for a countable undirected graph @xmath40 we can define a _ depth - first spanning tree _ to be a spanning tree @xmath148 of @xmath40 where no edge of @xmath40 connects disjoint subtrees of @xmath148 ( see @xcite.08em ; in @xcite such trees are called _ normal _ ) . to generalise this definition to uncountable graphs we have to admit trees of arbitrary ordinal height . such trees are necessarily _ order trees , _ i.e. , partial orders @xmath149 where @xmath150 is a _ tree order , _ that is , a partial order such that any two elements have an infimum and , for every element @xmath59 , the set of all elements below @xmath59 is well - ordered . unfortunately , we can not in general hope to have a spanning subgraph that is an order tree , since the partial order @xmath150 requires too many edges . therefore , we will use a hybrid between an ordinary tree and an order tree . the precise definition of a spanning tree @xmath148 of a graph @xmath40 is as follows . instead of requiring @xmath148 to be a subgraph of @xmath40 we consider trees @xmath148 such that 1 . for every vertex @xmath151 of @xmath148 with immediate predecessor @xmath152 , the edge @xmath153 belongs to @xmath40 , and 2 . for every vertex @xmath151 of @xmath148 without immediate predecessor , we can fix an increasing chain @xmath154 of predecessors of @xmath151 with limit @xmath151 and a family @xmath155 of paths from @xmath151 to @xmath156 . hence , every vertex @xmath151 of @xmath148 is attached to its predecessors via some auxiliary graph @xmath157 that is either a single edge or a tree with root @xmath151 whose leaves form an increasing sequence of predecessors of @xmath151 with limit @xmath151 . consider the complete graph @xmath158 , for an uncountable cardinal @xmath159 . we can enumerate the vertices of @xmath158 as @xmath160 where the index @xmath161 ranges over all ordinals less than @xmath159 . as depth - first spanning tree of this graph we can use a chain of length @xmath159 as follows . we set @xmath162 where @xmath163 is the set of all vertices and @xmath164 the first part of @xmath10 consists of the successor edges , whereas the second part contains the auxiliary graphs @xmath165 attaching a limit vertex @xmath166 to its predecessors . to generalise these ideas to hypergraphs we need a suitable replacement for the trees @xmath157 . unfortunately , not every hypergraph has a spanning tree . a typical example is the hypergraph instead , we will use certain tree - like hypergraphs called _ priority trees . _ let @xmath19 be a hypergraph . a _ hyperpath _ in @xmath20 is a sequence @xmath167 of edges such that @xmath168 if @xmath169 and @xmath170 then we say that the hyperpath _ connects @xmath171 and @xmath152 . _ let @xmath19 be a hypergraph of rank at most @xmath21 , @xmath172 be a subhypergraph of @xmath20 with @xmath173 , and suppose that there are partitions @xmath174 ( a)suppose that @xmath175 with @xmath176 and @xmath177 . we define by induction when such a tuple @xmath148 is a _ priority tree . _ the element @xmath152 is called the _ root _ of @xmath148 and @xmath178 is its set of _ leaf edges . _ we start the induction with the case where @xmath6 consists of a single hyperpath @xmath167 with @xmath179 , we have @xmath180 , @xmath181 , and @xmath182 . then @xmath148 is a priority tree . we also call @xmath148 a priority tree if it can be obtained from a priority tree @xmath183 with the same root @xmath152 by adding a hyperpath @xmath167 such that @xmath184 where @xmath0 is the minimal index such that @xmath185 . this is the successor case of the induction step . finally , we also have a limit case . suppose that @xmath186 is an increasing chain of priority trees . that is , the sequences @xmath187 , @xmath188 , @xmath189 , @xmath190 , and @xmath191 are all increasing , and all trees @xmath192 have the same root @xmath152 . then @xmath148 is a priority tree if it is the union of this chain , that is , if @xmath193 ( b)a _ branch _ of @xmath148 is a hyperpath @xmath194 satisfying the following conditions.08em : 1 . @xmath195 contains the root @xmath152 of @xmath196 . 2 . let @xmath197 be the index such that @xmath198 . we have @xmath199 , for every @xmath200 . furthermore , if @xmath201 then @xmath202 ( c)with each priority tree @xmath148 we associate two relations , an order @xmath150 on @xmath6 defined by @xmath203 and an equivalence relation @xmath204 on @xmath196 defined by @xmath205 consider the following priority tree with edges @xmath206 where we have labelled each vertex in @xmath207 by the index @xmath208 . the edge colours are given by @xmath209 , @xmath210 , @xmath211 . the ordering @xmath150 is displayed to the right . recall that a _ tree order _ is a partial order such that any two elements have an infimum and , for every element @xmath59 , the set of all elements below @xmath59 is well - ordered . a _ preorder _ is a reflexive and transitive relation . every preorder @xmath212 induces an equivalence relation @xmath213 . the equivalence classes of this relation are called _ @xmath212-classes . _ let @xmath148 be a priority tree . the order @xmath150 on the edges is a tree order . the proof consists of a straightforward but tedious induction following the construction of priority trees . [ lem : existence of priority trees ] let @xmath19 be a connected hypergraph of rank at most @xmath21 . for each vertex @xmath93 and every set @xmath214 of edges , there exists a priority tree @xmath215 with root @xmath152 such that @xmath216 and @xmath217 . for every @xmath1-formula @xmath218 ( possibly with parameters ) , there exists a @xmath1-formula @xmath219 ( with parameters ) such that , if @xmath220 defines a well - order on @xmath221 and @xmath148 is a priority tree as in then @xmath219 defines a linear order on @xmath196 . ( a)let @xmath222 be an enumeration of @xmath221 . for every @xmath223 , we fix a hyperpath @xmath224 connecting @xmath152 with @xmath225 . we construct @xmath148 by induction on @xmath208 . we start with the hyperpath @xmath226 . at step @xmath227 we determine the shortest suffix @xmath228 of the path @xmath229 that meets the tree constructed so far and we add this suffix to the tree . ( if @xmath230 we leave the tree unchanged . ) we choose the least index @xmath0 with @xmath231 and we put the new edges into @xmath232 and the new vertices into @xmath233 . the limit of this construction is the desired priority tree . ( b)the equivalence relation @xmath204 associated with @xmath148 is @xmath2-definable in @xmath20 with the help of the parameters @xmath196 , @xmath6 , @xmath234 , and @xmath207 . we denote the @xmath204-class of a vertex @xmath171 by @xmath235 $ ] . note that , by construction of @xmath148 , @xmath235 $ ] is a hyperpath and @xmath235 $ ] contains a unique leaf edge which , furthermore , is one of the ends of the hyperpath . we denote by @xmath236 the suffix of the hyperpath @xmath235 $ ] that connects @xmath171 to the leaf edge in @xmath235 $ ] . to define the desired order on @xmath196 we first construct a preorder on @xmath196 by setting @xmath237 if and only if one of the following conditions is satisfied.08em : 1 . @xmath238 and @xmath239 , for @xmath240 . 2 . @xmath241 and the leaf edge in @xmath242 $ ] is @xmath220-smaller than the leaf edge in @xmath243 $ ] . @xmath241 , @xmath242 = [ y]$ ] , and @xmath244 . note that we have @xmath237 and @xmath245 if and only if @xmath246 . in this case @xmath247 and @xmath248 belong to the same edge @xmath249 . hence , every @xmath212-class has size at most @xmath21 . adding @xmath21 additional unary predicates @xmath250 such that each @xmath251 contains at most one element of each @xmath212-class , we can define @xmath252 -- we have seen that every @xmath0-sparse graph has an @xmath2-definable orientation @xmath31 that is bounded by @xmath0 . if we want to encode sets of edges via sets of vertices we can try to encode each edge @xmath86 by a pair @xmath253 consisting of the vertex @xmath254 and a number @xmath240 . this idea requires a way to linearly order the sets @xmath255 . in @xcite courcelle uses depth - first spanning trees to obtain such linear orders . as remarked above one needs to adapt the definition of a depth - first spanning tree when one tries to extend these results to uncountable hypergraphs . let @xmath19 be a hypergraph of rank @xmath21 and suppose that @xmath256 is a structure where @xmath149 is a tree ( of ordinal height ) with @xmath257 and with every vertex @xmath177 we associate a set @xmath258 of edges . we assume that @xmath259 , for @xmath260 . ( a)the set of _ auxiliary nodes _ associated to a vertex @xmath177 is @xmath261 ( b)for @xmath48 , we define @xmath262 ( c)@xmath148 is a _ depth - first spanning tree _ of @xmath20 if it satisfies the following conditions.08em : 1 . for all @xmath260 , @xmath263 and @xmath264 . 2 . for each edge @xmath12 the set @xmath265 is nonempty and linearly ordered by @xmath150 . the vertices @xmath177 are partitioned into the following classes.08em : @xmath266 the root.08em ; @xmath267 a successor.08em ; @xmath268 a limit.08em ; where the successor and limit vertices are subdivided into @xmath21 subclasses . this partition satisfies the following conditions.08em : 1 . if @xmath152 has type @xmath57 then it is the root of @xmath196 and @xmath269 . if @xmath152 has type @xmath270 then it is the ( immediate ) successor of some vertex @xmath271 . we have @xmath272 with @xmath14 . futhermore , @xmath152 is the only vertex in @xmath265 of type @xmath270 and @xmath273 . if @xmath152 has type @xmath274 then it is the limit of an increasing sequence @xmath275 of vertices @xmath276 . @xmath277 is ( the set of edges of ) a priority tree with root @xmath152 . furthermore , @xmath278 is a cofinal subset of @xmath279 and @xmath152 is the only vertex in @xmath280 with type @xmath274 . every connected hypergraph @xmath20 has a depth - first spanning tree . if in the definition of a depth - first spanning tree we drop the condition that @xmath281 , for every edge @xmath86 , then we obtain a structure that we call a _ partial _ depth - first spanning tree . we construct an increasing sequence @xmath282 of such partial depth - first spanning trees with the property that , for every connected component @xmath26 of @xmath283 , the set @xmath284 is linearly ordered by @xmath150 . ( a _ connected component _ of @xmath285 is a maximal subset @xmath286 such that the subhypergraph @xmath287 is connected . ) the limit of this sequence will be the desired depth - first spanning tree of @xmath20 . we start by choosing an arbitrary element @xmath93 and setting @xmath288 and @xmath289 . for limit ordinals @xmath290 , we define @xmath291 . for the successor step , suppose that we have already defined @xmath292 . fix some connected component @xmath26 of @xmath285 . note that @xmath293 is nonempty since @xmath20 is connected . we distinguish two cases . 1 . if @xmath293 has a maximal element @xmath171 then we choose some edge @xmath86 with @xmath294 and @xmath295 , and we fix some vertex @xmath296 . we add @xmath152 to @xmath292 as immediate successor of @xmath171 and we set @xmath297 . it follows that @xmath298 . since @xmath299 contains at most @xmath300 vertices there is some @xmath301 such that @xmath299 contains no vertex of type @xmath270 . hence , in the new tree @xmath302 we can assign the type @xmath270 to @xmath152 . 2 . suppose that @xmath293 has no maximal element . we choose a sequence @xmath303 of edges with @xmath304 such that the sequence @xmath275 defined by @xmath305 is increasing and cofinal in @xmath293 . by taking a suitable subsequence we may assume that the set of types appearing in @xmath306 is the same for every @xmath307 . + for each edge @xmath308 , choose some edge @xmath309 with @xmath310 and set @xmath311 . we select a vertex @xmath312 and a priority tree @xmath313 such that @xmath314 . we define @xmath315 where @xmath152 is the limit of @xmath293 and we set @xmath316 . it follows that @xmath317 . it remains to show that the constructed tree @xmath302 is a partial depth - first tree where all sets @xmath318 are linearly ordered . we start by showing that each set @xmath319 with @xmath12 is linearly ordered . if @xmath320 then @xmath321 and we are done . otherwise , we have @xmath322 . note that @xmath323 implies @xmath295 . therefore , we have @xmath324 . since @xmath152 is larger than every element in @xmath293 the claim follows . let @xmath325 be a connected component of @xmath326 . we have to show that @xmath327 is linearly ordered . since @xmath328 there is some connected component @xmath329 of @xmath285 containing @xmath325 . if @xmath330 then @xmath331 implies that @xmath332 and the set @xmath333 is linearly ordered . if , on the other hand , @xmath334 then we have @xmath335 and the latter set is linearly ordered since @xmath152 is greater than every element of @xmath293 . ( a)if the hypergraph @xmath20 is countable then we can actually obtain a depth - first spanning tree of height at most @xmath336 as follows . in the above proof , if we are slightly more careful in choosing the vertex @xmath152 that is added to the partial tree , then we can ensure that every vertex is chosen already after finitely many steps . ( b)note that , strictly speaking , the above proposition is not a generalisation of theorem 1.4 of @xcite since we use a different notion of a depth - first spanning tree . we use depth - first spanning trees to encode orientations of a hypergraph . first , we show that each depth - first spanning tree can be encoded by finitely many @xmath1-parameters . [ lem : d - f spanning trees definable ] for every @xmath337 we can construct @xmath2-formulae @xmath338 , @xmath339 , and @xmath340 such that , for every connected hypergraph @xmath20 of rank at most @xmath21 and each depth - first spanning tree @xmath341 of @xmath20 , there are @xmath1-parameters @xmath342 such that @xmath343 we will use the following parameters.08em : 1 . unary predicates @xmath344 , @xmath345 , @xmath346 , for @xmath301 , containing all vertices of the corresponding type . 2 . @xmath347 , for every type @xmath348 . 3 . @xmath349 . @xmath350 , for every type @xmath348 . if @xmath152 is of type @xmath274 then the set @xmath277 forms a priority tree . we use additional parameters @xmath351 and @xmath207 , for @xmath200 , encoding the corresponding partition @xmath352 hence , for every vertex @xmath152 of type @xmath274 we have a priority tree @xmath353 ( 1)first , we construct the formula @xmath220 . to simplify our task we define separate formulae @xmath354 , for each type @xmath348 , such that @xmath355 then we can set @xmath356 . if @xmath152 has type @xmath57 then @xmath357 and we can set @xmath358 if the type of @xmath152 is @xmath270 then @xmath272 and @xmath359 , where @xmath86 is the unique edge in @xmath360 containing @xmath152 . hence , we can define @xmath361\,.\end{aligned}\ ] ] finally , if @xmath152 has type @xmath274 then @xmath362 is the least subset of @xmath363 satisfying the following conditions.08em : 1 . @xmath364 2 . if @xmath365 and @xmath366 then @xmath367 . hence , we can define @xmath368\,.\end{aligned}\ ] ] ( 2)next , we define a formula @xmath369 such that @xmath370 the formula @xmath369 should state that @xmath371 using the formulae @xmath372 we can write @xmath161 as @xmath373\,.\end{aligned}\ ] ] ( 3)with the help of @xmath161 we can write down the desired formulae @xmath75 and @xmath374 . @xmath375\ , , \\ \chi(x , y ) & : = \forall x[\varphi(x ) \lso ( xy \lso xx)]\,.\end{aligned}\ ] ] -- [ lem : preorder on auxiliary sets ] we can construct @xmath1-formulae @xmath376 , for @xmath337 , such that , for every depth - first spanning tree @xmath341 of a connected hypergraph @xmath20 of rank @xmath21 , there are @xmath1-parameters @xmath342 such that the formula @xmath377 defines a preorder @xmath378 with the following properties.08em : 1 . the restriction of @xmath378 to @xmath196 coincides with @xmath150 . 2 . @xmath378 linearly preorderes every set @xmath48 such that @xmath379 is linearly ordered by @xmath150 . each @xmath378-class has at most @xmath21 elements . let @xmath380 and @xmath381 be the formulae of lemma [ lem : d - f spanning trees definable ] . for each type @xmath348 , we define a formula @xmath382 linearly preordering each set @xmath362 where @xmath152 is of type @xmath348 . then the desired formula @xmath383 states that either 1 . @xmath384 and @xmath385 for @xmath386 , or 2 . @xmath387 , for some @xmath152 of type @xmath348 , and @xmath382 holds . if @xmath152 is of type @xmath57 or @xmath270 then @xmath362 contains at most @xmath21 elements and we can set @xmath388 for vertices of type @xmath274 we can use the formula from lemma [ lem : existence of priority trees ] . [ cor : order on auxiliary sets ] we can construct @xmath1-formulae @xmath376 , for @xmath337 , such that , for every depth - first spanning tree @xmath341 of a connected hypergraph @xmath20 of rank @xmath21 , there are @xmath1-parameters @xmath342 such that the formula @xmath377 defines a partial order @xmath212 with the following properties.08em : 1 . the restriction of @xmath212 to @xmath196 coincides with @xmath150 . 2 . @xmath212 linearly orderes every set @xmath48 such that @xmath379 is linearly ordered by @xmath150 . let @xmath378 be the preorder from lemma [ lem : preorder on auxiliary sets ] . since every @xmath378-class contains at most @xmath21 elements we can add @xmath21 new unary predicates @xmath389 such that @xmath390 and we have @xmath391 , for each @xmath378-class @xmath131 and all @xmath208 . then we can define @xmath392 -- we can construct @xmath1-formulae @xmath393 , for @xmath337 , such that for every hypergraph @xmath20 of rank @xmath21 , there are @xmath1-parameters @xmath342 such that , the formula @xmath394 defines an orientation of @xmath20 . suppose that @xmath20 has @xmath159 connected components @xmath395 , @xmath396 . for each component @xmath395 we fix a depth - first spanning tree @xmath397 . let @xmath398 be the parameters from lemma [ lem : d - f spanning trees definable ] and lemma [ lem : preorder on auxiliary sets ] . for every edge @xmath12 , there exists a unique component @xmath395 such that the intersection @xmath399 is finite and nonempty . furthermore , the set @xmath400 is linearly ordered by @xmath150 . using the ordering @xmath212 of corollary [ cor : order on auxiliary sets ] we can write down a formula @xmath401 stating that @xmath152 is the @xmath212-least element of this set @xmath131 . every hypergraph of rank @xmath337 is @xmath1-orientable . let us mention the following consequences of this result . for countable hypergraphs they are again due to courcelle @xcite . a formula @xmath402 defines an _ edge ordering _ of a hypergraph @xmath19 if , for every edge @xmath12 , the formula @xmath403 defines a linear ordering on the vertices of @xmath86 . a formula @xmath404 defines an _ neighbourhood ordering _ of a directed graph @xmath39 if , for every vertex @xmath93 , the formula @xmath405 defines a linear ordering on the set @xmath406 . [ lem : edge ordering ] there exist @xmath1-formulae @xmath407 , for @xmath337 , such that , for every hypergraph @xmath19 of rank @xmath21 , there are @xmath1-parameters @xmath342 such that the formula @xmath408 defines an edge ordering of @xmath20 . [ lem : neighbourhood ordering ] there exist @xmath2-formulae @xmath409 , for @xmath337 , such that , for every directed graph @xmath40 of indegree at most @xmath21 , there are @xmath2-parameters @xmath108 such that the formula @xmath410 defines a neighbourhood ordering of @xmath40 . we can apply lemma [ lem : edge ordering ] to the hypergraph @xmath411 where @xmath412 note that every subset @xmath413 can be encoded by the set @xmath414 hence , every @xmath1-formula over @xmath20 can be translated into an @xmath2-formula over @xmath40 . in @xcite courcelle has shown that we can translate every @xmath1-formula @xmath75 into an @xmath2-formula @xmath415 that is equivalent to @xmath75 on all countable @xmath0-sparse hypergraphs . using the results of the previous sections we can lift the restriction to countable hypergraphs . the proof in @xcite goes through unchanged since it relies only on the statements of lemma [ lem : edge ordering ] and lemma [ lem : neighbourhood ordering ] , and on local modifications of hypergraphs . for all numbers @xmath416 , there exists a monadic second - order interpretation ( with monadic parameters ) that maps a @xmath0-sparse hypergraph of rank @xmath21 to its incidence structure . for all @xmath416 and all formulae @xmath417 with first - order variables @xmath418 , monadic variables @xmath419 , and guarded second - order variables @xmath420 , there exists a formula @xmath421 with the following property.08em : for all @xmath0-sparse hypergraphs @xmath19 of rank @xmath21 and all parameters @xmath422 , @xmath35 , @xmath423 , there exist parameters @xmath424 such that @xmath425 the results so far concern ways to encode edges by vertices . in this last section we consider a more general problem . let @xmath39 be a graph . we denote by @xmath426 the set of all finite subsets of @xmath9 . we would like to encode a given subset @xmath427 by a set of vertices , that is , we would like to find a definable function @xmath428 that is injective . for @xmath429 this reduces to the problem considered in the preceding sections . for arbitrary @xmath6 , such a function @xmath89 does not always exist . but we will show that sometimes we can transform a given function @xmath430 into an injective one . these results are inspired by work of colcombet and lding @xcite on set interpretations . colcombet and lding consider a power set operation @xmath431 on structures . one of their main results in a commutation theorem for interpretations and the power set operation . they show that , given a tree @xmath148 and an @xmath432-interpretation @xmath433 such that @xmath434 is of the form @xmath435 , for some structure @xmath145 , then there exists a @xmath18-interpretation @xmath436 such that @xmath437 . on ingredient in the proof of this result is a method to encode , in a definable way , finite subsets of the tree @xmath148 by single vertices . suppose we are given a function @xmath430 that we want to transform into an injective function @xmath428 . let @xmath438 . the first step in the construction of @xmath89 consists in finding a definable function @xmath439 such that @xmath440 , for all @xmath152 . of course , this is not always possible . for instance , if the graph is finite and we have @xmath441 , for all vertices @xmath152 . therefore , we consider only functions @xmath290 that are _ sparse _ in the sense of the following definition . let @xmath39 be an undirected graph . the _ border _ of a subset @xmath442 is the set @xmath443 of all edges connecting a vertex in @xmath444 with a vertex outside of @xmath444 . a _ distribution _ of @xmath40 is a map @xmath445 . for @xmath48 , we define the shorthand @xmath446 let @xmath447 be an arbitrary mapping . the distribution _ induced _ by @xmath89 is the function @xmath445 with @xmath448 a distribution @xmath290 is _ @xmath0-sparse _ if @xmath449 given a @xmath0-sparse distribution @xmath290 we will construct the desired function @xmath439 by solving a network flow problem . let @xmath39 be an undirected graph . a _ flow _ of @xmath40 is a function @xmath450 such that , for all @xmath451 , 1 . @xmath452 and 2 . @xmath453 implies @xmath454 . a flow @xmath31 is _ acyclic _ if there is no cycle @xmath455 of @xmath40 such that @xmath456 and @xmath457 , for all @xmath200 . the _ defect _ of a flow @xmath31 is the distribution @xmath458 a flow @xmath31 is a _ @xmath290-flow _ if , for every @xmath93 , either @xmath459 a flow @xmath31 is _ edge - bounded _ by @xmath0 if @xmath460 , for all @xmath451 . we call @xmath31 _ vertex - bounded _ by @xmath0 if @xmath461 our aim is to show that , for every @xmath0-sparse distribution @xmath290 there is a bounded @xmath290-flow @xmath31 and a function @xmath439 inducing @xmath290 . furthermore , if @xmath290 is definable then @xmath68 should also be definable . let @xmath178 be a logic . a distribution @xmath290 is _ @xmath178-definable _ if there exist formulae @xmath462 , @xmath240 , such that @xmath463 similarly , a flow @xmath31 is _ @xmath178-definable _ if there exist formulae @xmath464 such that @xmath465 note that every edge - bounded flow can be encoded with the help of the @xmath1-parameters @xmath466 for trees the problem of encoding sets by vertices has been solved by colcombet and lding @xcite . in the general case proved below the function @xmath68 is only definable with the help of @xmath1-parameters , but for trees we can do without them . let @xmath467 be an infinite directed tree and @xmath290 a @xmath18-definable @xmath0-sparse distribution of @xmath148 . there exists a @xmath18-definable flow @xmath31 that is edge - bounded by @xmath468 and satisfies @xmath469 , for all @xmath152 . let @xmath467 be a directed tree and @xmath290 a @xmath18-definable @xmath0-sparse distribution of @xmath148 such that @xmath470 . there exists @xmath18-definable function @xmath471 such that @xmath290 is the distribution induced by @xmath68 . to prove our generalisation of these results we start with a few lemmas about bounded flows the first two follow immediately from the definitions . every flow that is vertex - bounded by @xmath0 is also edge - bounded by @xmath0 . suppose that @xmath40 is a graph with maximal degree @xmath472 . every flow of @xmath40 that is edge - bounded by @xmath0 is vertex - bounded by @xmath473 . for every @xmath290-flow @xmath31 there exists an acyclic @xmath290-flow @xmath474 such that , if @xmath31 is edge - bounded by @xmath0 or vertex - bounded by @xmath0 then so is @xmath474 . we repeat the following construction until the flow is acyclic . select a cycle @xmath455 such that @xmath475 we define @xmath474 by @xmath476 -- [ prop : existence of flows ] let @xmath39 be an undirected graph and @xmath290 a @xmath0-sparse distribution . then @xmath40 has a @xmath290-flow @xmath31 that is edge - bounded by @xmath0 . first , we assume that @xmath40 is finite . in this case we can reduce the task to a network flow problem . let @xmath20 be the graph obtained from @xmath40 by adding two new vertices @xmath477 and @xmath478 that are connected to every vertex of @xmath40 . we define the capacity @xmath479 of edges @xmath86 of @xmath20 as follows . for edges @xmath86 of @xmath40 we set @xmath480 . if @xmath481 with @xmath93 we set @xmath482 . finally , if @xmath483 with @xmath93 we define @xmath484 let @xmath31 be a maximal flow from @xmath477 to @xmath478 with respect to @xmath485 . we claim that its restriction to the edges of @xmath40 is the desired flow . according to the max - flow min - cut theorem , there is a set @xmath131 of vertices containing @xmath477 but not @xmath478 such that the maximal flow @xmath21 from @xmath477 to @xmath478 equals @xmath486 let @xmath487 and @xmath488 . since @xmath489 we have @xmath490 on the other hand , for the set @xmath491 , we have @xmath492 consequently , the maximal flow @xmath21 from @xmath477 to @xmath478 equals @xmath493 this implies that @xmath494 for each @xmath93 , we therefore have @xmath495 if @xmath496 this implies @xmath497 while , for @xmath498 , we have @xmath499 hence , either @xmath500 or @xmath501 . it remains to prove the lemma for infinite graphs . let @xmath502 consist of the elementary diagram of @xmath40 together with first - order formulae stating that @xmath31 is a @xmath290-flow on @xmath40 that is edge - bounded by @xmath0 . we will use the compactness theorem to show that @xmath502 is satisfiable . let @xmath503 be finite . there exists a finite induced subgraph @xmath504 of @xmath40 such that @xmath505 . let @xmath506 be an enumeration ( without repetitions ) of all edges @xmath507 with @xmath508 and @xmath509 . we construct a new graph @xmath510 by attaching to each vertex @xmath156 a path @xmath207 of length @xmath0 . let @xmath511 be the distribution on @xmath512 with @xmath513 , for @xmath514 , and @xmath515 , for @xmath516 . in order to show that @xmath517 is satisfiable it is sufficient to prove that @xmath512 has a flow of the desired form . consider an arbitrary set @xmath518 of vertices . let @xmath519 it follows that @xmath520 by the first part of the proof it follows that @xmath512 has a flow of the desired form . it remains to show how we can use the @xmath290-flow @xmath31 we have just constructed to define the desired function @xmath439 . we start by selecting a certain family of definable paths . note that we allow paths of length @xmath57 . such paths are uniquely determined by the vertex they start ( and end ) at . [ lem : paths coding a flow ] let @xmath40 be a countable undirected graph and @xmath31 an acyclic @xmath290-flow of @xmath40 . there exists a set @xmath431 of finite paths through @xmath40 satisfying the following conditions.08em : for every @xmath93 , there are exactly @xmath521 paths in @xmath431 starting at @xmath152 . for every @xmath93 there is at most one path in @xmath431 ending at @xmath152 . for every pair @xmath451 of vertices there are at most @xmath522 paths in @xmath431 containing the edge @xmath523 ( in this direction ) . fix an enumeration @xmath524 of the set @xmath525 for @xmath526 , we construct paths @xmath527 with the following properties.08em : 1 . @xmath527 starts at @xmath528 . if @xmath529 then the endpoints of @xmath530 and @xmath527 are different . 3 . for every edge @xmath523 there are at most @xmath522 paths @xmath527 containing the edge @xmath523 . by induction , suppose that we have already defined @xmath229 , for @xmath531 . let 1 . @xmath532 be the number of paths @xmath229 , @xmath531 , starting at @xmath152 , 2 . @xmath533 the number of paths @xmath229 , @xmath531 , ending at @xmath152 , and 3 . @xmath534 the number of paths @xmath229 , @xmath531 , containing the edge @xmath523 . we construct a path @xmath535 inductively starting with @xmath536 . for the induction step , suppose that we have already defined @xmath537 . if @xmath538 then we stop and set @xmath539 . otherwise , we claim that there is some neighbour @xmath151 of @xmath156 with @xmath540 . hence , we can set @xmath541 . to prove the claim , we distinguish two cases . if @xmath542 then @xmath543 implies that @xmath544 as desired . similarly , if @xmath227 then @xmath545 implies that @xmath546 note that the construction of @xmath527 must terminate after at most @xmath547 steps since the flow @xmath31 is acyclic and there are only @xmath45 vertices @xmath171 with @xmath548 . [ lem : encoding a set of paths ] there exist @xmath1-formulae @xmath549 , for @xmath337 , such that , for every graph @xmath40 and each set @xmath431 of finite paths such that every vertex and every edge of @xmath40 is contained in at most @xmath21 paths of @xmath431 , there exists a tuple @xmath342 of @xmath1-parameters such that @xmath550 for every edge @xmath523 of @xmath40 we fix a bijection @xmath551 \to \calp_e$ ] where @xmath552 is the set of all paths containing the edge @xmath523 ( in either direction ) and @xmath553 . we assume that @xmath554 . let @xmath555 be the set of all edges of @xmath40 contained in some path in @xmath431 . by lemma [ lem : neighbourhood ordering ] there exists an @xmath2-formula @xmath556 with parameters @xmath557 such that , for every @xmath93 , the formula @xmath558 linearly orders the set of all vertices that are connected to @xmath152 via an edge in @xmath555 . finally , we define unary predicates @xmath559 containing all vertices @xmath152 such that there exists a path @xmath560 containing edges @xmath523 , @xmath561 where 1 . @xmath562 , @xmath563 , 2 . @xmath171 is the @xmath208-th neighbour of @xmath152 ( in the order defined by @xmath374 ) , 3 . @xmath151 is the @xmath564-th neighbour of @xmath152 . it follows that a nonempty set @xmath565 of edges is a path in @xmath431 if and only if @xmath566 is a minimal nonempty subset of @xmath10 satisfying the following condition.08em : 1 . @xmath566 can be written as a union @xmath567 such that , for all vertices @xmath568 such that @xmath569 and @xmath171 and @xmath151 are , respectively , the @xmath208-th and @xmath564-th neighbour of @xmath152 , we have @xmath570 . this condition can be expressed in @xmath1 . note that the set of empty paths in @xmath431 is trivially definable with the help of the parameter @xmath571 using the family @xmath431 we can construct a formula @xmath75 defining the function @xmath68 . [ prop : function from a flow ] there exist @xmath1-formulae @xmath572 , for @xmath337 , with the following property.08em : for every graph @xmath39 and each acyclic @xmath290-flow @xmath31 of @xmath40 that is vertex - bounded by @xmath21 , there exist @xmath1-parameters @xmath342 such that @xmath573 defines on @xmath40 a partial function @xmath439 with @xmath574 let @xmath575 be the graph obtained from @xmath40 by removing every edge @xmath523 with @xmath576 . note that @xmath31 is also a @xmath290-flow of @xmath575 . since @xmath31 is vertex - bounded by @xmath21 it follows that every vertex of @xmath575 has degree at most @xmath337 . consequently , each connected component @xmath577 of @xmath40 is countable . let @xmath578 be the set of paths obtained by applying lemma [ lem : paths coding a flow ] to the restriction of @xmath31 to @xmath577 , and let @xmath431 be the union of all these sets @xmath578 corresponding to the connected components of @xmath575 . by lemma [ lem : encoding a set of paths ] , there exists a formula @xmath579 and a set @xmath342 of guarded relations such that @xmath580 with the help of @xmath415 we can define a partial function @xmath439 such that @xmath581 by construction of @xmath431 we have @xmath440 , for every @xmath93 . let @xmath39 be a graph of finite degree and @xmath582 a @xmath1-formula that defines a partial function @xmath583 such that the distribution @xmath290 induced by @xmath89 is @xmath0-sparse . suppose that there exists a @xmath1-formula @xmath584 such that , for every vertex @xmath93 , @xmath585 linearly orders the set @xmath586 . then there exist @xmath2-definable partial functions @xmath587 and @xmath439 such that @xmath588 and @xmath589 is injective . by proposition [ prop : existence of flows ] there exists a @xmath290-flow @xmath31 that is edge - bounded by @xmath0 . since @xmath40 has finite degree it follows that @xmath31 is vertex - bounded by some constant @xmath337 . hence , we can use proposition [ prop : function from a flow ] to find a definable function @xmath439 with @xmath590 . choose unary predicates @xmath591 such that we have @xmath592 whenever @xmath593 and @xmath594 are distinct vertices with @xmath595 . using these predicate we can define partial functions @xmath596 such that @xmath597 is the unique element of @xmath598 . we define @xmath587 by @xmath599 where the index @xmath208 is chosen such that @xmath131 is the @xmath208-th element of @xmath600 ( in the order defined by @xmath374 ) . it follows that @xmath601 and @xmath589 is injective . furthermore , the function @xmath589 is clearly @xmath1-definable . since the graph @xmath40 has degree at most @xmath0 it is @xmath0-sparse . hence , every @xmath1-definable function is already @xmath2-definable . recall that @xmath426 denotes the set of all finite subsets of @xmath9 . combining the preceding lemmas we obtain the main result of this section . [ thm : coding sets by vertices ] let @xmath39 be a graph of finite degree and @xmath582 a @xmath1-formula that defines a partial function @xmath602 such that the distribution @xmath290 induced by @xmath89 is @xmath0-sparse . then there exist @xmath2-definable partial functions @xmath603 and @xmath439 such that @xmath588 and @xmath589 is injective . by the preceding lemma it is sufficient to construct a formua @xmath584 ( with @xmath1-parameters ) such that @xmath585 linearly orders @xmath586 , for every @xmath93 . let @xmath604 be a spanning forest of @xmath40 and let @xmath605 be a set containing exactly one element of each connected component . using the parameters @xmath566 and @xmath344 we can define the tree ordering on @xmath9 by @xmath606 let @xmath607 be the set obtained from @xmath344 by orienting the edges according to this ordering . then @xmath196 is a directed forest . furthermore , since the degree of @xmath40 is bounded we can use lemma [ lem : neighbourhood ordering ] to linearly order the successors of every vertex in @xmath196 . we use these two orderings to define the lexicographic ordering @xmath608 on @xmath196 . finally , we obtain the desired ordering on @xmath426 by setting @xmath609 each of these definitions can be expressed in @xmath1 . we have presented several methods to encode sets of finite vertices as single vertices . in the first part , we used depth - first spanning trees to encode edges by vertices . as an application we were able to extend courcelle s result on the collapse of @xmath1 to @xmath2 on sparse hypergraphs from countable hypergraphs to hypergraphs of arbitrary cardinality . in the second part we used network flows to encode arbitrary finite sets by vertices . let us mention some open questions . considering the first part it would be interesting to find out whether sparse classes are the only examples where @xmath1 collapses to @xmath2 . is there a class @xmath26 that is not @xmath0-sparse , for any @xmath0 , such that over @xmath26 every @xmath1-sentence is equivalent to an @xmath2-sentence.08em ? the results of the second part are much less complete . it is unlikely that they are the best possible . improve theorem [ thm : coding sets by vertices ] by allowing more general classes of graphs or hypergraphs.08em ; more general classes of partial functions @xmath583 . our results were inspired by work of colcombet and lding @xcite . the question arises of whether we can also generalise the remaining results of that article . can we prove corollary 4.4 of for other graphs than trees.08em ? i like to thank bruno courcelle for his many comments on earlier versions of this paper .

we determine the large cardinal consistency strength of the existence of a-supercompact cardinal such that fails at . indeed , we show that the existence of a-supercompact cardinal such that is equiconsistent with the existence of a-supercompact cardinal that is also-tall . we also prove some basic facts about the large cardinal notion of tallness with closure .

1110.5365

woodin showed that the existence of a measurable cardinal at which @xmath2 fails is equiconsistent with the existence of a cardinal @xmath1 that is @xmath5-tall ( see @xcite , @xcite , or @xcite ) , where a cardinal @xmath1 is _ @xmath4-tall _ if there is a nontrivial elementary embedding @xmath6 with critical point @xmath1 such that @xmath7 and @xmath8 in @xmath9 . in this paper we extend woodin s result into the realm of partially supercompact cardinals . since @xmath1 is measurable if and only if @xmath1 is @xmath1-supercompact , one immediately sees several natural ways of doing this . let us consider the following questions for cardinals @xmath1 , @xmath0 , and @xmath4 . 1 . what is the strength of the hypothesis that @xmath1 is @xmath0-supercompact and @xmath2 fails at @xmath1 ? 2 . what is the strength of the hypothesis that @xmath1 is @xmath0-supercompact and @xmath2 fails at @xmath1 with @xmath10 ? 3 . what is the strength of the hypothesis that @xmath1 is @xmath0-supercompact and @xmath2 fails at @xmath0 ? 4 . what is the strength of the hypothesis that @xmath1 is @xmath0-supercompact and @xmath2 fails at @xmath0 with @xmath3 ? we note that woodin s theorem answers question ( 1 ) in the case that @xmath11 the following theorem , together with woodin s result , provides complete answers to questions ( 1 ) - ( 4 ) . suppose @xmath0 and @xmath4 are cardinals . 1 . for @xmath12 , the existence of a @xmath0-supercompact cardinal @xmath1 such that @xmath2 fails at @xmath1 is equiconsistent with the existence of a @xmath0-supercompact cardinal . 2 . the existence of a @xmath0-supercompact cardinal @xmath1 such that @xmath10 is equiconsistent with the existence of a @xmath0-supercompact cardinal that is also @xmath4-tall . 3 . the existence of a @xmath0-supercompact cardinal @xmath1 such that @xmath2 fails at @xmath0 is equiconsistent with the existence of a @xmath0-supercompact cardinal that is @xmath13-tall . 4 . the existence of a @xmath0-supercompact cardinal @xmath1 such that the @xmath3 is equiconsistent with the existence of a @xmath0-supercompact cardinal that is @xmath4-tall . in each case above , the term `` equiconsistent '' is intended to mean that , in the forward direction the same cardinal witnessing the hypothesis also witnesses the conclusion ; and in the reverse direction , the same cardinal witnessing the hypothesis witnesses the conclusion in a forcing extension . the details of cardinal preservation in the various forcing extensions in parts ( 1 ) - ( 4 ) of the main theorem will be worked out below . questions ( 1 ) - ( 4 ) above can be seen as a special case to a more general question : 1 . what kind of @xmath2 patterns are consistent with a @xmath0-supercompact cardinal from what type of large cardinal assumption ? there are some obvious resrtictions such as if @xmath2 fails at @xmath1 , a @xmath0-supercompact cardinal , then it must fail unboundedly often below @xmath1 . also , if @xmath0 is a strong limit and @xmath2 holds below and at @xmath1 then @xmath2 must hold up to @xmath0 . there are however some subtle issues in answering the general question which we will address in a forthcoming paper . let us note that friedman and thompson have an alternate approach to proving woodin s result . indeed they show in @xcite that from the hypothesis of the existence of a @xmath14-hypermeasurable cardinal @xmath1 , which is equiconsistent with the existence of a cardinal @xmath1 that is @xmath5-tall , after doing a preparatory forcing iteration one may use side - by - side sacks forcing to pump up the size of the power set of @xmath1 to @xmath5 while preserving the measurability of @xmath1 using what they refer to as the tuning fork method . the way we will establish math theorem ( 2)-(4 ) in this paper is by forcing that achieves @xmath15 , and hence @xmath16 , and preserves the @xmath0-supercompactness of @xmath1 , where @xmath12 is a cardinal . this suggests the question , can one force a violation of @xmath2 at @xmath0 while preserving @xmath2 in the interval @xmath17 and preserving the @xmath0-supercompactness of @xmath1 ? it seems as though the method of surgical modification of a generic , due to woodin , does not generalize to answer this question . however , friedman and honsik show in their forthcoming paper @xcite that the answer to this question is yes by using generalized sacks forcing . we also note that question ( 5 ) above is part of a larger program of attempting to describe what kind of easton functions can be realized as the continuum function by forcing while preserving large cardinals . for more on this see @xcite . here we provide an outline of the rest of the paper . in section [ sectionterminology ] we discuss some notational conventions . in section [ sectionpreliminaries ] we state some basic lemmas about lifting large cardinal embeddings that will be used throughout the paper . we will prove main theorem ( 1 ) in section [ easysection ] . in section [ preliminariessection ] , in order to prepare for the proof of main theorem ( 2 ) - ( 4 ) , we discuss the large cardinal concept of `` tallness with closure , '' which synthesizes the concepts of @xmath0-supercompactness and @xmath4-tallness . we prove main theorem ( 2 ) - ( 4 ) in section [ mainproofsection ] . let us now give a few remarks about our notational conventions and terminology , which are for the most part standard . for a forcing poset @xmath18 and conditions @xmath19 and @xmath20 in @xmath18 we write @xmath21 to indicate that @xmath19 extends @xmath20 . we say that @xmath18 is _ @xmath22-closed _ if every descending sequence of conditions of length less than or equal to @xmath1 has a lower bound . we say that @xmath18 is _ @xmath23-closed _ if every descending sequence of conditions of length less than @xmath1 has a lower bound . we say that @xmath18 is _ @xmath22-distributive _ if the intersection of @xmath1 open dense subsets of @xmath18 is an open dense subset of @xmath18 . we will assume that all forcing posets @xmath18 have a greatest element which we will denote by @xmath24 . suppose @xmath25 and @xmath26 are models of @xmath27 and @xmath18 is a forcing notion in @xmath25 . we say that @xmath28 is _ @xmath25-generic _ for @xmath18 if @xmath29 intersects every dense subset of @xmath18 that is in @xmath25 . we will write `` @xmath30 in @xmath26 '' or say that @xmath25 is closed under @xmath1-sequences in @xmath26 to mean that if @xmath31 is a sequence of elements of @xmath25 of length @xmath1 in @xmath26 , then @xmath32 . if @xmath33 is an elementary embedding between models of @xmath27 the _ critical point of @xmath34 _ is the least ordinal @xmath35 such that @xmath36 which we will denote as @xmath37 . for ordinals @xmath35 we write @xmath38 to denote the cofinality of @xmath35 . if @xmath39 is a set and @xmath40 is a function with @xmath41 we write @xmath42 to denote @xmath43 . the large cardinal properties we are interested in , namely partial supercompactness and partial tallness , are witnessed by embeddings @xmath33 and we will be interested in showing that these large cardinal properties are preserved in certain forcing extensions . here we collect , for conveinence , some terminology and several standard lemmas that will be used to lift embeddings of the form @xmath33 to forcing extensions @xmath44\to n[j(g)]$ ] . in order to avoid notational complexities we will make use of the standard abuse of notation by using the same symbol , namely @xmath34 , to denote both the ground model embedding @xmath34 and the lifted embedding @xmath45 , even though @xmath34 and @xmath45 may be classes in different models . after lifting an embedding we will explicitly state in which model the lift is a class . whenever we say that @xmath33 is an embedding in @xmath46 we will mean that @xmath34 is a nontrivial elementary embedding , @xmath25 , @xmath26 , and @xmath46 are models of @xmath27 , and @xmath34 is a class of @xmath46 . for a detailed discussion of lifting embeddings as well as a proofs of lemmas [ groundlemma ] [ lambdadist ] , the reader may consult @xcite or cummings article in @xcite . the following two lemmas are useful for building generic objects . [ groundlemma ] suppose that @xmath30 in @xmath9 and there is in @xmath9 an @xmath25-generic filter @xmath47 for some forcing @xmath48 . then @xmath49^\lambda\subseteq m[h]$ ] in @xmath9 . [ chainlemma ] suppose that @xmath50 is a model of @xmath27 , @xmath30 in @xmath9 and @xmath18 is @xmath51-c.c . if @xmath28 is @xmath9-generic , then @xmath52^\lambda\subseteq m[g]$ ] in @xmath53 $ ] . suppose @xmath33 is an embedding and @xmath54 a forcing notion . in order to lift @xmath34 to @xmath52 $ ] where @xmath29 is @xmath25-generic for @xmath18 , we will often use lemmas [ groundlemma ] and [ chainlemma ] to build an @xmath26-generic filter @xmath55 for @xmath56 satisfying condition ( 1 ) in lemma [ liftingcriterion ] . [ liftingcriterion ] let @xmath33 be an elementary embedding between transitive models of @xmath27 . let @xmath54 be a notion of forcing , let @xmath29 be @xmath25-generic for @xmath18 and let @xmath55 be @xmath26-generic for @xmath56 . then the following are equivalent . 1 . @xmath57 2 . there exists an elementary embedding @xmath44\to n[h]$ ] , such that @xmath58 and @xmath59 . we say that the embedding @xmath45 in condition ( 2 ) above is a @xmath60 of @xmath34 . suppose @xmath6 is an elementary embedding . we say that a set @xmath61 _ generates @xmath34 over @xmath9 _ if @xmath25 is of the form @xmath62^{<\omega}\to v , s\in [ s]^{<\omega } , h\in v\}.\label{seedrep}\end{aligned}\ ] ] where @xmath63 and @xmath64 . we will often make use of the following lemma which states that the above representation ( [ seedrep ] ) of the target model of an elementary embedding remains valid after forcing . [ seedpreservationlemma ] if @xmath6 is an elementary embedding generated over @xmath9 by a set @xmath61 then any lift of this embedding to a forcing extension @xmath65\to m[j^*(g)]$ ] is generated by @xmath66 over @xmath53 $ ] . we will often make use of the next standard lemma which states that embeddings witnessed by extenders are preserved by highly distributive forcing . [ lambdadist ] suppose @xmath6 is an elementary embedding with critical point @xmath1 and @xmath67^{<\omega } , \textrm { and } h\in v\}$ ] where @xmath68 and @xmath64 is some set in @xmath9 . if @xmath69 is @xmath70-distributive forcing then @xmath71 generates an @xmath25-generic filter for @xmath56 and @xmath34 lifts uniquely to @xmath65\to m[h]$ ] in @xmath53 $ ] where @xmath58 is the filter on @xmath56 generated by @xmath71 . our proof of the main theorem will use a preparatory forcing notion called the lottery preparation , which was introduced by hamkins in @xcite . the lottery preparation works uniformly as a generalized laver preparation in a variety of large cardinal contexts . here we give a brief introduction to the lottery preparation . the _ lottery sum _ of a collection of posets @xmath72 is @xmath73 where the ordering on the lottery sum is defined by ( 1 ) @xmath74 for all @xmath75 and @xmath76 and ( 2 ) @xmath77 if and only if @xmath78 and @xmath79 . as hamkins says , a generic for the lottery sum of a collection of posets chooses a poset and forces with it . for a detailed account of the lottery preparation see @xcite . we now define the lottery preparation of @xmath1 , which is a reverse easton support iteration of length @xmath1 . we say a poset @xmath80 is _ allowed _ at stage @xmath81 if @xmath80 is @xmath82-strategically closed ; note that `` @xmath82-strategic closure '' will not play a role in this paper so the reader that is unfamiliar with this concept may take this to simply mean @xmath82-closed . for a partial function @xmath83 we define the _ lottery preparation of @xmath1 with respect to @xmath40 _ to be the reverse easton support forcing iteration such that if @xmath84 is inaccessible and @xmath85 then the stage @xmath81 forcing is the lottery sum in @xmath86 of all allowed posets in @xmath87 and otherwise the stage @xmath81 forcing is trivial . suppose @xmath18 is the lottery preparation of @xmath1 with respect to a partial function @xmath83 and that @xmath80 is a poset that appears in the stage @xmath84 lottery sum . then the condition @xmath88 with @xmath89 for @xmath90 and @xmath91 _ opts _ for @xmath80 at stage @xmath81 where @xmath92 is a @xmath93-name for the top element of @xmath80 . the lottery preparation @xmath18 of some large cardinal @xmath1 is usually used with respect to a partial function @xmath83 with the menas property , such as a function added by fast function forcing ( see section [ fffsection ] ) . using the lottery preparation with respect to such a function insures that @xmath56 , where @xmath34 is an elementary embedding witnessing the large cardinal property at hand , has a tail with a high degree of closure . we will now show that given a @xmath0-supercompact cardinal @xmath1 , one may pump up the power set of @xmath1 to have size at least @xmath51 while maintaining the @xmath0-supercompactness of @xmath1 . this will establish main theorem ( 1 ) because if we assume @xmath12 then in a forcing extension we will have @xmath94 . we note that this result essentially follows from the methods of @xcite . [ easy ] if @xmath1 is @xmath0-supercompact then there is a forcing extension preserving this in which @xmath95 . since any @xmath0-supercompact cardinal is also @xmath96-supercompact we may assume without loss of generality that @xmath97 . we assume @xmath98 because otherwise the theorem is trivial . we may further assume that @xmath99 since the forcing to achieve this is @xmath70-distributive and hence preserves the @xmath0-supercompactness of @xmath1 . let @xmath6 witness that @xmath1 is @xmath0-supercompact . by lemma [ seedpreservationlemma ] , we may assume that each element of @xmath25 is of the form @xmath100 where @xmath101 is in @xmath9 . note that @xmath102 since @xmath25 is closed under @xmath0 sequences in @xmath9 . thus since @xmath103 is inaccessible in @xmath25 we have that @xmath104 . now let @xmath18 be the lottery preparation of @xmath1 relative to a partial function @xmath83 with the menas property @xmath105 . let @xmath29 be @xmath9-generic for @xmath18 . let @xmath106}$ ] and let @xmath55 be @xmath53$]-generic for @xmath80 . by elementarity , @xmath56 is the lottery preparation of @xmath103 defined relative to @xmath107 . since @xmath25 is closed under @xmath0-sequences in @xmath9 we know that the first @xmath1 stages of @xmath18 and @xmath56 agree . since @xmath108 it follows that @xmath109 and since @xmath1 is inaccessible in @xmath25 we see that the stage @xmath1 forcing in @xmath56 is the lottery sum in @xmath52 $ ] of all allowed posets in @xmath110 . since @xmath80 is in @xmath52 $ ] and also in @xmath111}$ ] it follows that @xmath80 appears in the stage @xmath1 lottery sum in @xmath56 . thus , we may factor @xmath56 below a condition @xmath19 that opts for @xmath80 at stage @xmath1 as @xmath112 where @xmath113 is a term for the forcing @xmath56 beyond stage @xmath1 . for example , @xmath19 could be the condition @xmath114 such that @xmath115 and @xmath89 for every other @xmath116 . since @xmath105 and nontrivial forcing occurs in @xmath56 only at closure points of @xmath107 , we see that the next stage of nontrivial forcing beyond @xmath1 in @xmath56 is indeed beyond @xmath0 and from this it follows that @xmath113 is a term for @xmath70-closed forcing . since @xmath50 we see that @xmath29 is @xmath25-generic for @xmath18 and @xmath55 is @xmath52$]-generic for @xmath80 . thus @xmath113 is @xmath70-closed in @xmath52[h]$ ] . furthermore , it follows from lemma [ chainlemma ] that @xmath52[h]$ ] is closed under @xmath0-sequences in @xmath53[h]$ ] because @xmath117 is @xmath118-c.c .. since in @xmath9 , @xmath18 has at most @xmath98-many dense subsets we see that @xmath113 has at most @xmath119-many dense subsets in @xmath52[h]$ ] where @xmath120 . thus in @xmath53[h]$ ] , by constructing a descending sequence of conditions , we may build an @xmath52[h]$]-generic for @xmath113 , call it @xmath121 . we may now lift the embedding to @xmath122\to m[j(g)]$ ] where @xmath123 and the lifted embedding is a class of @xmath53[h]$ ] . it follows from lemma [ groundlemma ] that @xmath124 $ ] is closed under @xmath0 sequences in @xmath53[h]$ ] . now we lift the embedding through @xmath80 . to do this we follow the method used in @xcite corollary 10 . let @xmath125 be a maximal antichain in @xmath124 $ ] . let @xmath126 be a condition that is compatible with every element of @xmath127 . we will show that there is a condition @xmath128 that decides @xmath39 that is still compatible with every element of @xmath127 . since @xmath129 is @xmath130-c.c . we know that @xmath131 in @xmath124 $ ] . since @xmath132 we have @xmath133 and this implies that @xmath134 for some @xmath135 . we fix such an @xmath35 so that also @xmath136 . let @xmath137 . since @xmath138 for @xmath139 we have @xmath140 and thus @xmath141 $ ] is a master condition in @xmath142 ( which is a complete subposet of @xmath143 ) . now since @xmath144 is compatible with every element of @xmath127 we see that @xmath144 and @xmath20 are compatible in @xmath142 . choose @xmath145 below @xmath144 and @xmath20 deciding @xmath39 . we will show that @xmath146 remains compatible with @xmath127 . consider @xmath147 for @xmath148 . we may split @xmath19 into two pieces : @xmath149 where @xmath150 and @xmath151 . then @xmath152 where the domain of @xmath153 is disjoint from the domain of any element of @xmath142 . thus , @xmath146 is compatible with @xmath153 in @xmath143 . furthermore , we have @xmath154 and hence @xmath146 is compatible with @xmath147 . since @xmath80 has @xmath51-many antichains we may iterate this to choose a decreasing sequence of conditions in @xmath53[h]$ ] meeting all the antichains of @xmath143 such that each element of the sequence is compatible with @xmath127 . let @xmath155 be the filter generated by this sequence . then @xmath155 is an @xmath124$]-generic for @xmath143 with @xmath156 . hence we may lift the embedding to @xmath122[h]\to m[j(g)][j(h)]$ ] in @xmath53[h]$ ] , which implies that @xmath1 is @xmath0-supercompact in @xmath53[h]$ ] . here we include some basic definitions and results about @xmath4-tall cardinals , and @xmath4-tall cardinals with closure @xmath0 , where @xmath0 is some cardinal and @xmath4 is an ordinal . a cardinal @xmath1 is called _ @xmath4-tall _ if there is a nontrivial elementary embedding @xmath6 with critical point @xmath1 such that @xmath7 and @xmath8 . woodin and gitik used such cardinals to determine the strength of the failure of @xmath2 at a measurable cardinal ( see @xcite ) , and hamkins has studied them in their own right in @xcite . hamkins says that @xmath1 is _ @xmath4-tall with closure @xmath0 _ if there is an elementary embedding @xmath6 with @xmath157 , @xmath7 , and @xmath30 in @xmath9 . by following a @xmath0-supercompactness embedding with a @xmath4-tallness embedding one may show that if @xmath1 is @xmath4-tall and @xmath0-supercompact , then @xmath1 is @xmath4-tall with closure @xmath0 . indeed , a cardinal @xmath1 is @xmath4-tall and @xmath0-supercopmact if and only if it is @xmath4-tall with closure @xmath0 . we will need the following lemma . [ generators ] if @xmath1 is @xmath4-tall with closure @xmath0 then there is an embedding witnessing this @xmath6 such that @xmath158 where @xmath159 . let @xmath160 witness the @xmath4-tallness with closure @xmath0 of @xmath1 and let @xmath161 where @xmath162 . it is routine to verify that @xmath163 . let @xmath164 be the mostowski collapse of @xmath165 and define an elementary embedding @xmath6 by @xmath166 an let @xmath167 . it follows that @xmath34 is the desired embedding . we will often make use of the easy fact that if @xmath1 is @xmath4-tall with closure @xmath0 , then it is @xmath168-tall with closure @xmath96 , which we demonstrate now . if @xmath1 is @xmath0-supercopmact it is easy to see that @xmath1 must be @xmath96-supercompact . by following a @xmath96-supercompactness embedding by a @xmath4-tallness embedding we obtain @xmath6 witnessing that @xmath1 is @xmath4-tall with closure @xmath96 , then since @xmath103 is inaccessible in @xmath25 and @xmath169 , we have @xmath170 . thus @xmath34 witnesses that @xmath1 is @xmath168-tall with closure @xmath96 . by the remarks in the previous paragraph , given that @xmath1 is @xmath4-tall with closure @xmath0 , in many arguments we will be able to assume without loss of generality that @xmath171 and @xmath97 . then by lemma [ generators ] there is an embedding @xmath6 witnessing that @xmath1 is @xmath4-tall with closure @xmath0 such that @xmath172 the goal of this section will be to prove that we can force to add a function with the menas property with respect to @xmath4-tallness with closure @xmath0 . in other words , we will show that if @xmath6 witnesses that @xmath1 is @xmath4-tall with closure @xmath0 , then we may force to add a partial function @xmath173 with the property @xmath174 . in fact , we can arrange any particular value for @xmath175 up to @xmath103 , the degree of tallness of @xmath1 . to accomplish this we will use a technique invented by woodin called fast function forcing . for a cardinal @xmath1 we define the _ fast function forcing _ poset @xmath176 as follows . conditions in @xmath176 are partial functions @xmath177 such that 1 . each @xmath178 is inaccessible and @xmath179 , 2 . for each inaccessible @xmath180 we have @xmath181 . the ordering on @xmath176 is given by @xmath21 if and only if @xmath182 . for a fixed condition of the form @xmath183 we may factor @xmath176 below @xmath19 as @xmath184 where @xmath0 is the next inaccessible beyond @xmath185 and @xmath186 . a generic @xmath29 for @xmath176 provides a partial function @xmath187 from @xmath1 to @xmath1 . since we will only be concerned with the function @xmath40 , and @xmath40 determines @xmath29 , we will write the forcing extension by the fast - function - forcing poset as @xmath188 $ ] from this point forward . for a more detailed account of fast - function - forcing see @xcite . [ menaslemma ] suppose @xmath6 is a @xmath4-tallness embedding with closure @xmath0 with critical point @xmath1 where @xmath189 ( or merely @xmath0 is less than the first inaccessible beyond @xmath4 ) . then there is a fast function forcing extension @xmath188 $ ] such that @xmath34 lifts to @xmath190\to m[j(f)]$ ] witnessing the @xmath4-tallness with closure @xmath0 in @xmath188 $ ] such that @xmath174 . furthermore , for any @xmath191 there is such a lift @xmath34 such that @xmath192 . as mentioned at the end of subsection [ tallness ] , we may assume without loss of generality that @xmath97 and @xmath171 . we may assume that @xmath193 since this can be accomplished using @xmath70-distributive forcing , which preserves the @xmath4-tallness with closure @xmath0 of @xmath1 by lemma [ lambdadist ] . let @xmath40 be @xmath9-generic for @xmath176 and let @xmath6 be a @xmath4-tallness embedding with closure @xmath0 such that @xmath172 let @xmath194 be an ordinal with @xmath195 and let @xmath196 . we may factor @xmath197 where @xmath81 is the next inaccessible cardinal above @xmath194 . we would like to build an @xmath198$]-generic filter for @xmath199 in @xmath188 $ ] . let @xmath200 be a dense subset of @xmath199 in @xmath198 $ ] . then @xmath200 has an @xmath176 name @xmath201 and @xmath202 for some @xmath203 and @xmath204 . since @xmath205 it follows that @xmath206 . using the @xmath207-closure of @xmath199 in @xmath198 $ ] we may find a single condition in @xmath199 meeting every dense set mentioned by @xmath208 . now we can assume without loss of generality that @xmath209 . since @xmath210 it follows that there are @xmath211-many nice names for dense subsets of a tail of @xmath176 . this implies that there are @xmath212-many functions @xmath213 with domain @xmath214 that represent nice names for dense subsets of a tail of @xmath176 . in @xmath9 we may enumerate such @xmath213 s as @xmath215 . since every dense subset of @xmath199 in @xmath198 $ ] has a nice name and each nice name is represented by a function @xmath216 on our list , we may build an @xmath198$]-generic for @xmath217 in @xmath188 $ ] as follows . at a successor stage @xmath218 , by using the @xmath207-closure of @xmath199 in @xmath198 $ ] we find a single condition @xmath219 below all previously constructed conditions meeting each dense set of the form @xmath220 for @xmath203 . at limit stages we use the fact that @xmath217 is @xmath221-closed in @xmath188 $ ] to find a condition below all previously constructed conditions . this defines a descending sequence of conditions in @xmath188 $ ] and we let @xmath222 be the @xmath198$]-generic filter for @xmath199 generated by the sequence . since @xmath223 we may lift @xmath34 to @xmath190\to m[j(f)]$ ] where @xmath224 and @xmath34 is a class of @xmath188 $ ] . since @xmath225 is @xmath118-c.c . and @xmath222 is in @xmath188 $ ] it follows by lemmas [ groundlemma ] and [ chainlemma ] that @xmath226 $ ] is closed under @xmath0-sequences in @xmath188 $ ] and hence that the lifted embedding witnesses that @xmath1 is @xmath4-tall with closure @xmath0 in @xmath188 $ ] . in @xcite , hamkins shows that the lottery preparation makes many large cardinals indestructible by a wide array of forcing notions . here we will extend the results in @xcite to include @xmath4-tallness with closure @xmath0 . [ indestructibility ] suppose @xmath1 is @xmath4-tall with closure @xmath0 where @xmath189 . then after the lottery preparation , the @xmath4-tallness with closure @xmath0 is indestructible by @xmath23-directed closed forcing of size @xmath70 and @xmath0 is preserved . suppose @xmath6 witnesses the @xmath4-tallness with closure @xmath0 of @xmath1 . without loss of generality we may assume that @xmath97 , @xmath171 , @xmath193 , and @xmath172 we remark that the forcing to obtain @xmath193 collapses cardinals in the interval @xmath227 $ ] to @xmath51 . by lemma [ menaslemma ] we may assume there is a fast function @xmath83 with @xmath174 . let @xmath18 be the lottery preparation defined relative to @xmath40 and let @xmath29 be @xmath9-generic for @xmath18 . let @xmath80 be any @xmath23-directed closed forcing of size @xmath70 in @xmath53 $ ] and let @xmath55 be @xmath53$]-generic for @xmath80 . since @xmath80 could be trivial forcing it will suffice to lift @xmath34 to @xmath53[h]$ ] in @xmath53[h]$ ] . we assume without loss of generality that @xmath228 . since @xmath229 it follows that @xmath52^\lambda\subseteq m[g]$ ] in @xmath53 $ ] and hence @xmath230 $ ] . by elementarity , @xmath56 is the lottery preparation of @xmath103 with respect to @xmath107 . since @xmath25 is closed under @xmath0-sequences in @xmath9 it follows that the first @xmath1 stages in @xmath18 and @xmath56 are the same . since @xmath189 we have @xmath231 and thus @xmath80 appears in the lottery sum at stage @xmath1 in @xmath56 . thus we may factor @xmath56 below a condition @xmath19 that opts for @xmath80 at stage @xmath1 as @xmath232 , where @xmath113 is a term for the iteration beyond stage @xmath1 . we know that @xmath113 is a term for @xmath207-closed forcing because @xmath174 . since @xmath233}\leq\lambda$ ] , it follows that @xmath52[h]^\lambda\subseteq m[g][h]$ ] in @xmath53[h]$ ] . we will now construct an @xmath52[h]$]-generic for @xmath113 in @xmath53[h]$ ] . let @xmath200 be a dense subset of @xmath113 in @xmath52[h]$ ] . let @xmath201 be a @xmath117-name for @xmath200 , that is @xmath234 , and let @xmath202 where @xmath204 , @xmath235 , and @xmath203 . since @xmath236 the sequence of names @xmath237 is in @xmath25 , and furthermore the sequence of dense subsets of @xmath113 , @xmath238 , is in @xmath52[h]$ ] . since @xmath113 is @xmath207-closed in @xmath52[h]$ ] we can find a single condition below every dense set mentioned by @xmath239 . without loss of generality we may assume that the range of @xmath240 is contained in the set of nice names for dense subsets of a tail of @xmath18 . now working in @xmath53[h]$ ] we put a bound on the number of functions @xmath241 since @xmath242 there are @xmath211-many nice names for subsets of a tail of @xmath18 . thus there are at most @xmath212-many such @xmath213 s . in @xmath53[h]$ ] we may enumerate all such @xmath213 s as @xmath243 since every dense subset of @xmath113 in @xmath52[h]$ ] has a nice @xmath117-name and each nice @xmath117-name is represented by one of the @xmath216 s on our list , we may construct an @xmath52[h]$]-generic descending sequence of conditions of @xmath113 as follows . at successor stages @xmath218 , we work in @xmath52[h]$ ] and use the fact that @xmath113 is @xmath207-closed in @xmath52[h]$ ] to find a condition of @xmath113 meeting every dense subset of @xmath113 which has a name on the list @xmath244 . at limits @xmath245 , since @xmath52[h]$ ] is closed under @xmath0-sequences in @xmath53[h]$ ] it follows that @xmath113 is @xmath70-closed in @xmath53[h]$ ] , and hence , in @xmath53[h]$ ] , we may find a condition of @xmath113 below all previously constructed conditions . this defines a descending @xmath51-sequence of conditions in @xmath113 and we let @xmath121 be the filter generated by this sequence of conditions . clearly @xmath246[h]$ ] is an @xmath52[h]$]-generic filter for @xmath113 . thus we may lift the embedding to @xmath122\to m[j(g)]$ ] where @xmath247 and @xmath34 is a class of @xmath53[h]$ ] . since @xmath117 is @xmath51-c.c . it follows from lemma [ chainlemma ] that @xmath52[h]^\lambda\subseteq m[g][h]$ ] in @xmath53[h]$ ] . furthermore , since @xmath246[h]$ ] , we see by lemma [ groundlemma ] that @xmath124 $ ] is closed under @xmath0-seqences in @xmath53[h]$ ] . we will now lift @xmath34 to @xmath53[h]$ ] . since @xmath248 $ ] we may build @xmath127 in @xmath124 $ ] . since @xmath129 is @xmath249-directed closed in @xmath124 $ ] it follows that in @xmath124 $ ] , there is a master condition @xmath126 below each element of @xmath127 . we will now construct an @xmath124$]-generic filter for @xmath129 in @xmath53[h]$ ] . let @xmath250 $ ] be a dense subset of @xmath129 . then by lemma [ seedpreservationlemma ] we may write @xmath251 where @xmath252 $ ] is a function from @xmath214 to the collection of dense subsets of @xmath80 and @xmath203 . now let @xmath253 . since @xmath254 $ ] we see that @xmath255 $ ] . since @xmath129 is @xmath249-directed closed in @xmath124 $ ] we can find , via an internal argument in @xmath124 $ ] , a single condition that meets every dense set mentioned by @xmath256 . in @xmath53 $ ] , @xmath257 and this implies that there are at most @xmath258-many functions @xmath259 $ ] that represent dense subsets of @xmath129 in @xmath124 $ ] . as before , we enumerate these functions as @xmath215 and define a descending sequence of conditions meeting every dense subset of @xmath129 . we start the descending sequence with the master condition , @xmath144 . if @xmath218 is a successor , we use the @xmath249-directed closure of @xmath129 in @xmath124 $ ] to meet all dense sets mentioned in @xmath244 with a single condition that is also below @xmath144 . at limit stages @xmath245 since @xmath124 $ ] is closed under @xmath0-sequences in @xmath53[h]$ ] it follows that @xmath129 is @xmath70-closed in @xmath53[h]$ ] , and hence we may find a condition of @xmath129 below all previously constructed conditions . this defines a descending @xmath51-sequence of conditions below the master condition @xmath144 . let @xmath155 be the generic filter generated by this sequence . since @xmath144 is stronger than every element of @xmath127 and @xmath260 we have @xmath156 and thus we may lift the embedding to @xmath122[h]\to m[j(g)][j(h)]$ ] where the lifted embedding is a class in @xmath53[h]$ ] . this shows that the @xmath4-tallness with closure @xmath0 of @xmath1 is indestructible by any @xmath23-directed closed forcing of size @xmath70 in @xmath53 $ ] . let us now give a quick application of theorem [ indestructibility ] . [ gchoninterval ] if @xmath1 is @xmath4-tall with closure @xmath0 where @xmath189 and @xmath0 is inaccessible then there is a forcing extension preserving the inaccessibility of @xmath0 in which @xmath1 is @xmath4-tall with closure @xmath0 and @xmath2 holds on @xmath261 $ ] . by theorem [ indestructibility ] we may assume that the @xmath4-tallness with closure @xmath0 is indestructible by @xmath23-directed closed forcing of size @xmath70 . we define a length @xmath0 forcing iteration @xmath262 with reverse easton support as follows . let the first @xmath1 stages of @xmath263 be trivial forcing . for @xmath264 we force at cardinal stages @xmath81 with @xmath265 . clearly @xmath262 is @xmath22-directed closed and thus @xmath22-distributive . let @xmath29 be @xmath9-generic for @xmath18 . it is routine to show that @xmath0 remains inaccessible in @xmath53 $ ] and is thus not collapsed by the forcing @xmath29 . since @xmath0 is inaccessible it follows that @xmath18 has size at most @xmath266 . furthermore , @xmath18 is @xmath23-directed closed . hence by theorem [ indestructibility ] it follows that in @xmath53 $ ] , @xmath1 is @xmath4-tall with closure @xmath0 and @xmath2 holds on @xmath17 . now we may force @xmath2 to hold at @xmath0 with @xmath70-distributive forcing @xmath267 , which clearly preserves the @xmath4-tallness with closure @xmath0 of @xmath1 by lemma [ lambdadist ] . let us now argue that the equiconsistencies in the forward directions in main theorem ( 2 ) - ( 4 ) are actually implications . for main theorem ( 2 ) , suppose @xmath6 witnesses that @xmath1 is @xmath0-supercompact and @xmath10 . since @xmath103 is inaccessible in @xmath25 we have @xmath268 . hence @xmath34 is a @xmath4-tallness embedding with closure @xmath0 . the forward directions in main theorem ( 3 ) and ( 4 ) are similar . it remains to prove the backward directions of main theorem ( 2 ) - ( 4 ) . to do this we will start with an embedding @xmath6 witnessing the @xmath4-tallness with closure @xmath0 of @xmath1 , force to violate @xmath2 at either @xmath1 or @xmath0 , and then lift the embedding to the forcing extension . in order to lift the embedding we will use woodin s method of surgery to modify a certain generic @xmath269 to obtain @xmath270 with the pullback property @xmath271 . the following lemma , due to woodin , will allow us to show that @xmath270 is a generic filter . [ surgery ] suppose @xmath26 and @xmath25 are transitive inner models of @xmath27 and @xmath272 is a nontrivial elementary embedding with critical point @xmath1 that is continuous at regular cardinals @xmath273 where @xmath274 . then if @xmath275 is such that @xmath276 then @xmath277 . first we will argue that it suffices to consider the case where @xmath39 is a set of ordinals . let @xmath279 be a sequence of length @xmath280 such that @xmath281 ; for example , @xmath282 could be an enumeration of some sufficiently large @xmath283 so that @xmath284 is an enumeration of @xmath285 . clearly @xmath284 is a sequence of length @xmath286 in @xmath25 , write @xmath287 . let @xmath288 . then @xmath289 and we have @xmath290 . clearly @xmath291 if and only if for some @xmath292 we have @xmath293 . in other words , @xmath294 if and only if @xmath295 . it follows that @xmath296 , hence we have reduced to the case in which @xmath39 is a set of ordinals . suppose @xmath275 is a set of ordinals with @xmath276 and @xmath297 . then @xmath39 contains @xmath51-many elements of the form @xmath298 for @xmath299 . that is , we may assume @xmath39 contains elements of the form @xmath300 where @xmath135 and @xmath301 is a strictly increasing sequence of ordinals which is not necessarily in @xmath26 since it was defined using @xmath275 . now let @xmath302 . furthermore , we know that @xmath303 and hence @xmath304 . this implies that @xmath305 . since @xmath34 is continuous at regular cardinals @xmath273 , and thus at @xmath306 , we know that @xmath39 contains unboundedly many @xmath300 less than @xmath307 . so in @xmath25 , @xmath39 is unbounded in @xmath307 , but this implies that @xmath308 which contradicts our assumption that @xmath276 . [ alllambda ] for any cardinals @xmath309 , if @xmath1 is @xmath0-supercompact and @xmath4-tall then there is a forcing extension in which @xmath1 is @xmath0-supercompact and @xmath10 ; and hence also @xmath3 . indeed , the forcing preserves cardinals on @xmath310\cup ( 2^\lambda,\infty)$ ] and assuming @xmath2 holds at @xmath0 , all cardinals @xmath311 are preserved . let @xmath1 be @xmath0-supercompact and @xmath4-tall . as before , by the remarks at the end of subsection [ tallness ] , we may assume without loss of generality that @xmath97 and @xmath171 . we may further assume that @xmath193 since the forcing to achieve this is @xmath70-distributive and thus preserves the @xmath0-supercompactness and @xmath4-tallness of @xmath1 . by lemma [ generators ] we have an elementary embedding @xmath6 with @xmath157 , @xmath7 , @xmath30 , and @xmath312 by lemma [ menaslemma ] we may assume without loss of generality that there is a partial function @xmath83 in @xmath9 such that @xmath174 . let @xmath18 be the lottery preparation relative to @xmath40 . let @xmath313 be @xmath9-generic and let @xmath314}$ ] . let @xmath47 be @xmath53$]-generic . notice that since @xmath18 has size @xmath1 and @xmath80 is @xmath118-c.c . it follows that @xmath117 is @xmath118-c.c . , and thus by lemma [ chainlemma ] that @xmath52[h]$ ] is closed under @xmath0-sequences in @xmath53[h]$ ] . by elementarity @xmath56 is the lottery preparation of length @xmath103 relative to @xmath107 as defined in @xmath25 . since @xmath25 is closed under @xmath0-sequences in @xmath9 it follows that the iterations @xmath18 and @xmath56 agree up to stage @xmath1 and since @xmath230 $ ] is @xmath23-closed it appears in the stage @xmath1 lottery in @xmath56 . hence we may factor @xmath56 below a condition @xmath315 that opts for @xmath80 at stage @xmath1 as @xmath232 . since @xmath174 it follows that the next nontrivial stage of forcing in @xmath56 is beyond @xmath4 and hence that @xmath113 is a term for @xmath207-closed forcing . as in the proof of theorem [ indestructibility ] we will construct a descending @xmath51-sequence of conditions in @xmath53[h]$ ] that meets every dense subset of @xmath113 in @xmath52[h]$ ] . let @xmath200 be a dense subset of @xmath113 in @xmath52[h]$ ] and let @xmath201 be a nice @xmath117-name for @xmath200 . then @xmath316 for some @xmath203 and some function @xmath240 with domain @xmath214 and range contained in the set of nice names for dense subsets of a tail of @xmath18 . since the sequence of names @xmath317 is in @xmath25 and @xmath113 is @xmath207-closed in @xmath52[h]$ ] we can find a condition in @xmath113 meeting every dense set mentioned by the sequence @xmath317 . since there are @xmath318-many functions from @xmath214 to the set of nice names for dense subsets of a tail of @xmath18 , it follows from the fact that @xmath52[h]$ ] is closed under @xmath0-sequences in @xmath53[h]$ ] that we can construct a descending @xmath51-sequence in @xmath53[h]$ ] that meets each dense subset of @xmath113 in @xmath52[h]$ ] . let @xmath121 be the @xmath52[h]$]-generic filter generated by this descending sequence . then we may lift the embedding in @xmath53[h]$ ] to @xmath122\to m[j(g)]$ ] where @xmath123 and since @xmath246[h]$ ] we may use lemma [ groundlemma ] to see that @xmath124 $ ] is closed under @xmath0-sequences in @xmath53[h]$ ] . let @xmath319 \textrm { where $ h\in v[g]$}\}$ ] . then @xmath320 $ ] and we let @xmath321 $ ] be the inverse of the mostowski collapse @xmath322 and let @xmath323\to m_0'$ ] be defined by @xmath324 . it follows that @xmath325 is the ultrapower embedding by the measure @xmath326 where @xmath327[h]$ ] . by elementarity , @xmath328 is of the form @xmath329 $ ] , where @xmath330 and @xmath331 is @xmath332-generic . furthermore , @xmath333 where @xmath334 is @xmath335$]-generic for @xmath336}$ ] and @xmath337 is @xmath52[h_0]$]-generic for the tail of the iteration @xmath338 above @xmath1 . the following diagram is commutative . \ar[r]^j \ar[rd]_{j_0 } & m[j(g ) ] \\ & m_0[j_0(g ) ] \ar[u]_k \\ } \ ] ] it follows that @xmath325 is a class of @xmath329 $ ] which is closed under @xmath0-sequences in @xmath53[h_0]$ ] and that @xmath340 . we would like to lift @xmath34 through the stage @xmath1 forcing , @xmath80 . this can not be accomplished using a master condition argument since @xmath341 . in order to lift the embedding we will force with @xmath342 over @xmath53[h]$ ] to obtain a generic @xmath343 for @xmath342 . in subsection [ obtainingsection ] we will argue that @xmath344 generates an @xmath124$]-generic @xmath269 for @xmath129 . however , we have no reason to expect that @xmath345 and thus we need to do more work in order to lift the embedding . in subsection [ surgerysection ] we will use woodin s method of surgery to modify the filter @xmath269 to obtain an @xmath124$]-generic @xmath270 for @xmath129 with @xmath271 . then the embedding lifts to @xmath346[h]\to m[j(g)][g^*]\label{surgerylift}\end{aligned}\ ] ] in @xmath53[h][g_0]$ ] where @xmath347 . the embedding ( [ surgerylift ] ) does not witness that @xmath1 is @xmath0-supercompact in @xmath53[h]$ ] because the embedding is a class of @xmath53[h][g_0]$ ] . under the assumption that we have lifted the embedding as in ( [ surgerylift ] ) we will now show that the embedding lifts further to our target model @xmath53[h][g_0]$ ] witnessing that @xmath1 is @xmath4-tall and @xmath0-supercompact in @xmath53[h][g_0]$ ] . furthermore , we will show in subsection [ preservescardinals ] that @xmath348[h][g_0]}.\label{gchresultsalllambda}\end{aligned}\ ] ] let us first argue , assuming we have @xmath270 as above , that @xmath124[g^*]$ ] is closed under @xmath0-sequences in @xmath53[h][g_0]$ ] . we now show that @xmath342 is @xmath70-distributive in @xmath53[h]$ ] . since @xmath349 it follows that @xmath342 is @xmath70-closed in @xmath329 $ ] . since @xmath329 $ ] is closed under @xmath0-sequences in @xmath53[h_0]$ ] it follows that @xmath342 is @xmath70-closed in @xmath53[h_0]$ ] and since @xmath70-closed forcing remains @xmath70-distributive in @xmath51-c.c . forcing extensions , it follows that @xmath342 is @xmath70-distributive in @xmath53[h]$ ] . since @xmath124 $ ] is closed under @xmath0-sequences in @xmath53[h]$ ] and @xmath342 is @xmath70-distributive in @xmath53[h]$ ] it easily follows that @xmath124 $ ] is closed under @xmath0-sequences in @xmath53[h][g_0]$ ] . since @xmath270 is constructed from @xmath343 we have @xmath350[h][g_0]$ ] and it follows that @xmath124[g^*]$ ] is closed under @xmath0-sequences of ordinals in @xmath53[h][g_0]$ ] . by using a well ordering of a sufficient initial segment of the universe @xmath124[g^*]$ ] , it follows that @xmath124[g^*]$ ] is closed under @xmath0-sequences in @xmath53[h][g_0]$ ] . now we show that the embedding ( [ surgerylift ] ) lifts through @xmath342 . every element of @xmath124[g^*]$ ] is of the form @xmath351 where @xmath352[h]$ ] is in @xmath53[h]$ ] and @xmath203 . in other words , @xmath124[g^*]$ ] is generated by @xmath353 over @xmath53[h]$ ] . since @xmath214 has size @xmath0 and @xmath342 is @xmath70-distributive , it follows by lemma [ lambdadist ] that @xmath354 generates an @xmath124[g^*]$]-generic filter @xmath355 for the poset @xmath356 . thus @xmath34 lifts in @xmath53[h][g_0]$ ] to @xmath357[h][g_0]\to m[j(g)][g^*][j(g_0)].\ ] ] since @xmath358[h][g_0]$ ] and @xmath124[g^*]$ ] is closed under @xmath0-sequences in @xmath53[h][g_0]$ ] it follows that @xmath124[g^*][j(g_0)]$ ] is closed under @xmath0-sequences in @xmath53[h][g_0]$ ] . since @xmath34 is a lift of the original embedding it still satisfies @xmath7 . hence @xmath34 witnesses that @xmath1 is @xmath4-tall with closure @xmath0 in @xmath53[h][g_0]$ ] . let @xmath343 be as in subsection [ outline ] ; that is , @xmath343 is @xmath53[h]$]-generic for @xmath342 . in this subsection we will argue that @xmath344 generates an @xmath124$]-generic for @xmath129 . each @xmath359 $ ] is of the form @xmath360 for some @xmath203 and some @xmath352 $ ] with @xmath259 $ ] . since @xmath361 it follows that each @xmath359 $ ] is of the form @xmath362 for some @xmath203 and some @xmath363 $ ] with @xmath364 $ ] ; in fact since @xmath365 , where @xmath366 is the collapse of @xmath4 , still has every @xmath203 in its domain , we may assume that @xmath367 . let @xmath200 be an open dense subset of @xmath129 in @xmath124 $ ] . then @xmath368 for some fixed @xmath203 where @xmath369 is a sequence of dense open subsets of @xmath342 . since @xmath342 is @xmath370-distributive we know that @xmath371 is open dense in @xmath342 . hence there is a condition @xmath372 . then @xmath373 . now @xmath374 for each @xmath375 and this implies @xmath376 for each @xmath377 . it follows that @xmath378 and hence @xmath379 . therefore @xmath344 generates an @xmath124$]-generic filter for @xmath129 . now that we have an @xmath124$]-generic @xmath269 for @xmath129 we use woodin s method of surgery to obtain an @xmath124$]-generic @xmath270 for @xmath129 with @xmath271 . we define @xmath270 in terms of @xmath269 and @xmath127 in the following way . let @xmath380 be the set of coordinates @xmath381 such that there is a @xmath148 such that @xmath382 and @xmath383 and let @xmath384 be the automorphism induced by flipping bits over coordinates in @xmath380 . then we let @xmath385 . in other words , we obtain the modified generic @xmath270 by using @xmath269 except that whenever @xmath269 and @xmath127 disagree , we change @xmath269 to match @xmath127 . since @xmath34 is continuous at regular cardinals @xmath273 the key lemma applies and we use this to show that @xmath270 is a generic filter on @xmath129 . first note that if @xmath387 then @xmath388}<j(\kappa)$ ] and so the set of coordinates on which @xmath389 has size @xmath70 by the key lemma and is thus in @xmath124 $ ] since @xmath124^\lambda\subseteq m[j(g)]$ ] in @xmath53[h][g_0]$ ] . this implies that @xmath390 $ ] and thus that @xmath270 defines a filter in @xmath124 $ ] . now we show that @xmath270 is @xmath124$]-generic for @xmath129 . let @xmath391 be a maximal antichain in @xmath124 $ ] . since @xmath129 has the @xmath392-c.c . we have @xmath393}\leq j(\kappa)$ ] . furthermore , each @xmath394 has @xmath388}<j(\kappa)$ ] . hence @xmath395}\leq j(\kappa)$ ] . by the key lemma , the set of coordinates mentioned by conditions in @xmath39 that were involved in the changes we made from @xmath269 to @xmath270 has size @xmath396 , call this set @xmath397 . in other words , @xmath398 . let @xmath399 be the automorphism induced by flipping bits over coordinates in @xmath397 . the coordinates of bits that get flipped by @xmath400 are contained in the domain of the antichain ( see the shaded region in the figure below ) . since @xmath401 we have @xmath402 $ ] and it follows that @xmath403 $ ] . then @xmath404 is a maximal antichain of @xmath129 and by genericity of @xmath269 there is a condition @xmath405 that decides @xmath404 . it follows that @xmath406 decides @xmath39 since @xmath407 . this establishes that @xmath270 is @xmath124$]-generic for @xmath129 . since we arranged @xmath271 by definition , we may use lemma [ liftingcriterion ] to lift the embedding to @xmath122[h]\to m[j(g)][j(h)]$ ] where @xmath347 . since we used @xmath343 to define @xmath270 , this lift is a class of @xmath53[h][g_0]$ ] . we argued above ( in the outline given in section [ outline ] ) that we can lift the embedding further through the @xmath343 forcing . so all that remains is to show that @xmath342 preserves cardinals and that @xmath10 in @xmath53[h][g_0]$ ] . we have already argued that @xmath342 is @xmath70-distributive in @xmath53[h]$ ] and we will now argue that @xmath342 is @xmath13-c.c . from this it follows that @xmath342 preserves cardinals over @xmath53[h]$ ] and @xmath10 in @xmath53[h]$ ] . each condition @xmath408 is in @xmath329 $ ] and is thus of the form @xmath409 for some @xmath410 with @xmath259 $ ] . for each @xmath408 , @xmath411 has size @xmath0 in @xmath53 $ ] and thus @xmath412 leads to a function @xmath413 , which can be viewed as a condition in the full support product of @xmath0-many copies of @xmath80 as defined in @xmath53 $ ] , which we denote by @xmath414 . we will show that @xmath342 is @xmath13-c.c . in @xmath53[h]$ ] by arguing that @xmath415 is @xmath13-c.c . in @xmath53[h]$ ] and that an antichain of @xmath342 of size @xmath13 in @xmath53[h]$ ] would lead to an antichain of @xmath415 of size @xmath13 in @xmath53[h]$ ] . by a delta system argument @xmath415 is @xmath13-c.c . in @xmath53 suppose @xmath416[h]$ ] is an antichain of @xmath415 with @xmath417 . we will show that @xmath39 leads to an antichain of size @xmath194 of @xmath418 in @xmath53 $ ] and thus that @xmath419 . let @xmath420 where @xmath421 and @xmath422 . for each @xmath423 let @xmath424 be such that @xmath425 where @xmath426 . we have @xmath427 in @xmath53 $ ] and we now show that @xmath428 is an antichain of size @xmath194 of @xmath429 in @xmath53 $ ] . clearly @xmath430 $ ] because in choosing the pairs @xmath431 in @xmath432 we only used the forcing relation @xmath433 . suppose for a contradiction that @xmath432 is not an antichain , i.e. that @xmath434 for some @xmath435 with @xmath436 and some @xmath437 . let @xmath438 be @xmath53$]-generic for @xmath80 with @xmath439 . since @xmath440 it follows that @xmath441 enumerates an antichain . furthermore we have @xmath442 , @xmath443 , and @xmath444 , a contradiction . hence we conclude that @xmath432 is an antichain of @xmath429 in @xmath53 $ ] and since @xmath427 we see that @xmath432 leads to an antichain of @xmath415 of size @xmath194 in @xmath53 $ ] . therefore , @xmath419 . hence we conclude that an antichain @xmath39 of @xmath415 in @xmath53[h]$ ] must have size @xmath445 . now we complete the proof of theorem [ alllambda ] by showing that @xmath342 is @xmath13-c.c . in @xmath53[h]$ ] . suppose that in @xmath53[h]$ ] , @xmath342 has an antichain @xmath39 of size @xmath194 . for each @xmath446 let @xmath410 be such that @xmath447 where @xmath448 $ ] . as above each @xmath412 yields a condition in @xmath415 , call it @xmath449 . for @xmath450 we have @xmath451 , and thus by elementarity we conclude that there is a @xmath452 such that @xmath453 . this implies that @xmath454 is an antichain in @xmath415 where @xmath455 . by claim [ qbarlemma ] , @xmath419 . thus @xmath342 is @xmath13-c.c . in @xmath53[h]$ ] . let us argue that cardinals in @xmath310\cup(2^\lambda,\infty)$ ] are preserved . we started with a model and forced @xmath193 which may have collapsed cardinals in @xmath456 $ ] . we then add a fast function using @xmath118-c.c . forcing which preserves cardinals @xmath311 . the remaining forcing is @xmath457 where @xmath117 is @xmath118-c.c . and @xmath342 preserves cardinals over @xmath53[h]$ ] . thus in the final model @xmath53[h][g_0]$ ] we have preserved cardinals in the interval @xmath310\cup ( 2^\lambda,\infty)$ ] .

we discuss the scaling properties of free branched polymers . the scaling behaviour of the model is classified by the hausdorff dimensions for the internal geometry : and , and for the external one : and . the dimensions and characterize the behaviour for long distances while and for short distances . we show that the internal hausdorff dimension is for generic and scale - free trees , contrary to which is known be equal two for generic trees and to vary between two and infinity for scale - free trees . we show that the external hausdorff dimension is directly related to the internal one as , where is the stability index of the embedding weights for the nearest - vertex interactions . the index is for weights from the gaussian domain of attraction and for those from the lvy domain of attraction . if the dimension of the target space is larger than one finds , or otherwise . the latter result means that the fractal structure can not develop in a target space which has too low dimension .

cond-mat0207459

in recent years the theory of random geometry @xcite has become a powerful method of investigating problems in many areas of research ranging from the statistical theory of membranes @xcite , branched polymers @xcite and complex networks @xcite to fundamental questions in string theory @xcite and quantum gravity @xcite . those problems have in common that they can be described by a dynamically alternating geometry which undergoes fluctuations of a statistical or quantum nature . the dynamics of such fluctuations can be modelled using the concepts of the statistical ensemble and the partition function in a similar way as it is done in particle physics by the methods of lattice field theory . contrary to lattice field theory where the partition functions run over field configurations on a rigid geometry , the geometry itself is variable here . since the geometry is dynamical many new features occur like for instance geometrical correlations or the influence of the random geometry on the fields living on it . similarly as in field theory where the concepts of universality , critical exponents , correlations , _ etc _ are independent of whether one discusses a field theoretical model of magnetism or a quantum theoretical model of particles , also in the theory of random geometry many questions are independent on details and may be addressed using general methods . general concepts can be best developed on an analytically treatable model . in field theory the role of a test bed is played by the ising model while in random geometry by the branched polymer model @xcite . despite its simplicity the branched - polymer model has a rich phase structure exhibiting different scaling properties of the fractal geometry and the correlation functions . the model has internal and external geometry sectors similarly as the polyakov string @xcite . as the polyakov string , it can also be interpreted as a model for quantum objects embedded into a @xmath9-dimensional target space or a model of quantum gravity interacting with @xmath9-scalar fields . the term quantum gravity refers to an euclidean feynman integral expressed by a sum over diagrams representing the nearest - neighbour relations between points of a discrete set . in more realistic models the sum runs over higher dimensional simplicial manifolds and can be interpreted as a regularized feynman integral over riemannian structures on the manifold @xcite . the intuition which one can gain from the analytic solution of the branched - polymer model is very helpful for considerations of more complicated models . in fact , the model has proven already many times to be extremly useful to test and develop various ideas concerning random geometry @xcite-@xcite . in addition to this general interest in this model there is a specific motivation which is related to the reduced super - symmetric yang - mills matrix model @xcite . this model was introduced as a non - perturbative definition for superstrings @xcite and referred to afterwards as the ikkt model . the one - loop approximation of this model leads to a model of graphs which have as a back - bone a branched - polymer with power - law weights for the link lengths . the ikkt model is believed to provide a dynamical mechanism for the spontaneous breaking of the lorentz symmetry from ten to four dimensions @xcite . if one tries to understand the breaking in terms of the one - loop level approximation one finds it to be related to the fractal properties of the branched - polymers , which have the hausdorff dimension equal to four @xcite . the question of the spontanous symmetry breaking was investigated also by many other methods with help of which one was able to gain an insight into underlying physical mechanisms @xcite-@xcite . in the simplest case one considers generic trees with the nearest - vertex interactions given by gaussian weights or weights which belong to the gaussian universality class @xcite . in this paper we also discuss the nearest - vertex interactions , but we extend the discussion to non - gaussian weights in particular to power - law weights @xcite . the energetical costs to generate long links on the polymer is then much smaller than for gaussian ones . in the extremal situation , when the power - law exponent @xmath6 of the link length distribution @xmath12 lies in the interval @xmath13 , very long links are spontanously generated on the tree and their presence shifts the model to a new universality class which can be called the class of lvy branched polymers that similarly as lvy paths exhibit a new scaling behaviour . the scaling properties and the universality class of the model depends also on the internal branching weights of the trees @xcite . under a change of the weights the model may undergo a transition from the phase of elongated trees with the internal hausdorff dimension @xmath14 , known as generic trees , to the phase of collapsed trees with @xmath15 which are localized around a singular vertex of high - connectivity @xcite . in between there is a phase of scale - free trees which may have any hausdorff dimension between @xmath14 and @xmath15 @xcite . we show that the internal properties decouple from the embedding in the target space but on the other hand that they strongly affect the embedding : the external hausdorff dimension is proportional to the internal one @xmath5 . in this paper we cover the entire classification of the universality classes of branched polymers with the nearest - vertex interactions . we hope it can be treated as a starting point in the discussion of models with more complicated interactions like for example those with correlations between neighbouring links , or self - avoidance in the embedding space . many pieces of this classification have been discussed for the gaussian branched - polymers already @xcite . several well - known results have been summarized within the appendix in which we present a systematical treatment of the internal geometry in terms of generating functions . the extension of this classification to weights with power - law tails and to the case when the internal geometry is non - generic , is presented in the main - stream of the text . emphasis is put on calcuations of the two - point functions . throughout the paper we also stress the factorization property of the internal and external geometry , which allows us to clearly separate the discussion of the internal geometry before considering the entire model . it also permits us to reveal many interesting relations between the correlation functions of the external space to those for the internal geometry . we consider a canonical ensemble of trees embedded in a @xmath9-dimensional target space . the partition function of the ensemble is defined as a weighted sum over all labeled trees with @xmath16 vertices . the set of such labeled trees , containing @xmath17 elements , will be denoted by @xmath18 . the statistical weight of a tree is given by a product of an internal weight @xmath19 which depends only on the internal geometry of the tree , and an external one which depends on the positions of the ( tree ) vertices in the target space . we shall consider trees with nearest - neighbour interactions for which the partition function reads : @xmath20 the external weight of a tree is a product of link weights @xmath21 which depend exclusively on the link vector @xmath22 . saying alternatively , the energy cost of the embedding of the tree in the target space is a sum of energy costs of the independent embedding of links . the second product in ( [ zn ] ) runs over the set of ( unoriented ) linked vertex pairs denoted by @xmath23 . the most natural choice of the internal weights is @xmath24 . we could entirely stick to this choice of weights , but since we want to discuss the problem of universality we also want to check whether a modification of the weights will change the scaling properties and hence the universality @xcite . here we will restrict our considerations to internal weights which can be written as a product of weights @xmath25 for the individual vertices : @xmath26 each vertex weight only depends on the degree of the vertex that means the number of links emerging from it . the internal properties of the model are determined when the whole set of branching weights @xmath27 for @xmath28 is specified . we demand that : @xmath29 for all @xmath30 and at least one @xmath31 . if @xmath32 were zero @xmath19 would vanish for all tree graphs while if all @xmath33 for @xmath34 were zero then the weights @xmath19 would vanish for all trees except chain - structures . note that the model is invariant with respect to translations in the external space : @xmath35 . because of the translational zero - mode the partition function ( [ zn ] ) is infinite . one can make it finite by dividing out the volume @xmath36 of the translational zero - mode : @xmath37 this can for example be realized by fixing the position of the center of mass of the trees . trees which can be obtained from each other by a permutation of the vertex labels contribute with the same statistical weight . for a tree with @xmath16 vertices there are @xmath38 such vertex permutations . in order to avoid overcountings one introduces the standard factor @xmath39 to the definition of the partition function ( [ zn ] ) . this factor divides out the volume of the permutation group of the vertex labels . the number of all labeled trees counted with this factor @xmath40 is exponentially bounded in the @xmath41 limit . if one defines the grand - canonical partition function @xmath42 one can see that it is well defined as long as the fugacity @xmath43 is smaller than the radius of convergence of the series , which in the particular case @xmath44 is equal @xmath45 . more generally , as long as @xmath46 grows only exponentially for large @xmath16 the grand - canonical partition function has a non - vanishing radius of convergence and hence one can savely define @xmath47 . the statistical average of a quantity @xmath48 defined on the ensemble ( [ zn ] ) is given by : @xmath49 for translationally invariant quantities the averages are proportional to the volume @xmath50 of the translational zero - mode . for such quantities one should rather speak of an average density per volume element of the target space : @xmath51 which is a finite number . in particular @xmath52 . we will frequently distinguish between the internal and external geometry of the trees . by the former we mean the connectivity of the corresponding tree graph , by the latter its embedding in the external space . for example , the internal ( geodesic ) distance between two vertices @xmath53 and @xmath54 of a graph is defined as the number of links of the shortest path connecting them , while the external distance is given by the length of the vector @xmath55 . note that the path between @xmath53 and @xmath54 is unique for tree graphs . thus the length of this path , i.e. the number of its links , determines the internal geodesic distance . the properties of the embedding in the external space depend on the link weight function @xmath56 ( see ( [ zn ] ) ) . we consider isotropic weights depending only on the link length . that means @xmath57 , where @xmath58 . we further assume that @xmath56 is a positive integrable function . without loss of generality , we can then choose the normalization to be : @xmath59 . this allows us to interprete @xmath56 as a probability density . the fundamental quantities which encode the information about the statistical properties of the system are the correlation functions . for the canonical ensemble with the partition function ( [ zn ] ) , the @xmath60-point correlation functions are defined as : @xmath61 where the brackets @xmath62 on the right hand side denote the average over the ensemble ( [ an ] ) . if one multiplies all sums in the product in ( [ gnn ] ) one obtains a sum of terms being products of delta functions . the prime in equation ( [ gnn ] ) means that all terms which contain two or more identical delta functions are skipped from this sum . this exclusion principle applies only to the situation when any two arguments of @xmath63 are identical . what we will show now is that the problem of determining the @xmath60-point correlation function can be devided into two sub - problems . the first step is to determine the internal two - point function . this can in general be done independently of a particular choice of the embedding weights @xmath56 . the second is to use the information encoded in the internal two - point function to determine the external properties of the trees . in order to see that the internal geometry of the trees does not depend on the choice of the link weight function , consider a tree graph and calculate the following two integrals for this tree : @xmath64 the first integral corresponds to the embedding weight factor for a tree whose @xmath65-th vertex is fixed at the position @xmath66 in the target space . since the model is translationally invariant the result of the integration does not depend on the position . this result can be obtained by changing the integration variables from position vectors of all vertices @xmath67 to link vectors @xmath68 for which the integration completely factorizes . the jacobian for such a change of the integration variables is equal one . the second integral ( [ t2 ] ) gives the weight factor for a tree whose vertices @xmath65 and @xmath69 are fixed at the positions @xmath66 and @xmath70 in the external space . the result depends only on the difference @xmath71 and the number of links , @xmath72 , of the path connecting the vertices @xmath65 and @xmath69 . if one now changes the integration variables from vertex positions to link vectors , as before , one can see that all integrations , except those for links on the path between @xmath65 and @xmath69 , factorize . the sum of the link vectors on the path is restricted to @xmath73 . if we label the links of the path by consecutive numbers from @xmath74 to @xmath72 , we can write : @xmath75^n \ i \vec{p } \vec{x } } \ : , \end{aligned}\ ] ] where @xmath76 is the characteristic function of the probability distribution @xmath56 : @xmath77 it is important that the results of the integrations ( [ t1 ] ) and ( [ t2 ] ) do not depend on the internal geometry of the underlying tree graph . in particular , using ( [ t1 ] ) and ( [ t2 ] ) we find the partition function ( [ zeromode ] ) , the one - point and two - point correlation functions to reduce to the following form : @xmath78 we have denoted the ( canonical ) two - point function of the internal geometry by @xmath79 : @xmath80 it is normalized to unity : @xmath81 . the normalized internal two - point function gives us the probability that two randomly chosen vertices on a random tree of size @xmath16 are separated by @xmath72 links . in the last formula the geodesic internal distance between the vertices @xmath65 and @xmath69 is denoted by @xmath82 . the expressions ( [ izn ] ) for @xmath46 , ( [ g1 ] ) for @xmath83 and ( [ g2 ] ) for @xmath84 are independent of the link weight factor @xmath56 . thus the properties of the internal geometry , as mentioned , can be considered independently of and prior to the embedding . the external two - point function @xmath85 can be interpreted as the probability density for two random vertices on a tree of size @xmath16 to be embedded in the external space with the relative position @xmath86 . the probability normalization condition reads : @xmath87 the right hand side of ( [ g2 ] ) can be understood as a conditional probability . first , we choose two random vertices on a tree of size @xmath16 and calculate the probability @xmath79 that there are @xmath72 links on the path connecting them . for this path , which is a random path in the embedding space consisting of @xmath72 links , we can calculate the probability ( density ) @xmath88 that its endpoints are located with the relative position @xmath73 . since the internal geometry decouples from the external one , the probability densities @xmath88 and @xmath79 are independent of each other and can hence be calculated separately . similarly , higher correlation functions can be obtained from the corresponding internal correlation functions . for example , the three - point correlation function is : @xmath89 the three paths @xmath90 , @xmath91 and @xmath92 between the vertices @xmath65 , @xmath69 and @xmath93 of the tree can be decomposed into three pieces , namely @xmath94 , @xmath95 , @xmath96 between them and the common middle point @xmath60 . the summation indices @xmath97 , @xmath98 , @xmath99 denote the internal lengths of these pieces , and @xmath100 denotes the position of the common vertex in the external space . the internal three - point function then reads : @xmath101 one could extend this construction further . note that the most important piece of information is already encoded in the two - point function and is inherited by the higher correlation functions . in fact , one can directly derive the higher correlation functions from the two - point function , using a simple composition rule for the tree graphs which enormously simplifies in the grand canonical ensemble . in the next section we shall thus concentrate on the two - point function . the canonical two - point correlation functions @xmath79 and @xmath85 contain the information about the fractal structure of the internal and external geometry , respectively . the average distance for the internal geometry , given by the average number of links between two vertices on the tree , is the first moment of the probability distribution ( [ g2 ] ) : @xmath102 one expects the following scaling behaviour for large @xmath16 : @xmath103 the exponent @xmath1 relates the systems average ( internal ) extent @xmath104 to its size @xmath16 and is thus called the internal hausdorff dimension . this exponent controls the behaviour for large distances growing with the system size @xmath16 . one can also introduce a local definition of the fractal dimension for distances in the scaling window @xmath105 . the scaling window contains distances between the scale of the ultraviolet cut - off and below the infrared scale set by the system size . this is a sort of thermodynamic definition which becomes valid locally for sufficently large @xmath16 . in a large system one can be interested in how the volume of a local ball ( or sphere ) depends on its radius @xmath72 . the volume of the sphere can be calculated as the number of vertices lying @xmath72 links apart from a given vertex : @xmath106 the definition @xmath0 is more practical for a local observer , for example , someone who lives in a fractal geometry and wants to determine its fractal dimension . the global definition @xmath1 is accessible only for an observer who can survey the whole system from outside @xcite . in a similar way we can define the external hausdorff dimension . in order to do this we first have to introduce a measure of the systems extent in the external space . such a measure is provided by the gyration radius : @xmath107 where @xmath108 is the target space position of the systems center of mass . the statistical average of the gyration radius is directly related to the two - point function , namely : @xmath109 since the gyration radius is a translationally invariant quantity we have to normalize it with the total volume of the target space and rather refer to its average density . the external hausdorff dimension @xmath110 can then be read off from the large @xmath16 behaviour : @xmath111 the dependence on the volume @xmath50 is hidden in an @xmath16-independent constant which is not displayed in the last formula . the symbol @xmath112 referes to the leading behaviour . for trees of sufficently large size @xmath16 one can also define a local fractal dimension @xmath2 @xcite of the external geometry by measuring the average number of vertices within a spherical shell of radius @xmath113 from the scaling window @xmath114 , which is defined above the ultraviolet cut - off scale and below the infrared scale . as follows from the definition ( [ gnn ] ) for the case of two - point function , the number of vertices within a spherical shell of width @xmath115 is given by the two - point function @xcite : @xmath116 here we have used the fact that the the two - point function is spherically symmetric , i.e. @xmath117 . the integral over the @xmath9-dimensional angular part is included in the proportionality constant . in the large @xmath16 limit one expects the existence of a window @xmath114 where the two - point function exhibits the scaling behaviour @xcite : @xmath118 if @xmath119 is negative then @xmath120 behaves as a slow varying function of @xmath121 , which can in a narrow range and with some corrections be viewed as a constant @xmath122 . thus depending on the value of @xmath119 we have : @xmath123 in the first case @xmath124 the number of points in a spherical shell @xmath125 grows with the power of the canonical dimension @xmath9 . only in the second one @xmath126 the fractal nature leaves traces in the calculation of @xmath2 . in this section we will shortly summarize results concerning the classification of the scaling behaviour according to the internal geometry of the trees @xcite . one defines the critical exponent @xmath127 of the grand - canonical susceptibility via : @xmath128 where @xmath129 controls the behaviour of the partition function at the radius of convergence @xmath130 of the series ( [ grandz ] ) . here @xmath131 is the critical value of the chemical potential . if @xmath127 is positive the susceptibility @xmath132 itself is divergent at @xmath133 . if it is negative , the right - hand - side of the last equation should be understood as the most singular part of the susceptibility , which after taking higher derivatives , will give the leading divergence . the primary classification of the universality classes for models of branched geometry is based on the value of the susceptibility exponent @xmath127 . the susceptibility exponent gives the subexponential behaviour of the canonical coefficents @xmath134 for large @xmath16 : @xmath135 . indeed , if one inserts this form to the definition of the partition function ( [ grandz ] ) , one obtains : @xmath136 the susceptibility @xmath137 is proportional to the first derivative of the grand - canonical partition function @xmath138 for planted rooted trees , defined by equation ( [ caphi ] ) in the appendix : @xmath139 . the reason , why this relation is useful is , that there exists a closed relation a so - called self - consistency relation ( [ sc ] ) for @xmath138 : @xmath140 which can be inverted for @xmath141 and from which one can extract the singular part of @xmath138 : @xmath142 and hence also of @xmath132 . note that the denominator of ( [ gofphi ] ) is nothing but the first derivative of the potential @xmath143 defined ( [ v ] ) within the appendix . one can invert the function @xmath144 in the region where it is strictly monotonous . generically @xmath144 grows monotonically from zero for @xmath145 to some critical value @xmath146 at @xmath147 where @xmath148 . clearly the inverse function @xmath138 has a square root singularity at @xmath146 : @xmath149 then . it follows that @xmath150 for the class of generic trees . the region of the monotonous growths of the function on the right hand side of ( [ gofphi ] ) may be limited by @xmath147 being the radius of convergence of the series in the denominator of @xmath144 . the inverse function @xmath151 is then singular at @xmath152 with a singularity inherited from the singular behaviour of @xmath144 at @xmath147 . it can be shown that in this case the susceptibility exponent is negative and the corresponding trees are collapsed . in the marginal situation the two conditions which limit the region of the monotonous growth of @xmath144 work collectively at a point @xmath147 being at the same time the radius of convergence of the series in the denominator of @xmath144 and the zero of the derivative @xmath148 . in this case the exponent @xmath127 can assume any value within the interval @xmath153 . trees which belong to this class are called scale - free @xcite . + the three classes correspond to different scaling behaviours of the two - point function as will be discussed in the following section . in this section we shall calculate the two - point function for the internal geometry . this function will enable us to determine the scaling and the fractal properties of the internal geometry of tree graphs . as we will show they are different for generic , collapsed and scale - free trees . in the appendix we deduce an explicit formula ( [ tpdef ] ) for the grand - canonical two - point function @xmath154 . this function is singular for @xmath155 , and singularity is related to the large @xmath16-behaviour of the canonical two - point function @xmath156 . the singularity of @xmath154 can be determined directly from the identity ( [ tpdef ] ) by inserting the most singular part of @xmath157 to @xmath158 and @xmath159 . here we will show an alternative way using a standard scaling argument from statistical mechanics @xcite . denote the singularity exponent of the two - point funtion by @xmath160 : @xmath161 where @xmath93 is a constant which only depends on the particular choice of the weights @xmath25 . the exponent @xmath160 is usually called mass exponent . summing over distances @xmath72 we obtain the susceptibility ( [ susc ] ) : @xmath162 according to the definition ( [ susc ] ) , the susceptibility exponent is @xmath127 . thus we have @xmath163 this relation is the fisher scaling relation for this case . since we have already determined @xmath127 we do not have to calculate @xmath160 additionally . the scaling argument given above holds only for positive @xmath127 , since in this case the susceptibility is divergent and the divergent part dominates the small @xmath164 behaviour . for negative @xmath127 the left hand side of ( [ susc ] ) is not divergent which means that it behaves as a constant as @xmath164 goes to zero . in this case , if one compares the result of the integration ( [ integ ] ) to the leading behaviour of ( [ susc ] ) , one shall effective see that @xmath165 . this is what happens in the collapsed phase . now , inserting the most singular part of the grand - canonical two - point function ( [ gsing ] ) to the inverse laplace transform ( [ git ] ) we can deduce the large @xmath16 behaviour of the canonical two - point function : @xmath166 where @xmath167 is the lvy distribution with the index @xmath160 , the maximal asymmetry and the range @xmath168 @xcite . the large @xmath16 asymptotic behaviour of @xmath169 with @xmath170 is given by the following series @xcite : @xmath171 for large @xmath16 and fixed @xmath72 , the first term dominates the behaviour of the series : @xmath172 we see from the formulae ( [ gl2 ] ) and ( [ levy ] ) that the two - point correlation function in the large @xmath16 limit is effectively a function of the argument @xmath173 . indeed , if one changes the integration variable @xmath174 in ( [ levy ] ) to @xmath175 one obtains : @xmath176 , where @xmath177 is a function of a single argument @xmath178 . for later convenience we also included @xmath160 into the definition of the universal argument . using the saddle point approximation to the integral ( [ levy ] ) one can find that for large @xmath178 the function @xmath177 leads to : @xmath179 where @xmath180 and @xmath181 . the average internal distance between two vertices can then be calculated : @xmath182 comparing the @xmath16-dependence on the right hand side of this equation to the @xmath16-dependence on the right hand side of equation ( [ ddh ] ) , which defines the internal hausdorff dimension @xmath1 , we see that @xmath1 is the inverse of @xmath160 : @xmath183 thus , the hausdorff dimension is @xmath14 for generic trees . for scale - free trees @xmath1 changes continuously from @xmath184 to @xmath185 since @xmath127 belongs to the interval @xmath186 then @xcite . on the other hand we see from ( [ lmu ] ) that for large @xmath16 and small @xmath72 the normalized two - point function grows linearly with @xmath72 , i.e. : @xmath187 the normalization coefficent behaves as @xmath188 in the large @xmath16 limit . since the sum over this function is proportional to the number of vertices in the distance @xmath72 from a given vertex , the last formula tells us that the local hausdorff dimension is @xmath4 . we see that locally for sufficently large @xmath16 , it is difficult to distinguish the scale - free trees from the generic ones by measuring short internal distances , since both classes have the same internal hausdorff dimension @xmath4 . one has to go to large distances to see different scaling properties depending on the type of the ensemble ( [ spoint ] ) . for collapsed trees the hausdorff dimension is infinite . in this case , the universal scaling argument @xmath189 of the two - point function is proportional to @xmath72 but it does not depend on @xmath16 . this is related to the fact discussed before that the effective value of the exponent @xmath160 is equal zero . the saddle point approximation ( [ spoint ] ) actually gives the exact result for @xmath190 for the whole range of @xmath189 . the reason for this is that in this case the integrand of the approximated expression ( [ levy ] ) is gaussian . for some specific values of @xmath160 one can express the laplace transform ( [ levy ] ) in terms of special functions . for example for @xmath191 : @xmath192 where @xmath193 . for large @xmath189 the saddle point formula ( [ spoint ] ) coincides with this one , while for small @xmath189 the two functions deviate a little from each other . now we can determine the properties of the external geometry of gaussian trees . in this case , the weight @xmath194 in the partition function ( [ zn ] ) for embedding a link @xmath195 is given by a gaussian function . the function has a vanishing mean : @xmath196 in other words , for gaussian trees the link vectors are independent identically distributed gaussian random variables . as a consequence , the probability density @xmath88 ( [ convo ] ) for the endpoints of the path of length @xmath72 on a tree to have the relative coordinate @xmath100 is given by : @xmath197 this follows from the stability of the gaussian distribution with respect to the convolution . inserting the function @xmath88 to the formulae ( [ g2 ] ) , ( [ g3 ] ) , _ etc . _ we can determine the multi - point correlation functions for gaussian trees . in particular , if we insert ( [ fnx ] ) to ( [ g2 ] ) , we obtain in the large @xmath16 limit : @xmath198 here we used the same approximation for the internal two - point function as in the discussion of equation ( [ ggn2 ] ) in the appendix . this is a good approximation for large @xmath16 . additionally , we substituted the upper limit @xmath199 of the summation over @xmath72 by infinity . this introduces small corrections which disappear exponentially in the large @xmath16 limit . in order to measure the external hausdorff dimension @xmath3 we have to determine the dependence of the expectation value @xmath200 of the gyration radius on the system size @xmath16 . the expectation value can be calculated by integrating the two - point function over @xmath100 as in equation ( [ gyrr2 ] ) . if one first integrates over @xmath100 before summing over @xmath72 one obtains : @xmath201 one can approximate the right - hand side by replacing the summation from @xmath74 to @xmath185 through an integration over the whole positive real axis : @xmath202 we see that the typical extent of the system @xmath203 grows as @xmath204 and hence the hausdorff dimension for generic gaussian trees is @xmath205 . more generally , in order to determine the dependence of the expectation value of the gyration radius on the system size for any type of trees one can first calculate the second moment of the function @xmath88 : @xmath206 which corresponds to the average extent of the path built out of @xmath72 links of the tree . the insertion of this result to ( [ gyrr2 ] ) yields : @xmath207 since for gaussian weights the second moment @xmath208 is proportional to @xmath72 , i.e. @xmath209 , the following relation holds @xcite : @xmath210 from which we conclude that the external and internal hausdorff dimensions are related by : @xmath211 for gaussian trees . this relation holds for generic , scale - free and collapsed trees . this for example means that the hausdorff dimension @xmath110 of collapsed trees is infinite , or in other words , that the target space extent of the system does not change with the number of vertices on the tree . let us come back to generic gaussian trees . we will calculate the local hausdorff dimension @xmath2 and compare it with @xmath205 . the starting point of this calculation is equation ( [ shell ] ) which relates the number of points within the shell of radius between @xmath121 and @xmath212 to the behaviour of the two - point function in the scaling window @xmath213 , where @xmath214 , @xmath215 . the two - point function ( [ gg2 ] ) is a decreasing function . it has a cut - off at @xmath216 as follows from the scaling arguments . the large @xmath72-part of the sum ( [ gg2 ] ) over @xmath72 : @xmath217 can be approximated by an integral over @xmath72 . this part of the sum has a significant contribution if @xmath218 . thus for @xmath219 the dominant dependence of the sum on @xmath121 can be approximated by : @xmath220 with some constants @xmath221 . the upper limit of the integral comes from the term @xmath222 . the exact shape of the integrand at large @xmath223 is unimportant for @xmath224 , because the dominating behaviour @xmath225 is due to the lower limit of the integration . this scaling form of @xmath120 breaks down for short distances @xmath121 of order @xmath226 and for large @xmath121 of the order of the infrared cut - off @xmath227 . when @xmath121 is of the order @xmath226 the integrand is a sum of gaussians of widths larger than @xmath121 and hence @xmath120 is a slow varying function . for @xmath228 the regime changes . the divergence at small @xmath72 disappears and the terms for large @xmath72 , @xmath223 , dominate in the sum . the sum ( [ gg2 ] ) , viewed as a function of @xmath121 , looks as a sum of gaussians whose arguments @xmath121 are maximally of the order of the widths . this is a slow varying function of @xmath121 for @xmath229 . hence we expect that as long as @xmath230 it is almost constant @xmath122 . as an example we performed the sum ( [ gg2 ] ) numerically for @xmath16 up to @xmath231 . the results presented in the figures [ d2fig ] and [ d10fig ] corroborate the anticipated behaviour of @xmath120 by the arguments given above @xcite . as a consequence we see that the number of points within the spherical shell ( [ shell ] ) depends on the radius @xmath121 as : @xmath232 this leads to the following result for the hausdorff dimension : @xmath233 in other words , the fractal dimension @xmath2 measured by the local observer , is equal to the global one , i.e. @xmath234 , if the dimension of the target space is large enough . if the target space dimensionality is too small , the fractal structure can not develop . one can understand this in the following way . for trees embedded in a @xmath235 dimensional target space , vertices of the tree lie in a ball with a radius proportional to @xmath204 . there are @xmath16 vertices within the ball while the volume of the ball is proportional to @xmath236 . this means that for large @xmath16 , the vertices deep inside the ball are densily and uniformly packed . a local observer who surveys only a small region far from the ball boundary will see uniformly distributed vertices in a @xmath9 dimensional space . as a consequence he or she will measure @xmath11 . the situation changes for @xmath224 , because then the volume of the ball is proportional to @xmath236 and hence grows much faster than the the number of the vertices . in the large @xmath16 limit the volume of the ball will therefore be large enough to let the system develop a loose fractal structure . similarly , we expect that for the scale - free trees , the local hausdorff dimension is @xmath237 since the hausdorff dimension @xmath110 is infinite for collapsed trees , one can not define a local hausdorff dimension @xmath2 in the same manner as above , because the infrared cut - off is a constant in this case . so far we have only considered gaussian embedding weights @xmath194 ( [ f1x ] ) . in this case links typically have the length @xmath226 and one can hardly find a link on a tree longer than @xmath238 . in other words the energy costs for the embedding of long links are so high that those links do not appear . one can , however , consider models with weights which allow long links . in the next section we will discuss this issue in a more general context , while in this section , as toy models , we will consider models of trees embedded in @xmath239 dimensional target space with the weigths given by a symmetric lvy distribution @xcite . despite their simplicity , the models with those weights already basically capture all interesting features of more complicated models . the weights read : @xmath240 with @xmath6 from the interval @xmath241 $ ] . in the limiting case @xmath7 is a gaussian distribution with the width @xmath242 , and for @xmath243 is a cauchy distribution . the weights are symmetric stable distributions with the stability index @xmath6 . here we are interested only in symmetric functions @xmath244 because the links are unoriented . this implies that any function defined on them has the property : @xmath245 . the distribution ( [ flevy ] ) is stable with respect to the convolution : @xmath246 where @xmath247 . if one repeats this for the convolution of @xmath72 identical terms to calculate @xmath248 ( [ charact ] ) one can see that the function @xmath248 is given by a rescaled version of the function for a single link @xmath249 : @xmath250 now we can combine this scaling with the scaling of the internal two - point function which , as we know , is a function @xmath251 of a scaling variable @xmath252 , to deduce the scaling of the external two - point function ( [ g2 ] ) : @xmath253 the result of the integration can be written as a function of an argument @xmath254 with some prefactor depending on @xmath16 . as a consequence one expects the external hausdorff dimension to be : @xmath255 the case @xmath7 was discussed before . despite similarities , the case @xmath256 is different from the gaussian one , since in this case the distribution @xmath257 has a fat tail for large @xmath258 : @xmath259 which according to the scaling ( [ ffn ] ) is equally important in @xmath248 for any @xmath72 . the second moment @xmath208 of the distribution @xmath248 is infinite . as a consequence , also the gyration radius is infinite . one has to find an alternative measure of the linear system extent in order to define the hausdorff dimension @xmath3 . a natural candidate for such a quantity is : @xmath260 for @xmath261 . the hausdorff dimension can now be calculated from the large @xmath16 behaviour of this quantity : @xmath262 using the same arguments as for the gaussian case , one can check that the following relations hold @xmath263 where @xmath264 it follows , that @xmath265 and hence @xmath5 , as already mentioned . we will continue the discussion of the one - dimensional case , i.e. @xmath239 . the extension to higher dimensions @xmath9 shall afterwards be straightforward . the embedding weights for links may in general be given by any normalizable non - negative symmetric function : @xmath266 such that : @xmath267 . we are interested in the emergence of the scaling properties for large @xmath16 . from the considerations of the internal geometry we know that the internal distance between two random vertices on the tree , @xmath268 , grows with @xmath16 unless the trees are collapsed . we also know that between those two random vertices we can draw a unique path on the tree . this path can be treated as a random path of @xmath72 links . so , in a sense , we are interested in the probability distribution that the remote ends of the random path with @xmath72 links have the the relative position @xmath121 in the embedding space . in particular we are interested in the limit @xmath269 . this probability distribution is given by @xmath248 . for large @xmath72 the function @xmath248 can be determined from the central limit theorem . roughly speaking , if the second moment of @xmath257 exists , @xmath248 approaches a gaussian distribution with the variance @xmath270 , otherwise @xmath248 approaches the lvy distribution ( [ flevy ] ) with the scale parameter @xmath271 . thus , if a distribution has a power - tail @xmath272 for large @xmath258 , the limiting distribution @xmath248 for large @xmath72 will approach the gaussian distribution if @xmath273 , or the lvy distribution ( [ flevy ] ) if @xmath256 @xcite . the limiting case @xmath274 belongs to the gaussian domain of attraction but it has a logarithmic anomaly of the variance which in this case does not grow as @xmath275 but faster , i.e. with some additional logarithmical factor of @xmath72 . for @xmath276 the approach of @xmath248 to the gaussian distribution for large @xmath72 is non - uniform and takes place in the central region of the distribution @xmath248 for @xmath277 , where @xmath278 scales with @xmath72 as @xmath279 here @xmath69 is some constant @xcite representing a scale of the distribution . outside this region @xmath248 deviates severe from the normal law , and in particular it preserves the power - law tail for @xmath280 : @xmath281 with a tail amplitude proportional to @xmath72 . in other words , for any finite @xmath72 the power - law tail is present in the distribution @xmath248 . therefore all absolute moments of order @xmath282 of this distribution are infinite : @xmath283 for finite @xmath72 , and as a consequence also @xmath284 for @xmath41 the sum is dominated by terms of large @xmath285 . for @xmath286 the distribution @xmath248 becomes normal in the whole region from @xmath287 to @xmath288 . indeed , the ends of the central region @xmath278 move to infinity faster than the variance @xmath289 and the contribution coming from the outside of the central region @xmath290 disappears as : @xmath291 thus the non - gaussian part including the tails becomes marginal and can be neglected . what is left over for @xmath292 is a gaussian distribution with all moments defined . for example , even integer moments @xmath293 are : @xmath294 . thus , after taking this limit the trees behave like gaussian ones . this limit is subtle , because as long as @xmath16 ( and @xmath72 ) is large but finite higher moments , @xmath282 , of the distribution are infinite . now we shall shortly discuss the model in @xmath295 dimensions . as before we consider spherically symmetric distributions @xmath296 , @xmath297 which have a power - law dependence for large lengths of the link vectors @xmath298 : @xmath299 where @xmath300 is the angular part of the measure . the main difference to the one dimensional case is that the effective power of the link length distribution changes by @xmath301 due to the angular measure @xmath302 . otherwise , the dependence of the scaling on @xmath6 goes in parallel to the one dimensional case that is the distribution belongs to the gaussian domain of attraction if @xmath303 and to the lvy one if @xmath304 . the characteristic function ( [ charact ] ) of the corresponding limiting distribution is spherically symmetric : @xmath305 , where @xmath306 and @xmath307 . for example , a distribution which has a power - law tail @xmath308 belongs to the domain of attraction of the cauchy distribution : @xmath309 a new effect which arises in higher dimensions is a possibility of a spontaneous breaking of the rotational symmetry . the limiting distributions for @xmath310 and the two - point function @xmath85 are spherically symmetric but the configurations which contribute to them are not . the effect is strong when @xmath311 and is known from the considerations of lvy paths @xcite . if we have such an ensemble of @xmath16 links , one can find a link whose length is @xmath312 larger than the sum of the lengths of the remaining links . this link makes the system look like a one dimensional system , since the extent of the system in the direction of this link is significantly larger than in the other directions . the effect becomes weaker when @xmath6 is larger than one . actually it is then seen for configurations which come from the large @xmath121-tail of the two - point function @xmath313 . this configurations become marginal for @xmath273 in the large @xmath16 limit . however , as long as @xmath16 is finite , the probability of large @xmath121 in @xmath313 is finite and it strongly influences the measurements of higher moments @xmath314 of the system extent . in other words the higher @xmath48 is , the stronger is the contribution from the large @xmath121 part of the two - point function and the more contribute systems , which are elongated , to this quantity . in the limiting case : @xmath315 the main contribution to the moments @xmath316 comes from one - dimensional configurations . as mentioned , the branched polymer model with power - law weight arises as the one - loop approximation of the reduced super - symmetric yang - mills model @xcite . in particular for @xmath317 dimensions , the embedding weights @xmath257 behave as @xmath318 for large link lengths @xmath258 . this is the marginal case @xmath7 which belongs to the gaussian domain of attraction . this means in particular that if one first takes the limit @xmath310 then the gaussian branched polymers emerge for which all correlation functions are well defined . on the other hand if one determines higher correlation functions before one takes @xmath310 one shall see they are divergent . in numerical simulations of the full matrix model one also observes power - law tails in the distribution of the system extent and one dimensional configurations @xcite-@xcite . it is possible @xcite , but not yet answered , that there also exists a gaussian limit at large @xmath16 in this model . we investigated the model of trees embedded freely in a @xmath9 dimensional target space . we classified the scaling properties of the model by determining the fractal dimensions for internal and external geometry for the ensembles of generic and exotic trees including those which have fat tails in the distributions of branching orders and of link lenghts in the embedding space . we showed that , for freely embedded trees , internal geometry is indepedent of the embedding as a result of the factorization ( [ gfg ] ) . on the other hand external geometry strongly depends on internal one : in particular the hausdorff dimension for external geometry is proportional to that for internal one @xmath319 . the proportionality coefficent is given by the stability index of the embedding weights . for gaussian trees , in particular , it is equal two . we pointed out that the finite effects related to the presence of fat - tails lead to singularities of higher order correlation functions before the inifinite large @xmath16-limit is taken . this is a similar effect to the one observed in the ikkt matrix model . the branched polymer model captures many features of the more complicated models of random geometry . despite its simplicity the model has a rich phase structure : a generic phase of gaussian trees which have the hausdorff dimensions @xmath205 , the phase of short trees with @xmath320 coming from the embedding of scale - free and crumpled tree graphs , and the the phase with elongated lvy branches , which has the hausdorff dimension @xmath321 . due to the simplicity , and the full control of the free case , the model of branched polymers which we discussed here , provides a good starting point for modelling effects of non - trivial embedding like those related to excluded volume effects or external curvature . such effects violate the factorization introducing correlations between the internal and external geometry . a sort of back - coupling occurs . the external geometry affects the internal one , which modifies and influences back the external one . for example , self - avoidance disfavours crumpled trees and hence changes the internal hausdorff dimension which in turn will change the external one . * acknowledgements * : we thank p. bialas for discussions . this work was partially supported by the ec ihp grant hprn - ct-1999 - 000161 and by the project 2 p03b 096 22 of the polish research foundation ( kbn ) for 2002 - 2004 . thanks the university of bielefeld for a graduate scholarship . thanks the alexander von humboldt foundation for a follow - up fellowship . this appendix summarizes important relations of the internal geometry of branched polymer models @xcite . it is intended to make this article more self - contained . as we already mentioned , the part of the model related to the internal geometry decouples from the problem of the embedding and can hence be solved independently . it is convenient to introduce several generating functions to ease the calculations . although many of the considerations made here are well - known , we deduce several relations for the generating functions which play important roles in the description of branched polymer models . graphical representations of the generating functions turn out to be effective tools for the just mentioned deductions . finally we will be able to calculate the partition function @xmath46 ( [ izn ] ) and the two - point function @xmath322 ( [ g2 ] ) of the internal geometry . note that we already discussed the scaling properties and the universality resulting from those calculations in the previous sections . recall that the internal properties of a branched polymer model ( [ zn ] ) only depend on the internal weight function @xmath323 . for internal weights of the form ( [ weightdef ] ) these properties are entirely determined when the whole set of branching weights @xmath324 , obeying the conditions ( [ wcond ] ) , is given . the information about the entire set @xmath27 of branching weights can alternatively be encoded in a single function of one real variable , namely ( [ v ] ) . @xmath325 as we shall show below , the scaling properties of the internal geometry are directly related to the analytic properties of this function . we will often refer to @xmath50 as a potential , since the most important generating functions can be written as derivates of @xmath50 . in the first section we defined the generating function for the canonical partition functions @xmath46 to be : @xmath326 which is nothing else but the grand - canonical partition function for the ensemble of trees with unrestricted size . one reason why it is convenient to introduce the generating function @xmath47 is that one can write a closed self - consistency equation for a first derivative of it , as we shall discuss below . the grand canonical partition function can be written as : @xmath327 where @xmath328 denotes the number of vertices of order @xmath329 and the sum begins with @xmath330 being the number of vertices on the smallest tree . note that each vertex introduces a factor @xmath331 to the ( internal ) weight of the tree in the grand - canonical ensemble . there are two derivatives of the generating function @xmath47 which will be useful : @xmath332 where @xmath333 clearly , @xmath334 is a generating function for the canonical partition functions @xmath335 of trees with @xmath16 vertices that have one marked vertex . intuitively the factor @xmath16 in the sum can be viewed as a factor which counts the possible choices of marking one vertex on a tree with @xmath16 vertices . the derivative @xmath138 is a generating function for the partition functions @xmath336 of branched polymers of size @xmath16 having one ( not counted ) additional marked vertex of order one . we will refer to those uncounted vertices as external lines . because we do not count the empty lines in ( [ caphi ] ) the sum starts with @xmath337 . we now introduce a graphical notation for the generating functions with the following conventions : whenever a vertex is represented by an empty circle , this means that this vertex neither introduces a weight @xmath25 corresponding to its order nor a fugacity @xmath43 . consequently those vertices are not counted . solid circles correspond to counted vertices and therefore introduce factors @xmath25 and @xmath43 . combinatorial factors which are due to certain symmetries of the represented object will always be displayed explicitly . links between vertices will be represented by solid lines . as can be seen in figure [ gfp ] the generating function @xmath47 will be represented by a bubble , its derivative @xmath334 by a bubble with a filled circle inside and its derivative @xmath138 by a bubble with a tail having an empty circle at the end . the tail corresponds to the external line of the tree . in a similar way one can also define higher derivatives of @xmath47 . each derivative @xmath338 introduces a new marked vertex and hence another filled circle in the bubble of the graphical representation . each derivative @xmath339 introduces a new external line with an empty circle at the end . one could also define derivatives @xmath340 which introduce an external uncounted vertex connected to the bubble via @xmath341 links . the most fundamental object among all these generating functions is the generating function @xmath138 for planted trees . with its help one can construct all the others . for example , the combination @xmath342 is a generating function for trees with one marked vertex of the order @xmath329 . if one sums over @xmath329 one obtains a generating function for trees which have just one marked vertex of any order . this is nothing else but @xmath334 itself . thus we have : @xmath343 if one adds a line with an empty end to this marked vertex one obtains the generating function for planted trees . the order of the marked vertex to which the line is added consequently increases by one , i.e. @xmath344 . thus the corresponding contribution to the sum over @xmath329 is @xmath345 : @xmath346 which is a self - consistency equation for @xmath138 from which @xmath138 can be calculated . having calculated @xmath138 one can insert it to ( [ nsc ] ) and determine @xmath334 and so on . as we have seen above @xmath347 generates trees with one marked vertex , @xmath348 trees with one marked vertex which is connected to an external line . one can easily check that the @xmath341-th derivate of @xmath143 : @xmath349 generates an ensemble of trees with one marked vertex connected to @xmath341 external lines . the graphical representations of @xmath350 for @xmath351 is depicted in figure [ phifig ] . let us now illustrate how the generating function machinery works solving the classical problem of the tree diagram enumeration @xcite . we shall calculate @xmath46 for the case @xmath24 . this is called the cayley problem . the number of all labeled trees with @xmath16 vertices is given by @xmath352 . the self - consistency equation reduces to : @xmath353 this equation can be solved for @xmath138 : @xmath354 for the weights @xmath355 equation ( [ nsc ] ) leads to a simple relation between @xmath334 and @xmath138 : @xmath356 from this relation one can calculate the canonical partition function : @xmath357 and the number of labeled tree diagrams to be @xmath17 . for large @xmath16 one can approximate @xmath46 by : @xmath358 where the last step is due to the comparison with ( [ susc ] ) . the critical value of the chemical potential and the susceptibility exponent take the values @xmath359 and @xmath150 , respectively . it turns out that the value @xmath150 is a generic one . it does not change for a wide class of the weights . in the section about the universality classes and singularity types two other universality classes of branched polymer models with @xmath360 will be discussed . we will close this appendix with the calculation of the internal two - point function . similarly as for the partition function , it is easier to work with the generating function . consider tree graphs which are weighted with the fugacity @xmath361 , and which have two marked vertices separated by @xmath362 links . the generating function @xmath363 defined as a sum over all such trees corresponds to the two - point function for the grand - canonical ensemble . figure ( [ twopf ] ) illustrates the decomposition of @xmath364 into the generating functions @xmath365 and @xmath366 depicted in figure ( [ phifig ] ) . the decomposition is unique , since the path connecting the marked vertices is unique . the two bubbles at the ends of the chain correspond to diagrams of the generating function @xmath365 , while the @xmath367 ones in between to @xmath366 . the decomposition leads to the following relation : @xmath368^{2 } \cdot \left [ v''\left ( \phi \right ) \right]^{\left ( n-1 \right ) } \ : . \label{tpdef}\ ] ] we can also define the internal grand - canonical two - point function for @xmath369 . in this case the two marked vertices lie on top of each other . thus the two - point function reduces to the one - point function : @xmath370 . relation ( [ tpdef ] ) allows us to find an explicit dependence of the grand - canonical two - point function on @xmath72 and @xmath371 , if we first solve the self - consistency equation ( [ gofphi ] ) for @xmath372 . we will be rather interested in the scaling behaviour of the two - point function near the critical point . let us first illustrate the calculation of the two - point function for the ensemble of trees having the natural weight @xmath24 . in this case @xmath373 , where @xmath372 is a solution of equation ( [ gofphi ] ) : @xmath374 in this case the two - point function simplifies to : @xmath375 the inversion of equation ( [ muphi ] ) for @xmath138 gives @xmath376 , when @xmath377 . thus the two - point function can be approximated in this limit by : @xmath378 we have neglected a term linear in @xmath164 in the exponent , because we are interested in the limit @xmath379 for which @xmath380 . what matters in this limit is the leading term in @xmath164 which is related to the large @xmath16-behaviour of the underlying canonical - ensemble : @xmath381 where @xmath382 is the two - point function for the canonical ensemble for trees of size @xmath16 . in the last formula we split into @xmath383 , where @xmath131 is the critical value of @xmath371 at which the partition function is singular . the leading terms in @xmath164 of @xmath384 are responsible for the scaling behaviour while the next - to - leading ones for finite - size corrections . formula ( [ lt ] ) is a discrete laplace transform . since we are interested in the large @xmath16 behaviour of @xmath382 , we can substitute the discrete by a continuous laplace transform . the inverse transform then yields : @xmath385 in particular , for the case discussed here ( see([g22 ] ) ) the exact result reads : @xmath386 we inserted @xmath359 in the last formula . the normalized two - point function can be approximated by : @xmath387 the constant @xmath93 in the last formula is equal @xmath74 . we displayed this constant , because the same formula holds for generic trees in general , but with a constant which depends on the choice of the weights . we used two approximations in the last formula . we substituted @xmath388 by @xmath72 . the difference between the function with @xmath72 and @xmath388 disappears in the large @xmath16-limit . secondly , the numerator and the denominator in the normalized two - point function ( [ ggn2 ] ) have a common part @xmath389 which does not depend on @xmath72 . it cancels out . the remaining part : @xmath188 is a normalization constant ensuring @xmath390 . for finite @xmath16 there will be some corrections to the normalization constant @xmath188 , but these disappear exponentially in the large @xmath16 limit . 99 jan ambjrn , b. durhuus and t. jonsson , _ quantum geometry _ , cambrige , 1997 . bowick and a. travesset , phys . rep . * 344 * ( 2001 ) 255 . wiese , in phase transition and critical phenomena 19 , ed . domb and lebowitz , academic press , new york , ( 2000 ) , 253 . j. ambjrn , b. durhuus , j. frhlich and p. orland , nucl . phys . * b270 * [ fs16 ] ( 1986 ) 457 . jan ambjrn , b. durhuus and t. jonsson , phys.lett . * b244 * ( 1990 ) 403 . r. albert and a.l . barabasi , rev.mod.phys . * 74 * ( 2002 ) 47 . z. burda , j.d . correia and a. krzywicki , phys.rev . * e64 * ( 2001 ) 046118 . m. polyakov , phys.lett . * b103 * ( 1981 ) 207 . f. david , nucl.phys . * b257 * ( 1985 ) 45 . kazakov , a.a . migdal and i.k . kostov , phys.lett . * b157 * ( 1985 ) 295 . j. ambjrn and j. jurkiewicz , phys.lett . * b278 * ( 1992 ) 42 . p. bialas , z. burda , a. krzywicki and b. petersson , nucl.phys . * b472 * ( 1996 ) 293 . j. ambjrn , j. jurkiewicz and r. loll , phys.rev . * d64 * ( 2001 ) 044011 . p. bialas , phys.lett . * b373 * ( 1996 ) 289 . j. jurkiewicz and a. krzywicki , phys.lett . * b392 * ( 1997 ) 291 . p. bialas , nucl.phys . * b575 * ( 2000 ) 645 . t. jonsson and j. wheater , nucl.phys . * b515 * ( 1998 ) 549 - 574 . p. bialas and z. burda , phys.lett . * b384 * ( 1996 ) 75 . w. krauth , h. nicolai , and m. staudacher , phys.lett . * b431 * ( 1998 ) 31 . n. ishibashi , h. kawai , y. kitazawa and a. tsuchiya , nucl . b498 * ( 1997 ) 467 . h. aoki , s. iso , h. kawai , y. kitazawa and t. tada , prog . * 99 * ( 1998 ) 713 . h. aoki , s. iso , h. kawai and y. kitazawa , phys . e62 ( 2000 ) 6260 j.p . bouchaud and a. georges , phys . * 195 * ( 1990 ) 127 . w. feller , _ an introduction to probability theory and its applications _ , wiley and sons , new york , 1957 . j. ambjrn , k.n . anagnostopoulos , w. bietenholz , t. hotta and j. nishimura , jhep * 0007 * ( 2000 ) 011 . j. ambjrn , k.n . anagnostopoulos , w. bietenholz , t. hotta and j. nishimura , jhep * 0007 * ( 2000 ) 013 . j. ambjrn , k.n . anagnostopoulos , w. bietenholz , f. hofheinz , and j. nishimura , phys.rev . * d65 * ( 2002 ) 086001 . z. burda , b. petersson and j. tabaczek , nucl.phys . * b602 * ( 2001 ) 399 . p. bialas , z. burda , b. petersson and j. tabaczek , nucl.phys . * b592 * ( 2001 ) 391 . p. austing and j.f . wheater , jhep * 0102 * ( 2001 ) 028 . p. austing and j.f . wheater , jhep * 0104 * ( 2001 ) 019 . g. vernizzi and j.f . wheater , het - th 0206226 j. nishimura , phys.rev . * d65 * ( 2002 ) 105012 . p. bialas , z. burda and d. johnston nucl.phys . * b493 * ( 1997 ) 505 . p. bialas , z. burda , b. petersson and j. tabaczek , nucl.phys . * b495 * ( 1997 ) 463 . j. ambjrn and y. watabiki , nucl . phys . * b445 * ( 1995 ) 129 . goulden and d.m . jackson , _ combinatorial enumeration _ , wiley and sons , new york , 1983 . like , for instance , a theoretical physicist performing analytical calculations . in the same way one can show that all moments of the two - point function scale uniformly for large @xmath16 : @xmath391 which means that the two - point function is indeed a function of the scaling argument @xmath392 for large @xmath16 . for the simulation we used the exact form of @xmath120 with the constants @xmath93 and @xmath226 set equal to one : + @xmath393 if all moments of @xmath257 exist , the size of the central region grows faster with @xmath72 . for example as @xmath394 for an even distribution . therefore the gaussian regime sets in faster .

we give a new construction of the minimal unitary representation of the exceptional groupon a hilbert space of complex functions in 29 variables . due to their manifest covariance with respect to thesubgroup ofour formulas are simpler than previous realizations , and thus well suited for applications in superstring and m theory . hep - th/0109005

hep-th0109005

in this paper , we present a new construction of the minimal unitary representation of @xmath0(_i.e . _ the maximally split real form of the exceptional lie group @xmath2 ) on a hilbert space of complex functions in 29 variables . the minimal realizations of classical lie algebras and of @xmath3 were given by joseph a long time ago @xcite . the existence of the minimal unitary representation of @xmath0was first proved by vogan @xcite who located it within the framework of langland s classification . later , the minimal unitary representations of all simply laced groups , including @xmath0 , @xmath1and @xmath4were constructed by kazhdan and savin @xcite , and brylinski and kostant @xcite by rather different methods . gross and wallach gave yet another construction of the minimal representation of @xmath0as well as for all exceptional groups of real rank four @xcite . for the exceptional group @xmath0 the minimal orbit is 58 dimensional , corresponding to its quotient by a distinguished parabolic subgroup . the representations of @xmath0 over the spaces of functions on the 128 dimensional coset space @xmath5 were studied in @xcite . while formulas for @xmath3 similar to the ones derived here for @xmath0can be found in @xcite , however , an explicit realization of the simple root ( chevalley ) generators in terms of pseudo - differential operators for the simply laced exceptional groups was given only very recently @xcite , together with the spherical vectors necessary for the construction of modular forms . we present here an alternative realization of @xmath0 , which has the advantage of yielding very compact formulas , in contradistinction to the rather complicated expressions obtained by multiple commutation of the simple root generators @xcite . the main reason for the relative simplicity of our final expressions ( [ e7gen ] ) and ( [ e7e8gen ] ) is their manifest @xmath6 covariance , in spite of the fact that @xmath6 is not realized linearly . the compactness of our formulas makes them especially suitable for applications in string and m theory which have primarily motivated the present work . although the present article is mainly addressed to a physicists audience , let us first summarize the main results in a somewhat more mathematical language . the minimal representation of a noncompact group @xmath7 corresponds , in general , to the quantization of its smallest co - adjoint orbit . in the case of @xmath0 the minimal co - adjoint orbit is 58-dimensional . starting from the 5-graded decomposition of the associated lie algebra @xmath8 given in eq . ( 4 ) below , it can be obtained simply as the co - adjoint orbit of the highest root generator ( designated by @xmath9 in eq . ( 6 ) ) . in this 5-graded decomposition of @xmath8 one can readily identify a maximal non - semisimple subalgebra @xmath10 as the annihilator of this generator , _ i.e. _ @xmath11=0\ ] ] the subalgebra @xmath10 is generated by the grade zero subalgebra @xmath12 together with the grade 1 and grade 2 elements of the full @xmath8 lie algebra ( _ i.e. _ the generators @xmath13 and @xmath9 in eq . ( 6 ) ) . the orthogonal complement of this parabolic subalgebra is a 58 dimensional nilpotent subalgebra @xmath14 . in the notation of section 2 , the latter is generated by the grade 0 element @xmath15 , the grade @xmath16 elements @xmath17 , and the grade @xmath18 element @xmath19 . acting with this nilpotent subalgebra on the generator @xmath9 corresponding to the highest root one obtains the generators corresponding to the minimal orbit . thus we see that the minimal orbit can be identified with the coset space @xmath20 where @xmath21 is the parabolic subgroup generated by @xmath10 . we were led to this minimal unitary representation rather naturally from the novel geometric realization of @xmath0 on @xmath22 found in our previous work @xcite . as shown there , the exceptional groups can be realized via conformal or quasiconformal transformations leaving invariant generalized `` light - cones '' ; these transformations are analogous to the non - linear action of the mbius group on the real line . our construction relied essentially on the connection with jordan algebras and freudenthal triple systems . to proceed from there to the unitary realization on a suitable ( infinite dimensional ) hilbert space of functions , we must first identify a phase space realization of this system and then quantize it . for the @xmath1subgroup this phase space realization involves 28 coordinates @xmath23 and 28 momenta @xmath24 , such that all @xmath1variations can be realized via the canonical action @xmath25 where @xmath26 is the appropriate @xmath1charge . for the canonical realization of the full @xmath0we have to add the momentum @xmath27 conjugate to the 57th coordinate @xmath28 , and replace the mbius action on @xmath28 by the symplectic realization acting on the two dimensional phase space spanned by @xmath28 and @xmath27 . the minimal unitary representation of @xmath0is then obtained by quantization , i.e. replacement of the classical momenta by differential operators . this requires in addition a prescription how to deal with the ordering ambiguities arising in the non - linear expressions for the lie algebra generators . the latter are then realized as self - adjoint operators on some dense subspace of @xmath29 . this coordinate space ( schrdinger ) representation is reformulated in section 3 as an oscillator realization in terms of annihilation and creation operators acting in a particle basis . by going to the corresponding coherent state basis of this oscillator realization labelled by 29 complex coordinates one obtains the fock - bargmann realization of the minimal unitary representation over an hilbert space of holomorphic functions . the irreducibility of the minimal representation is put in evidence by showing that the quadratic casimir operator of @xmath0 , when expressed in terms of either the coordinate or the oscillator basis reduces to a @xmath30-number ; a general argument shows that likewise the higher order casimirs reduce to @xmath30-numbers in our realization . when restricted to the subgroup @xmath31 the minimal representation decomposes into an infinite sum ( actually an integral , see @xcite ) of irreducible representations of @xmath31 . remarkably , the quadratic casimir of @xmath1 obtained from the minimal realization is identical with the quartic invariant of @xmath1 when expressed as a function of the 28 coordinate and 28 momentum operators ( or equivalently , the 28 annihilation and 28 creation operators in the oscillator basis ) . the irreducible representations of @xmath32 that occur in the decomposition of the minimal representation @xmath0 w.r.t . @xmath31 are all labelled by the eigenvalues of this quartic invariant . as already mentioned our chief aim with the present work is to present the results in such a way that they can be readily applied in the context of string and m theory ( for instance , readers familiar with @xcite should have no difficulties following our exposition ) . in particular , our results apply in the context of the ( super)conformal quantum mechanics description of quantum black holes @xcite . in that work , unitary representations of @xmath32 played a crucial role in the classification of physical states . it is therefore an obvious idea to extend these concepts to maximal supergravity . indeed , the candidate hamiltonian which we obtain as one of the @xmath0lie algebra generators reads @xmath33 \label{l0}\ ] ] which is precisely of the form studied by the authors of @xcite , with the only difference that they have a ( coupling ) constant instead of the differential operator @xmath34 . here @xmath35 is the quartic invariant of @xmath1expressed as a function of the 28 + 28 variables @xmath23 and @xmath24 ; when acting on one of the irreducible subrepresentations of @xmath1under its subgroup @xmath36 it yields the associated eigenvalue of the quadratic @xmath6 casimir . as shown for instance in @xcite in the context of the classical theory , the quartic invariant can assume both positive and negative values , and vanishes for @xmath37 or @xmath38 bps black hole solutions of @xmath39 supergravity @xcite . it is therefore tempting to interpret ( [ l0 ] ) as the effective hamiltonian describing ( in some approximation ) @xmath39 quantum black holes , such that every subrepresentation with a fixed eigenvalue of the operator @xmath35 is identified with the space of physical states associated with the corresponding black hole solution of @xmath39 supergravity . an interpretation along similar lines has also been suggested in @xcite . interestingly , for the vanishing eigenvalues of @xmath35 , the hamiltonian simplifies drastically , and the state space reduces to the well known singleton representation of @xmath32 . another ( and possibly related ) physical application of minimal representations has been outlined in @xcite . that work evolved from an earlier attempt to determine the @xmath40 corrections directly from the supermembrane and to understand them in terms of so - called `` theta - correspondences '' @xcite ; see also @xcite for a related attempt to come to grips with the non - linearities of the supermembrane . we first recall some basic features of the non - linear realization of @xmath0on @xmath22 coordinatized by 57 real variables @xmath41 ( @xmath42 ) , see @xcite for our notations and conventions , and further details . a key ingredient in that construction was the 5-graded decomposition of @xmath0w.r.t . its subgroup @xmath43 . denoting its lie algebra by @xmath8 we have @xmath44 an important property of this decomposition is the fact that the subspaces of grade @xmath45 and @xmath46 together form a maximal heisenberg subalgebra . the corresponding generators @xmath47 and @xmath48 obey the commutation relations @xmath49 } & = & 2{\mathrm{i}}\,\delta^{ij}_{kl } e \,.\ ] ] obviously , this algebra can be realized as a classical poisson algebra on a phase space with 28 coordinates @xmath23 and 28 momenta @xmath50 , and one extra real coordinate @xmath28 to represent the central term . we have @xmath51 it is then straightforward to determine the generators of the @xmath12 subalgebra in terms of these phase space variables ( for instance by requiring that they reproduce the @xmath1variations _ via _ ( [ poisson ] ) ) . they are realized by the following 63 + 70 bilinear expressions in @xmath23 and @xmath24 @xmath52 g^{ijkl } & : = & -{{\textstyle { \frac{1}{2 } } } } x^{[ij } x^{kl ] } + { { \textstyle { \frac{1}{48 } } } } \epsilon^{ijklmnpq } p_{mn } p_{pq } \,.\ ] ] to extend this canonical realization to the full @xmath0 , we need one more variable , the momentum @xmath27 conjugate to @xmath28 . combining the symplectic realization of @xmath32 with the non - linear variations ( 23 ) and ( 24 ) of @xcite we can then deduce the classical ( phase space ) analogs of the @xmath0generators . owing to the non - linearity of the realization , the resulting expressions are not only quadratic , but go up to fourth order in the phase space variables @xmath23 and @xmath24 , and moreover contain inverse powers of @xmath28 . thus all generators of @xmath0can be realized on a 58-dimensional phase space coordinatized by @xmath53 . the minimal unitary 29 dimensional representation of @xmath0is then obtained by quantization , i.e. by introducing the usual momentum operators obeying the canonical commutation relations @xmath54 \;=\ ; { \mathrm{i}}\,\delta^{ij}_{kl}\ , , \qquad [ y , p ] \;=\ ; { \mathrm{i}}\ , . { \nonumber}\ ] ] in elevating the @xmath0generators to quantum operators the only problem _ vis - - vis _ the classical phase space description is the non - commutativity of the coordinate and momentum operators , which requires some ordering prescription ( we note , that the @xmath1generators ( [ e7gen ] ) are insensitive to re - ordering the coordinates and momenta ) . since we are interested in finding a unitary representation we insist that all operators are hermitean w.r.t . to the standard scalar product on @xmath29 . we thus arrive at a unitary representation of the @xmath0lie algebra in terms of self - adjoint operators acting on ( some dense subspace of ) a hilbert space of complex functions in 29 real variables @xmath55 . we emphasize that this realization requires _ complex _ functions , for the same reason that the schrdinger representation of the one - dimensional point particle requires complex wave functionscan be rewritten in terms of bosonic annihilation and creation operators . in the corresponding coherent state basis the hilbert space will involve _ holomorphic _ functions in 29 variables . ] . before writing out the @xmath0generators , let us list the @xmath56commutation relations @xmath57 } & = & { \mathrm{i}}\,{\delta}{}_{j}^{k}\ , g^i{}_{l } -{\mathrm{i}}\,{\delta}{}_{l}^{i}\ , g^k{}_{j } \,,{\nonumber \\ } [ 2ex ] { [ g^i{}_j\ , , g^{klmn } ] } & = & -4{\mathrm{i}}\,{\delta}{}_{j}^{[k}\ , g^{lmn]i}_{\phantom{k } } -{{\textstyle { \frac{{\mathrm{i}}}{2 } } } } \ , { \delta}{}^i_j g^{klmn}_{\phantom{j}}\,,{\nonumber \\ } [ 2ex ] { [ g^{ijkl}\ , , g^{mnpq } ] } & = & { { \textstyle { \frac{{\mathrm{i}}}{36 } } } } \,\epsilon^{ijkls[mnp}\ , g^{q]}{}_{s } \,.\ ] ] observe that only the @xmath58 subgroup acts by linear transformations ; its maximal compact subgroup @xmath59 is generated by @xmath60 , where @xmath61 for later convenience , we also define @xmath62 and the selfdual and anti - selfdual combinations @xmath63 where the @xmath64 are compact and the @xmath65 non - compact . the following generators extend this representation to the full @xmath0(we use the same notation for the elements of @xmath8 as in @xcite , in accordance with its 5-graded structure ) : @xmath66 e^{ij } & : = & y\,x^{ij } \,,{\nonumber \\ } [ 2ex ] e_{ij } & : = & y\,p_{ij } \,,{\nonumber \\ } [ 2ex ] h & : = & { { \textstyle { \frac{1}{2 } } } } ( y\,p+p\,y ) \,,{\nonumber \\ } [ 2ex ] f^{ij } & : = & -p\,x^{ij } + 2{\mathrm{i}}y^{-1}\ , { [ x^{ij}\ , , i_4(x , p)]}{\nonumber \\ } & = & -4\,y^{-1 } x{\!\overbox{{}^{ik}p_{kl}x^{lj } } } - { { \textstyle { \frac{1}{2 } } } } \,y^{-1}(x^{ij}p_{kl}x^{kl}+x^{kl}p_{kl}x^{ij}){\nonumber \\ } & & + { { \textstyle { \frac{1}{12 } } } } y^{-1 } \epsilon^{ijklmnpq } p_{kl}p_{mn}p_{pq } -p\,x^{ij}\,,{\nonumber \\ } [ 2ex ] f_{ij } & : = & -p\,p_{ij } + 2{\mathrm{i}}y^{-1}\ , { [ p_{ij}\ , , i_4(x , p)]}{\nonumber \\ } & = & 4\,y^{-1 } p{\underbox{{}_{ik}x^{kl}p_{lj } } } + { { \textstyle { \frac{1}{2 } } } } \,y^{-1}(p_{ij } x^{kl}p_{kl}+p_{kl}x^{kl } p_{ij}){\nonumber \\ } & & -{{\textstyle { \frac{1}{12 } } } } y^{-1 } \epsilon_{ijklmnpq } x^{kl}x^{mn}x^{pq } -p\,p_{ij}\,,{\nonumber \\ } [ 2ex ] f & : = & { { \textstyle { \frac{1}{2 } } } } p^2 + 2 y^{-2 } i_4(x , p ) \,.\ ] ] the hermiticity of all generators is manifest . here @xmath34 is the fourth order differential operator @xmath67 as a function of @xmath23 and @xmath24 this operator represents the quartic invariant of @xmath1because @xmath68 } = { [ g^{ijkl}\ , , i_4(x , p ) ] } = 0 \,.\ ] ] we emphasize the importance of the ordering adopted in ( [ invgen ] ) for the vanishing of ( the second of ) these commutators . orderings that differ from ( [ invgen ] ) by a term containing the euler operator @xmath69 break @xmath1invariance . on the other hand , ( [ quartic ] ) is insensitive to re - orderings which differ from ( [ invgen ] ) only by a @xmath30-number , and in the absence of a preferred ordering there is thus no absolute significance to the additive constant appearing in the definition of @xmath35 . for instance , an admissible re - ordering is @xmath70 this relation confirms our previous assertion that , when acting on a given representation of @xmath1 , @xmath35 is just the quadratic @xmath1casimir invariant , up to an additive constant . for the derivation of ( [ e7e8gen ] ) , we note that , as already pointed out in @xcite , the crucial step is the determination of the operators @xmath9 and @xmath19 corresponding to the lowest and highest root of @xmath0 , respectively . the expressions for @xmath71 and @xmath72 then follow by commutation with @xmath73 and @xmath74 . while the derivation of @xcite relied on a generalization of the so - called `` capelli - identity '' , the form of our operators @xmath9 and @xmath19 follows directly from @xmath1invariance and a scaling argument , up to the additive constant in ( [ invgen ] ) . the latter originates from the re - ordering required to bring the commutator of @xmath72 and @xmath71 into the `` standard form '' defined by the r.h.s . of ( [ invgen ] ) as anticipated , all generators transform covariantly under the full @xmath1group . @xmath75 } & = & { \mathrm{i}}\,{\delta}{}^k_j\,e^{il } -{\mathrm{i}}\,{\delta}{}^l_j\,e^{ik } -{{\textstyle { \frac{{\mathrm{i}}}{4 } } } } { \delta}{}^i_j\,e^{kl } \,,{\nonumber \\ } { [ g^i{}_j\ , , e_{kl } ] } & = & { \mathrm{i}}\,{\delta}{}^i_k\,e_{lj } -{\mathrm{i}}\,{\delta}{}^i_l\,e_{kj } + { { \textstyle { \frac{{\mathrm{i}}}{4 } } } } { \delta}{}^i_j\,e_{kl } \,,{\nonumber \\ } [ 3ex ] { [ g^i{}_j\ , , f^{kl } ] } & = & { \mathrm{i}}\,{\delta}{}^k_j\,f^{il } -{\mathrm{i}}\,{\delta}{}^l_j\,f^{ik } -{{\textstyle { \frac{{\mathrm{i}}}{4 } } } } { \delta}{}^i_j\,f^{kl } \,,{\nonumber \\ } { [ g^i{}_j\ , , f_{kl } ] } & = & { \mathrm{i}}\,{\delta}{}^i_k\,f_{lj } -{\mathrm{i}}\,{\delta}{}^i_l\,f_{kj } + { { \textstyle { \frac{{\mathrm{i}}}{4 } } } } { \delta}{}^i_j\,f_{kl } \,.\ ] ] the remaining part of @xmath1acts as @xmath76 } & = & -{\mathrm{i}}\,{\delta}{}^{[ij}_{mn}\,e^{kl]}_{\vphantom{mn}}\,,{\nonumber \\ } [ 1ex ] { [ g^{ijkl}\ , , e^{mn } ] } & = & -{{\textstyle { \frac{{\mathrm{i}}}{24 } } } } \,\epsilon^{ijklmnpq}\,e_{pq } \ , , { \nonumber \\ } [ 1ex ] { [ g^{ijkl}\ , , f_{mn } ] } & = & -{\mathrm{i}}\,{\delta}{}^{[ij}_{mn}\,f^{kl]}_{\vphantom{mn } } \,,{\nonumber \\ } [ 1ex ] { [ g^{ijkl}\ , , f^{mn } ] } & = & -{{\textstyle { \frac{{\mathrm{i}}}{24 } } } } \,\epsilon^{ijklmnpq}\,f_{pq } \,.\ ] ] the grading of the generators is given by the dilatation generator @xmath15 @xmath77 } & = & -2\,{\mathrm{i}}\,e\ , , & & { [ h\ , , f ] } & = & 2\,{\mathrm{i}}\,f\ , , \\[1ex ] { [ h\ , , e^{ij } ] } & = & -{\mathrm{i}}\,e^{ij}\ , , & & { [ h\ , , f^{ij } ] } & = & { \mathrm{i}}\,f^{ij}\ , , \\[1ex ] { [ h\ , , e_{ij } ] } & = & -{\mathrm{i}}\,e_{ij}\ , , & & { [ h\ , , f_{ij } ] } & = & { \mathrm{i}}\,f_{ij}\ , . \end{array}\ ] ] the remaining non - vanishing commutation relations are @xmath78 } & = & 12\,{\mathrm{i}}\,g^{ijkl}\ , , & { [ e^{ij}\ , , f_{kl } ] } & = & 4\,{\mathrm{i}}\,\delta^{[i}_{[k}\ , g^{j]}{}_{l ] } -{\mathrm{i}}\,\delta_{kl}^{ij}\ , h \ , , \\[1ex ] { [ e_{ij}\ , , f_{kl } ] } & = & -12\,{\mathrm{i}}\,g_{ijkl}\ , , & { [ e_{ij}\ , , f^{kl } ] } & = & 4\,{\mathrm{i}}\,\delta^{[k}_{[i}\ , g^{l]}{}_{j ] } + { \mathrm{i}}\,\delta^{ij}_{kl}\ , h \,,\\[1ex ] { [ e^{ij}\ , , e_{kl } ] } & = & 2\,{\mathrm{i}}\,\delta^{ij}_{kl } e\ , , & { [ f^{ij}\ , , f_{kl } ] } & = & 2\,{\mathrm{i}}\,\delta^{ij}_{kl } f\ , , \\[1ex ] { [ e\ , , f^{ij } ] } & = & -{\mathrm{i}}\,e^{ij}\ , , & { [ f\ , , e^{ij } ] } & = & { \mathrm{i}}\,f^{ij}\ , , \\[1ex ] { [ e\ , , f_{ij } ] } & = & -{\mathrm{i}}\,e_{ij}\ , , & { [ f\ , , e_{ij } ] } & = & { \mathrm{i}}\,f_{ij}\ , , \\[1ex ] { [ e\ , , f ] } & = & { \mathrm{i}}\,h\ , . & & \\[1ex ] \end{array}\ ] ] the @xmath0commutation relations are the same as in @xcite , except that the structure constants carry an extra factor of @xmath79 . as can be verified by computation of the cartan killing form from the structure constants that can be extracted from the above commutation relations , the maximal compact subgroup @xmath80 is generated by the following linear combinations of @xmath0generators @xmath81 this confirms that we are indeed dealing the split real form @xmath0 . the quadratic @xmath0casimir operator ( in a convenient normalization ) is a sum of three terms @xmath82 = { \mathcal{c}}_2\big[sl(2,{\ensuremath{\mathbb{r}}\xspace})\big ] + { \mathcal{c}}_2\big[{\ensuremath{e_{7(7)}}\xspace}\big ] + { \mathcal{c}}_2'\ ] ] with @xmath83 & : = & { { \textstyle { \frac{1}{2 } } } } ( ef + fe - { { \textstyle { \frac{1}{2 } } } } h^2 ) \,,{\nonumber \\ } { \mathcal{c}}_2\big[{\ensuremath{e_{7(7)}}\xspace}\big ] & : = & -{{\textstyle { \frac{1}{2 } } } } \ , g^i{}_j g^j{}_i -3\,g^{ijkl } g_{ijkl } \,,{\nonumber \\ } { \mathcal{c}}_2 ' & = & { { \textstyle { \frac{1}{4 } } } } \left ( e^{ij } f_{ij } + f_{ij } e^{ij } - e_{ij } f^{ij } - f^{ij } e_{ij } \right ) \,.\ ] ] here the terms in the first and second line represent the casimir operators of the @xmath32 and @xmath1subalgebras , respectively . when substituting the explicit expressions in terms of coordinates and momenta for the generators , we obtain @xmath83 & = & i_4 - { { \textstyle { \frac{3}{16 } } } } \ , , { \nonumber \\ } { \mathcal{c}}_2\big[{\ensuremath{e_{7(7)}}\xspace}\big ] & = & 3\,i_4 - { { \textstyle { \frac{969}{16 } } } } \ , , { \nonumber \\ } { \mathcal{c}}_2 ' & = & -4\,i_4 - { { \textstyle { \frac{237}{4 } } } } \,.\ ] ] hence all terms containing the operators @xmath84 and @xmath27 actually cancel , leaving us with a constant value for the minimal representation @xmath85 = -120\ ] ] which is the @xmath0analog of the result @xmath83 = { { \textstyle { \frac{1}{4 } } } } g -{{\textstyle { \frac{3}{16 } } } } \,,\ ] ] familiar from conformal quantum mechanics except that there is no coupling constant any more for @xmath0that we can tune ! of course , unlike for the group @xmath32 have the same eigenvalue of the casimir . ] , this result does not by itself imply the irreducibility of the minimal representation of @xmath0 . to show that , we have to compute in addition the eigenvalues of the higher order @xmath0casimir invariants . however , it is almost self - evident that these , too , will collapse to @xmath30-numbers when the coordinates and momenta are substituted , for the simple reason that we can not build @xmath0invariants from the coordinate and momentum operators alone . this is in stark contrast to the singleton representation of @xmath1for which the coordinates and momenta do form a non - trivial ( linear ) representation of @xmath1 , permitting the construction of non - vanishing higher order @xmath1invariants . this provides an independent argument that the minimal representation of @xmath0is indeed irreducible . we conclude this section by giving the chevalley generators corresponding to the eight simple roots of @xmath0 , with the labeling indicated in the figure . they are @xmath86 e_2 & = & g^1{}_2 \quad , \quad & f_2 & = & g^2{}_1\ , , \\ e_3 & = & g^2{}_3 \quad , \quad & f_3 & = & g^3{}_2\ , , \\[1ex ] e_4 & = & g^3{}_4 \quad , \quad & f_4 & = & g^4{}_3\ , , \\[1ex ] e_5 & = & g^4{}_5 \quad , \quad & f_5 & = & g^5{}_4\ , , \\[1ex ] e_6 & = & g^5{}_6 \quad , \quad & f_6 & = & g^6{}_5\ , , \\[1ex ] e_7 & = & g^6{}_7 \quad , \quad & f_7 & = & g^7{}_6\ , , \\[1ex ] e_8 & = & e^{78 } \quad , \quad & f_8 & = & f_{78}\ , . \\[1ex ] \end{array}\ ] ] the generators of the cartan subalgebra are given by @xmath87 h_2 & = & g^1{}_1-g^2{}_2 \,,\\ h_3 & = & g^2{}_2-g^3{}_3 \,,\\[1ex ] h_4 & = & g^3{}_3-g^4{}_4 \,,\\[1ex ] h_5 & = & g^4{}_4-g^5{}_5 \,,\\[1ex ] h_6 & = & g^5{}_5-g^6{}_6 \,,\\[1ex ] h_7 & = & g^6{}_6-g^7{}_7 \,,\\[1ex ] h_8 & = & g^7{}_7+g^8{}_8-{{\textstyle { \frac{1}{2 } } } } h \ , . \end{array}\ ] ] comparison with the formulas given in the appendix of @xcite shows that the basis of coordinate vs. momentum variables used there differs from ours by the choice of polarization ( or `` fourier transformation '' ) . accordingly , the linearly realized subgroup exposed there is also different from ours . in the coordinate representation , the linearly realized @xmath58 subalgebra of @xmath1plays a distinguished role . as is well known , however , there is another basis of @xmath1 , which is equally important in supergravity and superstring theory , where @xmath58 is replaced by @xmath88 @xcite . in this section , we demonstrate that the full @xmath0lie algebra can be rewritten in terms of this complex basis . the change of basis is equivalent to the replacement of coordinates and momenta by creation and annihilation operators , and thus to the replacement of the coordinate representation by a holomorphic ( bargmann - fock ) representation . this will allow us to establish the connection with the oscillator realization of @xmath1discovered already some time ago @xcite . for this purpose , we introduce the creation and annihilation ( or raising and lowering ) operators @xmath89 the normalization has been chosen such that @xmath90 = \delta_{ab}^{cd}\ ] ] substituting these operators into ( [ e7gen ] ) and defining @xmath91}\,,\ ] ] we obtain the singleton representation of @xmath1 in the @xmath88 basis of @xcite @xmath92 } - { { \textstyle { \frac{1}{48 } } } } \epsilon^{abcdefgh } a_{ef } a_{gh}\ ] ] where @xmath93 now generates the @xmath88 subgroup of @xmath1 . in deriving this result , we made use of the formulas ( see e.g. the appendices of @xcite ) @xmath94 and @xmath95 } & = & -{{\textstyle { \frac{2}{3 } } } } \delta{\!\overbox{{}^{k[i } \gamma^{j]m}_{[ab } \gamma^{ml}_{cd]}\!\!\!}}\,\,\ , + \gamma^{[ij}_{[ab } \gamma^{kl]}_{cd ] } { \nonumber \\ } \gamma^{ik}_{[ab } \gamma^{kj}_{cd ] } & = & { { \textstyle { \frac{1}{24 } } } } \epsilon_{abcdefgh } \gamma^{ik}_{ef } \gamma^{kj}_{gh } { \nonumber \\ } \gamma^{[ij}_{[ab } \gamma^{kl]}_{cd ] } & = & - { { \textstyle { \frac{1}{24 } } } } \epsilon_{abcdefgh } \gamma^{[ij}_{ef } \gamma^{kl]}_{gh } { \nonumber \\ } & = & + { { \textstyle { \frac{1}{24 } } } } \epsilon^{ijklmnpq}\,\gamma^{mn}_{[ab } \gamma^{pq}_{cd]}\ ] ] we note the complex ( anti)self - duality relation @xmath96 and the commutation relations @xmath97 } & = & { \delta}{}_{b}^{c}\ , g^a{}_{d } -{\delta}{}_{d}^{a}\ , g^c{}_{b } \,,{\nonumber \\ } [ 2ex ] { [ g^a{}_b\ , , g^{cdef } ] } & = & -4\,{\delta}{}_{b}^{[c}\ , g^{def]a}_{\phantom{b } } -{{\textstyle { \frac{1}{2 } } } } \ , { \delta}{}^a_b g^{cdef}_{\phantom{b}}\,,{\nonumber \\ } [ 2ex ] { [ g^{abcd}\ , , g^{efgh } ] } & = & { { \textstyle { \frac{1}{36 } } } } \,\epsilon^{abcdi[efg}\ , g^{h]}{}_{i } \,.\ ] ] in order to render the remaining @xmath0generators @xmath88 covariant , we define @xmath98 e_{ab } & : = & { { \textstyle { \frac{1}{4\sqrt{2 } } } } } \gamma^{ij}_{ab } \big(e^{ij}+{\mathrm{i}}\,e_{ij}\big ) = y a_{ab}\,,\\[1ex ] \end{array}\ ] ] the computation is more tedious for the generators which are cubic and quartic in the oscillators , and most conveniently done by checking the @xmath0algebra again . we have @xmath99 with the @xmath88 invariant expression for @xmath35 in terms of oscillators @xmath100 note that the normal - ordered version of @xmath35 ( with all the annihilators to the right ) is _ not _ @xmath1invariant , because @xmath101 with the number operator @xmath102 which does not commute with @xmath103 . the remaining generators are now straightforwardly deduced by commuting @xmath19 with @xmath104 and @xmath105 : @xmath106 } { \nonumber \\ } & = & -p\ , a^{ab } + { { \textstyle { \frac{{\mathrm{i}}}{2 } } } } y^{-1}(a^{ab } a_{cd } a^{cd } + a^{cd } a_{cd } a^{ab } ) { \nonumber \\ } & & + 4{\mathrm{i}}y^{-1}\ , a\,{\!\overbox{{}\!{}^{ac } a_{cd } a^{db}\!}}\ , -{{\textstyle { \frac{{\mathrm{i}}}{12 } } } } y^{-1 } \epsilon^{abcdefgh } a_{cd } a_{ef } a_{gh } \,,{\nonumber \\ } f_{ab } & : = & { { \textstyle { \frac{1}{4\sqrt{2 } } } } } \gamma^{ij}_{ab } \big(f^{ij}+{\mathrm{i}}\,f_{ij } \big ) \;\equiv\ ; { \mathrm{i}}\,{[e_{ab}\ , , f ] } { \nonumber \\ } & = & -p\ , a_{ab } -{{\textstyle { \frac{{\mathrm{i}}}{2 } } } } y^{-1 } ( a_{ab } a^{cd } a_{cd } + a_{cd } a^{cd } a_{ab } ) { \nonumber \\ } & & - 4{\mathrm{i}}y^{-1}\ , a\,{\underbox{{}\!{}_{ac } a^{cd } a_{db}\!}}\ , + { { \textstyle { \frac{{\mathrm{i}}}{12 } } } } y^{-1}\epsilon_{abcdefgh } a^{cd } a^{ef } a^{gh } \,.\ ] ] these generators now transform covariantly under the @xmath88 group @xmath107 } & = & { \delta}{}^a_c\,e_{db } -{\delta}{}^a_d\,e_{cb } + { { \textstyle { \frac{1}{4 } } } } { \delta}{}^a_b\,e_{cd } \,,{\nonumber \\ } [ 3ex ] { [ g^a{}_b\ , , f_{cd } ] } & = & { \delta}{}^a_c\,f_{db } -{\delta}{}^a_d\,f_{cb } + { { \textstyle { \frac{1}{4 } } } } { \delta}{}^a_b\,f_{cd } \,,\ ] ] and the remaining part of @xmath1acts as @xmath108 } & = & -{\delta}{}^{[ab}_{ef}\,e^{cd]}_{\vphantom{ef}}\,,{\nonumber \\ } { [ g^{abcd}\ , , f_{ef } ] } & = & -{\delta}{}^{[ab}_{ef}\,f^{cd]}_{\vphantom{ef}}\,.\ ] ] furthermore , @xmath109 } & = & 2 \delta_{ab}^{cd}\,e \ , , { \nonumber \\ } { [ f_{ab}\ , , f^{cd } ] } & = & 2 \delta_{ab}^{cd}\,f \ , , { \nonumber \\ } { [ e^{ab}\ , , f^{cd } ] } & = & -12\,{\mathrm{i}}\,g^{abcd } \,,{\nonumber \\ } { [ e_{ab}\ , , f^{cd } ] } & = & -2\,{\mathrm{i}}\,\delta_{[a}^{[c } g^{d]}_{\vphantom{[]}}{}_{b ] } -\delta_{ab}^{cd}\ , h \,.\ ] ] it remains to discuss the @xmath32 subgroup , for which we likewise switch to a complex basis ( the @xmath110 basis ) @xmath111 in the absence of the term containing @xmath112 , it would again be convenient to employ creation and annihilation operators @xmath113 and @xmath114 to recover the well known singleton representation of @xmath110 , but for non - vanishing value of the quartic @xmath1invariant it is not possible to switch this term off . the presence of @xmath115 and @xmath112 makes it somewhat awkward to express all generators in terms of creation and annihilation operators , so one might prefer to keep the coordinate representation in this sector . the commutation relations are , however , not affected by the choice of variables . @xmath116 } & = & \mp l_{\pm 1}\,,\qquad { [ l_{+1}\ , , l_{-1 } ] } \;=\ ; 2\ , l_{0}\,.\ ] ] this basis is no longer hermitean , but @xmath117 diagonalizing the new hermitean generator @xmath118 instead of @xmath15 we obtain an alternative 5-graded decomposition @xmath119 such that @xmath120 we note that , strictly speaking , the elements of @xmath121 by themselves do not belong to the real lie algebra @xmath8 as defined in section 2 , but to its complexification . it is only the hermitean linear combinations which do . the subspaces of grade @xmath122 are one - dimensional with generators @xmath123 , the grade 0 space is spanned by the @xmath1generators ( [ e7gen ] ) together with @xmath124 and the subspaces of grade @xmath125 are generated by @xmath126 and @xmath127 defined by @xmath128 respectively . in terms of these generators all structure constants are real : @xmath129 } & = & 2\,{\mathcal{f}}^{ab } \ , , { \nonumber \\ } { [ l_{-1}\ , , { \mathcal{f}}^{ab } ] } & = & -2\,{\mathcal{e}}^{ab } \ , , { \nonumber \\ } { [ { \mathcal{e}}_{ab}\ , , { \mathcal{e}}^{cd } ] } & = & 2\,l_{-1 } \ , , { \nonumber \\ } { [ { \mathcal{f}}_{ab}\ , , { \mathcal{f}}^{cd } ] } & = & 2\,l_{+1 } \ , , { \nonumber \\ } { [ { \mathcal{e}}^{ab}\ , , { \mathcal{f}}^{cd } ] } & = & -12\,g^{abcd } \,,{\nonumber \\ } { [ { \mathcal{e}}_{ab}\ , , { \mathcal{f}}^{cd } ] } & = & -2\,\delta_{[a}^{[c } g^{d]}_{\vphantom{[]}}{}_{b ] } + 2\,\delta_{ab}^{cd}\ , l_0 \,.\ ] ] a non - compact group @xmath7 admits unitary representations of the lowest weight ( or highest weight ) type if and only if the quotient @xmath130 of @xmath7 with respect to its maximal compact subgroup @xmath131 is an hermitean symmetric space @xcite . from this theorem it follows that the simple non - compact groups that admit lowest ( highest ) weight unitary representations are @xmath132 , @xmath133 , @xmath134 , @xmath135 , @xmath136 , and @xmath137 . the unitary lowest ( highest ) weight representations belong to the holomorphic ( anti - holomorphic ) discrete series and within these representations the spectrum of , at least , one generator is bounded from below ( above ) . more generally , a non - compact group @xmath7 admits representations belonging to the discrete series if it has the same rank as its maximal compact subgroup . thus the non - compact group @xmath0as well as @xmath1admit discrete series representations . however , they are not of the lowest or highest weight type . in this subsection we will analyze the decomposition of the minimal representation of @xmath0with respect to its subgroup @xmath36 , but using the complex basis of the last section . the group @xmath110admits holomorphic and anti - holomorphic unitary representations , and they exhaust the list of discrete series representations for @xmath110 . ( this is not true for higher rank non - compact groups admitting such representations . ) as mentioned earlier , the realization of the @xmath110 subgroup within the minimal unitary representation of @xmath0is precisely of the form that arises in conformal quantum mechanics @xcite . this is perhaps not surprising since we obtained our realization from the geometric action of @xmath0as a quasi - conformal group in 57 dimensions @xcite . by comparison of the @xmath110 subgroup ( [ su11 ] ) with that of @xcite it follows that the coupling constant @xmath138 in conformal quantum mechanics is simply @xmath139 in our realization . the quadratic casimir of @xmath110 is @xmath140 & = & l_0 ^ 2 -{{\textstyle { \frac{1}{2 } } } } \ , ( l_{1}l_{-1}+l_{-1}l_{1 } ) = i_4(x , p ) - { { \textstyle { \frac{3}{16 } } } } \,.\ ] ] thus for a given eigenvalue of @xmath34 , we are led to a definite unitary realization of @xmath110 . as we showed above , @xmath34 is simply the quadratic casimir operator of @xmath1(cf . ( [ e7-casimir ] ) ) that commutes with @xmath110 , up to an additive normal ordering constant . hence classifying all the possible eigenvalues of @xmath34 within the minimal unitary realization of @xmath0is equivalent to giving the decomposition of the @xmath0representation with respect to @xmath141 . in such cases one needs to use additional labels such as the eigenvalues of @xmath124 . ] . unitarity requires the eigenvalues of the casimir operators to be real . therefore all the eigenvalues of @xmath35 must be real . as we showed above , the realization of @xmath1within @xmath0coincides with the singletonic oscillator realization of @xmath1 @xcite . the oscillator realization of @xmath1leads to an infinitely reducible unitary representation @xcite . hence the minimal representation of @xmath0will be infinitely reducible with respect to @xmath141 . denoting the eigenvalues of the quadratic casimir operator @xmath110 as @xmath142 = j(j-1)\ ] ] we find that @xmath143 depending on the eigenvalue of @xmath35 we will be led to one of the well - known series of representations of @xmath110 @xcite * continuous principal series : @xmath144 with the eigenvalues @xmath145 of @xmath124 unbounded from above and from below @xmath146 * continuous supplementary series : @xmath147 again with unbounded eigenvalues @xmath145 of @xmath124 in both directions . * holomorphic discrete series @xmath148 ( lowest weight irreps ) : @xmath149 * discrete series @xmath150 ( highest weight irreps ) : @xmath151 for vanishing quartic invariant @xmath35 we find @xmath152 corresponding to the two singleton irreps of @xmath110 . note that the eigenvalues of the casimir operator for the two singleton irreps coincide . they are distinguished by the value of @xmath153 which is the eigenvalue of @xmath124 on the corresponding lowest weight vector . to give the decomposition of the minimal representation of @xmath0w.r.t . its @xmath155 subgroup we need to determine all possible eigenvalues of @xmath34 within our realization . since this determination appears to be rather complicated , we shall study this question in a somewhat simpler setting by truncating @xmath0to a special subgroup . @xmath0has the subgroup @xmath156 , where @xmath157 has the maximal compact subgroup @xmath158 . it is instructive to truncate our realization to the @xmath154 subgroup , which is one of the minimal subgroups that admit a non - trivial 5-grading . this is achieved by introducing coordinate @xmath159 and momentum @xmath160 corresponding to the symplectic trace components of @xmath23 and @xmath24 , respectively : @xmath161 and throwing away the symplectic traceless components in our realization . the matrix @xmath162 is the symplectic metric defined by @xmath163 and @xmath164 is its inverse : @xmath165 . the generators @xmath159 and @xmath160 obey the canonical commutation relation @xmath166 } & = & { \mathrm{i}}\,.\ ] ] the resulting expressions for the generators of @xmath154 are @xmath167 e_{\uparrow } & : = & y\,x \,,{\nonumber \\ } [ 2ex ] e_{\downarrow } & : = & y\,p_x \,,{\nonumber \\ } [ 2ex ] h & : = & { { \textstyle { \frac{1}{2 } } } } ( y\,p_y+p_y\,y ) { \nonumber \\ } [ 2ex ] a & : = & -{{\textstyle { \frac{3}{4 } } } } ( x^2+p_x^2 ) { \nonumber \\ } [ 2ex ] f_{\uparrow } & : = & -x p_y -{{\textstyle { \frac{1}{2 } } } } y^{-1}(x p_x x + p_x^3 ) \,,{\nonumber \\ } [ 2ex ] f_{\downarrow } & : = & p_x p_y -{{\textstyle { \frac{1}{2 } } } } y^{-1}(p_x x p_x + x^3 ) \,,{\nonumber \\ } [ 2ex ] f & : = & { { \textstyle { \frac{1}{2 } } } } p_y^2 + { { \textstyle { \frac{1}{8 } } } } y^{-2 } ( x^2+p_x^2)^2 -{{\textstyle { \frac{1}{8 } } } } y^{-2}\,,\ ] ] the 5-grading in the above basis is with respect to @xmath15 which is the non - compact dilatation generator . to understand the resulting representation of @xmath154 it is useful to go to the basis in which the lie algebra of @xmath154 has a 3-grading w.r.t . the compact @xmath168 generator that commutes with the @xmath169 subgroup @xmath170 the compact @xmath168 generator is @xmath171 and the @xmath169 generators are @xmath172 with the commutation relations @xmath173 } & = & { \mathrm{i}}\,\epsilon_{ijk } j_{k}\ , , \quad ( i , j , k=1,2,3 ) { \nonumber \\ } { [ j_0\ , , j_i ] } & = & 0\,.\ ] ] the grade @xmath125 generators are @xmath174 satisfying @xmath175 } & = & \pm k_{\pm } \,,{\nonumber \\ } { [ j_0\ , , l_{\pm } ] } & = & \pm l_{\pm } \,.\ ] ] more explicitly we have @xmath176 \ } ( x\mp ip_x ) \ , , { \nonumber \\ } l_{\pm } & = & \ { { { \textstyle { \frac{1}{4 } } } } ( y\mp ip_y)^2 - { { \textstyle { \frac{1}{16 } } } } y^{-2 } [ -1 + ( x^2+p_x^2)^2 ] \}\ ] ] the generator @xmath177 that determines the 3-grading of @xmath154 w.r.t . maximal compact subgroup @xmath178 is manifestly positive definite . in terms of the annihilation and creation operators @xmath179 the generator @xmath177 takes the form : @xmath180 where @xmath181 is the number operator . this implies that the resulting unitary irreducible representations of @xmath154 must be of the lowest weight type ( positive energy ) . now the quadratic casimir operator of @xmath154 is given by @xmath182 & = & { { \textstyle { \frac{1}{2 } } } } ( ef+fe ) -{{\textstyle { \frac{1}{4 } } } } h^2 + { { \textstyle { \frac{1}{3 } } } } a^2 { \nonumber \\ } & & + { { \textstyle { \frac{1}{4 } } } } ( e_{\uparrow}f_{\downarrow } + f_{\downarrow } e_{\uparrow } ) + { { \textstyle { \frac{1}{4 } } } } ( e_{\downarrow}f_{\uparrow } + f_{\uparrow } e_{\downarrow})\ ] ] substituting the expressions for the generators we find that the the quadratic casimir of @xmath154 becomes simply a @xmath30-number , namely @xmath182 = -{{\textstyle { \frac{3}{16 } } } } \ ] ] which suggests that the representation may be irreducible . to have an irreducible representation the cubic casimir must likewise reduce to a @xmath30-number . however , we can study the irreducibility of the representation without having to calculate the cubic casimir . since we know that our representation is of the lowest weight type we can use the fact that a unitary irrep of the lowest weight type is uniquely determined by a set of states @xmath183 transforming irreducibly under the maximal compact subgroup @xmath184 and that are annihilated by all the generators of grade @xmath45 under the 3-grading . thus we need to find all such states that are annihilated by @xmath185 and @xmath186 : @xmath187 and that transform irreducibly under the maximal compact subgroup @xmath184 . the only normalizable state satisfying these conditions is the one particle excited state @xmath188 where the vacuum state @xmath189 is annihilated by both annihilation operators @xmath190 this state is a singlet of @xmath169 and has the @xmath168 charge 1 . this proves that the minimal realization of @xmath154 obtained by truncation of the minimal representation of @xmath0is also irreducible . the subgroup @xmath191 of @xmath0under truncation to @xmath154 reduces to @xmath192 . the generators of @xmath110 subgroup of @xmath154 are given by @xmath193 \ } { \nonumber \\ } l_{-}&= & { { \textstyle { \frac{1}{2 } } } } ( e - f+ih ) { \nonumber \\ } & = & \ { { { \textstyle { \frac{1}{4 } } } } ( y + ip_y)^2 - { { \textstyle { \frac{1}{16 } } } } y^{-2 } [ -1 + ( x^2+p_x^2)^2 ] \ } { \nonumber \\ } l_0 & = & { { \textstyle { \frac{1}{2 } } } } ( e+f ) { \nonumber \\ } & = & \ { { { \textstyle { \frac{1}{4 } } } } ( y^2 + p_y^2 ) + { { \textstyle { \frac{1}{16 } } } } y^{-2 } [ -1 + ( x^2+p_x^2)^2 ] \}\ ] ] they are therefore the analogs of the generators([su11 ] ) . the generator of @xmath168 that commutes with @xmath110 is simply @xmath194 . in this truncation the quartic @xmath1invariant reduces to @xmath195\ ] ] which can be written in terms of the annihilation and creation operators as @xmath196 with the ( obvious ) identifications @xmath197 and @xmath198 . since the eigenvalues of the number operator @xmath199 are non - negative integers n , the eigenvalues of @xmath35 are simply @xmath200 . in fact the realization of @xmath110 in this case leads to unitary lowest weight representations of the type studied in @xcite . for @xmath201 we get the singleton irreps of @xmath110 corresponding to the values @xmath202 and @xmath203 . thus the minimal unitary representation of @xmath154 decomposes into a discretely infinite set of irreps of @xmath110 labelled by the eigenvalues of the @xmath168 generator which is the analog of the @xmath1subgroup of @xmath0for the @xmath154 truncation . the finite dimensional conformal group @xmath110 is well known to possess an infinite dimensional extension ( the witt - virasoro group ) . one may therefore ask whether there exists a generalization of this fact to @xmath0 . in other words , does there exist an infinite dimensional lie algebra ( or lie superalgebra ) that contains the witt - virasoro algebra and @xmath0at the same time ? while there appears to be no linear lie algebra with this property ( and no finite dimensional lie superalgebra , either ) , an infinite dimensional non - linear algebra of @xmath204-type does exist . it is a nonlinear quasi - superconformal algebra denoted as @xmath205 @xcite . the quasi - superconformal algebras in two dimensions were first introduced in@xcite and further systematized in@xcite . they were generalized in@xcite where two infinite families of nonlinear quasi - superconformal algebras were introduced . a classification of complex forms of quasi - superconformal algebras was given in @xcite . in @xcite a complete classification and a unified realization of the real forms of quasi - superconformal algebras were given . in the infinite central charge limit the exceptional quasi - superconformal algebra @xmath205 has the lie algebra @xmath0as a maximal finite dimensional simple lie subalgebra . the realization of @xmath205 given in@xcite involves 56 dimension @xmath206 bosons and a dilaton , which leads to a realization of @xmath0 in the infinite central charge limit . it will be important to understand if the resulting realization can be related to the one given here . furthermore one would like to know if one could use the methods of @xcite to give a unified realization of the unitary representations of non - compact groups that act as quasi - conformal groups as formulated in @xcite , thus generalizing our minimal realization @xmath0to all such noncompact groups . * acknowledgments : * we are grateful to b. pioline and a. waldron for discussions , and for sending us an advance copy of ref . we would also like to thank r. brylinski for bringing the references @xcite and @xcite to our attention and for helpful discussions . j. adams , j. huang and d.a . functions on the model orbit in @xmath2 _ representation theory _ * 2 * , 224 ( 1998 ) . v. bargmann . irreducible unitary representations of the lorentz group _ ann . _ * 48 * , 568 ( 1947 ) . b. gross and n. wallach . a distinguished family of unitary representations for the exceptional groups of real rank = 4 , in _ lie theory and geometry : in honor of b. kostant _ , brylinski , r. brylinski , v. guillemin , v. kac , eds , progress in mathematics * 123 * , birkhauser , boston ( 1994 ) , 289 - 304 .

the simple method for the calculating of the anomalous dimensions of the composite operators up to order is developed . we demonstrate the effectiveness of this approach by computing the critical exponents of the and operators in the order in the nonlinear sigma model . the special simplifications due to the conformal invariance of the model are discussed . -12 mm 0.2 cm 0.2 cm the simple scheme for the calculation of the anomalous dimensions of composite operators in the expansion . + [ 8 mm ] s. . derkachov & a.n . manashov + [ 8 mm ] * _ nordita , blegdamsvej 17 , dk-2100 , copenhagen , denmark . _ * _ department of theoretical physics , state university of st . petersburg , + sankt petersburg , 198904 russia .

hep-th9710015

the large @xmath6 expansion provides the powerful tool for the investigating of the fields theoretic models with a internal symmetry . being nonperturbative in its nature this method has proved to be successful in revealing properties which are inaccessible in conventional perturbation theory . the @xmath3 approach relies on the expansion of the effective action in the saddle point approximation which determines the leading order structure in @xmath7 . however , beyond the lowest order the computations become quite untractable because the propagators of auxiliary fields depend on the mass in a very complicated fashion . to obviate these difficulties it had been suggested in refs . @xcite , where the nonlinear sigma model had been analyzed , to consider theory directly at critical point , that should ensure the masslessness of all propagators . the latter circumstance simplifies considerably the computation of the feynman diagrams . solving the skeleton dyson equation the authors of refs . @xcite have calculated the basic indices of sigma model with @xmath0 accuracy in the arbitrary space dimensions . whilst the initial application of self consistency method was to a bosonic model , the techniques have also been extended to examine the fermionic theories @xcite . the further progress in the calculation of the critical indices is related to the conformal bootstrap method . in the frame of this approach the most strong at the present moment results @xmath8 anomalous dimensions of the basic fields in the @xmath9 model @xcite and in the gross -neveu model @xcite had been obtained . if one is interested in the calculation of the critical indices related to the basic fields only , these methods self - consistency equations and the conformal bootstrap method are likely to be the most efficient . however , some specific features of the above methods make them inconvenient for the treatment of the composite operators , except for the simplest ones @xcite . at the same time , the calculations of this type became actual now . first of all , it is related with the attempts to understand in more details the structure of the conformal fields theories ( cft ) in @xmath10 dimensions ( see refs . @xcite and references therein ) . the basic examples of higher dimensional cft being bosonic and fermionic @xmath11 vector models , the @xmath3 expansion underlies the computational method used for analyzing of these models . second , the renewing of the interest to the @xmath3 calculations is conditioned by the recent progress in the higher order computations of the renormalization group ( * rg * ) functions in qcd . the analytical calculations in qcd in the high orders ( four loop @xmath12 - function @xcite , anomalous dimensions of twist - 2 operators @xcite ) are very complex ones because of the huge number of feynman diagrams to be evaluated . thus it is important to have methods which may ensure the independent checks of this results . the one of the possible approach of this type is the @xmath13 expansion @xcite . further , the necessity of the careful analysis of the anomalous dimensions of the composite operators of special type have arisen in connection with the so - called stability problem @xcite . recently , it was realized that the @xmath3 expansion is the more suitable approach for this purpose @xcite . as mentioned above , the calculations in @xmath3 scheme beyond the @xmath3 order become tractable only when theory is considered directly at the critical point . but in this case the corresponding model lost the property of the multiplicative renormalizability and can not be worked out with the help of the standard * rg * methods . it forces one to use other approaches applicable in this case . the methods of self - consistency equations and conformal bootstrap , discussed above , are applicable for the operators of the special type only . the approach based on the operator product expansion used in refs @xcite is too involved to have a practical use in higher order calculations and limited by the models possessing of the conformal symmetry . the most convenient approach for the calculating of the anomalous dimensions of the composite operators in the first order of @xmath3 expansion had been developed in ref @xcite . the authors of ref . @xcite have exploited the property of the scale invariance of the correlators at critical point to obtain the anomalous dimensions of arbitrary composite operators . the main result of the paper @xcite is the formula expressing @xmath3 order anomalous dimensions via the @xmath14 pole residues of the corresponding diagrams ( @xmath14 is the regularization parameter , analogous @xmath15 in dimensional regularization ) . certainly , this way is the most effective in the case of the operators of arbitrary type . at the straightforward generalization beyond the @xmath3 order ( see refs . @xcite ) the above method lost many its attractive features . in the present paper we shall take advantage of the other approach @xcite which allow to derive the simple formula for the anomalous dimensions in the @xmath0 order . to avoid the inconveniences connected with the absence of the multiplicative renormalizability we consider extended model which is reduced to the initial one by tuning of some parameters . the extended model can be analyzed by the standard renormalization group ( * rg * ) methods . to obtain the anomalous dimensions in the initial model we relate it with the corresponding * rg * - functions of the extended one . in the following we restrict ourselves to the nonlinear sigma model , the generalization on the other model being straightforward . the paper is organized as follows : in section 2 we review the basic features of the nonlinear sigma model and discuss the problem to be solved . in section 3 the basic formulae for the anomalous dimensions of composite operators are derived . some technical tricks useful in higher order calculations are collected in section 4 . the results of @xmath0 order calculations of anomalous dimensions of some composite operators are presented in section 5 . the conclusions are given in section 6 . the @xmath3 - expansion for the massless @xmath16 nonlinear sigma model in @xmath17-dimensions is derived in the standard manner from the following action : @xmath18 where @xmath19 is the @xmath16 vector field and @xmath20auxiliary field ; @xmath21 - regularization parameter , @xmath22 renormalization mass . the two first term in ( [ action ] ) form the free part of the action , whereas two last are treated as a interaction . the kernel @xmath23 is determined from the requirement of the cancellation of the simple @xmath24 loop insertion in @xmath20 line = 0.7 cm ( hsize=10)(6,0)(2,0)(1,0)*(0.3 ) ( 2,-0.15)@xmath25 ( 7,-0.15)@xmath26 ( 4.5,0)(0.95,0)(1,0 ) ( -0.75,0)(-1,0)(1.25 ) -0.5 cm and reads as : @xmath27 where @xmath28 . the regularized kernel @xmath29 is defined as @xmath30 the function @xmath31 is assumed to be regular at @xmath32 and @xmath33 at @xmath34 . the freedom in the choice of the form of @xmath31 will be discussed in sec.[extend ] . the propagators of @xmath35 and @xmath20 fields have form ( @xmath36 ) : @xmath37 in the momentum space and @xmath38 in the coordinate space . here @xmath39 , @xmath40 and the explicit expressions for the amplitudes @xmath41 can be easily deduced from the eq . ( [ propp ] ) . the renormalization of this theory carried out in the standard way . the divergences appears as @xmath14 - pole terms in diagrams and can be excluded by the standard @xmath42 - operation . the power counting shows that the theory with action ( [ action ] ) is renormalizable in the @xmath3 - expansion in the whole range @xmath43 and the renormalized action has the form : @xmath44 as distinct from the more usual field theoretic models , the theory under consideration is not multiplicatively renormalizable . indeed , the renormalized action ( [ actionr ] ) can not been brought into the form of that ( [ action ] ) by a redefinition of the fields ( there is not a parameter like the coupling constant in ( [ action ] ) ) . due to nonmultiplicativity , one fails to apply * rg * methods for the calculation of critical exponents and has to seek for another approaches . one of them , which is briefly sketched below relies on the property of the scale invariance of the theory . the latter can be deduced from the equivalence of the nonlinear sigma model to the @xmath45 theory in @xmath46 dimensions at the critical point . the more formal proof not appealing to such equivalence can be found in ref . now we shall discuss the basic points of the approach mentioned above . ( for more details see refs.@xcite . ) let us consider the 1-irreducible green function @xmath47 that depends on the one momentum @xmath48 only the inverse propagator of @xmath24 field or the vertex function with one zero momentum . we prefer to work with the dimensionless object : @xmath49 , @xmath50 being the canonical dimension of the function @xmath47 . due to the scale invariance , the form of the renormalized green function is fully determined ( up to some constant factor @xmath51 ) by its anomalous dimension @xmath52 : @xmath53 the differentiation of the eq.([fix ] ) with respect @xmath22 yields the equation on the anomalous dimension @xmath54 to calculate the left side of the eq.([scale ] ) we notice that the all dependence from @xmath22 in the diagrams contributing to @xmath55 results from the propagators of @xmath20 field ( see eq.([propp ] ) ) , i.e. the every diagram acquires the factor @xmath56 , where @xmath57 is the number of @xmath58-lines in a diagram . hence , the formula ( [ scale ] ) takes the form : @xmath59 where @xmath60 is the number of @xmath58-lines in diagram @xmath61 . the sum runs over all diagrams @xmath62 for the renormalized green function @xmath63 and , of course , the limit @xmath64 is implied in eq . ( [ m ] ) . in the first order of @xmath3 expansion eq . ( [ m ] ) becomes very simple . indeed , taking into account that the first order diagrams can develop the simple @xmath14 poles only , @xmath65 , and that @xmath66 one gets @xmath67 thus for the calculation of the anomalous dimensions in the lowest order one need know the pole parts of the corresponding diagrams only . at first sight this property is lost in the next order . indeed , the evaluating of the @xmath52 with @xmath0 accuracy requires the knowledge of the @xmath3 order corrections to the constant @xmath68 , which are determined by the finite parts of the @xmath3 order diagrams . this circumstance greatly reduces the effectiveness of the eq . ( [ m ] ) for the high order calculations . to avoid this difficulty one may try to pick out of the sum in the eq . ( [ m ] ) the diagrams reproducing the undesirable terms on the lhs of the latter . taking in mind the analogous formulae in the dimensional regularization scheme we regroup the terms in the sum ( [ m ] ) in the way shown on the fig [ fr ] . it can be checked , in these particular cases , that only the @xmath69 terms contribute to the anomalous dimensions , whereas the others ensure the cancellation of the extra terms on the lhs of the eq . ( [ m ] ) . ( the @xmath69 is the standard operation of the subtracting of the subdivergencies from diagram . ) eventually , the formula for the anomalous dimensions reads @xmath70 here the sum runs over all diagrams ( up to @xmath0 order ) generated from the action ( [ action ] ) , i.e. the nonrenormalized diagrams . the notation @xmath71 is used for the residue at the simple pole in laurent expansion of a function @xmath72 . the formula ( [ mkr ] ) gives the simple algorithm for the calculating of the critical indeces @xmath73 and @xmath74 defined as @xmath75 up to @xmath0 order : @xmath76 the advantages of the formula ( [ mkr ] ) for the practical calculations in the comparison with those ( [ m ] ) is self - evident . our next purpose is to derive the analogous formulae in the case of arbitrary composite operator . however , to do this we shall take the advantage of the another approach . = 0.4 cm ( hsize=10 ) ( 0.5,0)(-0.5,0)@xmath77(0.4,0)(3.5,0 ) ( 2.5,0)[+2.5](1,0)(2.5,0)[+2.5 ] ( 6,0)(-1.8,0)@xmath78(0.5,0)(2.5,0 ) ( 2.5,0)[+2.5]*(0.5 ) ( 10,0 ) ( -1.5,0)@xmath78(0.4,0)(2.5,0 ) ( 2.5,0)[+2.5](2.5,0)*(0.5 ) ( 14,0)@xmath79 ( 17,0)(0.5,0)(-2.5,0)@xmath80 ( 0.4,0)(3.5,0 ) ( 2.5,0)[+2.5](1,0)(2.5,0)[+2.5 ] ( 6.5,0)(-1.8,0)@xmath81(0.5,0)(2.5,0 ) ( 2.5,0)[+2.5]*(0.5 ) ( 10.5,0 ) ( -1.5,0)@xmath81(0.4,0)(2.5,0 ) ( 2.5,0)[+2.5](2.5,0)*(0.5 ) = 0.3 cm ( hsize=10 ) ( 0.5,0 ) ( -0.5,0)@xmath82(0.5,0)(4,0 ) ( 1,2)[+0.5](2,2)(0,1)(0,-2)(2 ) ( 2.5,2)(1,-2)[+0.5 ] ( 7,0 ) ( -2,0)@xmath83(0.5,0)(4,0 ) ( 1,2)[+0.5](2,2)(-1,0)*(0.5)(2 ) ( 2.5,2)(1,-2)[+0.5 ] ( 13.5,0 ) ( -2,0)@xmath83(0.5,0)(4,0 ) ( 1,2)[+0.5](2,2)(1,0)*(0.5)(2 ) ( 2.5,2)(1,-2)[+0.5 ] ( 19,0)@xmath79 ( 22.5,0 ) ( 0.5,0 ) ( -3.5,0)@xmath84(0.5,0)(4,0 ) ( 1,2)[+0.5](2,2)(0,1)(0,-2)(2 ) ( 2.5,2)(1,-2)[+0.5 ] ( 7,0 ) ( -2,0)@xmath81(0.5,0)(4,0 ) ( 1,2)[+0.5](2,2)(-1,0)*(0.5)(2 ) ( 2.5,2)(1,-2)[+0.5 ] ( 13.5,0 ) ( -2,0)@xmath81(0.5,0)(4,0 ) ( 1,2)[+0.5](2,2)(1,0)*(0.5)(2 ) ( 2.5,2)(1,-2)[+0.5 ] let us consider the model based on the action obtained from ( [ action ] ) by introducing two new independent couplings @xmath85 and @xmath86 : @xmath87 in the following we will refer to it as @xmath88 model . it had been firstly introduced and investigated in ref . obviously , the initial model ( [ action ] ) is recovered by the special choice of parameters @xmath89 . as distinct from the ordinary sigma model the propagator of @xmath20 field in the @xmath88 model has the more complicated form . indeed , at @xmath90 the last term in the action ( [ uvaction ] ) does not ensure the exact cancellation of the simple @xmath24 loop insertion in @xmath91 line and one should sum up all such insertions ( see fig . [ ins ] ) , that yields for the propagator of @xmath20 field : @xmath92 here @xmath93 . = 0.7 cm ( hsize=10)(1.5,0)(-0.6,-0.15 ) @xmath94(1.5,0 ) ( 1.6,-0.15)@xmath95 ( 16.8,-0.15)@xmath96 ( 5.1,0)(0.95,0)(1,0)(-0.6,0)(-1,0 ) ( 1.5 ) ( 7.2,-0.15)@xmath97 ( 11.4,0)(0.95,0)(1,0)(-0.6,0)(-1,0 ) ( 1.5 ) ( 13.9,0)(0.8,0)(1,0)(1.5 ) the theory is evidently renormalizable and the renormalized action takes form : @xmath98 where the renormalization constants depend on the coupling @xmath99 also ( @xmath100 ) . the redefinition of the fields and the couplings @xmath101 where @xmath102 , @xmath103 , @xmath104 and @xmath105 , brings the action ( [ uvraction ] ) into the initial form , hence , the @xmath88 model is the multiplicatively renormalizable . this implies that the model can be analyzed by the standard * rg * methods . the basic * rg*functions are defined as follows : @xmath106 the * rg * equation for the one particle irreducible ( @xmath107 ) green functions reads as : @xmath108 here we used the shorthand notations @xmath109 , @xmath110 and @xmath111 . for the set of the composite operators @xmath112 mixing under renormalization the corresponding equation has the matrix form : @xmath113\delta^{ik } + \gamma_{f}^{ik}\right ) \gamma_{k}(p;p_{1},\ldots , p_{n})=0 . \label{rg2}\ ] ] here @xmath114 is the @xmath115-points @xmath107 green function with insertion of the renormalized operator @xmath116_{r } $ ] . the latter is defined as @xmath117_{r}= q_{ik}f_{k } $ ] and @xmath118 is chosen to ensure finiteness of all green functions of operator @xmath119_{r}$ ] . as usual , the summation over repeated indexes is implied . at last the matrix of anomalous dimensions @xmath120 is defined as @xmath121 here @xmath122 and the multiindex @xmath123 shows the number of the @xmath24 and @xmath20 fields in the monomial @xmath124 . we remind , that our purpose is the calculation of the critical indices of composite operators in the ordinary sigma model , i.e. we are interested in the limit @xmath125 of the @xmath88 model . however , in this limit the beta functions @xmath126 are not zero . ( indeed , @xmath127 . ) thus , generally speaking the point @xmath89 is not the critical point of @xmath88 model and the * rg*functions @xmath128 calculated at @xmath89 are not coincide with the anomalous dimensions of the fields ( operators ) in the ordinary sigma model . however , it had been shown in ref . @xcite that at the particular choice of the renormalization scheme ( the so called scheme of subtraction at fixed momenta , we refer to it as the scheme i ) the renormalized green functions depend on the difference @xmath129 only . in this case the anomalous term @xmath130 is dropped out of the eq . ( [ rg2 ] ) and the * rg * functions @xmath128 calculated in this scheme are the true anomalous dimensions . the green functions calculated in any other renormalization scheme are related to those @xmath131 calculated in the scheme i by a finite renormalization : @xmath132 here @xmath133 and @xmath134 are the constants of the finite renormalization . taking into account that the bare couplings @xmath135 are independent from the renormalization scheme , we obtain from the eq . ( [ bare ] ) that @xmath136 and , hence , @xmath137 . then , from the eq . ( [ renorm ] ) , one immediately obtains that green functions calculated in the different regularization schemes at @xmath138 differ one from another only by a constant multiplier . this proves that the critical exponents calculated in the model based on action ( [ action ] ) are independent from the regularization scheme being used and , in particularly , from the choice of the function @xmath139 . further , inserting @xmath140 in the form ( [ renorm ] ) into eq . ( [ rg2 ] ) one obtains the following equation for the green functions of the ordinary sigma model : @xmath141\delta^{ik } + ( \gamma_{f}^{ik}+\gamma_{f}^{*,ik})\right ) \gamma_{k}(p;p_{1},\ldots , p_{n})=0.\ ] ] where @xmath142 . thus the anomalous dimensions are given by the eigenvalues of the matrix @xmath143 , but not of the @xmath144 alone . of course , the @xmath145 is zero in the scheme i , but the latter is not very suitable for the practical calculations . the most convenient scheme for the calculations of * rg * functions is the * ms * scheme , but then one has to know @xmath145 . fortunately , the matrix @xmath145 is appeared to be of @xmath146 in the * ms * scheme , i.e. for the determination of the critical exponents up to @xmath0 order it is sufficient to know @xmath147 only . henceforth we restrict ourselves to * ms * scheme only . the * rg * functions @xmath52 , then , are related to the simple @xmath14 poles of the corresponding renormalization constants : @xmath148 where @xmath149 , @xmath150 . the mixing matrix @xmath118 is calculated directly from the diagrams as the divergent part of the functional @xmath151 : @xmath152 here @xmath153 is the nonrenormalized ( i.e. one arising at averaging with action ( [ uvaction ] ) ) @xmath115-points @xmath107 green function with operator @xmath154 insertion and the operation @xmath155 selects the singular ( @xmath21 poles ) part of the diagrams . because there are not derivatives with respect @xmath86 in the formulae ( [ rg ] ) , we can put @xmath156 from the very beginning . in this case the propagator of @xmath20 field ( [ line ] ) is reduced up to factor @xmath157 to that in the ordinary sigma model : @xmath158 . then taking into account that the constants @xmath159 are determined by the divergent ( pole ) parts of the corresponding green functions and that operation @xmath160 gives simply the factor @xmath161 ( @xmath162 is the number of @xmath20 lines ) for each diagram , one concludes that all calculations needed for determining of anomalous dimensions ( [ rg ] ) can be done in the framework of the standard @xmath9 model . let us prove now that matrix @xmath163 vanishes in the @xmath0 order . for this it is sufficient to show that the anomalous term @xmath164 ( see eq . ( [ a ] ) ) is zero in this order . first of all , note , @xmath165 being of @xmath166 order , we need to prove that the remaining part is of @xmath167 order . let us consider the arbitrary first order diagram of the ordinary @xmath9 model with @xmath115 internal @xmath20 lines . regarding the regulator @xmath14 on each line as the independent variable and picking out the multiplier @xmath31 from the @xmath20 propagator we write the answer for the diagram as @xmath168 . then , taking into account the form of the @xmath20 propagator eq . ( [ line ] ) one obtains the following expression for this diagram in the extended model : @xmath169 as usual , the renormalized diagram is obtained from the nonrenormalized one by the @xmath170 operation . in the ref . @xcite it was shown that the @xmath171 operation based on the subtraction scheme i gives for each term in the sum ( [ full ] ) the same ( up to factor @xmath172 ) answer . ( then , the sum ( [ full ] ) becomes a trivial and results in the factor @xmath173 . ) the above property holds no longer in the * ms * scheme . the action of the @xmath170 operation on different terms in the sum ( [ full ] ) leads to the essentially different answer even in the first order of @xmath3 expansion . however , in the first order of the @xmath3 expansion the whole sum ( [ full ] ) after renormalization depends on the difference @xmath174 only too . in the case of a convergent diagram @xmath175 this statement is trivial . let us consider the superficially divergent diagram . in the @xmath3 order such diagram having no subdivergencies , we represent function @xmath176 as : @xmath177 where the function @xmath178 is regular in the neighborhood of zero in all its arguments . since in the * ms * scheme the @xmath179 operation is reduced to the removing of the @xmath14 poles one has for the renormalized diagram : @xmath180 taking into account that the terms like @xmath181 due to the evident symmetry of the rest of summand in @xmath182 can be replaced by @xmath183 , one obtains for the sum ( [ sum1 ] ) @xmath184 the case when a superficially convergent diagram has a divergent subgraph can be treated along these lines as well . since the anomalous term @xmath185 is proportional to @xmath186 we conclude that @xmath164 , and hence , @xmath187 vanish up to @xmath0 order . ( note also , that we proved the more strong statement @xmath188 at @xmath189 , than it was needed : @xmath188 at @xmath89 . ) unfortunately , the arguments used above do not work in the next orders of @xmath3 expansion , since the diagrams which give the nonzero contributions to the anomalous term @xmath164 can be easy found in the higher orders . though we can not to reject the possibility of the cancellation of the contributions from the different diagrams , it seems to be very unlikely . thus we have shown that up to @xmath0 order the anomalous dimensions of composite operators are simply related to the @xmath14 poles of the corresponding diagrams . the simplicity of the formulae ( [ rg ] ) in the combination with the uniqueness of the triple vertex ( at @xmath32 ) and masslessness of propagators makes the @xmath0 order calculations a rather straightforward task . the only thing which causes some problem is a large number of diagrams to be evaluated . in the next section we show how to overcome this difficulty . everyone being a familiar with the @xmath3 expansion knows that the number of diagrams to be evaluated increases drastically with a order of expansion . to receive the impression about the rate of the growth it is sufficient to compare the number of the first and the second order diagram for any operator . indeed , one has only one @xmath3 order diagram ( fig . [ fn1 ] ) and eight ones in @xmath0 order ( fig . [ fn2 ] ) for the @xmath190 operator . the analogous comparison for the @xmath191 operator gives 3 and 69 diagrams , respectively . however , it can be easy realized that the bulk of the second order diagrams are the nothing else as the first order ones with the insertion of the vertex or the propagator corrections . it is intuitively clear that taking into account of these corrections must lead to the dressing of the bare propagators and vertices . in what follows we shall show that this really occurs . for further discussion to be more transparent we will use @xmath190 operator as illustrative example . the second order diagrams for the two - point green function with operator @xmath190 insertion are drawn on fig . two of them ( c , d ) arise due to the vertex corrections to the first order diagram ( fig . [ fn1 ] ) ; one ( e ) due to the self - energy ( se ) insertion in @xmath24 line and three ( f , g , h ) due to the se insertions in @xmath20 line . we are interested in the contributions from the diagrams of this type ( c g on fig . [ fn2 ] ) to the matrix @xmath192 . the more precisely we need the simple @xmath14 pole term in the sum @xmath193 = 0.5 cm ( hsize=10 ) ( 3,0)*(0.5)(-3.1,3.1)(3.1,3.1 ) ( -2.2,2)(4,0 ) = 0.5 cm ( hsize=10 ) ( 3,0)*(0.5)(-3.1,3.1)(3.1,3.1 ) ( -3.3,3)(6,0)(-2.3,2)(4,0 ) ( 9.5,0)*(0.5)(-3.1,3.1)(3.1,3.1 ) ( -3.3,3)(4,-2)(2.7,3)(-4,-2 ) ( 16,0)*(0.5)(-3.1,3.1)(3.1,3.1 ) ( -2.3,2)(4,0)(-1.3,0.9)(-1.9,2)[+1 ] ( 23,0)*(0.5)(-3.1,3.1)(3.1,3.1 ) ( -3.3,3)(2,0)(-2.3,2)(1,0 ) ( -1.5,3)(0,-1)(1,-0.5)(0,-1)(1,0.5 ) ( 0.8,-0.5)(2.3,0 ) ( 1.4,-1,5)(a)(8.2,-1,5)(b ) ( 14.9,-1,5)(c)(22.1,-1,5)(d ) ( hsize=10 ) ( 3,0)*(0.5)(-3.1,3.1)(3.1,3.1 ) ( -3.3,3)(6,0)(-1.3,0.9)(-1.5,1.6)[+1 ] ( 9.5,0)*(0.5)(-3.1,3.1)(3.1,3.1 ) ( -2.3,2)(1,0)(3,0)(1,0 ) ( -0.3,0)(2,0)(2)(2,1)(0,-2 ) ( 16,0)*(0.5)(-3.1,3.1)(3.1,3.1 ) ( -2.9,2.6)(1,0)(1,0)(1,0.5)(1,-0.5 ) ( 0.7,0.5)(0,-1)(1,0)(0,-1)(1,0 ) ( 2.5,0)(-1,0.5)(-1,-0.5)(-1,0.5)(0,-1 ) ( 2.2,0)(1,0 ) ( 23,0)*(0.5)(-3.1,3.1)(3.1,3.1 ) ( -2.3,2)(1,0)(3,0)(1,0 ) ( -0.3,0)(2,0)(2)(1.1,0.4)(1.8,0 ) ( 1.4,-1,5)(e)(8.2,-1,5)(f ) ( 14.9,-1,5)(g)(22.1,-1,5)(h ) here sum runs over the @xmath0 order nonrenormalized diagrams @xmath194 which arise from the @xmath3 order diagrams after taken into account the vertex and the propagator corrections ( the diagrams c - h on the fig . [ fn2 ] ) . for later convenience we rewrite the sum ( [ sum1 ] ) in the following form : @xmath195 here @xmath196 is the set of the counterterm diagrams corresponding to the ones from the set @xmath197 , i.e @xmath198 ; @xmath199 is the number of @xmath20 lines in the diagram @xmath200 , while @xmath201 is the number of @xmath20 lines in the divergent subgraph . it is appeared that the last sum in the eq . ( [ sum2 ] ) has no simple @xmath14 pole , provided that the relation @xmath202 between the number of the internal lines and the triple vertices holds for the @xmath3 order base diagram . indeed , one can see that the divergent subgraphs in the diagrams with the @xmath203 insertions in @xmath20 line ( fig . [ fn2 ] f h ) are the same as those in the diagrams with the vertex corrections and the @xmath203 insertion in @xmath24 line ( fig . [ fn2 ] c e ) . we combine the diagrams into three different groups ( c , f ) , ( d , g ) , ( e , h ) in accordance with the type of the divergent subgraph . obviously , the diagrams appearing after reducing of the divergent subgraph in the diagrams from the same group , i.e. those from the set @xmath204 , differ one from another only by the simple @xmath24 loop insertion in @xmath91 line . for example , for @xmath205 being the pole part of the divergent subgraph , the contribution of the ( e , h ) pair into the sum in question reads : 0.3 cm = 0.4 cm ( hsize=10 ) ( 3.8,1.5)@xmath206 ( 10,0)*(0.5)(-3.1,3.1)(3.1,3.1 ) ( -2.3,2)(4,0 ) ( 14.0,1.5)@xmath207 ( 19,0)*(0.5)(-3.1,3.1)(3.1,3.1 ) ( -2.3,2)(1,0)(3,0)(1,0 ) ( -0.3,0)(2,0)(2 ) ( 23,1.5)@xmath208 0.3 cm ( the factor @xmath209 here is due to the symmetry coefficient of diagram ( h ) . ) the analytical expression for the above sum can be represented as @xmath210 @xmath178 being the regular function of @xmath14 . ( to derive ( [ example ] ) it is sufficient to remember the definitions ( [ kp]),([propp]),([propx ] ) ) . it is seen the difference ( [ example ] ) does not contain the simple @xmath14 pole . obviously , the same arguments work and in a general case , the condition @xmath211 ensuring the cancellation of the simple pole contributions among each group separately . henceforth we shall only deal with the diagrams satisfying this condition . moreover , we confine our consideration by those diagrams for which the only subdivergencies are due to the vertex and se subgraphs . in spite of that these conditions restrict the class of the admissible operators it appears to be wide enough to justify the necessity of the present discussion . we start with the analyzing of the diagrams with the @xmath212 insertions , say , in @xmath24 lines ( fig . [ fn2 ] e ) . for later convenience we will regard the line indices as the variables and distinguish the regulators @xmath21 on the @xmath58 - lines being internal ( @xmath213 ) and external ( @xmath214 ) with respect to the divergent subgraph . also , we introduce the regular function @xmath178 related to the base diagram ( fig . [ fn1 ] ) with the line indeces differing from their canonical values @xmath215 and @xmath216 by @xmath217 as follows @xmath218 then the answer for the diagram with the se insertion in @xmath24 line can be written as @xmath219 where the factor @xmath220 result from the integration over loop momenta in the subgraph . for the corresponding counterterm diagram one has @xmath221 their contribution to the sum ( [ sum2 ] ) reads ( @xmath222 and @xmath223 in this case ) @xmath224 evaluating of the above difference at the `` symmetric '' point @xmath225 yields for the simple pole residue @xmath226 noticing that the constant @xmath227 is nothing else as the anomalous dimension of @xmath24 field : @xmath228 we relate the above expression to the base diagram with one bare @xmath24 propagator being replaced by the dressed one : @xmath229 ( @xmath230 and @xmath231 ) . namely , the expression ( [ contr ] ) can be obtained as the @xmath0 - order term in the expansion of the following object @xmath232 it is not more hard to see that taking into account the se insertions in @xmath58 line also leads to the dressing of this line , i.e. the contributions of these diagrams to the sum ( [ sum2 ] ) can be received from the expansion of the object : @xmath233 the following necessary step is the inclusion into consideration of the vertex corrections . indeed , having restricted by the propagators corrections only one loses the more attractive feature of @xmath3 - expansion the uniqueness of the triple vertex . the analysis of the vertex corrections is a bit more complicated than those of the @xmath203 insertions . indeed , the diagram with insertion in a line has , in fact , the same topology as the initial one , and hence , ( that is crucial ) they both can be described by one and the same function . to achieve this in the case in question , where the topology of the diagrams is essentially different , we proceed in the following way . = 0.5 cm ( hsize=10 ) ( 8,0)(0,-1.5)(a ) ( -3.4,3.4)(3.4,3.4 ) ( -2.75,2.5)(5,0 ) ( 20,0)(0.5,-1.5)(b ) ( -1.5,1.5)(1.5,1.5 ) ( -1.77,1.5)(-1.5,1.5)(1.25,1.5)(1.5,1.5 ) ( -3.5,2.95)(5.95,0)(-1.0,1.0)(6,0)(1,1 ) ( -2.2,1.5)(3,0 ) first of all , let us consider the vertex function itself . up to terms of @xmath167 order it reads @xmath234 here @xmath235 , @xmath236 are the contributions of the diagrams shown on the fig . [ ver]a , b , respectively . ( in the further we shall not show explicitly the momentum dependence in the functions @xmath237 . note only , that at @xmath32 the functions @xmath238 are independent from @xmath239 . ) due to conformal invariance of the theory , the form of the _ full _ vertex @xmath240 is fixed up to the constant factor . in coordinate space it reads @xmath241 where @xmath242 , @xmath243 . further , taking in mind the pole structure of the diagrams ( c , d ) on fig . [ fn2 ] we write for the latter @xmath244 } { m\delta_1(m\delta_1+\delta_2)},\ ] ] where @xmath245 , respectively . the function @xmath246 is regular with respect the variables @xmath247 and is defined as follows : @xmath248 = ( m\delta_{1}+\delta_{2 } ) \int \frac{d^{2\mu}p}{(2\pi)^{2\mu } } \gamma_{m}(\delta_{1},p , p+q ) { ( g/\gamma_{1})}(p, ... ,\delta_{2 } ) . \label{fun}\ ] ] here the function @xmath249 is related to the diagram with the vertex subgraph removed . ( for the diagram on fig . [ fn2]c , d it is given by the product of three propagators ) . the evaluation of the contribution of diagrams ( c , d ) on fig . [ fn2 ] together with the corresponding counterterm diagrams yields for the simple pole residue in the sum ( [ sum2 ] ) @xmath250 + \frac12{\widehat}f[\gamma_2(\delta);0]\right)\bigl |_{\delta=0}. \label{vert}\ ] ] looking on the eq . ( [ fun ] ) one can see that the renormalized vertex @xmath240 should appear in the natural way if it were be possible to carry out differentiating with respect @xmath14 under the sign of integral . however , at @xmath251 the integral evolves the pole in @xmath213 , therefore this operation is not allowed . to obviate this difficulty , one can , taking into account the regularity of the function @xmath252 in @xmath247 , carry out differentiation at the nonzero second argument @xmath214 and then take the limit @xmath253 . at first step we obtain for ( [ vert ] ) @xmath254 where @xmath255 ( see eq . ( [ gr ] ) ) . to proceed further we represent @xmath256 as @xmath257 and , again , change the order of the differentiation and integration . at last , taking into account that scaling dimension @xmath258 ( [ chieta ] ) of the function @xmath240 is equal to @xmath259 at first @xmath3 order , so that the integral in ( [ ver1 ] ) evolves the pole in @xmath260 , we find eventually for ( [ vert ] ) the following representation @xmath261 in other words the contribution of the diagrams in question is obtained from the @xmath3 expansion of the diagram with the dressed vertex . basing on these results ( eqs . ( [ 1]),([2]),([3 ] ) ) we can formulate the simple algorithm for the calculating of the contribution of the diagrams with the propagator and vertex corrections to the matrix of the anomalous dimensions . let g to be a @xmath3 order diagram and the diagrams @xmath262 are the @xmath0 order ones from the class under consideration . then to obtain the contribution of the diagrams @xmath262 to the anomalous dimension one need : = 0.5 cm ( hsize=5 ) ( 3,1)(-2.1,2.1)(2.1,2.1)(0,-2 ) ( 8,1.5)(1,0)(4,0 ) ( 15,1.5)@xmath263 ( 18,0)(-3.1,3.1)(3.1,3.1)(0,-1 ) ( -2.75,2.5)(5,0 ) ( -2.5,1)@xmath264(+1,1)@xmath264 ( -1.2,2.8)@xmath265 1 . replace in the base diagram @xmath175 the bare propagators by dressed ones and the pointlike triple vertex by the conformal triangle ( see fig . [ triangle ] ) : @xmath266 here @xmath242 , @xmath243 and @xmath267 are the full scaling dimensions of the fields @xmath268 , respectively . @xmath269 @xmath270 2 . introduce a regulator @xmath14 in anyone line ( it is needed because after the above changes one obtains a superficially divergent diagram ) . 3 . evaluate the @xmath14 pole residue in the resulting diagram and pick out the @xmath0 order term @xmath271 . then the required contribution is given by @xmath272 . the explicit expressions for the amplitudes @xmath273 , @xmath274 , @xmath263 are given in appendix a. following these prescriptions one need calculate only one ``conformal '' diagram ( see fig . [ conf ] ) instead of the six @xmath0 order diagrams shown on fig . [ fn2 ] ( c - h ) . it should be noted also , that the resulting diagram , unlike the original ones , has superficial divergence only , that , in its turn , gives the definite advantages at the calculations . we end this section with remark , that the above procedure being useful even if the condition @xmath275 is not fulfilled . indeed , the only changes which will occur are that the last sum in the eq . ( [ sum2 ] ) will not be zero any longer and will contribute to the anomalous dimension on the equal foot with the first ones . in this section we apply the technique developed above to calculate the anomalous dimensions of the operators @xmath276 in the @xmath0 order . here , the symbol @xmath277 implies the symmetrization over the internal indeces as well as the spatial ones . the operators @xmath278 are multiplicatively renormalized , whereas @xmath279 does not . however , the operators admixing to @xmath279 being the total derivatives , they are irrelevant for our purposes . the diagrams to be evaluated for the determination of the anomalous dimensions of operators @xmath280 are those shown on fig . [ fn2 ] and fig . [ fn3 ] , the latter being obviously relevant only for the @xmath278 operators . because the diagrams @xmath281 on fig . [ fn2 ] can be replaced by one `` conformal '' diagram ( fig [ conf ] ) in accordance with the rules formulated in the previous section , one has to calculate @xmath282 diagrams for the @xmath278 operators and @xmath283 ones for the @xmath279 operators . = 0.5 cm ( hsize=10 ) ( 15,0)(-2.4,2.5)*(0.7)(2.75,2.5)*(0.7 ) * ( 0.5)(-3.1,3.1)(3.1,3.1 ) ( -2.75,2.5)(5,0 ) ( -3.8,1)@xmath284 ( 1,1)@xmath285 ( -1.7,3)@xmath286 we do not dwell upon the technique of the calculating of the diagrams , since it is discussed in very details in the papers @xcite and give the final results for the anomalous dimensions only . ( for completeness , we give the explicit expressions for the pole terms of each diagram in appendix [ ab ] and the some useful formulae are collected in appendix [ ac ] . ) the anomalous dimensions @xmath287 and @xmath288 of the operators @xmath289 and @xmath290 it is convenient to represent as @xmath291 where @xmath292 is the anomalous dimension of the field @xmath24 known up to @xmath8 order @xcite . then representing @xmath85 in the following form @xmath293 we have for the two first terms : @xmath294 \bigr\},\end{aligned}\ ] ] @xmath295 where @xmath296 the critical exponents of the operators @xmath297 are known with the high accuracy in the @xmath298 as well as in the @xmath299 expansions . expanding our result for @xmath287 in the @xmath300 series near @xmath301 we have obtained the full agreement with the four loop result of wegner @xcite . the same exponents are known up to @xmath302 order in the @xmath298 expansion @xcite . in this case we find out the discrepancy of our result with those of wallace and zia @xcite in term @xmath303 ( in our answer this term enters with the coefficient @xmath304 , against the coefficient @xmath305 in the paper @xcite ) . further , using the results of the paper @xcite , where the four loop anomalous dimensions of @xmath306 operators had been calculated in the scalar @xmath307 model in the @xmath298 expansion , we reconstructed the answer for the @xmath16 vector model by restoring the @xmath16 coefficients for all diagrams . comparing the expressions for the critical exponents of @xmath279 operators derived in the both ( @xmath3 and @xmath298 ) approaches we found that they are in the full agreement up to @xmath308 terms . = 0.5 cm ( hsize=10 ) ( 10,0)*(0.5)(-3.2,3.2)(3.2,3.2)(0,3.2 ) ( -3.3,3)(3,0 ) ( -0.3,2)(2,0)(-0.5,-1.5)(a ) ( 17.5,0)*(0.5)(-3.2,3.2)(-4,1)(4,2 ) ( -2.3,2)(1,0 ) ( -0.3,0)(2,0)(2 ) ( 1.5,0.95)(-1.1,0.92 ) ( -3.1,0.5 ) ( 1.15,0.92 ) ( -1,-1.5)(b ) in the present paper we have developed the simple method for the calculating of the anomalous dimensions ( up to @xmath0 order ) in the @xmath3 expansion . in the comparison with the other well known methods of the conformal bootstrap and the selfconsistency equations this one has the advantage of being not restricted to the special class of the operators . in this approach the formulae relating the anomalous dimensions with the divergencies of the corresponding feynman diagrams are very simple and in the one to one correspondence with their counterparts in the dimensional regularization scheme . these features together with the uniqueness of the triple vertex of the nonlinear sigma model make the @xmath0 order calculations of the rather straightforward task . in some cases ( see for example the sec . [ ff ] ) the calculations are not much more complicated than the two loop ones in the @xmath16 model in the @xmath298 expansion . however , in a general situation , beyond @xmath3 order , the number of the diagrams to be evaluated is usually large enough . we have shown that this problem is partially removed after going on to the diagrams with the dressed propagators and vertices . the number of the diagrams is considerably reduced by this trick and only slightly exceeds the number of those in the conformal bootstrap method , which is considered to be the most economic in this sense . we conclude with remark , that though so far we consider the sigma model only , it is evident that the same technique can be apply to the many other models admitting @xmath3 expansion @xmath309 , gross neveu and thirring models , etc . * acknowledgments * 0.5 cm it is a pleasure to thank prof . a. n. vasilev for the valuable discussions and dr . j. gracey for critical remarks . the work by s.d . was supported by nordita as a part of the baltic fellowship program funded by the nordic council of ministers . this work was also supported in part by the russian fond of basic research ( grant 970101152 ) and by intas grant 932492ext and carried out under the research program of the international center for fundamental physics in moscow . [ aa ] in this appendix we give the explicit expressions for the amplitudes @xmath311 , @xmath312 , @xmath313 introduced into consideration in the sec . [ confors ] . obviously , these quantities depend on the our choice of the function @xmath31 ( see eqs . ( [ propp]),([propx ] ) ) . however , it is easy to understand , that in the first @xmath3 order this dependence is almost trivial and can be taking into account by the redefinition of the parameter @xmath22 . we specify @xmath31 by the condition for the propagator of @xmath20 field in coordinate space ( eq.([propx ] ) ) to have the amplitude independent from @xmath14 , except from the factor @xmath314 . then one has for the amplitude @xmath41 of the bare propagators in the coordinate space : @xmath315 evaluating the @xmath3 order diagrams for @xmath107 green functions @xmath316 , @xmath317 @xmath318 shown on fig . [ apd ] we obtain for the amplitudes @xmath311,@xmath312,@xmath313 : = 0.7 cm ( hsize=10 ) ( 3,0 ) @xmath319 ( 0.2,0.1)(2,0)(0.1,0)(1.8,0)[+1.5 ] ( 2.5,0 ) @xmath320 ( 0.1,0.1)(1,0.5)(1,-0.5)(0.9,0.4)(0,-0.8 ) ( 1.1,0)@xmath321 ( 1.6,0.1 ) ( 1,0.5)(1,-0.5)(1,0.5)(0,-1 ) ( 0.8,0.5)(1,0)(0.8,-0.5)(1,0 ) ( 1.6,0.5)(0,-1)(0.1,0.05 ) ( 1.6,-0.5)(0.1,-0.05 ) ( hsize=10 ) ( 3,0 ) @xmath322 ( 0,0.1 ) ( 0.05,0)(0.15,0 ) ( 0.5,0)(1)(0,0.5)(0,-1 ) ( 1,0)(0.15,0 ) ( 1.5,0 ) @xmath321(0.05,0.1)(0.15,0 ) ( 0.5,0.1)(1)(-0.45,0.2)(0.9,0 ) ( 0.8,0.1)(0.15,0 ) ( 3.6,0)@xmath323 ( 4.6,0.1 ) ( 0.05,0)(0.15,0 ) ( 1,0.5)(1,-0.5)(1,0.5)(0,-1 ) ( 0.8,0.5)(1,0)(0.8,-0.5)(1,0 ) ( 1.6,0.5)(0,-1)(1,-0.5)(1.6,-0.5)(1,0.5 ) ( 2.4,0)(0.15,0 ) @xmath324 , \\ { \widehat}b & = & bm^{-2\gamma_{\psi } } \left [ 1 + \frac{\eta_{1}}{n } \left ( \frac{2\mu^2 - 3\mu+2}{\mu-2}\cdot r - \frac{4\mu^5 - 19\mu^4 + 25\mu^3-\mu^2 - 14\mu+4}{\mu(\mu-1)(\mu-2)^2 } \right ) \right],\\ { \widehat}z & = & -m^{-2\chi}\frac{\chi}{2}\frac{h^2(1 ) h(\mu-2 ) \gamma(\mu)}{\pi^{2\mu } } \cdot \left [ 1 + \frac{\eta_{1}}{n } \frac{(\mu-3)(6\mu^2 - 9\mu+2)}{(\mu-2)^2}\right].\end{aligned}\ ] ] here @xmath325 are given by eq . ( [ eta1]),([rr ] ) and @xmath326 are the known anomalous dimensions of the fields @xmath327 @xmath328 the index @xmath74 is defined as @xmath329 [ ab ] in this appendix we give the explicit expressions for the contributions to the anomalous dimensions @xmath331 of the corresponding feynman integrals . we will use the labels @xmath332 for the diagrams on the fig . [ fn2 ] a , b , fig . [ conf ] and fig . [ fn3 ] a , b , resp . , as well as for the values of the latters . in the case of the @xmath333 operators we have obtained for the simple @xmath14 pole residue of the corresponding integrals in the units @xmath334 : @xmath335,\\ & & \nonumber \\ d&= & { s(s-1)(s-2 ) } \frac{\mu^2}{8(\mu-1)(\mu-2)^2},\\ & & \nonumber \\ e&= & \frac{s(s-1)(s-2)}{6}\frac{3\mu^2(\mu-1)}{2(\mu-2)^2}c.\end{aligned}\ ] ] for diagrams with the insertion of the @xmath336 operators we derived @xmath337 each of these quantities enters into expression for the anomalous dimension with the coefficient ( @xmath338 ) , if we accept @xmath339 for diagram @xmath340 . here we collect some rules ( see refs . @xcite ) which form the basis for the calculations of diagrams in the massless theories . as usual , we will denote the propagator @xmath341 by the line with the index @xmath342 . further , we consider the chain of propagators , which is illustrated on fig . [ apchain ] . the point where the ends of propagators are joined is the vertex of integration . the result of this integration is a new line with index @xmath348 and amplitude proportional to @xmath349 , where @xmath350 . = 0.7 cm ( hsize=10)(5.,0 . ) ( 1.7,0 ) ( 0.5,0.1)@xmath264 ( 1.4,0)(1.5,0)*(0.1 ) ( 0.5,0.1)@xmath265 ( 3.1,-0.1)@xmath351 ( 9,0)(3,0)(0.5,0.1)@xmath352 ( hsize=10)(7.,0 . ) ( 0.2,0)*(0.1)(0,1.4)(0.1,0.5)@xmath353 ( -1,-1)(-0.9,-0.5)@xmath264(1,-1)(0.6,-0.5)@xmath265 ( 1.2,0)@xmath354 ( 6.5,1.4)(-1,-2.4)(1,-2.4)(-1,-2.4)(2,0 ) ( 7,0)@xmath355(5.2,0)@xmath356 ( 6,-1.5)@xmath357 ( 1,-2.5 ) @xmath358 the other useful relation is the so - called uniqueness relation . in the case if the sum of the exponents of the lines meeting at the vertex is equal to the dimension of space - time it can be related to a triangle graph ( see fig . [ apchain ] ) . the corresponding generalization of these rules needed for the evaluation of the the diagrams with the insertion of the tensor operators , are given in refs . we give here only that which is relevant for the treatment of the traceless and symmetric operators , like operator @xmath279 in our case . it is convenient to swamp the the spatial indices @xmath359 of a operator by contracting with a constant vector @xmath360 , such that @xmath361 , and consider scalar operator @xmath362 . in the course of calculations one should often to integrate the chains of the propagators like @xmath363 . then , the corresponding rule for the integration of the chains reads @xmath364 where @xmath365 . vasilev , yu.m . pismak & j.r . honkonen , theor . phys . * 46 * , ( 1981 ) 104 . vasilev , yu.m . pismak & j.r . honkonen , theor . phys . * 47 * , ( 1981 ) 465 . gracey , j.phys . * a24 * , ( 1991 ) l431 ; int.j . of mod.phys . * a6 * , ( 1991 ) 395 ; j.phys . * a25 * , ( 1992 ) l109 ; mod.phys.lett . * a7 * , ( 1992 ) 1945 . vasilev , yu.m . pismak & j.r . honkonen , theor . phys . * 50 * , ( 1982 ) 127 . vasilev , s.e . derkachov , n.a . kivel & a.s . stepanenko , theor . phys . * 94 * , ( 1993 ) 127 . gracey , int.j . of mod.phys . * a9 * , ( 1994 ) 727 ; a.n . vasilev , & a.s . stepanenko , theor . * 97 * , ( 1993 ) 364 . broadhurst , j.a . gracey , d. kreimer , z. phys . * c75 * , ( 1997 ) 559 ; j.a . gracey , nucl . * a389 * , ( 1997 ) 361 . h. osborn and a. petkou , ann.phys . * * , ( 1994 ) 311 . a. petkou , ann.phys . * * , ( 1996 ) 180 . k. lang and w. rhl , nucl . phys . * b[fs ] 400 * , 597 ( 1993 ) . s.a.larin , t. van ritbergen , j.a.m . vermaseren , phys . lett . * b400 * , ( 1997 ) 379 . s.a.larin , t. van ritbergen , j.a.m . vermaseren , nucl . phys . * b427 * , ( 1994 ) 41 ; s.a.larin , p. nogueira , t. van ritbergen , j.a.m . vermaseren , nucl . phys . * b492 * , ( 1997 ) 338 . gracey , phys . lett . * b322 * , ( 1994 ) 141 ; phys . lett * b373 * , ( 1996 ) 178 . kravtsov , i.v . lerner and v.i . yudson , zh . eksp . teor . fiz . * 94 * , ( 1988 ) 255 . f. wegner , z. phys . * b 78 * , ( 1990 ) 33;nucl . phys . * b[fs ] 354 * , ( 1991 ) 141 . e. brezin and s. hikami , phys.rev . * b55 * , ( 1997 ) 10169r . derkachov and a.n . manashov , phys.rev.lett . * 79 * , ( 1997 ) 1423 . vasilev and a.s . stepanenko , teor . fiz . * 95 * , ( 1993 ) 160 ; theor . phys . * 95 * , ( 1993 ) 471 . gracey , phys . lett . * b297 * , ( 1992 ) 293 a.n . vasilev and m.yu . nalimov , teor . mat . fiz . * 55 * , ( 1983 ) 163 ; theor . * 55 * , ( 1983 ) 423 . kazakov , theor . * 58 * , ( 1984 ) 223 ; phys . b133 * , ( 1983 ) 406 . kazakov & a.v . kotikov , theor . phys . * 73 * , ( 1988 ) 1264 . f. wegner , nucl.phys . * b[fs ] 280 * , ( 1987 ) 193 . wallace and r.k.p . zia , j.phys . * c8 * , ( 1975 ) 839 . derkachov , j.a . gracey , a.n . manashov , four loop anomalous dimensions of gradient operators in @xmath366 theory ; preprint * spbu ip977 , lth391 * ; hep - ph/9705268 .

we present results from a survey ofco ( _ j _ = 10 ) spectra obtained for the central regions of 68 nearby galaxies at an angular resolution of 16 using the nobeyama radio observatory 45 m telescope , aimed at characterizing the properties of star forming molecular gas . combined with similar resolution observations in the literature , the compiled sample set of 166 galaxies span a wide range of galactic properties . ngc 4380 , which was previously undetected in co , was detected . this initial paper of a series will focus on the data and the gaseous properties of the samples , and particularly on the degree of central concentration of molecular gas in a range of morphological types , from early ( s0/sa ) to late ( sd / sm ) galaxies with and without bars . the degree of molecular central concentration in the central kiloparsec , compared to the central several kiloparsecs of galaxies , is found to vary smoothly with hubble type , so that early type galaxies show larger central concentration . the comparison of barred and non - barred galaxies within early and late type galaxies suggest that difference in hubble type , representing the effect of bulges , is the more important factor in concentrating gas into the central regions than bars .

0805.4048

molecular gas is the driving element of star formation in galaxies , and its quantitative relation to the current star formation rate is essential to our understanding of galactic evolution . it has become increasingly clear that the dynamical properties of galaxies are responsible for redistributing molecular gas on large scales , and hence the star formation which follows is also affected by the dynamical properties . these dynamical properties are expressed as large scale features ( morphology ) of the galaxies , i.e. , the presence bars and difference between hubble type . these two features ( presence of bars and hubble type ) are most prominent and characteristic in the central kiloparsecs of galaxies , which is also where most of the molecular gas in galaxies is present . therefore , the distribution of gas within the central region of various morphologies and its relation with star formation has gained attention in recent years . bars on kpc scales at the centers of galaxies are known to redistribute angular momentum and transfer molecular gas to the centers . @xcite showed that molecular gas in barred galaxies has a higher degree of concentration in its central 1 kpc compared to gas inside the optical disk , by using interferometric data from the nobeyama millimeter array ( nma ) and owens valley radio observatory ( ovro ) at @xmath2 resolution . similarly , @xcite showed by using the berkeley - illinois - maryland - array ( bima ) at @xmath3 resolution that the central 1 kpc of barred galaxies has four times more molecular gas compared to the global average . some barred galaxies in their sample were not detected , and they attribute it to the quenching of gas due to efficient star formation in barred galaxies . single dish observations have also proved effective . by using the nobeyama radio observatory@xmath4 ( nro ) 45 m telescope , @xcite mapped out 40 nearby galaxies at @xmath5 resolution . they defined the `` inner '' central concentration @xmath6 as the ratio between molecular mass in @xmath7 of the k band radius ( defined at 20 @xmath8 in the @xmath9 band ) and that within @xmath10 of the k band radius . they similarly defined the `` outer '' central concentration @xmath11 as the ratio between molecular mass in @xmath10 of the k band radius and that within the whole of the k band radius . they found that barred galaxies have a significantly higher @xmath6 , whereas the @xmath11 did not show much dependence . furthermore , they found a correlation between @xmath6 and the bar strength . it seems that molecular gas concentration due to bars is a well established phenomenon , and this has also been explored in terms of numerical simulations(@xcite ; @xcite ; @xcite ; @xcite ) . molecular gas is also known to be distributed differently between early and late type galaxies . molecular gas mass in the centers of galaxies have been found to decrease in later hubble types ( e.g. , @xcite , @xcite ) . this has been attributed to the existence of a stellar bulge in early type galaxies ( @xcite ) creating a deep potential well in which gas is accumulated . even in the absence of stellar bulges , possible correlations with nuclear star clusters and stellar surface brightness have been found ( @xcite ) . sheth et al . ( 2005 ) report a four - fold central molecular mass increase in barred early type galaxies compared to barred late type galaxies . early type galaxies , however , have been found to host stronger bars than late types , with larger @xmath12 and @xmath13 fourier amplitudes ( @xcite , but see also @xcite ) . this points to the possibility that early type galaxies may show larger central concentration simply because they host stronger bars . it is important , therefore , to quantify which factor ( the presence of bars or hubble type ) plays the larger role in ridistributing gas to the central regions of galaxies . mapping observations in co are optimum for the study of molecular gas distribution in nearby galaxies , but available galaxy samples are limited by the vast observing time needed for such projects ( e.g. , sakamoto et al . 1999 ; helfer et al . 2001 ; sofue et al . 2003 ) . in turn , single beam measurements at the centers of a large sample of galaxies lack the precision in terms of spatial distribution of molecular gas , but can compensate for the lack of number of sample galaxies . in this respect , quantifying the co content of nearby galactic centers for a large sample using single dish telescopes , is still rewarding . in order to complement existing studies on molecular gas concentration in galaxies of various morphology , we have conducted a large survey of the central region of nearby galaxies using nro 45 m telescope . our data provides a unique opportunity to study the gas distribution in the central region , with several advantages : 1 ) samples are distributed mostly at the distance of 16mpc , and therefore the single beam subtends a similar scale of the central 1 kpc of galaxies . 2 ) many of our samples are also observed by young et al . ( 1995 ) using the fcrao 14 m dish , with an angular resolution of 45@xmath1 or a linear scale of 3 kpc . therefore , the ratio of co intensity measured by the different surveys can be interpreted as a degree of central concentration of molecular gas . the sample galaxies have been selected from those observed and detected in the fcrao catalog ( @xcite and @xcite ) of nearby galaxies . the fcrao survey is the largest co survey of nearby galaxies ( 300 galaxies ) , spanning a wide range of parameter space for spirals , but with a large beamsize of 45@xmath1 . all of our observing samples , mainly spirals , come from galaxies observed in this survey . since one of the secondary aims of our survey is compare the co data with star formation ( to be presented in forthcoming papers ) , galaxies in the fcrao survey were selected on whether they had available @xmath14 data as listed in @xcite , and also with measured @xmath15 ratio ( to be used to correct for internal dust extinction ) in @xcite . ninety - one galaxies meet this criteria , of which 14 were already observed at either the nro 45 m or the iram 30 m telescope which has a similar beamsize as the nro 45 m telescope . therefore , the final selected sample consisted of 77 galaxies . galaxies in the fcrao survey were selected to span a wide variety of galaxies , and were not magnitude or volume limited . our sample also follows this idea . in the observation ( see following section ) , 38 of the 77 were actually observed , due mainly to weather conditions . an additional five , with @xmath14 data from @xcite but without @xmath15 , and 25 with observations from the fcrao survey but not @xmath14 , were observed . these galaxies without @xmath14 measurements were simply selected by whether they were observable at that certain perioud . thus , our observed sample consists of 68 galaxies . table [ 45msample ] lists the observed samples . llll ngc 253 & 00 47 35.2 & -25 17 20 & 2.2 + ngc 520 & 01 24 34.7 & + 03 47 49.8 & 30.3 + ngc 772 & 01 59 20.3 & + 19 00 22 & 34.1 + ngc 877 & 02 17 58.7 & + 14 32 50.3 & 54.9 + ngc 908 & 02 23 04.8@xmath16 & -21 14 06 & 19.6 + ngc 1022 & 02 38 32.9@xmath16 & -06 40 29 & 20.1 + ngc 1482 & 03 54 39.4@xmath16 & -20 30 08.5 & 20.5 + ngc 1569 & 04 30 50.3 & + 64 50 48 & 3.1 + ngc 1961 & 05 42 04.3 & + 69 22 46 & 54.1 + ngc 2276 & 07 27 14.4@xmath4 & + 85 45 16 & 34.4 + ngc 2336 & 07 27 04.1@xmath4 & + 80 10 41 & 31.9 + ngc 2339 & 07 08 20.5@xmath4 & + 18 46 49 & 31.1 + ngc 2403 & 07 36 54.5 & + 65 35 58.4 & 2.2 + ngc 2559 & 08 17 07.5@xmath16 & -27 27 34 & 17.3 + ngc 2683 & 08 52 41.3@xmath4 & + 33 25 19 & 3.2 + ngc 2976 & 09 47 15.6 & + 67 54 49 & 2.3 + ngc 3034 & 09 55 54.0 & + 69 40 57 & 2.2 + ngc 3147 & 10 16 53.6@xmath4 & + 73 24 03 & 38.4 + ngc 3556 & 11 11 30.9 & + 55 40 27 & 10.3 + ngc 3593 & 11 14 37.0@xmath4 & + 12 49 04 & 7.3 + ngc 3623 & 11 18 56.0@xmath4 & + 13 05 32.0 & 4.5 + ngc 3631 & 11 21 02.9@xmath4 & + 53 10 11 & 16.6 + ngc 3675 & 11 26 08.6@xmath4 & + 43 35 09 & 9.8 + ngc 3690 & 11 28 33.6 & + 58 33 51.3 & 41.4 + ngc 3893 & 11 48 39.1 & + 48 42 40 & 13.8 + ngc 3938 & 11 52 49.8 & + 44 07 26 & 11.2 + ngc 4038 & 12 01 53.0@xmath16 & -18 51 48 & 19.2 + ngc 4039 & 12 01 54.2@xmath16 & -18 53 06 & 19.2 + ngc 4041 & 12 02 12.2@xmath17 & + 62 08 14 & 17.6 + ngc 4088 & 12 05 35.3 & + 50 32 31 & 16.4 + ngc 4096 & 12 06 01.0 & + 47 28 31 & 11.2 + ngc 4157 & 12 11 04.4@xmath4 & + 50 29 05 & 12.1 + ngc 4178 & 12 12 46.4@xmath4 & + 10 51 57 & 16.1 + ngc 4212 & 12 15 39.3@xmath4 & + 13 54 06 & 16.1 + ngc 4293 & 12 21 12.9@xmath4 & + 18 22 57 & 16.1 + ngc 4298 & 12 21 32.8@xmath4 & + 14 36 22 & 16.1 + ngc 4302 & 12 21 42.4 & + 14 36 05 & 16.1 + ngc 4321 & 12 22 54.9@xmath4 & + 15 49 20 & 16.1 + ngc 4380 & 12 25 22.2@xmath4 & + 10 01 00 & 16.1 + ngc 4394 & 12 25 55.6@xmath4 & + 18 12 50 & 16.1 + ngc 4418 & 12 26 54.2 & -00 52 45 & 25.5 + ngc 4419 & 12 26 56.4@xmath4 & + 15 02 51 & 16.1 + ngc 4424 & 12 27 11.6@xmath4 & + 09 25 14 & 16.1 + ngc 4527 & 12 34 08.8 & + 02 39 13 & 16.1 + ngc 4536 & 12 34 27.1@xmath4 & + 02 11 16 & 16.1 + ngc 4548 & 12 35 55.2@xmath4 & + 14 29 47 & 16.1 + ngc 4567 & 12 36 32.7@xmath4 & + 11 15 29 & 16.1 + ngc 4568 & 12 36 34.7 & + 11 14 15 & 16.1 + ngc 4647 & 12 43 32.3@xmath4 & + 11 34 55 & 16.1 + ngc 4666 & 12 45 08.9 & -00 27 38 & 18.6 + ngc 4689 & 12 47 45.5@xmath4 & + 13 45 46 & 16.1 + ngc 4691 & 12 48 14.1@xmath16 & -03 19 51 & 13.1 + ngc 4698 & 12 48 23.5 & + 08 29 16 & 16.1 + ngc 4710 & 12 49 38.9@xmath4 & + 15 09 56 & 16.1 + ngc 4818 & 12 56 48.7@xmath16 & -08 31 25 & 13.5 + ngc 4845 & 12 58 01.4 & + 01 34 30 & 16.1 + ngc 4984 & 13 08 57.3@xmath16 & -15 30 59 & 14.7 + ngc 5236 & 13 36 59.4 & -29 52 04 & 5.9 + ngc 5247 & 13 38 03.5 & -17 53 02 & 20.1 + ngc 5713 & 14 40 11.5 & -00 17 26 & 24.6 + ngc 5861 & 15 09 16.5@xmath16 & -11 19 24 & 24.1 + ngc 6000 & 15 49 49.4 & -29 23 13 & 27.1 + ngc 6814 & 19 42 40.1@xmath16 & -10 19 27 & 21.1 + ngc 6946 & 20 34 51.9 & + 60 09 15 & 6.7 + ngc 6951 & 20 37 15.2 & + 66 06 22 & 21.7 + ngc 7479 & 23 04 57.1 & + 12 19 18 & 34.7 + ngc 7541 & 23 14 43.0 & + 04 32 05 & 38.1 + ic 342 & 03 46 49.7 & + 68 05 45 & 3.0 + the observations of the @xmath0co ( _ j _ = 10 ) line ( rest frequency 115.27 ghz ) were conducted with the 45 m telescope at the nobeyama radio observatory ( nro ) of the national astronomical observatory of japan , during three observing runs in may 2005 , december 2005 , and january 2006 , under moderate weather conditions . two superconductor - insulator - superconductor ( sis ) receivers , s100 and/or s80 , were used for the observations . the backends were the 1024 channel digital auto - correlator with a frequency coverage of 512 mhz , or the 2048 channel acousto - optical spectrometer with a frequency coverage of 250 mhz . the co line was observed in the upper sideband . typical system temperatures ranged from 700 to 2000 k in the may 2005 run , and 450 to 1000 k ( all in single - sideband scale ) in the december 2005 and january 2006 run . the beamwidth was 16@xmath1 . pointing was checked using sio masers or continuum sources at 43 ghz every 0.51 hour , and found to be accurate within 5@xmath1 unless otherwise noted where wind prevented pointing at such accuracy . each of the target galaxies was observed with a single beam at the center . the obtained data were reduced with the newstar software used commonly at nro , which is based on the astronomical image processing system ( aips ) package , developed by the national radio astronomy observatory . after flagging bad spectra , first or second order baselines were subtracted , and then the antenna temperature @xmath18 was converted to main beam temperature @xmath19 through @xmath20 where @xmath21 is the main beam efficiency , taken to be 0.38 based on november 2004 observations of the saturn and 3c279 . the typical r.m.s noise ranged from 10 to 20 mk , after reduction . the spectra were then binned into velocities which best showed the emission features ( typically 10 km / s or 15 km / s ) . the integrated intensity @xmath22 is calculated by @xmath23 , were the range of velocity for integration was chosen with reference to spectra from the fcrao calatog ( @xcite , or kenney and young ( 1988 ) ) . the errors for the line intensities were calculated as @xmath24 \label{eqerror}\ ] ] where @xmath25 is the r.m.s . noise in @xmath19 , @xmath26 the full line width , and @xmath27 the velocity resolution ( 5 , 10 , 15 , or 20 @xmath28 ) . for spectra with signal to noise ratio of less than 3 , the upper limits were calculated by @xmath29\ ] ] for those galaxies which were detected in the fcrao survey but not ours , the same @xmath30 was assumed . for galaxies which were detected in neither sample , an arbitraray @xmath30 of 100 km / s was assumed . the observational errors presented in the following section are only for the r.m.s . errors of the baseline . in general , other sources of error ( pointing error , baseline subtraction and calibration ) which are difficult to quantify , are dominant sources of error for millimeter observations . three galaxies ( ngc 6951 , ngc 6946 and ic 342 ) were observed both in this study and the survey by @xcite at the same telescope . the integrated intensities of ngc 6951 and ic 342 are consistent to within 30 % . for ngc 6946 , @xcite give a value which is @xmath31 % of this study , but may be attributed to the bad observing condition in our study ( system temperature 1400 k ) . for most of the samples , therefore , readers should assume a typical error of @xmath32 % for the obtained integrated intensities . we show the spectra of the observed galaxies in figure 1 ( detected galaxies followed by tentative detections and non - detections ) . of the 68 galaxies observed , 54 were detected at a signal to noise ratio ( s / n ) of more than 5 , and 60 galaxies with s / n more than 3 . all analysis in this paper will be limited towards galaxies with s / n over 5 . [ cols="^,^ " , ] [ spectra2_3 ] in addition to the data gathered in this study at the nro 45 m telescope , other galaxies which were observed in other previous studies either at nro 45 m or the iram 30 m telescope were compiled . the iram 30 m telescope has a beamwidth of 22@xmath1 , similar to the nro45 m telescope . after compilation , we have a total sample of 166 galaxies observed in co at the central region . furthermore , 117 of these were observed by young et al . ( 1995 ) or kenney & young ( 1988 ) with 45@xmath1 resolution . table [ tbl-1 ] lists the compiled data , along with the derived integrated intensities from this study . lllllll ngc 157 & [email protected] & 2 & 23.3 & 9.02 & 4.0 & sab(rs)bc ngc 253 & 248@xmath3311 & 1 & 2.2 & 467 & 5.1 & sab(s)c ngc 278 & [email protected] & 2,6 & 11.8 & 14.2 & 2.9 & sab(rs)b ngc 337a & < 0.46 & 5 & 13.3 * & 35.5 & 7.9 & sab(s)dm ngc 520 & [email protected] & 1 & 20.2 & 30.2 & 0.8 & pec ngc 628 & [email protected] & 2,6 & 10.6 & 4.15 & 5.2 & sa(s)c ngc 660 & [email protected] & 2 & 13.1 & 38.2 & 1.3 & sb(s)a pec ngc 772 & [email protected] & 1 & 34.1 & 16.1 & 3.0 & sa(s)b ngc 864 & [email protected] & 2 & 21.8 & 1.31 & 5.1 & sab(rs)c ngc 877 & [email protected] & 1 & 54.9 & 12.9 & 4.8 & sab(rs)bc ngc 891 & 96@xmath335 & 2 & 9.4 & 26.2 & 3.0 & sa(s)b sp ngc 908 & [email protected] & 1 & 19.6 & 18.9 & 5.1 & sa(s)c ngc 925 & [email protected] & 2 & 9.5 & 1.18 & 7.0 & sab(s)d ngc 1022 & [email protected] & 1 & 20.1 & 9.67 & 1.1 & sb(s)a ngc 1042 & [email protected] & 2 & 19.1 * * & - & 6.0 & sab(rs)cd ngc 1055 & [email protected] & 2 & 14.3 & 41.1 & 3.2 & sbb sp ngc 1068 & 218@xmath337 & 3 & 15.1 & 99.8 & 3.0 & sa(rs)b ngc 1084 & [email protected] & 2 & 18.7 & 11.5 & 4.9 & sa(s)c ngc 1087 & 15@xmath331 & 2 & 24.6 & 6.86 & 5.2 & sab(rs)c ngc 1482 & [email protected] & 1 & 20.5 & 20.0 & -0.9 & sa0 + pec sp ngc 1569 & < 14.5 & 1 & 3.1 & 1.82 & 9.6 & ibm ngc 1637 & [email protected] & 2 & 9.5 * * & - & 5.0 & sab(rs)c ngc 1961 & [email protected] & 1 & 54.1 & 29.5 & 4.2 & sab(rs)c ngc 2146 & [email protected] & 2 & 13.7 & 68.4 & 2.3 & sb(s)ab pec ngc 2276 & [email protected] & 1 & 34.4 & 11.4 & 4.0 & sab(rs)c ngc 2336 & < 3.3 & 1 & 31.9 & < 1.01 & 5.4 & sab(r)bc ngc 2339 & [email protected] & 1 & 31.1 & 16.8 & 4.0 & sab(rs)bc ngc 2403 & < 7.0 & 1 & 2.2 & 4.91 & 6.0 & sab(s)cd ngc 2559 & < 6.5 & 1 & 17.3 & 32.1 & 9.0 & sb(s)bc pec ngc 2681 & [email protected] & 2 & 10.1 & 7.6 & 3.1 & sab(rs)0/a ngc 2683 & [email protected] & 1 & 3.2 & 5.07 & 0.4 & sa(rs)b ngc 2715 & [email protected] & 2 & 20.3 * * & - & 5.2 & sab(rs)c ngc 2805 & [email protected] & 5 & 26.2 * & - & 6.9 & sab(rs)d ngc 2820 & [email protected] & 2 & 22.9 * * & - & 5.4 & sb(s)c pec sp ngc 2841 & [email protected] & 2 & 9.3 & 1.46 & 3.0 & sa(r)b ngc 2903 & [email protected] & 6 & 6.2 & 24 & 4.0 & sb(s)d ngc 2964 & [email protected] & 2 & 16.8 & 11.4 & 4.1 & sab(r)bc ngc 2976 & [email protected] * * * & 1 & 2.3 & 2.07 & 5.3 & sac pec ngc 2985 & [email protected] & 2 & 19.1 & 9.55 & 2.3 & sa(rs)ab ngc 3034 & [email protected] & 1 & 2.2 & 289 & 8.0 & i0 ngc 3079 & [email protected] & 2 & 16.1 & 69.3 & 6.6 & sb(s)c ngc 3147 & [email protected] & 1 & 38.4 & 10.5 & 3.9 & sa(rs)bc ngc 3184 & 7.8@xmath331 & 6 & 7.9 & 5.8 & 5.0 & sab(rs)cd ngc 3187 & [email protected] & 2 & 20.0 * * & - & 5.9 & sb(s)c pec ngc 3198 & [email protected] & 2 & 9.3 * * & - & 5.2 & sb(rs)c ngc 3227 & [email protected] & 2 & 14.3 * * & - & 1.4 & sab(s ) pec ngc 3310 & [email protected] & 2 & 14.2 & 2.15 & 4.0 & sab(r)bc pec ngc 3344 & [email protected] & 2 & 6.9 & 4.69 & 4.0 & sab(r)bc ngc 3346 & [email protected] & 5 & 17.5 * & - & 6.0 & sb(rs)cd ngc 3351 & [email protected] & 2 & 9.0 & 14.6 & 3.0 & sb(r)b ngc 3359 & 3@xmath331 & 2 & 15.0 & 4.95 & 5.2 & sb(rs)c ngc 3368 & 35@xmath331 & 2 & 10.3 & 24.4 & 1.8 & sab(rs)ab ngc 3423 & [email protected] & 5 & 13.6 * & - & 6.0 & sa(s)cd ngc 3445 & [email protected] & 5 & 30.0 * & - & 8.9 & sab(s)m ngc 3486 & 7@xmath331 & 2 & 9.0 & 0.76 & 5.2 & sab(r)c ngc 3504 & [email protected] & 6 & 19.7 & 13.7 & 2.1 & sab(s)ab ngc 3521 & [email protected] & 6 & 8.5 & 22.7 & 4.0 & sab(rs)bc ngc 3556 & [email protected] & 1 & 10.3 & 14.4 & 6.0 & sb(s)cd ngc 3593 & [email protected] & 1 & 7.3 & 18.6 & -0.4 & sa(s)0/a ngc 3623 & [email protected] & 1 & 4.5 & < 1.90 & 1.0 & sab(rs)a ngc 3627 & 90@xmath334 & 2 & 4.5 & 34 & 3.0 & sab(rs)a ngc 3628 & [email protected] & 2 & 4.5 & 69.5 & 3.1 & sab pec sp ngc 3631 & [email protected] & 1 & 16.6 & 7.66 & 5.2 & sa(s)c ngc 3675 & [email protected] & 1 & 9.8 & 12.2 & 3.0 & sa(s)b ngc 3690 & [email protected] & 1 & 41.4 & 19.5 & 8.7 & ibm pec ngc 3782 & [email protected] & 5 & 12.6 * & - & 6.6 & sab(s)cd ngc 3810 & 24@xmath332 & 1,6 & 11.5 & 14.3 & 5.2 & sa(rs)c ngc 3893 & < 5.4 & 1 & 13.8 & 9.72 & 5.1 & sab(rs)c ngc 3906 & [email protected] & 5 & 15.6 * & - & 6.8 & sb(s)d ngc 3913 & [email protected] & 5 & 15.9 * & - & 6.6 & sa(rs)d ngc 3938 & [email protected] & 1 & 11.2 & 8.86 & 5.1 & sa(s)c ngc 4038 & [email protected] & 1 & 19.3 & 29.3 & 8.9 & sb(s)m pec ngc 4039 & [email protected] & 1 & 19.3 & 29.6 & 8.9 & sa(s)m pec ngc 4041 & [email protected] & 1 & 17.6 & 20.4 & 4.0 & sa(rs)bc ngc 4088 & [email protected] & 1 & 10.9 & 15.6 & 4.8 & sab(rs)bc ngc 4096 & [email protected] & 1 & 7.5 & 5.07 & 5.3 & sab(rs)c ngc 4102 & 115@xmath339 & 6 & 13.1 & 17.5 & 3.0 & sab(s)b ngc 4157 & [email protected] & 1 & 12.1 & 21.9 & 3.3 & sab(s)b sp ngc 4178 & < 3.4 & 1 & 16.1 & 1.64 & 7.1 & sb(rs)dm ngc 4204 & [email protected] & 5 & 12.9 * & - & 7.9 & sb(s)dm ngc 4212 & [email protected] & 1 & 16.1 & 7.46 & 4.9 & sac ngc 4254 & [email protected] & 6 & 16.1 & 19.3 & 5.2 & sa(s)c ngc 4274 & [email protected] & 2 & 11.9 * * & - & 1.7 & sb(r)ab ngc 4293 & [email protected] & 1 & 16.1 & 11.5 & 0.3 & sb(s)0/a ngc 4298 & [email protected] & 1 & 16.1 & 10.9 & 5.2 & sa(rs)c ngc 4299 & [email protected] & 5 & 16.1 & 1.11 & 8.3 & sab(s)dm ngc 4302 & [email protected] & 1 & 16.1 & 8.91 & 5.4 & sc sp ngc 4303 & [email protected] & 6 & 16.1 & 21.6 & 4.0 & sab(rs)bc ngc 4314 & [email protected] & 2 & 10.9 * * & - & 1.0 & sb(rs)a ngc 4321 & [email protected] & 1,2 & 16.1 & 27.3 & 4.0 & sab(s)bc ngc 4380 & [email protected] & 1 & 16.1 & < 0.1 & 2.3 & sa(rs)b ngc 4394 & [email protected] & 1 & 16.1 & 3.27 & 8.8 & sb(r)b ngc 4411 & [email protected] & 5 & 17.8 * & - & 5.4 & sb(rs)c ngc 4414 & [email protected] & 2,6 & 9.6 & 31.6 & 5.1 & sa(rs)c ngc 4416 & [email protected] & 5 & 19.3 * & - & 5.9 & sb(rs)cd ngc 4418 & [email protected] & 1 & 25.5 & 4.91 & 1.0 & sab(s)a ngc 4419 & [email protected] & 1 & 16.1 & 18.6 & 1.1 & sb(s)a ngc 4424 & [email protected] & 1 & 16.1 & 1.45 & 3.0 & sb(s)a ngc 4438 & 70@xmath337 & 4 & 16.1 & 5.27 & 0.7 & sa(s)0/a pec ngc 4487 & [email protected] & 5 & 13.6 * & - & 5.9 & sab(rs)cd ngc 4496a&[email protected] & 5 & 23.6 * & - & 7.5 & sb(rs)m ngc 4501 & [email protected] & 6 & 16.1 & 21.8 & 3.4 & sa(rs)b ngc 4527 & [email protected] & 1 & 16.1 & 28.7 & 4.0 & sab(s)bc ngc 4535 & [email protected] & 6 & 16.1 & 8.55 & 5.0 & sab(s)c ngc 4536 & [email protected] & 1 & 16.1 & 18.4 & 4.2 & sab(rs)bc ngc 4540 & [email protected] & 5 & 16.1 & 1.50 & 6.1 & sab(rs)cd ngc 4548 & [email protected] & 1 & 16.1 & 6.73 & 3.1 & sbb(rs ) ngc 4565 & [email protected] & 2 & 15.7 * * & - & 3.2 & sa(s)b sp ngc 4567 & [email protected] & 1 & 16.1 & 8.55 & 4.0 & sa(rs)bc ngc 4568 & [email protected] & 1 & 16.1 & 19.1 & 4.1 & sa(rs)bc ngc 4569 & [email protected] & 6 & 16.1 & 26.6 & 2.4 & sab(rs)ab ngc 4618 & [email protected] & 5 & 10.0 * & - & 8.6 & sb(rs)m ngc 4625 & [email protected] & 5 & 10.9 * & - & 8.8 & sab(rm)m pec ngc 4647 & [email protected] & 1 & 16.1 & 9.09 & 5.2 & sab(rs)c ngc 4651 & 6@xmath331 & 2 & 16.1 & 7.64 & 5.2 & sa(rs)c ngc 4654 & [email protected] & 2 & 16.1 & 6.0 & 5.9 & sab(rs)cd ngc 4666 & [email protected] & 1 & 18.6 & 30.2 & 5.0 & sabc ngc 4688 & [email protected] & 5 & 13.9 * & - & 6.0 & sb(s)cd ngc 4689 & [email protected] & 1 & 16.1 & 6.55 & 4.7 & sa(rs)bc ngc 4691 & [email protected] & 1 & 13.1 & 4.8 & 0.4 & sb(s)0/a pec ngc 4698 & < 3.0 & 1 & 16.1 & < 0.73 & 1.4 & sa(s)ab ngc 4701 & [email protected] & 5 & 10.3 * & - & 5.8 & sa(s)cd ngc 4710 & [email protected] & 1 & 16.1 & 7.82 & -0.8 & sa(r)0 sp ngc 4736 & [email protected] & 6 & 4.0 & 17.5 & 2.4 & sa(r)ab ngc 4775 & [email protected] & 5 & 20.9 * & - & 6.9 & sa(s)d ngc 4818 & [email protected] & 1 & 13.5 & 16.8 & 2.0 & sab(rs)ab pec ngc 4826 & [email protected] & 6 & 5.0 & 44.7 & 2.4 & sa(rs)ab ngc 4845 & < 3.1 & 1 & 16.1 & 6.07 & 2.3 & sa(s)ab sp ngc 4984 & [email protected] & 1 & 14.7 & 12.3 & -0.8 & sab(rs)0 ngc 5005 & [email protected] & 2 & 14.3 & 30.4 & 4.0 & sab(rs)bc ngc 5033 & 59@xmath334 & 2,6 & 12.8 & 13.1 & 5.2 & sa(s)c ngc 5112 & [email protected] & 2 & 13.4 * * & - & 5.8 & sb(rs)cd ngc 5236 & [email protected] & 1 & 5.9 & 88.7 & 5.0 & sab(s)c ngc 5247 & 31@xmath334 & 1 & 20.1 & 7.67 & 4.0 & sa(s)bc ngc 5248 & [email protected] & 6 & 14.7 & 19.5 & 4.1 & sb(rs)bc ngc 5364 & 5@xmath331 & 2 & 18.0 & 1.00 & 4.0 & sa(rs)bc pec ngc 5668 & [email protected] & 5 & 22.2 * & - & 6.0 & sa(s)d ngc 5669 & [email protected] & 5 & 19.8 * & - & 6.9 & sab(rs)cd ngc 5713 & [email protected] & 1 & 24.6 & 11.0 & 4.0 & sab(rs)bc pec ngc 5725 & [email protected] & 5 & 22.8 * & - & 7.0 & sb(s)d ngc 5861 & < 12 & 1 & 24.1 & 10.2 & 5.0 & sab(rs)c ngc 5907 & [email protected] & 2 & 10.4 & 7.49 & 5.4 & sa(s)c sp ngc 5921 & [email protected] & 2 & 19.0 * * & - & 4.0 & sb(r)bc ngc 5964 & [email protected] & 5 & 20.7 * & - & 6.9 & sb(rs)d ngc 6000 & [email protected] & 1 & 27.1 & 39.1 & 3.9 & sb(s)bc ngc 6015 & [email protected] & 2 & 13.6 * * & - & 6.0 & sa(s)cd ngc 6217 & [email protected] & 2 & 21.1 & 10.2 & 4.0 & sb(rs)bc ngc 6384 & [email protected] & 2 & 24 & 3.27 & 3.6 & sab(r)bc ngc 6503 & [email protected] & 6 & 4.2 & 3.78 & 5.9 & sa(s)cd ngc 6509 & [email protected] & 5 & 25.7 * & - & 6.6 & sbcd ngc 6643 & [email protected] & 6 & 23.1 & 13.8 & 5.2 & sa(rs)c ngc 6814 & [email protected] & 1 & 21.1 & 4.71 & 4.0 & sab(rs)bc ngc 6946 & [email protected] & 1,6 & 6.7 & 79.8 & 5.9 & sab(rs)cd ngc 6951 & [email protected] & 1,6 & 21.7 & 15.2 & 3.9 & sab(rs)bc ngc 7217 & [email protected] & 2,6 & 16.3 & 4.13 & 2.5 & sa(r)ab ngc 7331 & [email protected] & 2,6 & 14.7 & 6.98 & 3.9 & sa(s)b ngc 7479 & [email protected] & 1 & 34.7 & 12.9 & 4.4 & sb(s)c ngc 7541 & [email protected] & 1 & 38.1 & 14.8 & 4.7 & sb(rs)bc pec ngc 7640 & [email protected] & 2 & 8.5 & 1.33 & 5.3 & sb(s)c ngc 7741 & [email protected] & 2,5 & 11.7 * & - & 6.0 & sb(s)cd ic 342 & [email protected] & 1,6 & 3.0 & 65.5 & 5.9 & sab(rs)cd ugc 3574 & [email protected] & 5 & 21.9 * & - & 5.9 & sa(s)cd ugc 8516 & [email protected] & 5 & 15.4 * & - & 5.9 & scd mcg1 - 3 - 85 & [email protected] & 5 & 13.6 * & - & 7.0 & sab(rs)d although the beamwidth of the iram 30 m telescope at 22@xmath1 is similar to that of the nro 45 m telescope at 16@xmath1 , it is important that co detected at both telescopes can be compared on a same scale . in order to confirm this point , figure [ comparison ] plots the derived integrated intensity @xmath34 for galaxies which overlapped between observations at nro 45 m and iram 30 m telescope . a clear proportionality with a slope of order unity can be seen , despite the small number of samples . the zero point of this relation is effectively 0 , despite the difference in beam size , enabling a direct comparison of observed intensity . three galaxies deviate from the relation in [ comparison ] by more than a factor of two ( ngc 3810 , ngc 5033 , and ngc 7331 ) . all of these galaxies are from @xcite and @xcite for the nro 45 m and iram 30 m measurements , respectively . ngc 3810 has a starbursting nuclei , and its @xmath34 at a single beam away from its nuclei drops to nearly half @xcite . the observing coordinate in @xcite and @xcite are offset by 4@xmath1 , possibly explaining the excess of @xmath34 in the nro 45 m measurement . the @xmath34 for ngc 5033 also drops by more than 50% @xmath35 from the center in @xcite , making the co intensity consistent with that of @xcite . the offset in observing coordinate ( @xmath36 ) coupled with pointing errors may account for the discrepancy . the observing coordinate for ngc 7331 differ by @xmath37 . since the coordinate in @xcite is offset from the true center , the difference in @xmath34 may reflect the central co hole which explains the smaller co intensity in @xcite . nevertheless , it is important to bear in mind that when comparing co intensities from the different telescopes , an error of typically a factor 2 may infact be involved . a secondary check can be implemented by using the 45@xmath1 resolution data from the fcrao data from young et al . the ratio of co intensity at nro 45 m or iram 30 m telescope to the fcrao intensity expresses a measure of the central condensation of gas ; if the angular resolution difference of the nro 45 m and iram 30 m telescope has a significant effect on the derived intensity , this ratio should display a systematic difference between the two telescopes . figure [ condensation ] plots the co intensity observed at the fcrao @xmath38 versus the intensity observed either at nro 45 m or iram 30 m telescope , including those only with more than a 3 @xmath25 detection . a clear correlation can be seen , and a least squares fit to the co intensity from nro45 m telescope @xmath39 gives @xmath40 where as the same fit to iram30 m telescope will give @xmath41 both fits are consistent within the range of error , and we conclude that the data from these two telescopes with different beamsizes can be compared on a same scale without correction . for galaxies which were observed with both the iram 30 m and nro 45 m telescope . the galaxies are ngc 278 , ngc 628 , ngc 3810 , ngc 4321 , ngc 4414 , ngc 5033 , ngc 7217 , ngc 7331 . the line expresses @xmath42.,width=566,height=566 ] resolution , and the ordinate is same for telescopes iram 30 m ( open circles ) and nro 45 m ( filled circles ) with 22@xmath1 and 16@xmath1 resolution , respectively . the dashed line represents a least squares fit to the nro 45 m sample , and the solid line iram 30 m . the dash - dotted line represents where the ratio of absissa and ordinate becomes unity.,width=566,height=566 ] figure [ distancedistribution ] shows the distance distribution of our compiled samples from the nro 45 m and iram 30 m dishes . the samples are strongly peaked at around 16 mpc , where the virgo cluster members lie . the distances here are taken from young et al . ( 1995 ) scaled to @xmath43 except for virgo cluster members , where a more recent measurement ( @xcite ) gives 16.1 mpc using cepheid variables . the observing beam of 16@xmath1 and 22@xmath1 for 16 mpc correspond to a linear scale of 1.2 kpc and 1.8 kpc , respectively . these both correspond to radii within 1 kpc of the center , and considering that most of our samples lie around this distance , we will simply refer hereafter to these data as gas in the `` central kpc '' and its corresponding co intensity @xmath39 or @xmath44 as @xmath45 . co measured in young et al . ( 1995 ) and @xcite with the 45@xmath1 beamsize will be referred to as gas in the `` central 3 kpc '' , denoted @xmath46 . the reader should refer to table [ tbl-1 ] for the actual linear scale of the measurement for individual galaxies . figure [ morph_distri ] shows the distribution of morphology for all sample galaxies detected in co at the central kpc . the hubble types are taken from the hyperleda database ( paturel et al . most galaxies lie on hubble type 4 - 5 , corresponding to sbc - sc on de vaucouleur s scale . no apparent biases are seen in regard to the fraction of barred / non - barred galaxies in certain hubble type , although we have significantly less sb galaxies compared to sa or sab galaxies in total . a weak bias may exist that more barred galaxies are observed in the latest types , but may also be representative of the general galaxy population ( e.g. , @xcite ) . with detected co in the central kpc.,width=377,height=302 ] the inclination ( @xmath47 ) of the target galaxies result in a variation in the projected area of the observing beam . this is especially important in our case because we observed the galaxies with only one beam . it is common to multiply the observed co intensity by @xmath48 to correct for this effect . figure [ inclination ] shows the variation of co intensity on @xmath47 . the observed co intensity depends very weakly on @xmath47 . if the true co distribution within the observing beam is disk - like , requiring a correction for inclination , the observed co intensity must increase as the inclination becomes edge - on ; the effective observing area of the subtended beam becomes larger , scaling as @xmath49 . this will result in a slope of @xmath50 in the @xmath51 vs. @xmath48 figure . a least squares fit to figure [ inclination ] results in a slope of @xmath52 . although a variation of co intensity on inclination does exist , the intrinsic dispersion in the co intensities are far larger . this can be explained if the effect is of geometrical quality , where @xmath47 in the literature is physically not well - defined in the central regions , owing to thickness of the disk . the molecular gas distribution in the 16@xmath1 beam may not be represented well by a disk structure , in which case a correction by factor @xmath48 will be artificial . the view is also supported by the fact that the trend in figure [ inclination ] is same at different wavelengths ( @xcite ) , which can not be explained if the bias results from issues with optical depth . ( 0 being face - on ) . the line is a least squares fit with a slope of @xmath53 . galaxies those which do not have normal morphology ( peculiars , irregulars , and the like ) have not been included . , width=377,height=377 ] in the succeeding sections , we either use the co intensity alone ( without correcting for inclination ) , or use intensity ratios in different beamsizes to cancel any effects from inclinations . figure [ bars_iratio_all ] shows the number distribution of the ratio of molecular gas surface mass density in the central 1 kpc to the central 3 kpc , or the central concentration , normalized to sample size . the central concentration of gas in both barred and non - barred galaxies peak at about 1 to 2 , but non - barred galaxies quickly drop off above 5 . barred galaxies , however , seem to show a tail of high central molecular concentration up to 9 . above 5 , barred galaxies dominate the distribution . barred galaxies from @xcite also show a higher central concentration , and figure [ bars_iratio_all ] confirms this using a large number of samples . the probability that the distribution of barred galaxies and that of non - barred galaxies being derived from a same parent population ( the kolmogorov - smirnov ( ks ) test ) is @xmath54 . this indicates that the barred samples can not be said to be significantly different from non - barred samples . to check for any artificial effects by using samples at different distances ( therefore probing different regions ) , figure [ bars_iratio_virgo ] shows the same figure but only for samples within the distance range of 13 mpc to 19 mpc . the trend of barred galaxies having a high central concentration tail is still seen , but with @xmath55 ( see discussion for why the barred samples may not be observed to have higher central concentration as in previous studies ) . it is interesting that the central concentration in barred galaxies peak at around the same value as non - barred galaxies , but the median of the central concentration is higher for the barred galaxies . this can not be explained by the classical picture of bars simply helping the inflow of molecular gas towards the center , because then we should expect barred galaxies to have a peak at a higher degree of central concentration . figure [ gas_type ] shows the co intensity along the hubble types , for different beamsizes . molecular gas is significantly more massive in the central 1 kpc of early type galaxies ( upper figure ) , where the large stellar bulge can be assumed to create a deep potential well . on the other hand , the lower figure of figure [ gas_type ] shows co intensity observed with the 45@xmath1 beam , or the central 3 kpc . it is apparent that molecular gas is not as well correlated with hubble type than the central 1 kpc . the typical detection threshold in the fcrao sample is @xmath56 ( @xmath57 on the logscale ) , shown in figure [ gas_type ] with the horizontal dashed line . many of the co intensities in the late type galaxies are close to this threshold , but a weak correlation with hubble type is still observed . this indicates that the trend in figure [ gas_type](top ) is not created by the overall low molecular gas mass in the central regions of late type galaxies , but that the gas distribution is more extended in late type galaxies . figure [ gas_concentration ] shows the ratio of co intensity observed with different beamsizes , and we see that the distribution of molecular gas is a smooth function of hubble type , in that late type galaxies have more extended gas in the center . for the earliest galaxies ( hubble type @xmath58 , corresponding to s0/s0a ) , the intensity ratio @xmath59 is close to 1 , where virtually all the gas is confined to the central 1 kpc . for the latest types , however , @xmath59 falls to 0 indicating that there is no apparent central concentration of gas comparing the central 1 kpc and 3 kpc , so that the molecular gas disk is likely extended beyond the central 3 kpc . the horizontal dashed line in figure [ gas_concentration ] is the threshold below which non - detections in the fcrao sample increase , assuming a typical @xmath60 for the latest types and @xmath61 for the fcrao sample . in case a bias exists in the fcrao sample that weak latest type galaxies are not observed , the corresponding galaxies will be plotted above this line . the absence of early type galaxies with low central concentration , however , still points to the reality of this trend of lower central concetration towards later type galaxies . the previous sections have revealed that , hubble type can play an important role in centrally concentrating molecular gas . bars are also known to centrally concentrate gas , although not found to be statistically significant based on our current data ( see the discussion section for possible reasons ) . the main setbacks of previous studies is that they have not been able to solve the degeneracy between the effect of hubble type and bars because of small sample size . it is of interest , therefore , which ( bars or hubble type ) play the dominant role in concentrating molecular gas . our sample enables the categorization and comparison of different hubble types and presence of bars , simultaneously . figure [ ico_dist_earlylate ] shows the co intensity ( molecular gas mass within central 1 kpc ) for early and late type galaxies , respectively , each for barred and non - barred galaxies . the bottom figure shows the same figure as the top and the middle , but for barred and non - barred galaxies combined . the samples where categorized into early type ( type -1 to 4.5 ) and late type ( 4.5 to 9 ) somewhat arbitrary , so that both samples had significant numbers of galaxies . if bars concentrate gas into the central region regardless of hubble type , we can expect @xmath34 to be higher in barred galaxies to non - barred galaxies , both in early and late types . infact , figure 11 ( top ) seems to suggest that barred early type galaxies have a top - heavy molecular mass distribution compared to non - barred early types with a slightly higher mean co intensity . this trend is not statistically siginificant , however , with @xmath62 . for barred and non - barred late type galaxies , @xmath63 is @xmath64 . this is not siginificant either , and it is clear from figure 12 ( middle ) that the apparent difference in the two distribution , results from a wider distribution of co intensity in barred late type galaxies . four barred late type galaxies have high molecular gas mass ( @xmath65 ) . all of these are known starburst galaxies ( ngc 253 , ic 342 , ngc 3079 and ngc6946 ) . except for these four , barred galaxies in late types do not show a high gas density distribution ; barred galaxies in late types apparently do not have the high gas density expected from gas angular momentum redistribution . in cases where the gas is concentrated , they have extremely concentrated gas and an associated starburst . this leads us to speculate that these high central concentration late type starbursts , may either be on the verge of evolving to early type galaxies if the starburst is strong enough ( in the context of secular evolution ) , or just using up the molecular fuel to move its place in figure 12 ( middle ) to the most less massive barred late types , which also shows a distinctive tail from non - barred late types . if the presence of a bulge and its stellar potential drives molecular gas into the central region regardless of whether they have bars , we can expect early type non - barred galaxies to have higher @xmath45 than late type non - barred galaxies . infact , these two are likely to be significantly different , with @xmath66 . the average co intensity in early non - barred galaxies ( @xmath67 ) is more than two times larger than late non - barred galaxies ( @xmath68 ) . the bulge potential seems very effective in concentrating molecular gas into the central 1 kpc . barred early type galaxies also siginificantly differ from barred late types , with @xmath69 . in the same way , with the barred and non - barred galaxies combined , early type galaxies and late type galaxies are , again , significantly different ( @xmath70 ) . from these analysis , it seems that the bulge potential ( hubble type ) is the more effective parameter in changing @xmath34 in the central 1 kpc , than bars . the possibility of early type galaxies showing more massive molecular gas in the central 1 kpc only because of stronger bars than late types , is rejected . it is also important , however , to normalize the co intensity in the central 1 kpc with co data in a larger area to more accurately address the central concentration of molecular gas . figure [ conc_dist_earlylate ] shows the ratio of molecular mass in the central 1 kpc to the central 3 kpc , for early and late type galaxies . the bottom figure is the same as the top and middle , but with barred and non - barred galaxies combined . the high central concentration tail ( see section 8) in the distribution can be seen for early type galaxies , but not in late types ( meaning , that early type galaxies constitute the high concentration tail in figure [ bars_iratio_all ] ) . the ks test shows that the probability of barred galaxies and non - barred galaxies for the early type group derived from the same parent group is @xmath71 . molecular gas in barred early galaxies are not significantly more centrally concentrated than non - barred early types , but the tail of higher concentration in barred galaxies can still be seen , the median concentration being 1.6 times higher in barred early types . the central concentration in late type galaxies in figure 12 ( middle ) seem much more insensitive to the presence of bars . the tail of higher concentration that was seen in early types are not seen . the only exception of a highly centrally concentrated barred late type galaxy is ngc 3486 , with practically all of the gas within its central 1 kpc . this sabc galaxy is only weakly barred , with a bar strength ( ratio of maximum tangential force and average radial force ) of 0.2 ( @xcite ) . the central concentration of molecular gas in this galaxy is extraordinary in terms of bar and hubble type , within the context of this paper . the ks - test for barred and non - barred galaxies in the late type group show that the two morphologies can be derived from a same distribution by @xmath72 probability . comparing non - barred early types and non - barred late types give @xmath73 . similarly , barred early types and barred late types give @xmath74 . from the bottom figure of [ conc_dist_earlylate ] , we see that almost all early type galaxies have a ratio @xmath75 of more than 1 , indicating that the central 1 kpc is almost always more dense than its surrounding . late type galaxies , on the other hand , actually peak at less than 1 , meaning that many late type galaxies are less dense in the central kpc compared to its surrounding disk . early type galaxies are more likely to be centrally concentrated than late types ( @xmath76 ) . central concentration of molecular gas from bars in early type ( sbc and earlier ) galaxies had already been found by sheth et al . ( 2005 ) , where early barred galaxies showed concentration enhancement of a factor 4 compared to early non - barred galaxies . the median central concentration for our sample of early type bars ( @xmath774.1 ) is also higher than early type non - bars ( @xmath772.6 ) , although not by a factor of 2 . our definition of central concentration @xmath75 employs the molecular mass in the central 45@xmath1 for normalization whereas the @xcite sample uses the global molecular mass . since the central 45@xmath1 ( @xmath78 kpc ) would likely not contain all of the galaxy s molecular gas , any trends in the central concentration would be mitigated compared with normalizations using the whole molecular content . kuno et al . ( 2007 ) , similarly , used co maps to define the central concentration of molecular gas in a sample of 40 galaxies . they found barred galaxies are significantly ( @xmath79 ) more centrally concentrated than non - barred galaxies . therefore , our result that barred galaxies are not significantly more concentrated ( @xmath55 ) , is likely to be a result of using the central 45@xmath1 for normalization rather than the central concentration being not as strong . errors in the measurements , which can change the derived co intensity by factor two when comparing different telescopes , can also reduce the power of comparison with the fcrao measurements . these sources of error apply to the whole sample , however , and the observed result that hubble type affects the central concentration more effectively than bars , will likely not be changed . ( or 22@xmath1 ) and 45@xmath1 data . numbers in the parenthesis are the sample size . note the tail of high concentration for barred galaxies . one non - barred s0 galaxy , ngc 4438 , shows @xmath80 , which is more than the formal upper limit determined by the area of the two observing beams . this can be attributed to the uncertainties of the observations.,width=566,height=377 ] , but for a subsample of galaxies within the distance range of 13 mpc to 19 mpc . most are members of the virgo cluster . barred galaxies dominate the central concentration above 5.,width=566,height=377 ] for early ( top ) and late ( middle ) type galaxies . the high central concentration tail seen on the top figure is not seen on the middle . the bottom figure shows the histogram for the early and late types , without regard to the presence of bars.,title="fig:",width=377,height=264 ] for early ( top ) and late ( middle ) type galaxies . the high central concentration tail seen on the top figure is not seen on the middle . the bottom figure shows the histogram for the early and late types , without regard to the presence of bars.,title="fig:",width=377,height=264 ] for early ( top ) and late ( middle ) type galaxies . the high central concentration tail seen on the top figure is not seen on the middle . the bottom figure shows the histogram for the early and late types , without regard to the presence of bars.,title="fig:",width=377,height=264 ] this paper outlined a co survey of nearby galactic centers conducted with the nro 45 m telescope . combined with published data at similar and lower resolution , we are able to infer on the molecular gas distribution in barred and unbarred galaxies of various hubble types . the number of galaxies used ( @xmath81 ) is far more than that used in previous interferometric and single dish mapping studies . the main results of this study are ; 1 . molecular gas is denser in non - barred early type galaxies compared to non - barred late type galaxies . therefore , possible stronger bars in early types can not be the dominant reason for massive molecular gas observed in early types . central concentration of molecular gas in early type galaxies is confirmed . the ratio of molecular gas in the central kiloparsec to the central 3 kiloparsecs declines smoothly with hubble type . 3 . barred early type galaxies have a higher median of central concentration than non - barred early type galaxies , suggesting a tail distribution of high central concentration in barred galaxies , although not statistically significant based on the current data . this trend is not observed in late type galaxies . the authors thank the anonymous referee for valuable comments and careful reading of the manuscript . the authors also wish to thank the nro staff for generous allocation of telescope time . , f.e . , and k.m . were financially supported by a research fellowship from the japan society for the promotion of science for young scientists . we acknowledge the usage of the hyperleda database ( http://leda.univ-lyon1.fr ) , and the nasa / ipac extragalactic database(ned ) , which is operated by the jet propulsion laboratory , caltech , under contract with the national aeronautics and space administration .

for certain orientations of josephson junctions between two-wave or two-wave superconductors , the subgap andreev bound states produce a-periodic relation between the josephson current and the phase difference : . consequently , the ac josephson current has the fractional frequency , where is the dc voltage . in the tunneling limit , the josephson current is proportional to the first power ( not square ) of the electron tunneling amplitude . thus , the josephson current between unconventional superconductors is carried by single electrons , rather than by cooper pairs . the fractional ac josephson effect can be observed experimentally by measuring frequency spectrum of microwave radiation from the junction . we also study junctions between singlet-wave and triplet-wave , as well as between chiral-wave superconductors .

cond-mat0210148

in many materials , the symmetry of the superconducting order parameter is unconventional , i.e. not @xmath8-wave . in the high-@xmath10 cuprates , it is the singlet @xmath11-wave @xcite . there is experimental evidence that , in the quasi - one - dimensional ( q1d ) organic superconductors @xmath12 @xcite , the symmetry is triplet @xcite , most likely the @xmath0-wave @xcite , with the @xmath13 axis along the conducting chains . experiments indicate that @xmath14 has the triplet chiral @xmath9-wave pairing symmetry @xcite . the unconventional pairing symmetry typically results in formation of subgap andreev bound states @xcite on the surfaces of these superconductors @xcite . for @xmath1-wave cuprate superconductors , the midgap andreev states were predicted theoretically in ref . @xcite and observed experimentally as a zero - bias conductance peak in tunneling between normal metals and superconductors ( see review @xcite ) . for the q1d organic superconductors , the midgap states were theoretically predicted to exist at the edges perpendicular to the chains @xcite . in the chiral superconductor @xmath14 , the subgap surface states are expected to have a chiral energy dispersion @xcite . their contribution to tunneling is more complicated @xcite than a simple zero - bias conductance peak found for the midgap andreev states . various ways of observing electron edge states experimentally are discussed in ref . @xcite . when two unconventional superconductors are joined together in a josephson junction , their andreev surface states hybridize to form andreev bound states in the junction . these states play an important role in the josephson current through the junction @xcite . andreev bound states in high-@xmath10 junctions were reviewed in ref . the josephson effect between two q1d @xmath0-wave superconductors was studied in refs . andreev reflection @xcite at the interfaces between the a and b phases of superfluid @xmath15he was studied in ref . however , andreev bound states were not mentioned in this paper . in the present paper , we predict a new effect for josephson junctions between unconventional nonchiral superconductors , which we call the fractional ac josephson effect . suppose both superconductors forming a josephson junction have surface midgap states originally . this is the case for the butt - to - butt junction between two @xmath0-wave q1d superconductors , as shown in fig . [ fig : jj](a ) , and for the @xmath16 in - plane junction between two @xmath1-wave superconductors , as shown in fig . [ fig : d - wave](a ) . ( the two angles indicate the orientation of the junction line relative to the @xmath17 axes of each @xmath11 superconductor . ) we predict that the contribution of the hybridized andreev bound states produces a @xmath2-periodic relation between the supercurrent @xmath3 and the superconducting phase difference @xmath4 : @xmath5 @xcite . consequently , the ac josephson effect has the frequency @xmath6 , where @xmath18 is the electron charge , @xmath7 is the applied dc voltage , and @xmath19 is the planck constant . the predicted frequency is a half of the conventional josephson frequency @xmath20 originating from the conventional josephson relation @xmath21 with the period of @xmath22 . qualitatively , the predicted effect can be interpreted as follows . the josephson current across the two unconventional superconductors is carried by tunneling of _ single electrons _ ( rather than cooper pairs ) between the two resonant midgap states . thus , the cooper pair charge @xmath23 is replaced the single charge @xmath18 in the expression for the josephson frequency . this interpretation is also supported by the finding that , in the tunneling limit , the josephson current is proportional to the first power ( not square ) of the electron tunneling amplitude @xcite . possibilities for experimental observation of the fractional ac josephson effect are discussed in sec . [ sec : experiment ] . a summary of this work is published in the conference proceedings @xcite . the predicted current - phase relation @xmath5 is quite radical , because every textbook on superconductivity says that the josephson current must be @xmath22-periodic in the superconducting phase difference @xmath4 @xcite . to our knowledge , the only paper that discussed the @xmath2-periodic josephson effect is ref . @xcite by kitaev . he considered a highly idealized model of spinless fermions on a one - dimensional ( 1d ) lattice with superconducting pairing on the neighboring sites . the pairing potential in this case has the @xmath0-wave symmetry , and midgap states exist at the ends of the chain . they are described by the majorana fermions , which kitaev proposed to use for nonvolatile memory in quantum computing . he found that , when two such superconductors are brought in contact , the system is @xmath2-periodic in the phase difference between the superconductors . our results are in agreement with his work . however , we formulate the problem as an experimentally realistic josephson effect between known superconducting materials . for completeness , we also calculate the spectrum of andreev bound states and the josephson current between a singlet @xmath8-wave and a triplet @xmath24-wave superconductors , as well as between two chiral @xmath24-wave superconductors @xcite . in agreement with previous literature @xcite , we find that a josephson current is permitted between singlet and triplet superconductors , contrary to a common misconception that it is forbidden by the symmetry difference . however , we do not find the fractional josephson effect in these cases . the spin symmetry of the cooper pairing is classified as either singlet @xmath25 or triplet @xmath26 @xcite . here @xmath27 is the annihilation operator of an electron with the spin @xmath28 and momentum @xmath29 ; @xmath30 is the antisymmetric metric tensor and @xmath31 are the pauli matrices acting in the spin space ; @xmath32 is a unit vector characterizing polarization of the triplet state . in this paper , we consider only the class of triplet superconductors where the spin - polarization vector @xmath32 has a uniform , momentum - independent orientation . everywhere in the paper , except in sec . [ sec : nonparallel ] , we select the spin quantization axis @xmath33 along the vector @xmath32 . then the cooper pairing takes place between electrons with the opposite @xmath33-axis spin projections @xmath28 and @xmath34 : @xmath35 . because the fermion operators @xmath36 anticommute , the pairing potential has the symmetry @xmath37 , where the upper and lower signs correspond to the singlet and triplet cases . we select the coordinate axis @xmath13 perpendicular to the josephson junction plane . we assume that the interface between the two superconductors is smooth enough , so that the electron momentum component @xmath38 , parallel to the junction plane , is a conserved good quantum number . electron states in a superconductor are described by the bogoliubov operators @xmath39 , which are related to the electron operators @xmath36 by the following equations @xcite @xmath40 , \label{gamma_n } \\ & & \hat c_{\sigma k_y}(x ) = \sum_{n } [ u_{n\sigma k_y}(x ) \ , \hat\gamma_{n\sigma k_y } + v_{n\bar\sigma\bar k_y}^*(x ) \ , \hat\gamma_{n\bar\sigma\bar k_y}^\dag ] , \label{canon}\end{aligned}\ ] ] where @xmath41 , and @xmath42 is the quantum number of the bogoliubov eigenstates . the two - components vectors @xmath43 $ ] are the eigenstates of the bogoliubov - de gennes ( bdg ) equation with the eigenenergies @xmath44 @xmath45 where @xmath46 is the @xmath13 component of the electron momentum operator , and @xmath47 is a potential . in eq . ( [ eq : bdg ] ) and below , we often omit the indices @xmath28 and @xmath38 to shorten notation where it does not cause confusion . in this section , we consider junctions between two q1d superconductors , such as organic superconductors @xmath12 , with the chains along the @xmath13 axis , as shown in fig . [ fig : jj](a ) . for a q1d conductor , the electron energy dispersion in eq . ( [ eq : bdg ] ) can be written as @xmath48 , where @xmath49 is an effective mass , @xmath50 is the chemical potential , @xmath51 and @xmath52 are the distance and the tunneling amplitude between the chains . the superconducting pairing potentials in the @xmath8- and @xmath0-wave cases have the forms @xmath53 where @xmath54 is the fermi momentum , and @xmath28 is treated as @xmath55 for @xmath56 and @xmath57 for @xmath58 . the index @xmath59 labels the right ( @xmath60 ) and left ( @xmath61 ) sides of the junction , and @xmath62 acquires a phase difference @xmath4 across the junction : @xmath63 the potential @xmath64 in eq . ( [ eq : bdg ] ) represents the junction barrier located at @xmath65 . integrating eq . ( [ eq : bdg ] ) over @xmath13 from 0 to + 0 , we find the boundary conditions at @xmath65 : @xmath66 where @xmath67 is the transmission coefficient of the barrier . a general solution of eq . ( [ eq : bdg ] ) is a superposition of the terms with the momenta close to @xmath68 , where the index @xmath69 labels the right- and left - moving electrons : @xmath70 , \label{l0s}\ ] ] where @xmath71 for @xmath72 and @xmath73 . ( [ l0s ] ) describes a subgap state with an energy @xmath74 , which is localized at the junction and decays exponentially in @xmath13 within the length @xmath75 . the coefficients @xmath76 in eq . ( [ l0s ] ) are determined by substituting the right- and left - moving terms separately into eq . ( [ eq : bdg ] ) for @xmath77 , where @xmath78 . in the limit @xmath79 , we find @xmath80 where @xmath81 is the fermi velocity , and @xmath82 with @xmath62 given by eq . ( [ gapfn ] ) . the @xmath38-dependent fermi momentum @xmath83 in eq . ( [ l0s ] ) eliminates the dispersion in @xmath38 from the bdg equation . substituting eq . ( [ l0s ] ) into the boundary conditions ( [ bdy ] ) , we obtain four linear homogeneous equations for the coefficients @xmath84 and @xmath85 . these equations are compatible if the determinant of the corresponding @xmath86 matrix is zero . this compatibility condition has the following form : @xmath87 using the variables @xmath88 defined in eq . ( [ kappa - eta ] ) , eq . ( [ self - uv ] ) can be written in a simpler form @xmath89 substituting eq . ( [ kappa - eta ] ) into eq . ( [ self ] ) , we obtain an equation for the energies of the andreev bound states . for a given @xmath28 , there are two subgap states with the energies @xmath90 labeled by the index @xmath91 , where @xmath92 -wave superconductors . ( b ) the energies ( left panel ) and the currents ( right panel ) of the andreev subgap states in the @xmath8-@xmath8 junction as functions of the phase difference @xmath4 for @xmath93 ( thin lines ) and @xmath94 ( thick lines ) . ( c ) the same as ( b ) for the @xmath0-@xmath0 junction at @xmath95 . ] the energies ( [ e_s ] ) and ( [ e_p ] ) are plotted as functions of @xmath4 in the left panels ( b ) and ( c ) of fig . [ fig : jj ] . without barrier ( @xmath93 ) , the spectra of the @xmath8-@xmath8 and @xmath0-@xmath0 junctions are the same and consist of two crossing curves @xmath96 , shown by the thin lines in the left panel of fig . [ fig : jj](b ) . a nonzero barrier ( @xmath97 ) affects the energies of the andreev bound states in the @xmath8-@xmath8 and @xmath0-@xmath0 junctions in different ways . in the @xmath8-@xmath8 case , the two energy levels repel near @xmath98 and form two separated @xmath22-periodic branches shown by the thick lines in the left panel of fig . [ fig : jj](b ) . this is well known for the @xmath8-@xmath8 junctions @xcite . in contrast , in the @xmath0-@xmath0 case , the two energy levels continue to cross at @xmath98 , and they detach from the continuum of states above @xmath99 and below @xmath100 at @xmath101 and @xmath22 , as shown in the left panel of fig . [ fig : jj](c ) . the absence of energy levels repulsion at @xmath98 indicates that there is no matrix element between these levels in the @xmath0-@xmath0 case , unlike in the @xmath8-@xmath8 case . as shown in sec . [ sec : d - wave ] , the @xmath16 junction between two @xmath1-wave superconductors is mathematically equivalent to the @xmath0-@xmath0 junction . ( [ e_p ] ) was derived for the @xmath16 junction in refs . @xcite . it is well known @xcite that the current carried by a quasiparticle state @xmath102 is @xmath103 the two subgap states carry opposite currents , which are plotted vs. @xmath4 in the right panels ( b ) and ( c ) of fig . [ fig : jj ] for the @xmath8-@xmath8 and @xmath0-@xmath0 junctions . in thermodynamic equilibrium , the total current is determined by the fermi occupation numbers @xmath104 of the states at a temperature @xmath105 : @xmath106 for the @xmath8-@xmath8 junction , substituting eq . ( [ e_s ] ) into eq . ( [ thermal ] ) , we recover the ambegaokar - baratoff formula @xcite in the tunneling limit @xmath107 @xmath108 and the kulik - omelyanchuk formula @xcite in the transparent limit @xmath109 @xmath110 taking into account that the total current is proportional to the number @xmath111 of conducting channels in the junction ( e.g. the number of chains ) , we have replaced the transmission coefficient @xmath67 in eq . ( [ i_s ] ) by the junction resistance @xmath112 in the normal state . substituting eq . ( [ e_p ] ) into eq . ( [ thermal ] ) , we find the josephson current in the @xmath0-@xmath0 junction in thermodynamic equilibrium : @xmath113 the temperature dependences of the critical currents for the @xmath8-@xmath8 and @xmath0-@xmath0 junctions are shown in fig . [ fig : ic ] . they are obtained from eqs . ( [ i_s ] ) and ( [ i_p ] ) assuming the bcs temperature dependence for @xmath114 . in the vicinity of @xmath10 , @xmath115 and @xmath116 have the same behavior . with the decrease of temperature , @xmath116 quickly saturates to a constant value , because , for @xmath107 , @xmath117 ( [ e_s ] ) , thus , for @xmath118 , the upper subgap state is empty and the lower one is completely filled . in contrast , @xmath115 rapidly increases with decreasing temperature as @xmath119 and saturates to a value enhanced by the factor @xmath120 relative to the ambegaokar - baratoff formula ( [ e_s ] ) at @xmath121 . this is a consequence of two effects . as eqs . ( [ i_s ] ) and ( [ i_p ] ) show , @xmath122 and @xmath123 , thus @xmath124 in the tunneling limit @xmath107 . at the same time , the energy splitting between the two subgap states in the @xmath0-@xmath0 junction is small compared to the gap : @xmath125 . thus , for @xmath126 , the two subgap states are almost equally populated , so the critical current has the @xmath119 temperature dependence analogous to the curie spin susceptibility . -@xmath8 ( dashed line ) and @xmath0-@xmath0 ( solid line ) josephson junctions as functions of temperature for @xmath127 . ] eq . ( [ i_p ] ) was derived analytically for the @xmath16 junction between two @xmath1-wave superconductors in refs . @xcite , and a similar result was calculated numerically for the @xmath0-@xmath0 junction in ref . notice that eq . ( [ i_p ] ) gives the josephson current @xmath128 that is a @xmath22-periodic functions of @xmath4 , both for @xmath121 and @xmath129 . this is a consequence of the thermodynamic equilibrium assumption . at @xmath121 , this assumption implies that the subgap state with the lower energy is occupied , and the one with the higher energy is empty . as one can see in fig . [ fig : jj ] , the _ lower _ energy is always a @xmath22-periodic functions of @xmath4 . the assumption of thermodynamic equilibrium was explicitly made in ref . @xcite and was implicitly invoked in refs . @xcite by using the matsubara diagram technique . in ref . @xcite , temperature dependence of the josephson critical current was measured in the ybco ramp - edge junctions with different crystal angles and was found to be qualitatively consistent with the upper curve in fig . [ fig : ic ] . the calculations of the previous section apply in the static case , where a given phase difference @xmath4 is maintained for an infinitely long time , so the occupation numbers of the subgap states have enough time to relax to thermodynamic equilibrium . now let us consider the opposite , dynamical limit . suppose a small voltage @xmath130 is applied to the junction , so the phase difference acquires dependence on time @xmath131 : @xmath132 . in this case , the state of the system is determined dynamically starting from the initial conditions . let us consider the @xmath0-@xmath0 junction at @xmath121 in the initial state @xmath101 , where the two subgap states ( [ e_p ] ) with the energies @xmath133 are , correspondingly , occupied and empty . if @xmath134 changes sufficiently slowly ( adiabatically ) , the occupation numbers of the subgap states do not change . in other words , the states shown by the solid and dotted lines in fig . [ fig : jj](c ) remains , correspondingly , occupied and empty . the occupied state ( [ e_p ] ) produces the current ( [ eq : i_a ] ) : @xmath135 the frequency of the ac current ( [ i_t ] ) is @xmath6 , a half of the conventional josephson frequency @xmath20 . the fractional frequency can be traced to the fact that the energies eq . ( [ e_p ] ) and the corresponding wave functions have the period @xmath2 in @xmath4 , rather than conventional @xmath22 . although at @xmath136 the spectrum in the left panel of fig . [ fig : jj](c ) is the same as at @xmath101 , the occupation numbers are different : the lower state is empty and the upper state is occupied . only at @xmath137 the occupation numbers are the same as at @xmath101 . the @xmath2 periodicity is the consequence of the energy levels crossing at @xmath98 . ( in contrast , in the @xmath8-wave case , the levels repel at @xmath98 in fig . [ fig : jj](b ) , thus the energy curves are @xmath22-periodic . ) as discussed at the end of sec . [ sec : bound ] , there is no matrix element between the crossing energy levels at @xmath98 . thus , there are no transitions between them , so the occupation numbers of the solid and dotted curves in fig . [ fig : jj](c ) are preserved . in order to show this more formally , we can write a general solution of the time - dependent bdg equation as a superposition of the two subgap states with the time - dependent @xmath134 : @xmath138 $ ] . the matrix element of transitions between the states is proportional to @xmath139 . we found that it is zero in the @xmath0-wave case , thus there are no transitions , and the initial occupation numbers of the subgap states at @xmath101 are preserved dynamically . as one can see in fig . [ fig : jj](c ) , the system is not in the ground state when @xmath140 , because the upper energy level is occupied and the lower one is empty . in principle , the system might be able to relax to the ground state by emitting a phonon or a photon . at present time , we do not have an explicit estimate for such inelastic relaxation time , but we expect that it is quite long . ( the other papers @xcite that _ assume _ thermodynamic equilibrium for each value of the phase @xmath4 do not have an estimate of the relaxation time either . ) to observe the predicted ac josephson effect with the fractional frequency @xmath6 , the period of josephson oscillations should be set shorter than the inelastic relaxation time , but not too short , so that the time evolution of the bdg equation can be treated adiabatically . controlled nonequilibrium population of the upper andreev bound state was recently achieved experimentally in an @xmath8-wave josephson junction in ref . @xcite . ( [ i_t ] ) can be generalized to the case where initially the two subgap states are populated thermally at @xmath101 , and these occupation numbers are preserved by dynamical evolution @xmath141 \over \partial\phi}\ , f[e_a(\phi=0 ) ] \label{idyn } \\ & = & \sin\left({evt\over\hbar}\right ) { \pi\delta_0\over\sqrt{d}er } \tanh\left({\delta_0\sqrt{d}\over2t}\right ) . \label{i_t - t}\end{aligned}\ ] ] notice that the periodicities of the dynamical equation ( [ i_t - t ] ) and the thermodynamic eq . ( [ i_p ] ) are different . the latter equation assumes that the occupation numbers of the subgap states are in instantaneous thermal equilibrium for each @xmath4 . in the infinite barrier limit @xmath142 , the energies @xmath143 of the two subgap states ( [ e_p ] ) degenerate to zero , i.e. they become midgap states . the wave functions ( [ l0s ] ) simplify as follows : @xmath144 since at @xmath145 the josephson junction consists of two semi - infinite uncoupled @xmath0-wave superconductors , @xmath146 and @xmath147 are the wave functions of the surface midgap states @xcite belonging to the left and right superconductors . let us examine the properties of the midgap states in more detail . if @xmath148 is an eigenvector of eq . ( [ eq : bdg ] ) with an eigenvalue @xmath149 , then @xmath150 for @xmath8-wave and @xmath151 for @xmath24-wave are the eigenvectors with the energy @xmath152 . it follows from these relations and eq . ( [ gamma_n ] ) that @xmath153 with @xmath154 . notice that in the @xmath8-wave case , because @xmath148 and @xmath150 are orthogonal for any @xmath155 and @xmath156 , the states @xmath42 and @xmath157 are always different . however , in the @xmath24-wave case , the vectors @xmath148 and @xmath151 may be proportional , in which case they describe the same state with @xmath158 . the states ( [ l0 ] ) and ( [ r0 ] ) indeed have this property : @xmath159 substituting eq . ( [ uv0 ] ) into eq . ( [ gamma_n ] ) , we find the bogoliubov operators of the left and right midgap states @xmath160 operators ( [ conjugate ] ) correspond to the majorana fermions discussed in ref . @xcite . in the presence of a midgap state , the sum over @xmath42 in eq . ( [ canon ] ) should be understood as @xmath161 , where we identify the second term as the projection @xmath162 of the electron operator onto the midgap state . using eqs . ( [ uv0 ] ) , ( [ conjugate ] ) , and ( [ canon ] ) , we find @xmath163 let us consider two semi - infinite @xmath0-wave superconductors on a 1d lattice with the spacing @xmath164 , one occupying @xmath165 and another @xmath166 . they are coupled by the tunneling matrix element @xmath167 between the sites @xmath168 and @xmath164 : @xmath169 . \label{ht}\ ] ] in the absence of coupling ( @xmath170 ) , the subgap wave functions of each superconductor are given by eqs . ( [ l0 ] ) and ( [ r0 ] ) . using eqs . ( [ c_0 ] ) , ( [ uv0 ] ) , ( [ l0 ] ) , and ( [ r0 ] ) , the tunneling hamiltonian projected onto the basis of midgap states is @xmath171 \ , ( \hat\gamma_{l0\uparrow}^\dag \hat\gamma_{r0\uparrow } + { \rm h.c . } ) \nonumber \\ & & = \delta_0\sqrt{d}\ , \cos(\phi/2)\ , ( \hat\gamma_{l0\uparrow}^\dag \hat\gamma_{r0\uparrow } + \hat\gamma_{r0\uparrow}^\dag \hat\gamma_{l0\uparrow } ) , \label{pht}\end{aligned}\ ] ] where @xmath172 is the transmission amplitude , and we omitted summation over the diagonal index @xmath38 . notice that eq . ( [ pht ] ) is @xmath2-periodic in @xmath4 @xcite . hamiltonian ( [ pht ] ) operates between the two degenerate states of the system related by annihilation of the bogoliubov quasiparticle in the right midgap state and its creation in the left midgap state . in this basis , hamiltonian ( [ pht ] ) can be written as a @xmath173 matrix @xmath174 the eigenvectors of hamiltonian ( [ two - level ] ) are @xmath175 , i.e. the antisymmetric and symmetric combinations of the right and left midgap states given in eq . ( [ pm0 ] ) . their eigenenergies are @xmath176 , in agreement with eq . ( [ e_p ] ) . the tunneling current operator is obtained by differentiating eqs . ( [ pht ] ) or ( [ two - level ] ) with respect to @xmath4 . because @xmath4 appears only in the prefactor , the operator structures of the current operator and the hamiltonian are the same , so they are diagonal in the same basis . thus , the energy eigenstates are simultaneously the eigenstates of the current operator with the eigenvalues @xmath177 in agreement with eq . ( [ i_t ] ) . the same basis @xmath175 diagonalizes hamiltonian ( [ two - level ] ) even when a voltage @xmath7 is applied and the phase @xmath4 is time - dependent . then the initially populated eigenstate with the lower energy produces the current @xmath178 with the fractional josephson frequency @xmath6 , in agreement with eq . ( [ i_t ] ) . in the tunneling limit , the transmission coefficient @xmath67 is proportional to the square of the electron tunneling amplitude @xmath167 : @xmath179 . ( [ i_t ] ) and ( [ i_pm ] ) show that the josephson current in the @xmath0-@xmath0 junction is proportional to the first power of the electron tunneling amplitude @xmath167 . this is in contrast to the @xmath8-@xmath8 junction , where the josephson current ( [ i_s ] ) is proportional to @xmath180 . this difference results in the big ratio @xmath181 between the critical currents at @xmath121 in the @xmath0- and @xmath8-wave cases , as shown in fig . [ fig : ic ] and discussed in sec . [ sec : dc ] . the reason for the different powers of @xmath167 is the following . in the @xmath0-wave case , the transfer of just one electron between the degenerate left and right midgap states is a real ( nonvirtual ) process . thus , the eigenenergies are determined from the secular equation ( [ two - level ] ) already in the first order of @xmath167 . in the @xmath8-wave case , there are no midgap states , so the transferred electron is taken from below the gap and placed above the gap , at the energy cost @xmath182 . thus , the transfer of a single electron is a virtual ( not real ) process . it must be followed by the transfer of another electron , so that the pair of electrons is absorbed into the condensate . this gives the current proportional to @xmath180 . this picture implies that the josephson supercurrent across the interface is carried by single electrons in the @xmath0-@xmath0 junction and by cooper pairs in the @xmath8-@xmath8 junction . because the single - electron charge @xmath18 is a half of the cooper - pair charge @xmath23 , the frequency of the ac josephson effect in the @xmath0-@xmath0 junction is @xmath6 , a half of the conventional josephson frequency @xmath20 for the @xmath8-@xmath8 junction . these conclusions also apply to a junction between two cuprate @xmath1-wave superconductors in such orientation that both sides of the junction have surface midgap states , e.g. to the @xmath16 junction ( see sec . [ sec : d - wave ] ) . in both the @xmath0-@xmath0 and @xmath8-@xmath8 junctions , electrons transferred across the interface are taken away into the bulk by the supercurrent of cooper pairs . in the case of the @xmath0-@xmath0 junction , a single transferred electron occupies a midgap state until another electron gets transferred . then the pair of electrons becomes absorbed into the bulk condensate , the midgap state returns to the original configuration , and the cycle repeats . in the case of the @xmath8-@xmath8 junction , two electrons are simultaneously transferred across the interface and become absorbed into the condensate . clearly , electric charge is transferred across the interface by single electrons at the rate proportional to @xmath167 in the first case and by cooper pairs at the rate proportional to @xmath180 in the second case , but the bulk supercurrent is carried by the cooper pairs in both cases . in this section , we consider the josephson effect between two @xmath0-wave superconductors with nonparallel spin - polarization vectors @xmath32 forming an angle @xmath183 . this problem was studied in ref . @xcite using a tunneling hamiltonian approach . here we analyze the problem using the bdg formulation . there are experimental indications that the spin - polarization vector @xmath32 is parallel to the crystal axis @xmath184 in the @xmath12 compounds @xcite . then the considered junction can be realized in the geometry shown in fig . [ fig : jj](a ) where the @xmath184 axes of the two superconductors are rotated relative to each other by the angle @xmath183 around the common @xmath185 axis along the chains . let us select the spin quantization axis @xmath33 perpendicular to both vectors @xmath32 , and the @xmath13 axis in the spin space parallel to the vector @xmath32 of the left superconductor . then the vector @xmath32 of the right superconductor lies in the @xmath186 plane at the angle @xmath183 to the @xmath13 axes : @xmath187 . in this representation , the superconducting pairing takes place between electrons with parallel spins : @xmath188 then , the josephson effect can be considered separately for the spin up and down sectors having the phase differences @xmath189 , correspondingly . using eq . ( [ e_p ] ) for the @xmath0-@xmath0 junction , we find the energies of the andreev bound states for each spin sector @xmath190 the total josephson current is obtained by adding the currents carried by the two spin sectors @xcite . for simplicity , below we consider only the case of zero temperature . in the dynamical limit , assuming that the states ( [ e_up ] ) and ( [ e_down ] ) with @xmath191 are occupied initially and the occupation numbers are preserved dynamically and using eq . ( [ i_t ] ) , we find a @xmath2-periodic current : @xmath192 \nonumber \\ & = & { \sqrt{d}e\delta_0 \over \hbar}\ , \sin\left({\phi(t ) \over 2}\right)\ , \cos\left({\theta \over 2}\right ) . \label{i - theta - t}\end{aligned}\ ] ] in the static thermodynamic limit , using eq . ( [ i_p ] ) at @xmath121 , we find the dc josephson current : @xmath193 \right . \nonumber \\ & & + \left . \sin\left({\phi-\theta \over 2}\right)\ , { \rm sgn}\left [ \cos\left({\phi-\theta \over 2}\right ) \right ] \right\}. \label{i - theta}\end{aligned}\ ] ] for completeness , let us also consider the josephson effect between two @xmath194-wave or two @xmath195-wave superconductors , where the @xmath196 and @xmath33 axes are parallel to the junction plane . in these junctions , midgap states are absent in the @xmath142 limit , thus the current - phase relation is conventional @xmath197 . for nonparallel vectors @xmath32 , the total josephson current is the sum of the spin up and down sectors : @xmath198 \nonumber \\ & = & i_c\cos\theta\sin\phi = i_c(\bm{n}_l\cdot\bm{n}_r)\sin\phi . \label{cos - theta}\end{aligned}\ ] ] eq . ( [ cos - theta ] ) is consistent with ref . @xcite . in the case where the two vectors @xmath32 are perpendicular ( @xmath199 ) , the josephson current ( [ cos - theta ] ) for the superconductors without midgap states vanishes , but , according to eqs . ( [ i - theta - t ] ) and ( [ i - theta ] ) , it is not zero if the midgap states are present . in this section , we consider a junction between a singlet @xmath8-wave and a triplet @xmath0-wave superconductors . the junction geometry is the same as in fig . [ fig : jj](a ) , where one of the superconductors is taken to be a conventional @xmath8-wave superconductor and another one a q1d triplet @xmath0-wave superconductor , such as @xmath12 . -@xmath0 junction . here @xmath200 and @xmath201 . ( a ) the energies ( [ sp - e ] ) of the andreev bound states . the solid and dashed lines correspond to @xmath202 and @xmath203 . ( b ) the josephson current in the static thermodynamic limit , where the states with @xmath204 are occupied . ( c ) the josephson current in the dynamical limit , where the central branch with @xmath203 and the lower branches touching @xmath100 are occupied . ] we choose the spin quantization axis @xmath33 along the polarization vector @xmath32 of the triplet superconductor , so the spin projection @xmath28 on the @xmath33 axis is a good quantum number . in both triplet and singlet superconductors , the cooper pairing takes place between electrons with opposite spins . however , the pairing potential has the same sign for @xmath28 and @xmath34 in the triplet superconductor and the opposite signs in the singlet superconductor . thus , the phase difference across the josephson junction is @xmath4 for quasiparticles with @xmath202 and @xmath205 for @xmath203 . the energies of the andreev bound states can be found for each @xmath28 from eq . ( [ self ] ) together with eq . ( [ kappa - eta ] ) , where we should use the upper line of eq . ( [ delta_bsa ] ) for the left superconductor and the lower line for the right superconductor . to simplify calculations , we consider the case where the magnitudes of the gaps are equal for the @xmath8- and @xmath0-wave superconductors : @xmath206 . the energies of the andreev bound states are @xmath207 for each value of the spin index @xmath208 , eq . ( [ sp - e ] ) gives two andreev states labeled by the index @xmath91 . in the tunneling limit @xmath107 , we have @xmath209 the energies ( [ sp - e ] ) are plotted in fig . [ fig : spj](a ) vs. @xmath4 by the solid lines for @xmath202 and by the dashed lines for @xmath203 . we observe that the branches ( [ eq : e+ ] ) with @xmath191 touch the gap boundaries @xmath210 at @xmath101 and @xmath211 , whereas the branches ( [ eq : e- ] ) with @xmath212 stay in the center of the gap . in the limit @xmath107 , the central branches with @xmath212 dominate the energy dependence on @xmath4 , and energy minima are achieved at @xmath213 or @xmath214 . notice that if the system selects the energy minimum at @xmath213 , then the spin down states , shown by the dashed lines in fig . [ fig : spj](a ) , are populated , and the spin up states are empty , so the junction accumulates the spin @xmath215 per conducting channel @xcite . if the system selects the energy minimum at @xmath216 , then the junction accumulates the spin @xmath217 per conducting channel . in the limit @xmath107 , we can neglect the energies ( [ eq : e+ ] ) and obtain the josephson current by differentiating the energies ( [ eq : e- ] ) with respect to @xmath4 using eq . ( [ eq : i_a ] ) @xcite . in the dynamical limit , the occupation numbers of the andreev states ( [ eq : e- ] ) are preserved , and the josephson current has the @xmath22-periodicity , as shown in fig . [ fig : spj](c ) : @xmath218 in the static thermodynamic limit , the system occupies the branch of the minimal energy for each @xmath4 , and the josephson current is @xmath211-periodic , as shown in fig . [ fig : spj](b ) : @xmath219 the thermodynamic assumption implies that the spin accumulation at the @xmath8-@xmath0 junction changes sign when the phase @xmath4 crosses @xmath211 . -wave superconductor closed in a loop by an @xmath8-wave superconductor . no current is circulating in the loop in equilibrium . however , there is accumulation of spins up in one @xmath8-@xmath0 junction and spins down in another junction . the sketch at the top illustrates the fermi surface of a q1d metal with the opposite signs of the superconducting @xmath0-wave pairing potential on the two sheets of the fermi surface . ] now let us consider the circuit shown in fig . [ fig : sp - loop ] , where a @xmath0-wave superconductor has josephson junctions at both ends with an @xmath8-wave superconductor closed in a loop . because the sign of the @xmath0-wave pairing potential is opposite for the @xmath220 and @xmath221 sheets of the fermi surface , the two junctions have the relative phase shift @xmath211 . naively , one might expect a spontaneous current in this circuit by analogy with the corner squid in the cuprates @xcite . however , the system shown in fig . [ fig : sp - loop ] can accommodate the phase shift @xmath211 by selecting the energy minimum at @xmath213 for one junction and the energy minimum at @xmath216 for another junction . then , no current circulates in the loop . however , one junction accumulates spins up and another junction spins down , which might be possible to detect experimentally . the results of the this section clearly show that a josephson current is possible between singlet and triplet superconductors , in agreement with the earlier findings by yip @xcite . recently , the josephson current was calculated for the @xmath8-@xmath0 junction in ref . @xcite , but spin accumulation at the junction was not recognized in this paper . the @xmath8-@xmath0 junction considered in this section is mathematically equivalent to the @xmath222 @xmath1-@xmath1 junction and the @xmath223 junction between an @xmath8-wave and a @xmath1-wave superconductors ( see sec . [ sec : d - wave ] ) . ( [ sp - e ] ) was obtained for that case in refs . @xcite . however , there is no spin accumulation in junctions between singlet @xmath8- and @xmath1-wave superconductors , unlike in the @xmath8-@xmath0 junction . in this section , we study junctions between quasi - two - dimensional ( q2d ) superconductors such as nonchiral @xmath1-wave cuprates and chiral @xmath9-wave ruthenates . for simplicity , we use an isotropic electron energy dispersion law @xmath224 in the @xmath186 plane . as before , we select the coordinate @xmath13 perpendicular to the junction line and assume that the electron momentum component @xmath38 parallel to the junction line is a conserved good quantum number . then , the 2d problem separates into a set of 1d solutions ( [ l0s ] ) in the @xmath13 direction labeled by the index @xmath38 . the fermi momentum @xmath225 and velocity @xmath226 are replaced by their @xmath13-components @xmath227 and @xmath228 . the transmission coefficient of the barrier ( [ zdg ] ) becomes @xmath38-dependent @xmath229 where @xmath230 . the total josephson current is given by a sum over all occupied subgap states labeled by @xmath38 . for the cuprates , let us consider a junction parallel to the @xmath231 $ ] crystal direction in the @xmath232 plane and select the @xmath13 axis along the diagonal @xmath233 $ ] , as shown in fig . [ fig : d - wave](a ) . in these coordinates , the @xmath1-wave pairing potential is @xmath234 where the same notation as in eq . ( [ hat - delta ] ) is used . direct comparison of eqs . ( [ hat - delta - d ] ) and ( [ hat - delta ] ) demonstrates that the @xmath1-wave superconductor with the @xmath223 junction maps to the @xmath0-wave superconductor by the substitution @xmath235 . thus , the results obtained in sec . [ sec : q1d ] for the @xmath0-@xmath0 junction apply to the @xmath16 junction between two @xmath1-wave superconductors with the appropriate integration over @xmath38 . the energies of the subgap andreev states are given by eq . ( [ e_p ] ) with the @xmath38-dependent parameters @xmath114 and @xmath67 , and the energies and the wave functions are @xmath2-periodic functions of @xmath4 . thus , the ac josephson current in the dynamical limit is @xmath2-periodic and has the fractional frequency @xmath6 , as in eqs . ( [ i_t ] ) , ( [ i_t - t ] ) , and ( [ i_pm ] ) . the energies ( [ e_p ] ) of the subgap states @xcite and the dc josephson current ( [ i_p ] ) in the thermodynamic limit @xcite were calculated for the @xmath16 @xmath1-@xmath1 junction before . however , these papers did not recognize the fractional , @xmath2-periodic character of the josephson effect in the dynamical limit . junction ( panel a ) and @xmath236 junction ( panel b ) between two @xmath1-wave superconductors . the thick line represents the junctions line . the circles illustrate the fermi surfaces , where positive and negative pairing potentials @xmath237 are shown by the solid and dotted lines . the points a , b , c , and d in the momentum space are connected by transmission and reflection from the barrier . ] on the other hand , if the junction is parallel to the @xmath238 $ ] crystal direction , as shown in fig . [ fig : d - wave](b ) , then @xmath239 . this pairing potential is an even function of @xmath240 , thus it is analogous to the @xmath8-wave pairing potential in eq . ( [ hat - delta ] ) . then , the @xmath236 junction between two @xmath1-wave superconductors is analogous to the @xmath8-@xmath8 junction . it should exhibit the conventional @xmath22-periodic josephson effect with the frequency @xmath20 . for a generic orientation of the junction line , the @xmath1-wave pairing potential is @xmath0-like for some momenta @xmath38 and @xmath8-like for other @xmath38 . thus , the total josephson current is a sum of the unconventional and conventional terms : @xmath241 where @xmath242 and @xmath243 are some coefficients . we expect that both terms in eq . ( [ i - phi/2 ] ) are present for any real junction between @xmath1-wave superconductors because of imperfections in junction orientation . however , the ratio @xmath244 should be maximal for the junction shown in fig . [ fig : d - wave](a ) and minimal for the junction shown in fig . [ fig : d - wave](b ) . in general , whenever the superconductors on both sides of the junction have surface midgap states , we expect to observe the @xmath2-periodic fractional ac josephson effect . in principle , the effect may be spoiled by the gapless quasiparticles that exist near the gap nodes in a @xmath1-wave superconductor . however , they would affect only a small portion of the fermi surface near the nodes , and the @xmath2-periodic josephson effect should survive on the other parts of the fermi surface , where the gap is big . the @xmath16 junction shown in fig . [ fig : d - wave](a ) should not be confused with the @xmath222 @xmath1-@xmath1 junction @xcite or the @xmath223 @xmath8-@xmath1 junction @xcite , much discussed in literature . none of the papers @xcite treated the problem correctly , because they did not take into account the andreev bound states in the junction properly . the correct energy spectrum of the andreev bound states was obtained in refs . @xcite . in the @xmath222 @xmath1-@xmath1 and @xmath223 @xmath8-@xmath1 junctions , only one superconductor has midgap states , thus these junctions are mathematically analogous to the @xmath8-@xmath0 junction considered in sec . [ sec : s - p ] . the spectrum of the andreev bound states is given by eq . ( [ sp - e ] ) without the factor @xmath28 , because both superconductors are singlet . the energy levels are plotted vs. @xmath4 in fig . [ fig : spj](a ) , where the solid and dashed lines represent not spin , but positive and negative momenta @xmath38 . the junction has two energy minima at @xmath213 or @xmath214 , where the states with only negative or positive momenta @xmath38 are filled , thus there are persistent currents along the junction line @xcite . ( on the other hand , there is no spin accumulation , unlike in the @xmath8-@xmath0 junction discussed in sec . [ sec : s - p ] . ) in the thermodynamic limit , the current - phase relation shown in fig . [ fig : spj](b ) is @xmath211-periodic ; however , it requires reversing the currents along the junction line when @xmath4 passes through 0 or @xmath211 . in the dynamical limit , the current - phase relation shown in fig . [ fig : spj](c ) is @xmath22-periodic . the first two harmonics @xmath245 have been recently observed experimentally in the @xmath222 @xmath1-@xmath1 junction @xcite . in this section , we study junctions between the chiral @xmath246-wave superconductors @xmath14 , where the pairing potential is assumed to be @xmath247 @xcite , and the two signs correspond to opposite chiralities . we assume a uniform orientation of the spin - polarization vector @xmath32 across the junction . this problem was investigated in ref . @xcite using the eilenberger equation for green s functions . it was found that the chiral subgap states at the junctions enhance the low - temperature critical josephson current in symmetric junctions . here we use the bdg equation to obtain the spectrum of the andreev bound states . as before , we assume that the momentum component @xmath38 parallel to the junction is conserved . thus , the problem separates into a set of 1d solutions in the @xmath13 direction perpendicular to the junction plane , and we can use the method of sec . [ sec : bound ] . first we consider a junction between two superconductors with opposite chiralities , as illustrated in the first column of fig . [ fig : chiral](a ) . in this case , @xmath248 and @xmath249 . when the barrier is not transparent ( @xmath145 ) , each superconductor has chiral andreev surface states with the same energy dispersion @xmath250 @xcite . the electron tunneling amplitude @xmath251 produces a matrix element mixing the two states in the first - order degenerate perturbation theory . thus , the two energy spectra repel with the splitting proportional to @xmath252 . from eq . ( [ self ] ) , we find the following subgap energies : @xmath253 the energy levels splitting oscillates with the period @xmath2 as a function of @xmath4 : @xmath254 . the splitting depends on @xmath38 through the square - root prefactor and through the dependence of @xmath67 on @xmath38 in eq . ( [ zdg-2d ] ) , and vanishes at @xmath255 . the energy dispersion ( [ opposite - ch ] ) is plotted vs. @xmath38 in the second and third columns of fig . [ fig : chiral](a ) for several values of @xmath4 . the spectrum of excitations is gapless because of the chiral dispersion in @xmath38 . thus , it is reasonable to assume that the occupation numbers of the subgap states are in instantaneous thermodynamic equilibrium for any phase @xmath4 . then , the josephson current is a @xmath22-periodic function of @xmath4 , as illustrated at zero temperature in the fourth column of fig . [ fig : chiral](a ) , even though the energy levels ( [ opposite - ch ] ) are @xmath2-periodic functions of @xmath4 . now let us consider the case of two superconductors with opposite chiralities , as illustrated in the first column of fig . [ fig : chiral](b ) . when the two superconductors are disconnected ( @xmath145 ) , their chiral andreev surface states have opposite dispersions @xmath256 , thus they are nondegenerate . the electron tunneling amplitude @xmath251 repels the energy levels around the intersection point @xmath257 . from eq . ( [ self ] ) , we find the following subgap energies : @xmath258 the energy dispersion ( [ same - ch ] ) is plotted vs. @xmath38 in the second and third columns of fig . [ fig : chiral](b ) for several values of @xmath4 . the energy splitting around @xmath257 is a @xmath22-periodic function of @xmath4 and vanishes at @xmath98 . the josephson current is a @xmath22-periodic function of @xmath4 , as illustrated at zero temperature in the fourth column of fig . [ fig : chiral](b ) . now let us consider a junction between an @xmath8-wave and a @xmath259-wave superconductors shown in the first column of fig . [ fig : chiral](c ) . the josephson current was calculated in this case in ref . @xcite using the method of green s functions . however , the energies of the andreev bound states were not written explicitly . the subgap states in this junction are obtained by solving eq . ( [ self ] ) in the manner similar to the 1d @xmath8-@xmath0 junction . for simplicity , we assume that the magnitudes of the pairing potentials in both superconductors are the same : @xmath206 . the square of the subgap energies is given by the following expression @xmath260 where @xmath261 is the reflection coefficient , and @xmath262 . the signs of the energies are @xmath263 for a given @xmath28 , there are two branches of energies labeled by the index @xmath91 . the energy dispersions @xmath264 are shown in the second and third columns of fig . [ fig : chiral](c ) for several phases @xmath4 . in the limit of impenetrable barrier @xmath142 , the energy branch with @xmath191 approaches to the gap edges @xmath265 , whereas the branch with @xmath212 approaches to the energy dispersion @xmath266 of the chiral surface states in the @xmath259-wave superconductor @xcite the energy @xmath267 for a given @xmath28 is a @xmath22-periodic function of @xmath4 . the energy @xmath268 is obtained from @xmath267 by the shift @xmath269 , as discussed in sec . [ sec : s - p ] . thus , the josephson current ( [ thermal ] ) in the static thermodynamic limit , obtained by summation over @xmath28 and @xmath38 , is a @xmath211-periodic function of @xmath4 , as shown in the fourth column of fig . [ fig : chiral](c ) , in agreement with ref . @xcite . similarly to the @xmath8-@xmath0 junction considered in sec . [ sec : s - p ] , the @xmath8-(@xmath259 ) junction has two equal energy minima @xcite at @xmath213 and @xmath214 accompanied by accumulation of the down spin for @xmath213 and the up spin for @xmath216 . conceptually , the setup for experimental observation of the fractional ac josephson effect is straightforward . one should apply a dc voltage @xmath7 to the junction and measure frequency spectrum of microwave radiation from the junction , expecting to detect a peak at the fractional frequency @xmath6 . higher harmonics , such as @xmath20 , may also be present because of eq . ( [ i - phi/2 ] ) and circuit nonlinearities , but an observation of the 1/2 subharmonic of the conventional josephson frequency @xmath20 would be the signature of the effect . josephson radiation at the conventional frequency @xmath20 was first observed experimentally almost 40 years ago in kharkov @xcite , followed by further work @xcite . in ref . @xcite , the spectrum of microwave radiation from tin junctions was measured , and a sharp peak at the frequency @xmath20 was found . without any attempt to match impedances of the junction and waveguide , dmitrenko and yanson @xcite found the signal several hundred times stronger than the noise and the ratio of linewidth to the josephson frequency less than @xmath270 . more recently , a peak of josephson radiation was observed in ref . @xcite in indium junctions at the frequency 9 ghz with the width 36 mhz . in ref . @xcite , a peak of josephson radiation was observed around 11 ghz with the width 50 mhz in @xmath271 single crystals with the current along the @xmath184 axis perpendicular to the layers . to observe the fractional ac josephson effect predicted in this paper , it is necessary to perform the same experiment with the @xmath16 cuprate junctions shown in fig . [ fig : d - wave](a ) . for control purposes , it is also desirable to measure frequency spectrum for the @xmath236 junction shown in fig . [ fig : d - wave](b ) , where a peak at the frequency @xmath6 should be minimal . it should be absent completely in a conventional @xmath8-@xmath8 junction , unless the junction enters a chaotic regime with period doubling @xcite . the high-@xmath10 junctions of the required geometry can be manufactured using the step - edge technique . bicrystal junctions are not appropriate , because the crystal axes @xmath185 and @xmath17 of the two superconductors are rotated relative to each other in such junctions . as shown in fig . [ fig : d - wave](a ) , we need the junction where the crystal axes of the two superconductors have the same orientation . unfortunately , attempts to manufacture josephson junctions from the q1d organic superconductors @xmath12 failed thus far . the most common way of studying the ac josephson effect is observation of the shapiro steps @xcite . in this setup , the josephson junction is irradiated by microwaves with the frequency @xmath272 , and steps in dc current are detected at the dc voltages @xmath273 . unfortunately , this method is not very useful to study the effect that we predict . indeed , our results are effectively obtained by the substitution @xmath274 . thus , we expect to see the shapiro steps at the voltages @xmath275 , i.e. we expect to see only _ even _ shapiro steps . however , when both terms are present in eq . ( [ i - phi/2 ] ) , they produce both even and odd shapiro steps , so it would be difficult to differentiate the novel effect from the conventional shapiro effect . notice also that the so - called fractional shapiro steps observed at the voltage @xmath276 corresponding to @xmath277 have nothing to do with the effect that we propose . they originate from the higher harmonics in the current - phase relation @xmath278 . the fractional shapiro steps have been observed in cuprates @xcite , but also in conventional @xmath8-wave superconductors @xcite . another method of measuring the current - phase relation in cuprates was employed in ref . @xcite , but connection with our theoretical results is not clear at the moment . in this paper , we study suitably oriented @xmath0-@xmath0 or @xmath1-@xmath1 josephson junctions , where the superconductors on both sides of the junction originally have the surface andreev midgap states . in such junctions , the josephson current @xmath3 , carried by the hybridized subgap andreev bound states , is a @xmath2-periodic function of the phase difference @xmath4 : @xmath5 , in agreement with ref . thus , the ac josephson current should exhibit the fractional frequency @xmath6 , a half of the conventional josephson frequency @xmath20 . in the tunneling limit , the josephson current is proportional to the first power of the electron tunneling amplitude , not the square as in the conventional case @xcite . thus , the josephson current in the considered case is carried across the interface by single electrons with charge @xmath18 , rather than by copper pairs with charge @xmath23 . the fractional ac josephson effect can be observed experimentally by measuring frequency spectrum of microwave radiation from the junction and detecting a peak at @xmath6 . in @xmath0-@xmath0 junctions with nonparallel orientation of the spin - polarization vectors @xmath32 , the josephson current depends on the relative angle between the vectors @xmath32 @xcite . the josephson current is permitted between singlet and triplet superconductors , but , in the static thermodynamic limit , the current - phase relation is @xmath211-periodic @xcite . the @xmath8-@xmath24 junction has two equal minima in energy at @xmath213 and @xmath214 @xcite , characterized by accumulation of the up or down spins ( oriented relative to the vector @xmath32 ) in the junction . in josephson junctions between chiral @xmath9-wave superconductors , the andreev bound states are also chiral . in the static thermodynamic limit , the current - phase relation has the period of @xmath22 in the chiral @xmath24-@xmath24 junctions @xcite and the period of @xmath211 in the chiral @xmath8-@xmath24 junctions @xcite . _ note added in proof . _ the fractional josephson effect discussed in our paper is similar to the fractional quantum hall effect @xcite . both involve existence of several equivalent ground states , whose energy levels cross : compare fig . [ fig : jj]c of our paper and fig . 2a of ref . @xcite . i. j. lee , s. e. brown , w. g. clark , m. j. strouse , m. j. naughton , w. kang , and p. m. chaikin , , 017004 ( 2002 ) ; i. j. lee , m. j. naughton , g. m. danner , and p. m. chaikin , _ ibid . _ * 78 * , 3555 ( 1997 ) ; i. j. lee , p. m. chaikin , and m. j. naughton , , r14669 ( 2000 ) . the current - phase relation @xmath5 that we propose should not be confused with another unconventional current - phase relation @xmath278 with the period @xmath211 , which was predicted theoretically for junctions between @xmath8- and @xmath1-wave superconductors @xcite , @xmath8- and @xmath24-wave superconductors @xcite , and for the @xmath222 junctions between two @xmath1-wave superconductors @xcite . it is well - known that using the bdg formalism for the spin up and down sectors constitutes double - counting @xcite . thus , the total energy and current are obtained by adding the contributions of the spin up and down sectors and dividing the result by two . -wave superconductors , calculated for @xmath283 . ( a ) junction between two @xmath24-wave superconductors with opposite chiralities . ( b ) junction between two @xmath24-wave superconductors with the same chirality . ( c ) junction between @xmath8-wave and chiral @xmath24-wave superconductors . in the second and third columns of row ( a ) , the solid and dashed lines show the @xmath91 branches of eq . ( [ opposite - ch ] ) for different values of @xmath4 . in rows ( b ) and ( c ) , the energies ( [ same - ch ] ) and ( [ e - sp ] ) are shown by the solid and dashed lines for @xmath101 and @xmath284 in the second column and for @xmath98 and @xmath214 in the third column . the energies of quasiparticles with @xmath203 are the same as with @xmath202 for rows ( a ) and ( b ) , and they can be obtained by the shift @xmath269 for row ( c ) . the fourth column shows the josephson current in the static thermodynamic limit , normalized by @xmath285 , where @xmath286 is the length of the junction . ]

we present a framework for approximating the metric tsp based on a novel use of matchings . traditionally , matchings have been used to add edges in order to make a given graph eulerian , whereas our approach also allows for the removal of certain edges leading to a decreased cost . for the tsp on graphic metrics ( graph - tsp ) , the approach yields a-approximation algorithm with respect to the held - karp lower bound . for graph - tsprestricted to a class of graphs that contains degree three bounded and claw - free graphs , we show that the integrality gap of the held - karp relaxation matches the conjectured ratio . the framework allows for generalizations in a natural way and also leads to a-approximation algorithm for the traveling salesman path problem on graphic metrics where the start and end vertices are prespecified .

1104.3090

the traveling salesman problem in metric graphs is one of most fundamental np - hard optimization problems . in spite of a vast amount of research several important questions remain open . while the problem is known to be apx - hard and np - hard to approximate with a ratio better than @xmath3 @xcite , the best upper bound is still the 1.5-approximation algorithm obtained by christofides @xcite more than three decades ago . a promising direction to improve this approximation guarantee , has long been to understand the power of a linear program known as the held - karp relaxation @xcite . on the one hand , the best lower bound on its integrality gap ( for the symmetric case ) is @xmath1 and indeed conjectured to be tight @xcite . on the other hand , the best known analysis @xcite is based on christofides algorithm and gives an upper bound on the integrality gap of @xmath4 . in the light of this difficulty of even determining the integrality gap of the held - karp relaxation , a reasonable way to approach the metric tsp is to restrict the set of feasible inputs . one promising candidate is the _ graph - tsp , _ that is , the traveling salesman problem where distances between cities are given by any graphic metric , i.e. , the distance between two cities is the length of the shortest path in a given ( unweighted ) graph . equivalently , graph - tspcan be formulated as the problem of finding an eulerian multigraph within an unweighted input graph so as to minimize the number of edges . in contrast to tsp on euclidean metrics that admits a ptas @xcite , the graph - tsp seems to capture the difficulty of the metric tsp in the sense that , as stated in @xcite , it is apx - hard and the lower bound 4/3 on the integrality gap of the held - karp relaxation is established using a graph - tspinstance . the tsp on graphic metrics has recently drawn considerable attention . in 2005 , gamarnik et al . @xcite showed that for cubic 3-edge - connected graphs , there is an approximation algorithm achieving an approximation ratio of @xmath5 . this result was generalized to cubic graphs by boyd et al . @xcite , who obtained an improved performance guarantee of 4/3 . for subcubic graphs , i.e. , graphs of degree at most @xmath6 , they also gave an 7/5-approximation algorithm with respect to the held - karp lower bound . in a major achievement , gharan et al . @xcite recently presented an approximation algorithm for graph - tspwith performance guarantee strictly better than 1.5 . the approach in @xcite is similar to that of christofides in the sense that they start with a spanning tree and then add a perfect matching of those vertices of odd - degree to make the graph eulerian . the main difference is that instead of starting with a minimum spanning tree , their approach uses the solution of the held - karp relaxation to sample a spanning tree . although the proposed algorithm in @xcite is surprisingly simple , the analysis is technically involved and several novel ideas are needed to obtain the improved performance guarantee @xmath7 for an @xmath8 of the order @xmath9 . [ [ our - results - and - overview - of - techniques . ] ] our results and overview of techniques . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we propose an alternative framework for approximating the metric tsp and use it to obtain an improved approximation algorithm for graph - tsp . [ thm : approximationratio ] there is a polynomial time approximation algorithm for graph - tspwith performance guarantee @xmath10 . the result implies an upper bound on the integrality gap of the held - karp relaxation for graph - tspthat matches the approximation ratio . for the restricted class of graphs , where each block ( i.e. , each maximally 2-vertex - connected subgraph ) is either claw - free or of degree at most @xmath6 , we use the framework to construct a polynomial time @xmath1-approximation algorithm showing that the conjectured integrality gap of the held - karp relaxation is tight for those graphs . in fact , the techniques allow us to prove the tight result that any @xmath11-vertex - connected graph of degree at most @xmath6 has a spanning eulerian multigraph with at most @xmath12 edges , which settles a conjecture of boyd et al . @xcite affirmatively . our framework is based on earlier works by frederickson & jaja @xcite and monma et al . @xcite , who related the cost of an optimal tour to the size of a minimum @xmath11-vertex - connected subgraph . more specifically , monma et al . showed that a @xmath11-vertex - connected graph @xmath13 always has a spanning eulerian multigraph with at most @xmath14 edges , generalizing a previous result of frederickson & jaja who obtained the same result for the special case of planar @xmath11-vertex - connected graphs . one interpretation of their approaches is the following . given a @xmath11-vertex - connected graph @xmath13 , they show how to pick a random subset @xmath15 of edges satisfying : ( i ) an edge is in @xmath15 with probability @xmath16 and ( ii ) the multigraph @xmath17 with vertex set @xmath18 and edge set @xmath19 is spanning and eulerian . from property @xmath20 of @xmath15 , the expected number of edges in @xmath17 is @xmath14 yielding their result . although the factor @xmath1 is asymptotically tight for some classes of graphs ( one example is the family of integrality gap instances for the held - karp relaxation described in section [ sec : prelim ] ) , the bound rapidly gets worse for @xmath11-vertex - connected graphs with significantly more than @xmath21 edges . the novel idea to overcome this issue is the following . instead of adding all the edges in @xmath15 to @xmath22 , some of the edges in @xmath15 might instead be removed from @xmath22 to form @xmath17 . as long as the removal of the edges does not disconnect the graph , this will again result in a spanning eulerian multigraph @xmath17 . to specify a subset @xmath23 of edges that safely may be removed we introduce , in section [ sec : tspframework ] , the notion of a `` @xmath24 '' . the framework is then completed by theorem [ thm : main ] , where we show that a @xmath11-vertex - connected graph @xmath13 with a set @xmath23 of removable edges has a spanning eulerian multigraph with at most @xmath25 edges . in order to use the framework , one of the main challenges is to find a sufficiently large set of removable edges . in section [ sec : circulation ] , we show that this problem can be reduced to that of finding a min - cost circulation in a certain circulation network . to analyze the circulation network we then ( in section [ sec : algorithms ] ) use several properties of an extreme point solution to the held - karp relaxation to obtain our main algorithmic result . the better approximation guarantees for special graph classes follows from that the circulation network has an easier structure in these cases , which in turn allows for a better analysis . finally , we note that the techniques generalize in a natural way . our results can be adapted to the more general traveling salesman path problem ( graph - tspp ) with prespecified start and end vertices to improve on the approximation ratio of @xmath26 by hoogeveen @xcite when considering graphic metrics . more specifically , we obtain the following . [ thm : approximationratiohpp ] for any @xmath27 , there is a polynomial time approximation algorithm for graph - tsppwith performance guarantee @xmath28 . if furthermore each block of the given graph is degree three bounded , there is a polynomial time approximation algorithm for graph - tsppwith performance guarantee @xmath29 , for any @xmath30 . the generalization to the traveling salesman problem is presented in section [ sec : tspp ] . [ [ held - karp - relaxation . ] ] held - karp relaxation . + + + + + + + + + + + + + + + + + + + + + the linear program known as the held - karp ( or subtour elimination ) relaxation is a well studied lower bound on the value of an optimal tour . it has a variable @xmath31 for each pair of vertices with the intuitive meaning that @xmath31 should take value @xmath32 if the edge @xmath33 is used in the tour and @xmath34 otherwise . letting @xmath13 be the complete graph on the set of vertices and @xmath35 be the distance between vertices @xmath36 and @xmath37 , the held - karp relaxation can then be formulated as the linear program where we wish to minimize @xmath38 subject to @xmath39 where @xmath40 denotes the set of edges crossing the cut @xmath41 and @xmath42 for any @xmath43 . goemans & bertsimas @xcite proved that for metric distances the above linear program has the same optimal value as the linear program obtained by dropping the equality constraints . moreover , when considering a graph - tspinstance @xmath13 we only need to consider the variables @xmath44 . indeed , any solution @xmath45 to the held - karp relaxation without equality constraints such that @xmath46 for a pair of vertices @xmath47 can be transformed into a solution @xmath48 with no worse cost and @xmath49 by setting @xmath50 for each edge on the shortest path between @xmath36 and @xmath37 , and @xmath51 for the other edges . the held - karp relaxation for graph - tspon a graph @xmath13 can thus be formulated as follows : @xmath52 we shall refer to this linear program as @xmath53 and denote the value of an optimal solution by @xmath54 . its integrality gap was previously known to be at most @xmath55 and at least @xmath1 for graphic instances . the lower bound is obtained by a claw - free graphic instance of degree at most @xmath6 that consists of three paths of equal length with endpoints @xmath56 and @xmath57 that are connected so as @xmath58 and @xmath59 form two triangles ( see figure [ fig : intgap ] ) . we end our discussion of @xmath53 with a useful observation . when considering graph - tsp , it is intuitively clear that we can restrict ourselves to _ @xmath11-vertex - connected _ graphs , i.e. , graphs that stay connected after deleting a single vertex . indeed , if we consider a graph with a vertex @xmath37 whose removal results in components @xmath60 with @xmath61 then we can recursively solve the graph - tsp problem on the @xmath62 subgraphs @xmath63 induced by @xmath64 . the union of these solutions will then provide a solution to the original graph that preserves the approximation guarantee with respect to the linear programming relaxation since one can see that @xmath65 . we summarize this observation in the following lemma ( see appendix [ app:2conntsp ] for a fullproof ) . [ lemma:2conntsp ] let @xmath22 be a connected graph . if there is an @xmath66-approximation algorithm for graph - tsp on each @xmath11-vertex - connected subgraph @xmath17 of @xmath22 ( with respect to @xmath67 ) then there is an @xmath66-approximation algorithm for graph - tspon @xmath22 ( with respect to @xmath54 ) . [ [ matchings - of - cubic-2-edge - connected - graphs . ] ] matchings of cubic @xmath11-edge - connected graphs . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + edmonds @xcite showed that the following set of equalities and inequalities on the variables @xmath68 determines the perfect matching polytope ( i.e. , all extreme points of the polytope are integral and correspond to perfect matchings ) of a given graph @xmath69 : @xmath70 the linear description is useful for understanding the structure of the perfect matchings . for example , naddef and pulleyblank @xcite proved that @xmath71 defines a feasible solution when @xmath22 is _ cubic _ and _ @xmath11-edge connected _ , i.e. , every vertex has degree @xmath6 and the graph stays connected after the removal of an edge . they used that result to deduce that such graphs always have a perfect matching of weight at least @xmath16 of the total weight of the edges . standard algorithmic versions of carathodory s theorem ( see e.g. theorem @xmath72 in @xcite ) say that , in polynomial time , we can decompose a feasible solution to the perfect matching polytope into a convex combination of polynomially many perfect matchings ( see also @xcite for a combinatorial approach for the matching polytope ) . combining these results leads to the following lemma ( see @xcite for closely related variants that also have been useful for the graph - tspproblem ) . [ lem : matching ] given a cubic @xmath11-edge - connected graph @xmath22 , we can in polynomial time find a distribution over polynomially many perfect matchings so that with probability @xmath16 an edge is in a perfect matching picked from this distribution . note that all @xmath11-vertex - connected graphs except the trivial graph on @xmath11 vertices are @xmath11-edge connected . we can therefore apply the above lemma to cubic @xmath11-vertex - connected graphs . lemma [ lemma:2conntsp ] says that the technical difficulty in approximating the graph - tspproblem lies in approximating those instances that are @xmath11-vertex connected . as alluded to in the introduction , we shall generalize previous results @xcite that relate the cost of an optimal tour to the size of a minimum @xmath11-vertex - connected subgraph . the main difference is the use of matchings . traditionally , matchings have been used to add edges to make a given graph eulerian whereas our framework offers a structured way to specify a set of edges that safely may be removed leading to a lower cost . to identify the set of edges that may be removed we use the following definition . [ def : pairing ] given a @xmath11-vertex - connected graph @xmath22 we call a tuple @xmath73 consisting of a subset @xmath23 of removable edges and a subset @xmath74 of pairs of edges a _ @xmath24 _ if * an edge is in at most one pair ; * the edges in a pair are incident to a common vertex of degree at least @xmath6 ; * any graph obtained by deleting removable edges so that at most one edge in each pair is deleted stays connected . the following theorem generalizes the corresponding result of @xcite ( their result follows from the the special case of an empty removable pairing ) . [ thm : main ] given a @xmath11-vertex - connected graph @xmath13 with a @xmath24 @xmath73 , there is a polynomial time algorithm that returns a spanning eulerian multigraph in @xmath22 with at most @xmath75 edges . the proof of the theorem is presented after the following lemma on which it is based . [ lemma : sample ] given a @xmath11-vertex - connected graph @xmath13 with a @xmath24 @xmath73 , we can in polynomial time find a distribution over polynomially many subsets of edges such that a random subset @xmath15 from this distribution satisfies : * each edge is in @xmath15 with probability @xmath16 ; * at most one edge in each pair is in @xmath15 ; and * each vertex has an even degree in the multigraph with edge set @xmath76 . we shall use lemma [ lem : matching ] and will therefore need a cubic @xmath11-edge - connected graph . in the spirit of @xcite , we replace all vertices of @xmath22 that are not of degree three by gadgets to obtain a cubic graph @xmath77 as follows ( see also figure [ fig : degreplace ] ) : * a vertex @xmath37 of degree 2 with neighbors @xmath36 and @xmath78 is replaced by a cycle consisting of four vertices @xmath79 , @xmath80 , @xmath81 , @xmath82 with the chord @xmath83 . the gadget is then connected to the neighbours of @xmath37 by the the edges @xmath84 and @xmath85 . * a vertex @xmath37 with @xmath86 is replaced by a tree @xmath87 that has @xmath88 leaves , a binary root if @xmath89 is odd , and otherwise only degree @xmath6 internal vertices . each leaf is connected to two neighbours of @xmath37 such that the edges incident to @xmath37 that form a pair in @xmath90 are incident to the same leaf . if @xmath89 is odd , one of the neighbors is left and connected to the binary root . the above gadgets guarantee that the graph @xmath91 is cubic and it is @xmath11-vertex connected since @xmath22 was assumed to be @xmath11-vertex connected . we can therefore apply lemma [ lem : matching ] in order to obtain a random perfect matching @xmath92 . each edge of @xmath91 is in @xmath92 with a probability of exactly 1/3 . let @xmath15 be the set of edges obtained by restricting @xmath92 to the edges of @xmath22 in the obvious way . now @xmath15 contains each edge of @xmath22 with probability 1/3 . we complete the proof by showing that @xmath15 also satisfies properties @xmath93 and @xmath94 . as each pair of edges in @xmath90 is incident to a vertex of degree at least @xmath6 , we have , by the construction of the gadgets , that they are incident to a common vertex in @xmath91 and hence at most one edge of each pair is in @xmath15 . finally , property @xmath94 follows from that @xmath95 is clearly a spanning eulerian multigraph of @xmath91 and compressing a set of even - degree vertices results in one vertex of even degree . equipped with the above lemma we are now ready to prove the main result of this section . theorem [ thm : main ] pick a random subset @xmath96 of edges that satisfies the properties of lemma [ lemma : sample ] . let @xmath97 be the set of those edges of @xmath15 that are removable and let @xmath98 be the set of the remaining edges of @xmath15 . consider the multigraph @xmath17 on vertex set @xmath18 and edge set @xmath99 . observe that both adding an edge and removing an edge swaps the parity of the degree of an incident vertex . we have thus from property @xmath94 of lemma [ lemma : sample ] that the degree of each vertex in @xmath17 is even . moreover , as @xmath73 is a removable pairing , property @xmath93 of lemma [ lemma : sample ] gives that @xmath17 is connected . alltogether we have that @xmath17 is an eulerian graph , i.e. , a graph - tsp solution . we continue to calculate its expected number of edges , which is latexmath:[\[\label{eq : nredges } \mathbb{e}[|e| + |\bar m_r| - @xmath16 , we have , by linearity of expectation , that equals @xmath101 to conclude the proof , we note that the selection of @xmath15 can be derandomized since there are , by lemma [ lemma : sample ] , polynomially many edge subsets to choose from ; taking the one that minimizes the number of edges of @xmath17 is sufficient . in order to use our framework , one of the main challenges is to find a @xmath24 that is sufficiently large . in the following , we show how to obtain a useful @xmath24 based on circulations . consider a @xmath11-vertex connected graph @xmath22 and let @xmath102 be a spanning tree of @xmath22 obtained by depth - first search ( starting from some arbitrary root @xmath66 ) . then each edge in @xmath22 connects a vertex to either one of its predecessors or one of its successors . we call the edges in @xmath102 _ tree - edges _ and those in @xmath22 but not in @xmath102 _ back - edges_. we shall now define a circulation network @xmath103 . we start by introducing an orientation of @xmath22 : all tree - edges become tree - arcs directed from the root to the leaves and all back - edges become back - arcs directed towards the root . to distinguish the circulation network and the original graphs , we use the names @xmath104 and @xmath105 for the network versions of @xmath22 and @xmath102 . in order to ensure connectivity properties of subnetworks obtained from feasible circulations , we replace some of the vertices by gadgets . for each vertex @xmath37 except the root that has @xmath62 children @xmath106 in the tree , we introduce @xmath62 new vertices @xmath107 , @xmath108 , @xmath109 , @xmath110 and replace the tree - arc @xmath111 by the tree - arcs @xmath112 and @xmath113 for @xmath114 . then we redirect all incoming back - arcs of @xmath37 from the subtree rooted by @xmath115 to @xmath116 . for an illustration of the gadget see figure [ fig : circreplace ] and for an example of a complete network see figure [ fig : circreplace_appendix ] . this way , all back - arcs start in old vertices and lead to new vertices or the root . in the following , we call the new vertices and the root _ in - vertices _ and the remaining old ones _ out - vertices_. we also let @xmath117 be the set of all in - vertices . , introduces a new vertex ( depicted in white ) and redirects back - arcs.,width=377 ] we now specify a lower bound ( demand ) and an upper bound ( capacity ) on the circulation . for each arc @xmath118 in @xmath105 , we set the demand of @xmath118 to 1 and for all other arcs to 0 . the capacity is @xmath119 for any arc . finally , the cost of a circulation @xmath120 in @xmath103 is the piecewise linear function @xmath121 $ ] , where @xmath122 is the set of incoming back - arcs of @xmath37 . one can think of the cost as the total circulation on the back - arcs except that each in - vertex accepts a circulation of @xmath32 for free . note that algorithmically there is no considerable difference whether we use our cost function or define a linear cost function on the arcs : for any in - vertex @xmath37 we can redirect all back - arcs of @xmath37 to a new vertex @xmath123 and introduce two arcs @xmath124 , one of cost @xmath34 and capacity @xmath32 and the other of cost @xmath32 and capacity @xmath119 . all remaining arcs then have a cost of @xmath34 . the following lemma shows how to use a circulation in @xmath103 to approximate graph - tsp . [ lemma : costcirc ] given a @xmath11-vertex connected graph @xmath22 and a depth first search tree @xmath102 of @xmath22 let @xmath125 be the minimum cost circulation to @xmath103 of cost @xmath126 . then there is a spanning eulerian multigraph @xmath91 in @xmath22 with at most @xmath127 edges . we first note that , for any arc of @xmath103 , the demand and the capacity is integral . therefore , applying hoffman s circulation theorem ( see @xcite , corollary 12.2a ) , we can assume the circulation @xmath125 to be integral . let @xmath128 be the support of @xmath125 in @xmath103 , i.e. , the induced subgraph of the arcs with non - zero circulation in @xmath125 , and let @xmath91 be the subgraph of @xmath22 obtained from @xmath128 by compressing the gadges of the circulation network in the obvious way . to prove the lemma , we shall first prove that graph @xmath91 is @xmath11-vertex connected and then define a removable pairing @xmath73 on @xmath91 in order to apply theorem [ thm : main ] . that @xmath91 is @xmath11-vertex connected follows from flow conservation , that each arc @xmath118 in @xmath105 has demand @xmath32 , and the design of the gadgets . indeed , if @xmath91 would have a cut vertex @xmath37 with children @xmath129 in @xmath102 then one of the subtrees , say the one rooted by @xmath115 , has no back - edges to the ancestors of @xmath37 which in turn , by flow conservation , would contradict that the tree - arc @xmath130 in @xmath105 carries a flow of at least @xmath32 . ( recall that the edge @xmath131 in @xmath102 is replaced by tree - arcs @xmath130 and @xmath113 in @xmath105 . ) we now determine a @xmath24@xmath73 on @xmath91 . for ease of argumentation we shall first slightly abuse notation and define a @xmath24@xmath132 on @xmath128 . the set @xmath133 consists of all @xmath134 such that @xmath135 is a back - arc of cost zero in @xmath128 , @xmath37 has at least two incoming arcs , and @xmath136 is a tree - arc . note that each such @xmath37 is an in - vertex , the number of incoming back - arcs of cost zero is at most one , @xmath137 is the unique outgoing tree - arc of @xmath37 , and the only possible vertex @xmath37 with only one incoming back - arc and no other incoming arc is the root . the set @xmath138 contains all edges from @xmath133 and additionally all remaining back - arcs of @xmath128 . in other words , each edge of @xmath128 that is neither in @xmath105 nor in @xmath90 is a back - arc with integer non - zero cost in the circulation or a back - arc to the root . hence , @xmath139 if the root has more than one incoming back - arc and @xmath140 otherwise . the @xmath24@xmath73 on @xmath91 is now obtained from @xmath132 , by mereley compressing the gadgets used to form @xmath103 and by dropping the orientations of the arcs . as all edges in @xmath138 are either back - arcs or they are tree - arcs starting from an in - vertex , no arc in @xmath138 is removed by the compression and thus @xmath141 and @xmath142 . moreover , @xmath91 has @xmath143 edges and , assuming @xmath73 is a valid @xmath24 , theorem [ thm : main ] yields that @xmath91 ( and thus @xmath144 has a spanning eulerian multigraph with at most @xmath145 edges . the last inequality followed from that @xmath146 is at most @xmath147 . therefore , we can conclude the proof by showing that @xmath73 is a valid @xmath24 . it is easy to verify that @xmath73 satisfies the first two conditions of definition [ def : pairing ] , that is , each edge is contained in at most one pair and the edges in each pair are incident to one common vertex of degree at least three . the third condition follows from that , for any vertex @xmath37 of @xmath91 , the vertices in the subtree @xmath87 of @xmath102 rooted by @xmath37 form a connected subgraph of @xmath91 even after removing edges according to @xmath73 . to see this we do a simple induction on the depth of @xmath37 . in the base case , @xmath37 is a leaf and the statement is clearly true . for the inductive step , consider a vertex @xmath37 with @xmath62 children @xmath148 in @xmath102 . by the inductive hypothesis , the vertices in @xmath149 for @xmath150 stay connected after the removal of edges according to @xmath73 . to complete the inductive step it is thus sufficient to verify that @xmath37 is connected to each @xmath149 after the removal of edges . if @xmath151 is not in @xmath23 this clearly holds . otherwise if @xmath152 then by the definition of @xmath153 there is an edge @xmath154 such that @xmath155 and @xmath154 is incident to @xmath37 and a vertex in @xmath149 . since at most one edge in each pair is removed we have that @xmath37 also stays connected to @xmath149 in this case , which completes the inductive step . we have thus proved that @xmath73 satisfies the properties of a @xmath24which completes the proof of the statement . we first show how to apply our framework to restricted graph classes for which we obtain a tight bound on the integrality gap of the held - karp relaxation . we then show how to use our framework to obtain an improved approximation algorithm for general graphs . we consider the class of graphs that have a degree bounded by three . [ lem : boundeddeg ] given a @xmath11-vertex - connected graph @xmath22 with @xmath21 vertices , there is a polynomial time algorithm that computes a spanning eulerian multigraph @xmath17 in @xmath22 with at most @xmath12 edges . if @xmath22 has one or two vertices , we obtain an eulerian multigraph of zero or two edges . otherwise , we compute a depth - first search tree @xmath102 in @xmath22 and determine the circulation network @xmath103 . we now show that this network has a feasible circulation @xmath120 of cost at most one . let us assign a circulation of one to each back - arc @xmath154 in @xmath103 and push it through the path in @xmath105 that is incident to both the start and end vertex of @xmath154 . by the construction of @xmath103 and from the assumption that @xmath22 is @xmath11-vertex connected , each tree - arc is in a directed cycle that contains exactly one back - arc . therefore , all demand constraints are satisfied . due to the degree - bounds , no vertex but the root has more than one incoming back - arc . the cost @xmath156 $ ] of the circulation is therefore at most one and zero if the root has only one back - arc . if the circulation cost is zero , by lemma [ lemma : costcirc ] we obtain a spanning eulerian multigraph @xmath17 in @xmath22 with at most @xmath157 edges . for those circulations where the cost is one , the proof of lemma [ lemma : costcirc ] allows to save an additional constant of @xmath158 ( since then the root has more than one incoming back - arc ) and we obtain the same bound on the number of edges . note that it is sufficient to find a 2-vertex - connected degree three bounded spanning subgraph ( a 3-trestle ) and thus , using a result from @xcite , we can apply lemma [ lem : boundeddeg ] also to claw - free graphs . applying lemma [ lemma:2conntsp ] , we obtain an upper bound of 4/3 on the integrality gap for the held - karp relaxation for the considered class of graphs . in addition , along the lines of the proof of lemma [ lemma:2conntsp ] , one can see that the above arguments imply that any connected graph @xmath22 decomposed into @xmath159 blocks , i.e. , maximal @xmath11-connected subgraphs , such that each block is either degree three bounded or claw - free , has a spanning eulerian multigraph with at most @xmath160 edges . we now apply our framework to graphs without degree constraints . we start with an algorithm that achieves an approximation ratio better than @xmath161 for graphs for which the linear programming relaxation has a value close to @xmath21 . let @xmath162 be an @xmath21-vertex graph . the support @xmath163 of an extreme point @xmath164 of @xmath53 is known to contain at most @xmath165 edges ( see theorem @xmath166 in @xcite ) . moreover , if we let @xmath164 be an optimal solution , then any @xmath66-approximate solution to graph @xmath167 with respect to @xmath168 is an @xmath66-approximate solution to @xmath22 with respect to @xmath54 , because @xmath169 and @xmath170 . we can thus restrict ourselves to @xmath21-vertex graphs with at most @xmath165 edges and , by lemma [ lemma:2conntsp ] , we can further assume the graph to be @xmath11-vertex connected . a 2-vertex - connected graph @xmath22 with @xmath21 vertices and at most @xmath165 edges . obtain an optimal solution @xmath164 to @xmath53 . obtain a depth - first - search tree @xmath102 of @xmath22 by starting at some root and in each iteration pick , among the possible edges , the edge @xmath154 with maximum @xmath171 . solve the min cost circulation problem @xmath103 to obtain a circulation @xmath125 with cost @xmath126 . apply lemma [ lemma : costcirc ] to find a spanning eulerian multigraph with less than @xmath172 edges . to analyze the approximation ratio achieved by algorithm [ alg : allgraphs ] , we bound the cost of the circulation . [ lemma : circcost ] we have @xmath173 . for notational convenience , when considering an arc @xmath118 in the flow network we shall slightly abuse notation and use @xmath174 to denote the value of the corresponding edge in @xmath22 according to the optimal lp - solution @xmath164 . we prove the statement by defining a fractional circulation @xmath120 of cost at most @xmath175 . the circulation @xmath120 will in turn be the sum of two circulations @xmath176 and @xmath177 . we obtain the circulation @xmath176 as follows : for each back - arc @xmath118 we push a flow of size @xmath178 $ ] along the cycle formed by @xmath118 and the tree - arcs in @xmath105 . we shall now define the circulation @xmath177 so as to guarantee that @xmath120 forms a feasible circulation , i.e. , one that satisfies the demands @xmath179 for each @xmath180 . as out- and in - vertices are alternating in @xmath105 and in - vertices have only one child in @xmath105 and no outgoing back - edges , a sufficient condition for @xmath120 to be feasible can be seen to be @xmath179 for each @xmath181 that is from an out - vertex to an in - vertex . to ensure this , we now define @xmath177 as follows . for each vertex @xmath37 of @xmath22 that is replaced by a gadget consisting of an out - vertex @xmath37 and a set @xmath182 of in - vertices , we push for each @xmath183 a flow of size @xmath184 $ ] along a cycle that includes the arc @xmath185 ( and one back - arc ) . note that such a cycle is guaranteed to exist since @xmath22 was assumed to be @xmath11-vertex connected . from the definition of @xmath177 , we have thus that @xmath186 defines a feasible circulation . we proceed by analyzing the cost of @xmath120 , i.e. , @xmath187 $ ] , where @xmath117 is the set of all in - vertices and @xmath122 is the set of incoming back - arcs of @xmath188 . note that the cost is upper bounded by @xmath189 + \sum_{v\in \mathcal{i } } f''(b(v))$ ] and we can thus analyze these two terms separately . we start by bounding the second summation and then continue with the first one . if @xmath190 then one can see that @xmath191 . moreover , [ claim : firstcost ] we have @xmath192 . when considering a vertex @xmath37 as done above in the definition of @xmath177 , the flow pushed on back - arcs is @xmath193 $ ] which equals @xmath194 , where @xmath195 . letting @xmath196 be the set of vertices of @xmath22 in the subtree of the undirected tree @xmath102 rooted by the child of @xmath197 , we have , by the definition of @xmath176 , @xmath198 = x^*(\delta(t_w ) \setminus \delta(v)).\ ] ] the second equality follows from that if @xmath199 for some @xmath200 then @xmath201 and hence @xmath202 . we have thus @xmath203 as we are considering a depth - first - search tree ( see figure [ fig : circcostola ] ) , @xmath204 : both the left - hand - side and the right - hand - side of the equality express two times the value of the fat edges.,width=264 ] since by the feasibility of @xmath164 each of the sets corresponds to a cut of fractional value at least @xmath11 we use @xmath205 as a lower bound on . summarizing the above calculations yields @xmath206 repeating this argument for each @xmath37 we have @xmath207 , which equals @xmath208 since @xmath209 . we proceed by bounding @xmath189 $ ] from above . [ claim : secondcost ] we have @xmath189 \leq ( 7 - 6\sqrt{2})n + 4(\sqrt{2}-1){\ensuremath{opt_{lp}(g)}}$ ] to analyze this expression we shall use two facts . first @xmath22 has at most @xmath165 edges , and therefore the number of back - arcs is at most @xmath210 . second , as the depth - first - search chooses ( among the available edges ) the edge @xmath118 with maximum @xmath174 in each iteration , we have that @xmath211 for each @xmath212 where @xmath213 is the outgoing tree - arc of @xmath214 . moreover , as @xmath215 $ ] for each back - arc , the number of back - arcs in @xmath122 is at least @xmath216}\right\rceil$ ] . combining these two facts gives us that @xmath217}\right\rceil \leq n.\ ] ] for @xmath188 , we partition @xmath218 into @xmath219 $ ] and @xmath220 . furthermore , let @xmath221 . with this notation we can upper bound @xmath222 $ ] by @xmath223 + u^*\ ] ] and relax inequality to @xmath224 the cost ( where we ignore @xmath225 ) subject to can now be interpreted as a knapsack problem of capacity @xmath226 that is packed with an item of profit @xmath227 $ ] and size @xmath228 for each @xmath214 . consequently , we can upper bound by considering the fractional knapsack problem with capacity @xmath229 and infinitely many items of a maximized profit to size ratio . associating a variable @xmath230 with @xmath231 and @xmath102 with @xmath232 this ratio is @xmath233 for any @xmath102 the ratio is maximized by letting @xmath234 and we can thus restrict our attention to items with profit to size ratio @xmath235 . a simple analysis ( see appendix [ sec : maxsizratio ] ) shows that the maximum is achieved when @xmath236 . therefore , the profit is upper bounded by @xmath237 as the fractional degree of a vertex @xmath37 that is replaced by a gadget with a set @xmath182 of in - vertices is at least @xmath238 , we have @xmath239 . hence , @xmath240 which equals @xmath241 . finally , by summing up the bounds given by claim [ claim : firstcost ] and claim [ claim : secondcost ] we bound the cost of @xmath120 and hence @xmath126 from above by @xmath242 which equals @xmath175 . having analyzed algorithm [ alg : allgraphs ] , we are ready to prove our main algorithmic result . * [ thm : approximationratio ] * ( restated ) _ there is a polynomial time approximation algorithm for graph - tspwith performance guarantee @xmath243 . _ by lemma [ lemma:2conntsp ] and the discussion before algorithm [ alg : allgraphs ] , we can restrict ourselves to @xmath21-vertex graphs that are @xmath11-vertex connected and have at most @xmath165 edges . the statement now follows by using algorithm [ alg : allgraphs ] if @xmath244 is close to @xmath21 and otherwise by using christofides algorithm . on the one hand , since christofides algorithm returns a solution with at most @xmath245 edges ( see @xcite for an analysis of christofides algorithm in terms of @xmath54 ) , it has an approximation guarantee of at most @xmath246 on the other hand , by lemma [ lemma : circcost ] , the approximation guarantee of algorithm [ alg : allgraphs ] is at most @xmath247 in particular , the approximation guarantee of algorithm [ alg : allgraphs ] for a graph @xmath22 with @xmath248 is @xmath249 but deteriorates as @xmath54 increases . the approximation guarantee of christofides algorithm on the other hand is getting better and better as @xmath54 increases . and christofides algorithm depending on the ratio @xmath250.,width=377 ] comparing these two ratios , one gets that the worst case happens when @xmath251 ( see figure [ fig : ratios ] ) and , by using simple arithmetics , the approximation guarantee can be seen to be @xmath252 . [ sec : tspp ] in this section , we describe a sequence of generalizations and modifications of the techniques that we previously presented for graph - tspand conclude with improved approximation algorithms for the traveling salesman path problem on graphic metrics , graph - tspp . we can obtain a natural generalization of @xmath253 to graph - tsppby distinguishing whether the end vertices @xmath254 and @xmath255 are in the same set of vertices . to this end , let @xmath256 . then the relaxation can be written as @xmath257 \min & \sum_{e\in e } x_e \\[2 mm ] x(\delta(s ) ) & \geq 2 , & \emptyset \neq s \subset v , s \in \phi\\[2 mm ] x(\delta(s ) ) & \geq 1 , & \emptyset \neq s \subset v , s \notin \phi\\[2 mm ] x & \geq 0 . \end{aligned}\ ] ] we denote the optimum of this generalized linear program by @xmath258 . it is not hard to see that @xmath259 . the graph on the right - hand - side in figure [ fig : intgap ] has a fractional solution such that the integrality gap of @xmath260 is lower bounded by 1.5 . and @xmath4 , respectively.,width=453 ] for a given graph @xmath13 , let @xmath261 be the graph obtained from @xmath22 by inserting @xmath262 . note that , given any solution @xmath45 to @xmath260 , we can obtain a feasible solution to @xmath263 by adding 1 to @xmath264 . this way , for each of the cuts where @xmath265 , we have @xmath266 and thus @xmath267 . in the following , we will generalize our results for graph - tspby using @xmath268 as lower bound . similar to graph - tsp , we observe that the difficulty in approximating graph - tspplies in approximating those instances that are @xmath11-vertex connected . the proof of this lemma can be found in appendix [ app:2conntsp ] . * [ lemma:2conntsp ] * ( generalized ) _ let @xmath22 be a graph and let @xmath269 be an algorithm that , given a @xmath11-vertex - connected subgraph @xmath17 of @xmath22 and @xmath270 , returns a graph - tspp solution to @xmath271 with cost at most @xmath272 . then there is an algorithm @xmath273 that returns a graph - tspp solution to @xmath274 for any @xmath275 with cost at most @xmath276 . furthermore , the running time of @xmath273 is a polynomial in the running time of @xmath269 . _ we generalize the framework to the problem graph - tspp . we obtain an approximation ratio that depends on @xmath277 , the distance of @xmath254 and @xmath255 . therefore we can see the variant of theorem [ thm : main ] for graph - tspas a special case where @xmath254 and @xmath255 have the distance 0 . * [ thm : main ] * ( generalized ) _ given a @xmath11-vertex connected graph @xmath13 with a @xmath24 @xmath73 and @xmath278 , there is a polynomial time algorithm that returns a spanning subgraph @xmath17 of @xmath22 with an eulerian path between @xmath254 and @xmath255 with at most @xmath279 edges . _ a graph has an eulerian path between @xmath254 and @xmath255 if and only if it is connected and the multigraph obtained by adding the edge @xmath262 is a spanning eulerian subgraph . therefore , we basically want to apply ( the original ) theorem [ thm : main ] and swap the degree of @xmath254 and @xmath255 . to this end we create the graph @xmath167 from @xmath22 by adding the edge @xmath137 to @xmath280 if it is not already present in @xmath22 . then we apply theorem [ thm : main ] to @xmath91 with the removable pairing @xmath153 to obtain the spanning eulerian subgraph @xmath281 . if the eulerian graph @xmath281 contains exactly one copy of @xmath137 , we simply remove it to obtain @xmath17 . this case appears if and only if @xmath137 was not chosen during the sampling , which happens with a probability of @xmath158 . note that the 2-edge - connectedness ensures that the removal does not disconnect @xmath281 . otherwise , with probability @xmath16 , @xmath281 contains either two copies of @xmath137 if @xmath282 or none if @xmath283 . in either case we obtain @xmath17 from @xmath281 by removing all copies of @xmath137 and adding a shortest path of length exactly @xmath277 to @xmath281 . if @xmath284 , we add a path with probability @xmath16 and apart from that we only remove edges ; the claimed result follows immediately . if @xmath285 , it is also not in @xmath23 and thus the path is added if and only if two edges are removed . furthermore , with probability @xmath158 , one edge is removed . then the expected number of edges in @xmath17 is @xmath286 both the removal of @xmath137 and adding the shortest path swaps the parities of @xmath254 and @xmath255 , but of no other vertex . by using the generalized theorem [ thm : main ] within the proof of lemma [ lemma : costcirc ] , we obtain immediately the following generalization . * [ lemma : costcirc ] * ( generalized ) _ given a @xmath11-vertex connected graph @xmath22 , two vertices @xmath287 in @xmath22 , and a depth first search tree @xmath102 of @xmath22 , let @xmath125 be the minimum cost circulation to @xmath103 of cost @xmath126 . then there is a spanning multigraph @xmath17 of @xmath22 that has an eulerian path between @xmath254 and @xmath255 with at most @xmath288 edges . _ we are now equipped with the right tools to obtain algorithmic results for graph - tspp . * [ thm : approximationratiohpp ] * ( restated ) _ for any @xmath27 , there is a polynomial time approximation algorithm for graph - tsppwith performance guarantee @xmath289 _ _ if furthermore each block of the given graph is degree three bounded , there is a polynomial time approximation algorithm for graph - tsppwith performance guarantee @xmath29 , for any @xmath30_. by the generalized variant of lemma [ lemma:2conntsp ] , it is sufficient to show the theorem assuming that @xmath22 is 2-vertex connected . if @xmath22 is degree three bounded , we apply lemma [ lem : boundeddeg ] on @xmath22 , but use the generalized version of lemma [ lemma : costcirc ] to obtain a solution to graph - tsppthat has at most @xmath290 edges . additionally we may replace @xmath277 by @xmath291 , since in 2-vertex - connected graphs with more than two vertices there are two vertex - disjoint paths between @xmath254 and @xmath255 . to obtain the claimed approximation ratio , we use the trivial lower bound @xmath292 of @xmath293 . for any @xmath294 , we determine a constant @xmath295 such that , for all @xmath296 , the approximation ratio is bounded from above by @xmath297 . if the graph has fewer than @xmath295 vertices , we compute an optimal solution in constant time . we continue with the case of general unweighted graphs . as in the previous subsections , @xmath262 . we apply algorithm [ alg : allgraphs ] to obtain a circulation @xmath298 of @xmath261 such that , by lemma [ lemma : circcost ] , @xmath299 . using this circulation , we apply the generalized version of lemma [ lemma : costcirc ] . however , if @xmath285 and it is used in the solution ( i.e. , it was added as a shortest path ) , we have to replace @xmath137 by a shortest path between @xmath254 and @xmath255 in @xmath22 . this is equivalent to using @xmath277 from @xmath22 instead of @xmath91 in lemma [ lemma : costcirc ] . therefore , in the following @xmath277 always refers to the distance in @xmath22 and we obtain a solution to graph - tsppof at most @xmath300 edges . in the following , let @xmath301 and @xmath302 . then , using the lower bound @xmath268 on @xmath293 , the approximation ratio achieved by our algorithm is at most @xmath303 where @xmath304 and @xmath305 . in the following calculations , we omit @xmath306 , since it decreases with the input size . similarly , we assume @xmath307 . we will consider the deviation , however , in the final result . since ( [ eqn : hppapprox ] ) depends on @xmath308 , similar to the case of graph - tspwe employ a second algorithm to obtain an upper bound independent of @xmath308 . let @xmath269 be the following simple approximation algorithm for graph - tsppwhich can be considered folklore . first , @xmath269 computes a spanning tree @xmath102 of cost @xmath292 in @xmath22 . then @xmath269 doubles all edges but those on the unique path between @xmath254 and @xmath255 in @xmath102 . the output of @xmath269 is clearly a valid solution to graph - tsppand it computes a solution of at most @xmath309 edges . similar to ( [ eqn : hppapprox ] ) , this results in an approximation ratio of at most @xmath310 note that for @xmath311 and @xmath312 , disregarding @xmath294 , ( [ eqn : hppapproxtwo ] ) is the approximation ratio we are aiming for . any increase of @xmath308 or @xmath313 can only improve this ratio . therefore we may restrict the analysis to values of @xmath313 in the range @xmath314 $ ] . we will first analyze the approximation ratio depending on @xmath313 and determine afterwards the value of @xmath313 where the minimum of the two approximation ratios is maximized . for any fixed @xmath313 within the considered range , ( [ eqn : hppapprox ] ) is monotonically increasing with respect to @xmath308 , whereas ( [ eqn : hppapproxtwo ] ) is monotonically decreasing . since we are interested in the minimum of the ratios , in the worst case both ratios are equal . this happens when @xmath315 we now replace @xmath308 by ( [ eqn : equalhpp ] ) in ( [ eqn : hppapproxtwo ] ) to obtain the worst case approximation ratio depending on @xmath313 @xmath316 since this ratio can be seen to be monotonically increasing with respect to @xmath313 within the considered range , the worst case appears when @xmath317 , and thus we obtain as upper bound on the approximation ratio @xmath318 to conclude the proof , we still have to consider @xmath319 and the case where @xmath320 . for any @xmath30 , we determine an @xmath295 based on @xmath319 and @xmath308 similar to the degree bounded case and solve graph - tsppon graphs with fewer than @xmath295 vertices exactly . altogether , we obtain an approximation ratio of at most @xmath321 ( see also figure [ fig : ratioshpp ] ) . ) and ( [ eqn : hppapproxtwo ] ) depending on @xmath313 and @xmath308.,width=377 ] we have introduced a framework of removable pairings to find eulerian multigraphs . this framework proved to be useful to obtain an approximation algorithm for graph - tspwith an approximation ratio smaller than @xmath0 and to obtain a tight upper bound on the integrality gap of the held - karp relaxation for a restricted class of graphs that contains degree three bounded and claw - free graphs . in particular , we showed that in subcubic @xmath11-vertex - connected graphs we can always find a solution to graph - tspof at most @xmath12 edges , which settles a conjecture from @xcite affirmatively . our framework is not restricted to graph - tsp . with the same techniques and a more detailed analysis , our result translates to the traveling salesman path problem on graphic metrics with prespecified start and end vertex . in this way , one is guaranteed to obtain an approximation ratio smaller than @xmath2 and , for the degree three bounded case , the approximation ratio gets arbitrarily close to @xmath4 . we note that the framework of removable pairings is straightforward to generalize to general metrics , but the problem of finding a large enough removable pairing in such graphs in order to improve on christofides algorithm remains open . 10 sanjeev arora . polynomial time approximation schemes for euclidean traveling salesman and other geometric problems . 45:753782 , 1998 . francisco barahona . fractional packing of t - joins . , 17:661669 , 2004 . sylvia boyd , rene sitters , suzanne van der ster , and leen stougie . on cubic and subcubic graphs . in _ proc . of the 15th conference on integer programming and combinatorial optimization ( ipco 2011 ) _ , 2011 . to appear . nicos christofides . worst - case analysis of a new heuristic for the travelling salesman problem . technical report 388 , graduate school of industrial administration , carnegie - mellon university , 1976 . grard cornujols , jean fonlupt , and denis naddef . the traveling salesman problem on a graph and some related integer polyhedra . , 33:127 , 1985 . jack edmonds . maximum matching and a polyhedron with @xmath322 vertices . , 69:125130 , 1965 . greg n. frederickson and joseph jaja. on the relationship between the biconnectivity augmentation and travelling salesman problems . , 19(2):189 201 , 1982 . david gamarnik , moshe lewenstein , and maxim sviridenko . an improved upper bound for the tsp in cubic 3-edge - connected graphs . , 33(5):467474 , 2005 . shayan oveis gharan , amin saberi , and mohit singh . a randomized rounding approach to the traveling salesman problem . preprint , 2011 . michel x. goemans and dimitris j. bertsimas . on the parsimonious property of connectivity problems . in _ proceedings of the 1st annual acm - siam symposium on discrete algorithms ( soda 1990 ) _ , pages 388396 , 1990 . michel x. goemans . worst - case comparison of valid inequalities for the tsp . , 69:335349 , 1995 . michelangelo grigni , elias koutsoupias , and christos h. papadimitriou . an approximation scheme for planar graph tsp . in _ proc . of the 36th annual symposium on foundations of computer science ( focs 1995 ) _ , pages 640645 , 1995 . martin grtschel , lszlo lovsz , and alexander schrijver . , volume 2 of _ algorithms and combinatorics_. springer , 1988 . michael held and richard m. karp . the traveling - salesman problem and minimum spanning trees . , 18:11381162 , 1970 . j. a. hoogeveen . analysis of christofides heuristic : some paths are more difficult than cycles . , 10(5):291295 , 1991 . atsushi kaneko , alexander kelmans , and tsuyoshi nishimura . on packing 3-vertex paths in a graph . , 36(4):175197 , 2001 . joseph s. b. mitchell . guillotine subdivisions approximate polygonal subdivisions : a simple polynomial - time approximation scheme for geometric tsp , k - mst , and related problems . , 28:12981309 , march 1999 . c. l. monma , b. s. munson , and w. r. pulleyblank . minimum - weight two - connected spanning networks . , 46:153171 , 1990 . denis naddef and wiliam r. pulleyblank . matchings in regular graphs . , 34(3):283291 , 1981 . christos h. papadimitriou and santosh vempala . on the approximability of the traveling salesman problem . , 26(1):101120 , 2006 . alexander schrijver . . springer , 2003 . david b. shmoys and david p. williamson . analyzing the held - karp tsp bound : a monotonicity property with application . , 35(6):281285 , 1990 . laurence a. wolsey . heuristic analysis , linear programming and branch and bound . in _ combinatorial optimization ii _ , volume 13 of _ mathematical programming studies _ , pages 121134 . springer , 1980 . of a graph @xmath22 with depth - first - tree @xmath102 . in - vertices and out - vertices of the circulation network is depicted in white and black , respectively.,width=453 ] * [ lemma:2conntsp ] * ( restated ) _ let @xmath22 be a graph and let @xmath269 be an algorithm that , given a @xmath11-vertex - connected subgraph @xmath17 of @xmath22 and @xmath270 , returns a graph - tspp solution to @xmath271 with cost at most @xmath272 . then there is an algorithm @xmath273 that returns a graph - tspp solution to @xmath274 for any @xmath275 with cost at most @xmath276 . furthermore , the running time of @xmath273 is a polynomial in the running time of @xmath269 . _ 1 . if @xmath22 is @xmath11-vertex connected then return the graph - tspp solution obtained by running @xmath269 on @xmath274 . 2 . otherwise , let @xmath37 be a cut vertex whose removal results in components @xmath324 with @xmath325 . recursively run @xmath323 on the @xmath326 sub - instances @xmath327 and return the union of the obtained solutions , where @xmath328 denotes the subgraph of @xmath22 induced by @xmath329 , @xmath330 as a vertex is selected to be a cut vertex at most once , @xmath323 terminates in time bounded by a polynomial in the running time of @xmath269 . it remains to verify that it returns a graph - tspp solution to @xmath274 with cost at most @xmath331 . we do so by induction on the depth of the recursion . in the base case no recursive calls are made so the solution is that returned by @xmath269 which by assumption is a graph - tspp solution to @xmath274 with cost at most @xmath331 . now consider the inductive step when a cut vertex @xmath37 of @xmath22 is selected whose removal results in components @xmath332 with @xmath333 . let @xmath334 be the multiset of edges of the obtained graph - tspp solution to @xmath335 . with this notation the edge set returned by @xmath273 is @xmath336 and we need to prove that we start by proving ( a ) . by the induction hypothesis , the edge set @xmath340 forms a spanning eulerian subgraph of @xmath328 and , consequently , @xmath341 forms a spanning eulerian subgraph of @xmath22 . that @xmath342 is a spanning eulerian subgraph of @xmath22 now follows from that the endpoints of @xmath343 can be partitioned so that one is @xmath254 , one is @xmath255 and the remaining @xmath344 endpoints are @xmath37 ( possibly not different from @xmath254 and @xmath255 ) . we proceed by proving ( b ) . by the induction hypothesis , @xmath345 and it is thus sufficient to prove @xmath346 . to this end , let @xmath45 be an optimal solution to @xmath260 and let @xmath347 denote its restriction to the subgraph @xmath328 with start vertex @xmath348 and end vertex @xmath349 . by the definition of @xmath350 and the fact that @xmath37 is a cut vertex , it is easy to see that each constraint in @xmath351 has an identical constraint in @xmath260 . therefore , @xmath347 corresponds to a solution to @xmath352 and hence @xmath353 which completes the inductive step and the proof of the lemma . we verify that @xmath354 is obtained when @xmath355 . let @xmath356 and consider its first derivative @xmath357 from this it follows that @xmath358 when @xmath359 it is now easy to verify that the unique maximum of @xmath360 for @xmath361 is obtained when @xmath355 .

we study magnetism on the fcc lattice with tetragonal distortions , with general exotic directional magnetic interactions allowed by symmetry . we consider two models , corresponding to a uniform tetragonal distortion , or a two - sublattice model with a tetragonal screw axis . we establish their low temperature phase diagrams in the semi - classical limit using classical optimization and consideration of fluctuations both analytically and by monte carlo simulation . both order by disorder and exchange anisotropy mechanisms favor a easy axis for magnetization . we also show that spin - lattice coupling can give rise to an intermediate temperature paramagnetc nematic / orthorhombic phase , and discuss the transitions to / from this state . these results are relevant to a family of insulating magnetic double perovskites , and find immediate application to the ferromagnet banaoso .

1408.6785

spin-1/2 antiferromagnets are of considerable intrinsic interest because of their potential for strong quantum effects . of special value conceptually are `` pure '' spin systems without orbital degeneracy , where the magnetism can be isolated from other phenomena such as the jahn - teller effect and kugel - khomskii exchange . strong spin - orbit coupling ( soc ) , as is present in heavy 5d transition metal ions , provides a novel means of `` purifying '' orbitally degenerate spins by spin - orbital entanglement . this mechanism leads to novel directional - dependent exchange coupling of spins , as has discussed in many iridates @xcite and a wide family of insulating double perovskites . in this paper , we discuss a specific example of the latter , in which s=1/2 spins interact via novel anisotropic interactions on the geometrically frustrated face - centered cubic ( fcc ) lattice . we find a rich structure of magnetic phases , examples of thermal and quantum order - by - disorder , and the possibility for an intermediate nematic phase induced by spin - lattice coupling . with strong soc and ideal cubic symmetry , spins and orbitals in double perovskites with a single 5d electron per transition metal site combine to form an effective @xmath3 spin residing on a face - centered cubic ( fcc ) lattice . this `` large '' spin may still be rather quantum due to unusual multipolar interactions , but may also be reduced to an effective @xmath4 one by a structural transition or quadrupolar ordering , either of which may reduce the cubic symmetry . prior theoretical work in an @xmath3 model suggested a transition to tetragonal symmetry indeed may occur . if this transition occurs at a temperature large compared to magnetic scales , an @xmath4 hamiltonian can provide a more economical description of the latter . such a cubic to tetragonal transition has indeed been observed recently in the material ba@xmath1naoso@xmath2 at 320k . @xcite since the magnetic order sets in only around 10k in this material , it is a strong candidate for the @xmath4 description . with the general problem and this specific material as motiviation , we study here pure @xmath4 spins on a tetragonally distorted fcc lattice . we consider two models . in the first , the tetragonal state is produced by a simple expansion or contraction of the cubic @xmath5 axis . in the second , `` twisted '' model , motivated by the theory of ref . , we consider a different tetragonal state with an doubled unit cell along the @xmath5 axis . we study these models by a combination of classical analysis , thermal and quantum spin wave theory , and monte carlo simulation . we derive in this way phase diagrams over a broad parameter space , which should be applicable to a wide range of double perovskite materials . anisotropy and ( b ) canted ferromagnetic phase . the spheres denote os sites and the different colors in ( b ) indicate different ionic environments . see sec . [ sec : stagg - tetr - case ] for details . ] an interesting feature of the simple tetragonal model is a mean - field u(1 ) degeneracy of a set of ferromagnetic and antiferromagnetic states with spins ordered within the xy planes . in these states , spins may be rotated by an arbitrary global angle with no energy cost . this is an accidental degeneracy which is not required by the tetragonal @xmath6 symmetry of the model . hence one expects , and we indeed find , that the symmetry is broken by fluctuations , both classical thermal and zero point quantum ones . remarkably , fully independently of the hamiltonian parameters , these fluctuations select the four states with magnetization oriented along the @xmath0 axes [ see fig . [ fig : intro : fcc](a ) ] . this selection is in fact precisely what has been observed experimentally in ba@xmath1naoso@xmath2 as the ferromagnetic easy axis . we also show that the same @xmath0 axes are preferred in the twisted model , though in this case the selection occurs already at the classical level . in this case , the low temperature phase is a canted state with simultaneous and orthogonal staggered and uniform moments within the xy plane [ see fig . [ fig : intro : fcc](b ) ] . ) . the dot in the phase diagrams is the multicritical point connecting the paramagentic , nematic , and magnetically ordered phases . the solid diamond in ( b ) is a tricritical point ; the phase transition become first order for stronger coupling , while it remains mean - field - like for weaker coupling . ] the low temperature phase in both models has four possible domains , describable by a @xmath7 order parameter , breaking simultaneously both time - reversal symmetry and lattice rotation symmetry from tetragonal to orthorhombic . monte carlo simulations show that the transition to the @xmath7 state occurs in a single , continuous , transition . we argue this is also a consistent possibility according to renormalization group theory , and predict that the continuous transition is in the 3d xy universality class . however , if we consider the phonon degrees of freedom , another possibility emerges , since the lattice energy is sensitive to the orthorhombic distortion : spin - lattice coupling may induce an intermediate temperature phase breaking lattice rotations but not time - reversal . this would be a `` nematic '' state in modern parlance , or simply a paramagnetic orthorhombic state in more conventional terms . using the theory of critical phenomena , we argue that spin - lattice coupling may indeed induce a nematic phase , but that this occurs only beyond some non - zero threshold coupling strength . hence , the appearance of a nematic phase implies the presence of strong spin - lattice coupling . a system in which spin - lattice coupling is close to the threshold value for the onset of nematic order is governed by a universal _ multicritical _ point , as shown in figure [ fig : intro : phdiag ] . the remainder of the paper is organized as follows . in sec . [ sec : model ] , we present the uniform and staggered tetragonal models . [ sec : resa ] describes the analysis of the simpler uniform tetragonal model : its classical ground states , the effects of thermal and quantum fluctuations , and its phase transitions . next , in sec . [ sec : resb ] , we consider the staggered sublattice model , and show that it exhibits features relevant to ba@xmath1naoso@xmath2 already at the classical level , but does not show a nematic phase . spin - lattice coupling is described in sec . [ ssec : resb : gl ] . finally , we review the main results and discuss implications for experiments in sec . [ sec : discussions - summary ] . several appendices give additional details of calculations to support the main text . in this section , we introduce the models we study in this paper . in sec . [ ssec : model : cubic ] we first introduce a general model with nn interaction that is allowed by cubic symmetry . general models with symmetry - allowed nn interactions considering tetragonal symmetry are introduced in sec . [ ssec : model : tetra ] and sec . [ sec : stagg - tetr - case ] . when the fcc lattice has cubic symmetry , assuming pairwise interactions amongst @xmath4 spins , only two different interactions are allowed . the hamiltonian is given by @xmath8 here , @xmath9 is the spin operator for @xmath4 spins at site @xmath10 and @xmath11 is the vector connecting nearest - neighbor ( nn ) sites . the sum for @xmath10 is taken over all the sites and that for @xmath11 is over all nn sites @xmath12 when the symmetry is reduced to tetragonal e.g. , the lattice is shortened / elongated along @xmath5 axis the general spin model with nn interactions consist of seven interactions : @xmath13 with @xmath14 and @xmath15 here , the sums for @xmath16 in @xmath17 ( @xmath18 ) are taken only for the bonds in ( out of ) the @xmath19 planes , i.e. , @xmath20 ( @xmath21 ) with @xmath22 being the unit vector along @xmath5 axis . in @xmath18 , @xmath23 is the projection of @xmath11 on to the @xmath19 plane , @xmath24 with @xmath25 ( @xmath26 ) being the @xmath27 component of @xmath11 . the magnetic behavior of this model is studied in sec . [ sec : resa ] . tetragonal anisotropy may also occur via the presence of inequivalent sublattices . we consider the case shown in fig . [ fig : intro : fcc](b ) , where the local environment for ions alternates along the @xmath5 axis ; we call the two sublattices @xmath28 and @xmath29 . inspired by a previous study , @xcite in this two - sublattice model , we suppose there is no @xmath6 rotation symmetry ( i.e. no four - fold rotation symmetry in a single xy plane ) but instead only a @xmath6 screw axis parallel to @xmath5 axis . the hamiltonian for the two - sublattice model is given by @xmath30 where @xmath31 here , @xmath32 ( @xmath33 ) contains the bonds within one @xmath34 plane , and @xmath35 gives the interactions between the spins on different layers . the notation @xmath36 denote a term with same form as the preceding one but with the @xmath37 component replaced by the @xmath38 component . a notable difference of the hamiltonian in eq . ( [ eq : model : h2sub ] ) from the hamiltonian in eq . ( [ eq : model : htetra ] ) is the presence of ising - type interactions that alternate from layer to layer , e.g. , @xmath39 and @xmath40 . focusing on the role of these interactions , we mainly study a simplified model @xmath41 the results are presented in sec . [ sec : resb ] . in this section , we study the model in eq . ( [ eq : model : htetra ] ) . first , we investigate the classical ground state of the model by the luttinger - tisza method . @xcite we then turn to the effects of fluctuations , focusing on a particular ordered phase for concreteness . thermal and quantum fluctuations are studied in secs . [ ssec : resa : tfluc ] and sec . [ ssec : resa : qfluc ] , respectively . ) in the 2d limit obtained by the luttinger - tisza method . the phase boundaries are given in table [ tab : resa : gs ] . see the text for details . ] [ cols="^,^",options="header " , ] we first consider the magnetic phase diagram in absence of interlayer coupling , @xmath42 in eq . ( [ eq : model : htetra ] ) . the phase diagram for the independent layers is shown in fig . ( [ fig : resa:2ddiag ] ) and the phase boundaries in table [ tab : resa : gs ] . when @xmath39 is dominant and ferromagnetic , the ground state is ferromagnetic with spins pointing in @xmath34 plane . the phase is denoted as @xmath34fm in the phase diagram . on the other hand , when @xmath39 is dominant but antiferromagnetic , a neel state with spins pointing in @xmath34 plane become the ground state ( @xmath34afm in fig . [ fig : resa:2ddiag ] ) . an interesting feature of these phases is an accidental u(1 ) degeneracy : despite the absence of u(1 ) symmetry in the hamiltonian , the classical ground state energy is independent of the direction of the moments in the @xmath34 plane . the u(1 ) degeneracy is a feature of dominant @xmath39 interactions . if instead @xmath43 is dominant , spins align along the @xmath5 axis , either ferromagnetically ( @xmath44 ) or antiferromagnetically ( @xmath45 ) depending upon the sign of interaction . when @xmath46 , these four phases meet at @xmath47 . on the other hand , with infinitesimally small @xmath48 , a stripe order appears in the competition region where the four phases meet ( see fig . [ fig : resa:2ddiag ] ) . in this stripe phase , the moments align ferromagnetically along one of the bonds , and antiferromagnetically along the orthogonal bonds . they lie in the @xmath34 plane and point along the @xmath49 direction . in the stripe phase the axis of the moments and the wave - vector are related . when @xmath50 , the direction of moments and wave - vector are parallel , while when @xmath51 , the direction of moments and the wavevector are orthogonal . in either of these cases , the ground state is four - fold degenerate . we next consider how the interlayer couplings modify the phase diagram in fig . [ fig : resa:2ddiag ] . we here focus on the @xmath34 fm state and study how it changes with @xmath52 and @xmath53 . the phase diagrams for the other regions are described in appendix [ sec : appa ] . ) in the region where in - plane ferromagnetic ( @xmath34fm ) orders become the ground state in the 2d limit . the phase diagram is obtained by the luttinger - tisza method . the gray region denotes regions where we could not determine the ground state . see the text for details . ] figure [ fig : resa:3ddiag ] shows the ground state phase diagram of hamiltonian in eq . ( [ eq : model : htetra ] ) with ( @xmath52,@xmath53 ) in the regime so that the @xmath34fm state appears in the independent layer model of fig . [ fig : resa:2ddiag ] . in this region , introduction of @xmath54 selects ferromagnetic ordering of the 2d planes ( @xmath34fm / fm state ) , while antiferromagnetic stacking of the ferromagnetic layers is selected for @xmath55 ( @xmath34fm / afm state ) . the phase boundary of the two phases are located at @xmath56 . introducing @xmath53 gives rise to phase competition between ising and @xmath34 orders . for sufficiently large negative @xmath57 , spin align ferromagnetically and parallel to the @xmath5 axis ( @xmath5fm / fm state ) . as shown in the lower side of fig . [ fig : resa:3ddiag ] , this occurs when @xmath58 while the @xmath34fm orders become the ground state for larger but negative @xmath53 . for @xmath44 , the situation is somewhat more complicated . in particular , when @xmath59 the luttinger - tisza method can not verify either an antiferromagnetic or ferromagnetic ground state . a definitive identification of the ground state in this region requires further analysis . consequently , this region is left blank in figure [ fig : resa:3ddiag ] . in principle , the luttinger - tisza method gives the minimum region of stability for a given magnetic order . hence , it is expected that at least some portion of this area is , in reality , covered by the magnetic orders surrounding the region . for large positive @xmath60 , the ground state is given by antiferromagnetic stacking of the @xmath5fm layers ( @xmath5fm / afm ) , and otherwise the phase diagram for @xmath60 mirrors that for @xmath57 , as shown in fig . [ fig : resa:3ddiag ] . in sec . [ ssec : resa : gs ] , we found a wide region of magnetic phases with spins oriented within the @xmath34 plane , in several of which , despite the magnetic anisotropy of the hamiltonian , a u(1 ) degeneracy under rotations within the plane is recovered . however , as there is no symmetry that protects this degeneracy , it is expected to be lifted by perturbations . indeed , as we will show in the rest of this section , both thermal and quantum fluctuations lift this degeneracy selecting @xmath49 directions . in this section , taking the @xmath34fm / afm phases as a prototypical example , we consider how thermal fluctuations modify this classical degeneracy . in the classical limit , the stacking along @xmath5 axis does not matter , since the transformation @xmath61 changes the sign of @xmath52 in the model in eq . ( [ eq : model : htetra ] ) . here , @xmath62 is an integer . in sec . [ ssec : resa:1loop ] , we first consider the one - loop correction to the classical spin model due to thermal fluctuations at low temperature . we show that magnetic anisotropy is induced by @xmath63 . we then verify and study anisotropy in higher temperature by a classical mc simulation in sec . [ ssec : resa : mc ] . let us first consider the effect of thermal fluctuations in the region @xmath64 . in this limit , each spin is close to its ground state orientation . hence , we may expand in small fluctuations @xmath65 around their average magnetic moment @xmath66 , @xmath67 the distribution function at @xmath64 can be approximated as @xcite @xmath68 where @xmath69 is the partial distribution function with the net magnetic moment pointing along @xmath70 , and @xmath71 is the fourier transform of @xmath65 . similarly , @xmath72 is the fourier transform of the @xmath73 interaction matrix @xmath74 , where general form of hamiltonian is given by @xmath75 as we focus on the low @xmath76 limit , we may approximate @xmath77 calculating the gaussian integral in eq . ( [ eq:1loop_z ] ) with the orthogonal constraint for @xmath65 gives free energy for the ordered state with net magnetic moment @xmath70 as @xmath78 + \log { \cal j}_2({\bf k})\nonumber\\.\end{aligned}\ ] ] here , @xmath79 - 2(2j_1+j_3 + 4k_1)\\ { \cal j}_1^\prime({\bf k } ) & = & -4j_3\sin\frac{k_x}{\sqrt2}\sin\frac{k_y}{\sqrt2}\\ { \cal j}_2({\bf k } ) & = & 4j_2\cos\frac{k_x}{\sqrt2}\cos\frac{k_y}{\sqrt2 } + 4k_2 \cos\frac{k_z}{\sqrt2}[\cos k_x + \cos k_y ] - 4(j_2 + 2k_2)\end{aligned}\ ] ] and @xmath80 the result clearly shows that @xmath81 depends upon the angle @xmath82 , indicating that thermal fluctuations break the u(1 ) degeneracy of the ground state . the location of the minimum is determined by the vanishing derivative with respect to @xmath82 , which is given by @xmath83 as @xmath84 for @xmath85 , when @xmath86 , the sum over @xmath87 is always positive and finite . hence , @xmath88 gives the location of the maxima / minima ; @xmath89 gives the minimum of @xmath81 . thus , thermal fluctuations favor @xmath49 magnetic anisotropy irrespective of the sign of @xmath63 , reducing the u(1 ) symmetry of the ordered phase to @xmath7 . we observe that if @xmath46 , @xmath81 becomes independent of @xmath70 . this is natural as @xmath39 and @xmath43 do not break the u(1 ) symmetry . ) . temperature dependence of ( a ) @xmath90 and ( b ) @xmath91 . the inset in ( a ) shows the binder parameter for @xmath92 . ] this spin anisotropy , induced by thermal fluctuations , appears over a wide range of temperatures . this is confirmed by numerical simulation using classical mc method . [ fig : mc_tetra ] shows a result of the mc simulation for @xmath93 , @xmath94 , and @xmath95 ; the other parameters are set to zero . figure [ fig : mc_tetra](a ) shows temperature dependence of staggered magnetization @xmath96 calculated for different system sizes . here , @xmath97 with @xmath26 is the @xmath27th component of the staggered magnetization . the result shows an increase of @xmath98 below @xmath99 , indicating magnetic phase transitions . the critical temperature @xmath100 is estimated from the binder parameter @xcite for @xmath92 . the inset in fig . [ fig : mc_tetra](a ) shows the binder parameter calculated in the vicinity of @xmath100 . the result shows a monotonic decrease with increasing temperature with a crossing at @xmath101 . this is an indicative of second order phase transition . to study the magnetic anisotropy in the ordered phase at @xmath102 , we calculated @xmath103 the parameter @xmath91 becomes positive when the magnetic moments preferentially point along the @xmath104 directions while it becomes negative when the spins point along @xmath49 . on lowering temperature below @xmath100 , @xmath91 gradually deviates from zero to negative value @xmath105 , as shown in fig . [ fig : mc_tetra](b ) . this indicates that spins align along the @xmath0 directions , consistent with the one - loop analysis above . however , in contrast to @xmath98 , @xmath91 does not show a sharp increase at @xmath100 . also , @xmath91 shows a strong finite size effect even at temperatures @xmath64 . to understand this , we note that in the xy model , the cubic - anisotropy term is a dangerously irrelevant parameter , @xcite i.e. irrelevant at the critical point but relevant for @xmath102 . this implies the existence of a second length scale @xmath106 which is _ larger _ than the usual correlation length . hence , when the system size is small , the large length @xmath106 may obscure the anisotropic behavior . indeed , similar behavior was reported in a recent mc simulation of a 3d xy model with @xmath107 anisotropy . @xcite and ( b ) @xmath38 axes at @xmath108 , @xmath94 , and @xmath109 . the insets in ( a ) and ( b ) shows the field dependence of the binder parameter for staggered magnetization along the @xmath37 and @xmath38 axes , respectively . ] next , we study the effect of external magnetic field , which leads , we will see , to features that reflect the fluctuation - induced anisotropy . fig . [ fig : mc_mag ] shows the field dependence of the staggered magnetization along different crystal axes at @xmath110 . here , we used parameters @xmath108 , @xmath94 , and @xmath111 , and the external field was applied along the @xmath112 $ ] direction . with increasing field , the staggered magnetization along the @xmath37 axis vanishes first , which is seen by the rapid decrease of the magnetization to zero at @xmath113 [ fig . [ fig : mc_mag](a ) ] . this phase transition is also observed in the magnetic susceptibility @xmath114 . here , we calculated @xmath115 ( @xmath26 ) , by the fluctuation formula , @xmath116 with @xmath117 the result is also shown in fig . [ fig : mc_mag](a ) , which shows diverging behavior at @xmath118 . with further increase of the field , the @xmath38 component of staggered magnetization vanishes via a second transition at @xmath119 ; a similar divergence of susceptibility is also observed [ fig . [ fig : mc_mag](b ) ] . the critical fields are estimated from the binder parameters for @xmath120 and @xmath121 , which are shown in the insets . and @xmath122 for hamiltonian in eq . ( [ eq : model : htetra ] ) at @xmath108 , @xmath94 , and @xmath109 . the inset shows an enlarged view of the low field region . the solid lines are guides for the eye , and the dotted line is a sketch of the expected phase boundary in the presence of quantum fluctuations . ] figure [ fig : mc_diag ] shows the @xmath123 phase diagram for @xmath108 , @xmath94 , and @xmath109 . the upper boundary @xmath124 monotonically decreases with increasing temperature in a conventional fashion . on the other hand , @xmath125 shows a non - monotonic behavior with an increase with increasing temperature for @xmath64 and a maximum around @xmath126 . this behavior is characteristic of thermal - fluctuation induced anisotropy , since the spin anisotropy weakens at low temperature , hence @xmath125 approaches @xmath127 as @xmath128 . the non - monotonic behavior can therefore be used as an experimental diagnostic of thermal fluctuation - induced anisotropy . lastly , we briefly mention the behavior of the @xmath34fm / fm state in mangetic field . the above argument on magnetic anisotropy remains applicable since the @xmath34fm / afm case can be related using a transformation mentioned in the begining of this section . this correspondance is , however , no longer valid in the presence of the external magnetic field . in this case , it is expected that there is a single transition at @xmath129 , above which the net magnetic moment aligns parallel to the external field . as the anisotropy is driven by the fluctuations , the temperature dependence of @xmath129 is expected to be non - monotonic , similar to @xmath118 shown in fig . [ fig : mc_diag ] . we next focus on the effect of quantum fluctuations , which are expected to be substantial and dominate at low temperature for @xmath4 quantum spins . by a spin - wave analysis , we show that quantum fluctuations act similarly to thermal fluctuations , also giving rise to four - fold magnetic anisotropy favoring the @xmath49 directions . in the antiferromagnetic spin wave theory , the ground state energy is given by @xmath130 where @xmath131 is the classical ground state energy and @xmath132 is the spin - wave dispersion for @xmath27 mode ( @xmath133 ) . the latter two terms give the quantum correction to the energy . in the current model , @xmath134 is given by @xmath135 where @xmath136 the minimum of @xmath137 with respect to @xmath82 is determined from the derivative @xmath138 , which is given by @xmath139 expanding the last bracket in eq . ( [ eq : qcorr ] ) in @xmath140 gives @xmath141 with @xmath142 here , the sum is for all the odd natural numbers , @xmath143 and @xmath144 as @xmath145 and @xmath146 for arbitrary @xmath147 , it is shown that @xmath148 when @xmath48 . hence , the sign of the coefficient for @xmath149 in eq . ( [ eq : de_dth ] ) is @xmath150 this indicates that @xmath137 has a minimum at @xmath151 . hence , whenever @xmath48 , the quantum fluctuations give rise to four - fold magnetic anisotropy favoring the @xmath49 directions , i.e. with the same sign as the effect of thermal - fluctuations . as quantum fluctuations dominate as @xmath152 , it is expected that they will modify the @xmath118 in fig . [ fig : mc_diag ] to remain non - zero at @xmath128 . in this section , we study the magnetic behavior of the two - sublattice model in sec . [ sec : stagg - tetr - case ] . for simplicity , in this section , we mainly consider the hamiltonian in eq . ( [ eq : model : h2s ] ) . in sec . [ ssec : resb : gs ] , we study the classical ground state properties of the model . by using a variational method , we show that a wide range of the phase diagram is dominated by canted - ferromagnetic order . the thermodynamics of the canted - ferromagnetic state is studied in sec . [ ssec : resb : mc ] using classical monte - carlo simulation . ) obtained by the two sublattice variational calculation . here fm ( neel ) denotes a ferromagnetic ( antiferromagnetic ) state with spins pointing parallel to @xmath5 axis . the spins point in the @xmath34 plane in the canted ferromagnetic ( canted - fm ) state . ] we first present the ground state phase diagram of the model in eq . ( [ eq : model : h2s ] ) . to make progress , we assume a two - sublattice structure , and find the classical ground state within this space . this is equivalent to finding the best variational product state for the s=1/2 model with a two - sublattice structure . the resulting phase diagram is shown in fig . [ fig : gstwosub ] . we observe three different phases . when @xmath153 and @xmath154 , spins align ferromagnetically along the @xmath5 axis in spin space . on the other hand , when @xmath155 and @xmath154 , we obtain antiferromagnetic order with spins aligned along @xmath5 axis , antiparallel on the two sublattices . we note that , in this region of the phase diagram , an arbitrary spin state which satisfies the two - up two - down local constraint on each tetrahedron is a ground state . hence , the ground state retains quasi - macroscopic degeneracy , at the minimum . this degeneracy is of course unstable to additional interactions . on the other hand , if @xmath156 , the phase diagram is dominated by a canted ferromagnetic state . this is a two - sublattice magnetic order in which the spins on different sublattices are non - collinear ( see fig . [ fig : intro : fcc ] ) . spins lie within the @xmath34 plane with angles given by @xmath157 for @xmath155 and @xmath158 for @xmath153 . here , @xmath159 with @xmath160 . adding the two non - collinear moments , we see that this phase sustains a net uniform magnetic moment along the @xmath49 direction . the difference of the two moments gives the staggered magnetization perpendicular to the uniform magnetic moment . the ground state has a four - fold degeneracy , comprised of state related by time - reversal symmetry and a mirror reflection through the @xmath161 plane . we note that this state is similar to one found in a previous study on a @xmath162 model . @xcite $ ] and ( b ) @xmath112 $ ] direction . @xmath163 is used as the unit of energy . ] when an external magnetic field is applied , the canted ferromagnetic phase evolves with characteristic magnetization curves that depend upon the direction of the external field . [ fig : cfm_magproc](a ) shows the field dependence of uniform magnetic moment along @xmath164 $ ] direction for different values of @xmath165 . when @xmath166 , one finds a finite magnetic moment that is dependent on @xmath167 . when the magnetic field applied along @xmath164 $ ] , the magnetic moment increases monotonically without and sharp features , approaching @xmath168 at @xmath169 . this is a consequence of the @xmath165 interaction which induces moment canting for arbitrarily small @xmath170 . to see this , we consider the @xmath171 limit , in which the leading correction to the energy from @xmath172 appears in the form @xmath173 here , @xmath174 with @xmath175 and @xmath172 is the external field applied along @xmath164 $ ] axis , @xmath176 the presence of the linear @xmath177 term in eq . ( [ eq : cfm_h110 ] ) shows that the fully polarized state at @xmath178 is unstable to infinitesimally small @xmath165 . by contrast , for a field applied along the @xmath112 $ ] axis , a sharp saturation feature appears , as shown in fig . [ fig : cfm_magproc](b ) . the figure shows @xmath112 $ ] component of the net magnetic moment . we see that the magnetization reaches its saturation value at @xmath179 this feature is a characteristic which can distinguish the canted fm state . the presence of the canted - ferromagnetic phase is confirmed by the monte carlo simulation . [ fig : cfm_mc1 ] shows the temperature dependence of specific heat and magnetic structure calculated by classical monte carlo simulation . the temperature dependence of specific heat is shown in fig . [ fig : cfm_mc1](a ) . it shows a peak at @xmath180 indicating a phase transition which is associated with the rapid increase of @xmath181 and @xmath182 [ shown in fig . [ fig : cfm_mc1](b ) and ( c ) , respectively ] . the square of the @xmath34 compoment of net magnetization @xmath181 is defined in the same way as @xmath182 [ see eq . ( [ eq : mstg_xy ] ) ] . the results indicate a transition to the canted - fm state . the transition temperature @xmath183 was estimated from the binder parameter for @xmath184 , which is shown in the inset of fig . [ fig : cfm_mc1](c ) . with the application of an external field along @xmath112 $ ] axis , the four - fold degeneracy of the ground state is lifted . however , a two - fold degeneracy remains , related to mirror symmetry in the [ 100 ] plane . hence , at low temperature , the system is expected to show a phase transition associated with spontaneous symmetry breaking that selects one of the two ground states . this is observed as a peak at @xmath185 in the specific heat shown in fig . [ fig : cfm_mc2](a ) . associated with this peak , the @xmath38 component of the staggered magnetization shown in fig . [ fig : cfm_mc2](b ) shows a rapid increase . in addition to this phase transition , in the mc simulation in fig . [ fig : cfm_mc2 ] , a crossover from the high-@xmath76 paramagnetic state to a low-@xmath76 fm - like paramagnetic state was observed . this is not a phase transition but can be regarded as a rapid crossover related to the zero field transition . in fig . [ fig : cfm_mc2](a ) , the specific heat shows a shoulder around @xmath180 . in addition , a hump is observed in the monte carlo simulation of the susceptibility of the staggered moment [ see fig . [ fig : cfm_mc2](c ) ] . -@xmath186 phase diagram in the classical limit for the hamiltonian in eq . ( [ eq : model : h2s ] ) . the magnetic field is applied along the @xmath112 $ ] direction . the solid circles indicate the phase boundary between the paramagnetic phase and canted - ferromagnetic state ; the boundary was estimated from the binder parameter . the solid and open squares indicate the crossover temperature from the high-@xmath76 paramagnetic state to intermediate paramagnetic state . solid and open symbols are estimated from the maximum of the hump in specific heat and susceptibility of the staggered moment . ] the critical and the crossover temperatures estimated from the specific heat and the hump of the susceptibility are plotted in fig . [ fig : cfm_pd ] . at @xmath166 , both the transitions take place at the same temperature : namely , a single transition to the canted - fm state is observed . with increasing @xmath172 , the temperature of onset of @xmath187 and @xmath121 separate . under ambient magnetic field , the transition temperature to the canted - fm state shows a slight decrease around @xmath188 , but is nearly independent of the applied field through out the range of calculation . on the other hand , the crossover temperature is enhanced by @xmath172 . in this section , we consider the coupling of spins to lattices . in particular , we investigate the instability of lattices with tetragonal symmetry , assuming the spins to be of xy type with @xmath6 symmetry . in sec . [ ssec : resb : gl : formula ] , we introduce the ginzburg - landau ( gl ) theory we consider throughout this section . the effect of spin - lattice coupling in the weakly - coupled regime is studied in sec . [ sec : weak - spin - lattice ] and a mean - field argument for stronger coupling is presented in sec . [ sec : elastic - mean - field ] . the expected phase diagram based on these arguments are described in sec . [ sec : phase - transitions ] . to investigate the effect of spin - lattice coupling , we begin by writing an effective gl free energy as the sum of all three terms , @xmath189 the spin part is given by @xmath190 where @xmath191 is a complex order parameter field for the spins composed of the two components of the spins @xmath192 ( @xmath193 ) , and @xmath194 is the complex conjugate of @xmath195 . @xmath196 is the complex representation of the external magnetic field with @xmath197 being the external field along @xmath27 axis . the magnetic transition of @xmath198 is tuned by the reduced temperature @xmath199 , where @xmath200 is the mean field approximation to @xmath100 . the lattice free energy describes long wavelength phonons , corresponding to the uniform distortion of the lattice . the general form of the elastic free energy allowed by tetragonal symmetry is @xmath201 with @xmath202 ( @xmath203 ) the strain tensor @xmath204 and @xmath205 is the phonon mode that corresponds to lattice distortion along @xmath193 . for simplicity , we here assumed the lattice is rigid against distortion along @xmath5 axis . the constants @xmath206 follow standard definitions . structural transitions are described within elastic theory by vanishing eigenvalues of the elastic tensor , and have been discussed for example by cowley . @xcite in particular , tetragonal to orthorhombic transitions , which may be expected based on spin physics , arise when either @xmath207 or @xmath208 vanishes . the other elastic coefficients describe shears and other possible deformations of less interest here . by symmetry , the coupling of the spins and phonons is given by @xmath209 it may be convenient to define @xmath210,\ ] ] with @xmath211 and @xmath212 . then this can be rewritten as @xmath213 one may consider the perturbative regime of weak spin - lattice coupling . for zero coupling , @xmath214 , the spin and lattice sub - systems are independent . if we assume stability of the disconnected lattice subsystem , then there are three regimes to consider , dictated by the state of the spins : @xmath215 , @xmath216 , and @xmath102 . for @xmath215 , the spins are disordered , and @xmath195 undergoes small fluctuations about @xmath217 . the full tetragonal symmetry is maintained , and perturbation theory in @xmath218 will lead only to small corrections to the elastic theory . for @xmath102 , the spin system on its own breaks the tetragonal symmetry . depending upon the sign of @xmath219 , states with @xmath220 non - zero and real ( @xmath221 ) or imaginary ( @xmath222 ) are selected . this corresponds to magnetic order aligned along @xmath223 or @xmath0 axis , respectively . this non - zero expectation value reduces @xmath224 in eq . at first order to a linear term in the strains , which induces non - zero @xmath225 ( for @xmath221 ) or @xmath226 ( for @xmath222 ) . hence the lattice responds to the symmetry breaking dictated by the spins . for @xmath216 , nave perturbation theory is invalidated by the power - law correlations of the criticality of the spins . instead , we can analyze the stability of the critical point using renormalization group arguments . the decoupling spin system described by eq . in zero magnetic field has a @xmath7 symmetry under @xmath227 rotations of @xmath195 . it is established that the fourfold anisotropy ( @xmath219 ) is irrelevant at the critical point in three dimensions , so that the fourfold symmetry is enlarged to continuous rotational invariance in the @xmath34 plane . the universal properties of the system at the critical point are described by the scale invariant o(2 ) wilson - fisher fixed point field theory . this field theory contains two independent composite operators quadratic in the @xmath34 field . the first is the `` energy density '' operator @xmath228 , which has scaling dimension @xmath229=d-1/\nu$ ] , where @xmath230 is the dimensionality and @xmath231 is the correlation length exponent ( the fact that @xmath232 follows from hyperscaling , which requires @xmath233 , where @xmath27 is the specific heat exponent , which is known to be slightly negative for the 3d xy model ) . hence @xmath229 \gtrsim 3/2 $ ] . the second operator is the symmetric tensor field @xmath234 , which because it is affected by the fluctuations of the phase of @xmath195 , is expected intuitively to have a larger scaling dimension than the energy density . an estimate from the @xmath235-expansion @xcite gives @xmath236 \approx 9/5 $ ] . the scaling dimension of the strain is determined by the requirement that the total free energy be dimensionless , hence @xmath237=3/2 $ ] for all strain components in @xmath230 . the full scaling dimension of the perturbations in eq . is @xmath237+[|\phi|^2]$ ] and @xmath237+[\phi^2]$ ] . using the above results , we see that all these total scaling dimensions are _ larger _ than @xmath238 , which implies that the terms in @xmath224 are _ irrelevant _ at the xy fixed point . this demonstrates that spin - lattice coupling effects are _ irrelevant _ when weak , and a single direct transition in the xy universality class is maintained . by inspection , the terms in @xmath224 can lower the free energy of orthorhombically distorted states independent of whether they exhibit magnetic order . hence , despite the fact that , when they are weak spin - lattice coupling does not induce a nematic phase , we expect the nematic will occur when the appropriate @xmath218 is sufficiently large . to go beyond the perturbative approach , we consider a mean - field treatment of the elasticity , while continuing to include all fluctuation effects in the spins . formally , we define the effective free energy for @xmath239 ( @xmath240 ) obtained by integrating out @xmath195 completely : @xmath241 = f_u - \log \left [ \int [ d\phi ] \exp\left ( - f_\phi - f_{u\phi}\right ) \right],\ ] ] where @xmath242 $ ] indicates the full functional integral over @xmath195 . we treat @xmath243 $ ] in mean - field , i.e. just seeks its thermodynamic minimum . the coefficients @xmath244 play no essential role , as they do not couple symmetry breaking order parameters ( they instead just describe how changes in the lattice parameters of the tetragonal structure are coupled to the energy density of spins ) , and so they can be set to zero . in mean field , we assume that the tensor @xmath239 is a constant . then the free energy is simply the system volume times the free energy density , @xmath245 where in general , to consider the regime of non - zero orthorhombic distortions , we should augment @xmath246 by third and fourth order terms in @xmath239 for stability . if we focus on the case of zero field @xmath166 and assume we are above the order temperature for the magnetism , @xmath215 , then all the terms which are irrelevant at the magnetic critical point @xmath219 , @xmath247 , and @xmath248 are unimportant and can be neglected . then the isotropic gradient coefficient can be set to unity by rescaling coordinates : @xmath249 . thus finally , we see that @xmath250 is a function of @xmath251 , @xmath252 , and @xmath253 . using scaling around the gaussian fixed point where @xmath254 , we have ( using @xmath230 ) @xmath255 the free energy can depend only upon the absolute value of @xmath256 by u(1 ) symmetry . by choosing @xmath257 , we then obtain @xmath258 with @xmath259 . we used an overall minus sign in the definitions of the scaling functions in eqs . ( [ eq:6 ] ) and ( [ eq:7 ] ) , anticipating that @xmath250 is negative . to check the _ local _ stability of the tetragonal state , we expand the free energy to quadratic order in @xmath256 and check whether the resulting quadratic form is stable . we have @xmath260 the latter term may be absorbed into renormalized elastic moduli according to [ eq:10 ] @xmath261 we see that indeed @xmath262 and @xmath263 are reduced by the spin fluctuations , and if their bare values are small enough , the renormalized values become negative , signaling an elastic instability . we may consider two regimes . first , when @xmath264 , the first argument is small , and one can neglect the quartic interaction term . this is the gaussian regime . the inequality @xmath264 corresponds to the ginzburg criterion , under which mean field behavior ( for @xmath195 ) is expected . in the above calculation of the instability point , one may take @xmath265 as constant in this regime , and we observe that the corrections to the elastic modulo become arbitrarily large for small enough @xmath251 . this seems in conflict with the conclusion of the previous subsection that weak spin - lattice coupling does not destabilize the xy transition . we will return to this point . beyond just the instability analysis , in this case , the full free energy may be explicitly calculated , since the functional integral in eq . is quadratic . one finds simply @xmath266,\ ] ] which is well - defined only for @xmath267 . by taylor expansion , one obtains to fourth order @xmath268 this determines the instability points depending upon the values of @xmath207 , @xmath208 , and @xmath218 , according to the condition @xmath269 or @xmath270 . looking further , however , we see that the contribution from spin fluctuations to the fourth order term in the free energy as a function of @xmath271 is _ negative_. this is suggestive of a first order transition . to explore this further , let us assume for concreteness an instability in the @xmath225 channel . we take @xmath272 and @xmath273 and all other components of the strain tensor zero . then the elastic free energy is @xmath274 where we included a fourth order term for stability . including the second and fourth order terms from @xmath250 , we obtain then @xmath275 suppose @xmath276 is small , and @xmath207 is fixed . then in the gaussian regime , we see that the elastic instability occurs [ see eqs . ] at the critical reduced temperature @xmath277 . for this value @xmath278 the negative term in the coefficient of @xmath279 in eq . scales as @xmath280 , which is much larger than the constant @xmath281 for small @xmath276 . hence in this limit it seems that the fourth order term in the free energy @xmath282 is negative , and indeed a first order elastic transition obtains . however , this conclusion rests on the gaussian treatment of the @xmath195 field . this breaks down close to the xy critical point , when @xmath283 is no longer small . when @xmath283 becomes large , we expect that the universal critical exponents of the xy wilson - fisher fixed point should apply . therefore in this limit we can deduce that @xmath284 - 3)\nu ) } y),\ ] ] where @xmath285 is a new universal function . this is required so that the proper universal scaling behavior @xmath286 - 3)\nu } |\epsilon|)\ ] ] obtains . by expansion , we now see that in the xy regime , the renormalized elastic modulus takes the form @xmath287 - 3)\nu},\ ] ] where @xmath288 - 3)\nu \approx 2/5 $ ] is positive . we see that , contrary to the gaussian regime , the correction to the elastic modulus _ decreases _ with reducing @xmath251 on approaching close to the critical point . note furthermore that for small @xmath251 the xy regime is always entered , cutting off the apparent gaussian divergence . this means that for sufficiently small @xmath276 , where the instability criterion is not reached before the crossover to the xy regime , there is no instability and the xy critical point is stable , as claimed in the prior subsection . the above analysis implies that the renormalization of the elastic modulus for small @xmath276 is non - monotonic with temperature . according to scaling , for example @xmath289 where @xmath290 is a constant , and @xmath291 for @xmath292 . hence the ( negative ) renormalization of the elastic modulus is maximum for @xmath293 , and is of order @xmath294 . this sets a threshold value for an instability at @xmath295 . for @xmath296 , a nematic phase appears in an interval of temperature around @xmath297 . remarkably , the elastic mean field theory predicts at for @xmath276 close to @xmath298 , the nematic phase exists as an `` island '' _ within _ the tetragonal phase , i.e. there is a re - entrant tetragonal phase at _ lower _ temperature than the nematic one , before the xy transition occurs to the orthorhombic magnetic state . with further increase of @xmath276 , the extent of nematic phase grows until its lower boundary reaches the xy critical point . for larger @xmath276 , the re - entrant tetragonal phase no longer exists , and the system evolves through the simpler sequence of tetragonal to nematic ( orthrhombic ) to magnetic phases on reducing temperature . the above analysis predicts several different phase transitions when spin - lattice coupling is substantial . a tetragonal to orthorhombic ( which may also be called tetragonal to nematic ) transition may occur _ twice _ at intermediate @xmath276 , or once at larger @xmath276 . the elastic mean field theory predicts the higher temperature transition is first order , while the lower transition may be second order . a continuous second order transition is in fact known to be possible based on renormalization group analysis , @xcite and if this transition is continuous it is expected to display mean - field critical exponents owing to the influence of long - range elastic forces . the lower transition from nematic / orthorhombic to the magnetic phase is of ising type . finally , the phase diagram contains an interesting multicritical point where the nematic , magnetic , and tetragonal phases meet . to our knowledge , the universality class of this multicritical point has not been studied carefully , and is left as an interesting problem for future theory . in this paper , we studied the magnetic properties of two spin models on a fcc lattice with tetragonal lattice symmetry , using several analytical techniques and monte carlo simulations . the results should be applicable to insulating magnetic double perovskites with a single magnetic species , in which orbital degeneracy is broken . in the first half of the study , we considered the simplest tetragonal model corresponding to a uniform lattice distortion of the cubic fcc system along @xmath5 axis . we mapped out the ground state phase diagram in the classical limit , and showed that it is dominated by four magnetic phases , @xmath34fm , @xmath34afm , @xmath5fm , and @xmath5afm orders . the first two of these phases were shown to exhibit the phenomena of `` order by disorder '' : accidental classical degeneracy lifted by fluctuations . as a result , magnetic polarization along the @xmath49 directions was favored . in the latter half of the paper , we considered a more complex tetragonal model in which the tetragonal symmetry is realized through two inequivalent fcc sites , related by a screw axis . this is suggested by earlier theory as a result of orbital / quadrupolar order.@xcite after writing the general hamiltonian for this system , we focused upon a physically motivated parameter regime , and showed that the model exhibits a peculiar canted ferromagnetic state . in addition , we studied the effect of spin - lattice coupling , and pointed out the possibility of a nematic transition induced by the coupling . it might be interesting to measure a `` nematic susceptibility '' , as has been done by elasto - resistivity@xcite . for an insulator , the resistivity anisotropy may be too difficult to measure . hence we suggest this could be done instead by identifying the electronic nematic order parameter @xmath91 with the anisotropy of the magnetic susceptibility , i.e. @xmath299 then the nematic susceptibility may be defined as @xmath300 ( here we consider the nematic order corresponding to a @xmath301100@xmath302 deformation @xmath303 , but one can make a similar definition for a @xmath301110@xmath302 deformation @xmath304 by a 45 degree rotation ) . in the phase diagrams of fig . [ fig : intro : phdiag ] , we expect behavior of @xmath305 appropriate to a composite nematic order parameter @xmath306 on the lower boundary , and of a fundamental ising - nematic order parameter on the upper right boundary . our theory presented here is potentially applicable to wide range of materials in the double - perovskite family . @xcite it complements an earlier study,@xcite going into more depth with fewer theoretical assumptions , for the case of tetragonal symmetry . the most obvious application is to ba@xmath1naoso@xmath2 , which exhibits a ferromagnetic ground state with an unusual @xmath0 easy axis . @xcite this easy axis is readily explained if the cubic symmetry is broken to tetragonal , and we assume here that this occurs at a temperature high compared to the magnetic ordering . evidence of this appears to have been found recently by x - ray scattering.@xcite further comparison with structural data from x - rays at low temperature,@xcite and with nmr and nqr measurements that may discern details of the magnetic and tetragonal ordering,@xcite should be fruitful . more generally , we would like to emphasize that the combination of strong spin - orbit coupling , narrow electronic bandwidth , and varieties of structural motifs of double perovskites makes them potentially a rich realization of highly quantum frustrated spin-1/2 hamiltonians with exotic directional - dependent spin couplings . in this , they comprise another family to complement the honeycomb iridate family which has been much studied recently . the authors thank i. r. fisher , z. islam , v. mitrovic , and e .- moon for fruitful discussions . part of the calculations were done at center for scientific computing at california nanosystems institute and material research laboratory , university of california santa barbara : an nsf mrsec ( dmr-1121053 ) and nsf cns-0960316 . hi is supported by jsps postdoctoral fellowships for research abroad . lb was supported by the nsf through grant nsf - dmr-12 - 06809 . in this appendix , we elucidate the ground state phase diagrams for the model in eq . [ eq : model : htetra ] with interlayer couplings . the phase diagram in absence of interlayer coupling is already presend in sec . [ sec : independent - layers ] , and effect of interlayer couplings for @xmath34fm order is discussed in sec . [ sec : three - dimens - coupl ] . starting from the 2d phase diagram , we here present how the interlayer coupling @xmath52 and @xmath53 modifies the ground state . in sec . [ ssec : appa : zfm ] , we consider @xmath5fm case . the cases of two afm orders are considered in secs . [ ssec : appa : xyafm ] and [ ssec : appa : zafm ] , respectively . in the last , the stripe phase is studied in sec . [ ssec : appa : stripe ] . for the @xmath5fm case , a similar phase diagram to the @xmath34fm case is obtained . when @xmath57 , we obtained a simple ferromagnetic order with spins pointing along @xmath5 axis , as @xmath53 aligns the @xmath5fm layers ferromagnetically ( @xmath5fm / fm state ) . on the other hand , when @xmath57 , the @xmath5fm planes stacks alternatively forming an antiferromagnetic order ( @xmath5fm / afm state ) . these two phases meet at a boundary @xmath307 . introducing @xmath52 gives rise to competition between the @xmath5fm orders and magnetic states with spins pointing in the @xmath34 plane . when @xmath308 and @xmath309 , the phase diagram consists of four phases : @xmath5fm / fm , @xmath5fm / afm , @xmath34fm / fm , and @xmath34fm / afm orders . the two @xmath34fm states are separated by the @xmath5fm states that dominates the phase diagram around @xmath56 line . the phase boundaries between the @xmath5fm and @xmath34fm states are given by @xmath310 and/or @xmath311 , however , there remains a small region in which we could not find the ground state by the luttinger - tisza method . these regions are given by @xmath312 for @xmath50 , and @xmath313 for @xmath51 . the afm states in the 2d phase diagram in sec . [ ssec : resa : gs ] gives the 3d phase diagram which is quite different from the @xmath34fm and @xmath5fm states . in the case of @xmath34afm state , when @xmath314 is small , the energy contribution from @xmath52 bonds cancels out due to the in - plane afm pattern . hence , in the classical ground state , arbitrary stacking of the @xmath34afm planes are degenerate as the ground state . the @xmath34afm state with quasi - macriscopic degeneracy is stable as the ground state for @xmath315 and @xmath316 degrees of freedom . in addition , the @xmath317 degrees of freedom is expected to be further reduced by quantum fluctuations . in the classical limit with no further neighbor interactions , this disordered @xmath34afm states are taken over by @xmath34fm / fm ( @xmath34fm / afm ) order for @xmath318,\end{aligned}\ ] ] and @xmath54 ( @xmath55 ) . on the other hand , we could not determine the ground state for the regions in between the @xmath34fm and @xmath34afm orders , @xmath319.\label{eq : xyafm_incomm}\end{aligned}\ ] ] on the other hand , large @xmath322 stabilizes @xmath5fm orders ; @xmath5fm / fm for @xmath57 and @xmath5fm / afm for @xmath60 . the phase boundary between @xmath34fm and these orders are given by @xmath330 for @xmath50 and @xmath331 for @xmath51 . we could not determine the ground state in these regions . in the 2d phase diagram in sec . [ ssec : resa : gs ] , the stripe order appears in the region where @xmath332 and @xmath333 are relatively small compared to @xmath334 . in the luttinger - tisza method , however , the stripe phase is unstable to infinitesimally small @xmath335 and @xmath320 . the stripe phase is a fine - tuned case of the `` incommensulate '' ground states . this unstable region appears for latexmath:[\[\begin{aligned } 99 g. cao , y. xin , c. s. alexander , j. e. crow , p. schlottmann , m. k. crawford , r. l. harlow , and w. marshall , phys . b * 66 * , 214412 ( 2002 ) . y. singh and p. gegenwart , phys . b * 82 * , 064412 ( 2010 ) . x. liu , t. berlijn , w .- g . yin , w. ku , a. tsvelik , y .- j . kim , h. gretarsson , y. singh , p. gegenwart , and j. p. hill , phys . rev . b * 83 * , 220403 ( 2011 ) . s. k. choi , r. coldea , a. n. kolmogorov , t. lancaster , i. i. mazin , s. j. blundell , p. g. radaelli , y. singh , p. gegenwart , k. r. choi , s .- w . cheong , p. j. baker , c. stock , and j. taylor , phys . . lett . * 108 * , 127204 ( 2012 ) . m. j. omalley , h. verweij , and p. m. woodward , j. solid state chem . * 181 * , 1803 ( 2008 ) . y. singh , s. manni , j. reuther , t. berlijn , r. thomale , w. ku , s. trebst , and p. gegenwart , phys . lett . * 108 * , 127203 ( 2012 ) . k. a. modic , t. e. smidt , i. kimchi , n. p. breznay , a. biffin , s. choi , r. d. johnson , r. coldea , p. watkins - curry , g. t. mccandless , j. y. chan , f. gandara , z. islam , a. vishwanath , a. shekhter , r. d. mcdonald , and j. g. analytis , nature comm . * 5 * , 4203 ( 2014 ) . t. takayama , a. kato , r. dinnebier , j. nuss , h. takagi , preprint ( arxiv:1403.3296 ) . a. biffin , r. d. johnson , i. kimchi , r. morris , a. bombardi , j. g. analytis , a. vishwanath , and r. coldea , preprint ( arxiv:1407.3954 ) . j. xiang and m .- h . whangbo , phys . b * 75 * , 052407 ( 2007 ) . k .- w . lee and w. e. pickett , eur . phys . lett . * 80 * , 37008 ( 2007 ) . a. s. erickson , s. misra , g. j. miller , r. r. gupta , z. schlesinger , w. a. harrison , j. m. kim , and i. r. fisher , phys . lett . * 99 * , 016404 ( 2007 ) . j. m. luttinger and l. tisza , phys * 70 * , 954 ( 1946 ) . c. l. henley , j. appl . 61 * , 3962 ( 1987 ) . w. minor and t. m. giebultowicz , j. phys . colloq . c8 - 1551 ( 1988 ) . h. t. diep and h. kawamura , phys . b * 40 * , 7019 ( 1989 ) . m. v. gvozdikova and m. e. zhitomirsky , jetp letters * 81 * , 236 ( 2005 ) . g. chen , r. pereira , and l. balents , phys . b * 82 * , 174440 ( 2010 ) . a. j. steele , p. j. baker , t. lancaster , f. l. pratt , i. franke , s. ghannadzadeh , p. a. goddard , w. hayes , d. prabhakaran , s. j. blundell , phys . rev . b * 84 * , 144416 ( 2011 ) . k. binder , z. phys . b * 43 * , 119 ( 1981 ) . j. v. jose , l. p. kadanoff , s. kirkpatrick , and d. r. nelson , phys . b * 16 * , 1217 ( 1977 ) . d. blankschtein , m. ma , a. n. berker , g. s. grest , and c. m. soukoulis , phys . rev . b * 29 * , 5250 ( 1984 ) . m. caselle and m. hasenbusch , j. phys . a * 31 * , 4603 ( 1998 ) . m. oshikawa , phys . b * 61 * , 3430 ( 2000 ) . j. m. carmona , a. pelissetto , and e. vicari , phys . b * 61 * , 15136 ( 2000 ) . j. lou , a. w. sandvik , and l. balents , phys . * 99 * , 207203 ( 2007 ) . a. i. larkin and s. a. pikin , zh . * 56 * , 1664 ( 1969 ) ; sov . jetp * 29 * , 891 ( 1969 ) . h. a. jahn and e. teller , proc . lond . a * 161 * , 220 ( 1937 ) . z. islam and i. a. fisher , _ unpublished_. v. mitrovic , _ unpublished_. r. a. cowley , phys . b * 13 * , 4877 ( 1976 ) . chu , h .- h . kuo , j.g . analytis , and i.r . fisher , science * 337 * , 710 ( 2012 ) .

to explain the properties of the most massive low - redshift galaxies and the shape of their mass function , recent models of galaxy evolution include strong agn feedback to complement starburst - driven feedback in massive galaxies . using the near - infrared integral - field spectrograph spiffi on the vlt , we searched for direct evidence for such a feedback in the optical emission line gas around the z.16 powerful radio galaxy mrc1138 - 262 , likely a massive galaxy in formation . the kpc - scale kinematics , with fwhms and relative velocities and nearly spherical spatial distribution , do not resemble large - scale gravitational motion or starburst - driven winds . order - of - magnitude timescale and energy arguments favor the agn as the only plausible candidate to accelerate the gas , with a total energy injection of ergs or more , necessary to power the outflow , and relatively efficient coupling between radio jet and ism . observed outflow properties are in gross agreement with the models , and suggest that agn winds might have a similar , or perhaps larger , cosmological significance than starburst - driven winds , if mrc1138 - 262 is indeed archetypal . moreover , the outflow has the potential to remove significant gas fractions (% ) from a galaxy within a few 10 to 100 myrs , fast enough to preserve the observed [/fe ] overabundance in massive galaxies at low redshift . using simple arguments , it appears that feedback like that observed in mrc1138 - 262 may have sufficient energy to inhibit material from infalling into the dark matter halo and thus regulate galaxy growth as required in some recent models of hierarchical structure formation .

astro-ph0606530

models of structure formation and galaxy evolution have reached a state where the impact of large - scale baryonic feedback on galaxy evolution can no longer be neglected . however , the physics of such processes are complex , and as yet not very well understood , especially at high redshift , where the most massive galaxies formed most of their stars . therefore , simulations of the ensemble of galaxies typically incorporate these processes , which originate on scales beyond the resolution of typical large scale simulations , as simple relationships and parameterizations , which are in turn tuned to approximately match observed galaxy properties . in spite of these limitations , baryonic feedback plays a crucial role for our understanding of galaxy evolution , perhaps most dramatically at high redshift ( e.g. , * ? ? ? * ) the impact of feedback related to intense star - formation is now observationally well - established at low and high redshift ( e.g. , * ? ? ? * ; * ? ? ? * ) , and simulations improved considerably when including it . but significant and likely fundamental discrepancies remain , e.g. , when comparing the _ observed _ mass function of galaxies with the _ predicted _ mass function of dark matter halos at the upper end of the mass function @xcite . from low - redshift studies , starburst - driven `` superwinds '' are known to be efficient in driving out gas only from the comparably shallow gravitational potentials of low mass galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? . however , these winds are mainly energy driven , therefore susceptible to radiative losses , and they probably lack the power , efficiency , and velocities necessary to remove significant gas fractions from the deep potential wells of the most massive galaxies and their dark - matter halos . as a consequence , they can not explain the overabundance particularly of massive dark - matter halos compared to the co - moving galaxy density . therefore another powerful feedback mechanism is recently gaining in popularity among researchers modeling and simulating the ensemble properties of galaxies , which are vigorous outflows driven by powerful active galactic nuclei at high redshift ( agn ; e.g. , * ? ? ? * ; * ? ? ? energy- and momentum - driven agn winds may reach very high energy output through radiation pressure and particle ejection . the most powerful agn may eject enough energy and momentum to unbind significant fractions of the total interstellar medium ( ism ) of the host galaxy . as they slow and expand , such winds are able to efficiently heat gas in the inter - cluster medium ( icm ) and even the intergalactic medium ( igm ; e.g. , * ? ? ? however , due to the clumpy nature of the ism in galaxies , rather small working surface of the jet , and agn producing `` light jets '' , they are not generally thought to couple strongly with the ambient ism in their hosts @xcite . the observed characteristics of massive galaxies support the hypothesis that strong agn winds at high redshift may have a significant impact on galaxy evolution and their ensemble characteristics . spheroidal and black hole masses in nearby massive systems follow the tight m@xmath6-@xmath7 relation @xcite , which perhaps signals co - eval growth of smbh and the bulge self - regulated through negative agn feedback @xcite . the most massive galaxies , virtually all spheroidals , are `` old , red , and dead '' . they are metal - rich and have luminosity weighted ages consistent with massive star - formation at high - redshift and subsequent passive evolution , perhaps indicating that the gas was effectively removed during the most massive burst ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? maybe the tightest constraint comes from the metal abundance ratios , especially @xmath8 $ ] , which indicate intense star - formation truncated by strong feedback within less than few @xmath9 yrs ( e.g. , * ? ? ? * ) . in addition , the exponential cut - off in the bright end of the galaxy luminosity function may be a direct result of the efficiency of agn feedback @xcite . furthermore , @xcite find evidence that more moderate feedback of radio - loud agn might balance gas cooling in the halos of the most massive galaxies at low redshift . a promising way to identify the fingerprints of a highly efficient mode of agn feedback at high redshift is by observing the gas kinematics in rapidly growing , massive galaxies at high redshift which host agn . powerful high - z radio galaxies ( hzrgs ) fit these requirements : they reside in dense environments @xcite . their magnitudes , colors , and continuum morphologies are consistent with large stellar masses and passive evolution at @xmath10 @xcite . hzrgs at @xmath11 also contain large masses of molecular gas ( @xmath12 ; e.g. , * ? ? ? * ) , dust ( @xmath13 ; e.g. , * ? ? ? * ; * ? ? ? * ) , and uv / optical line - emitting gas ( @xmath13 ; e.g. , * ? ? ? at the highest redshifts , they are also likely to be forming stars at prodigious rates ( @xmath14 yr@xmath15 ; e.g. , * ? ? ? curiously , hzrgs are known from longslit - spectroscopy to often have spatially extended broad emission lines with fwhm@xmath16 , indicative of extreme kinematics ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? however , with the slits typically aligned along the radio axis , the global role of these kinematics could not be examined more closely , highlighting the need for integral - field observations of hzrgs , preferably at rest - frame optical wavelengths ( i.e. , in the observed near - infrared ) to investigate the overall gas dynamics . an excellent candidate for such a study is the @xmath17 mrc1138 - 262 , with @xmath18 and stellar mass @xmath19 . vla interferometry reveals 2 radio jets with complex morphologies @xcite . its nebulosity extends over @xmath20 kpc radius , @xmath21 , @xmath22 cosmololgy , size scales at @xmath17 are 8.3 kpc arcsec@xmath15 , the luminosity distance is 17070 mpc , the angular size distance is 1710 mpc . cosmic age at that redshift is 3.0 gyrs . ] , and has a significant surrounding over - density of emitters which indicates it might be in the center of a forming galaxy cluster @xcite . rest - frame uv spectroscopy implies star - formation rates of @xmath23 yr@xmath15 . by examining the spatially - resolved kinematics and spectral properties of the rest - frame optical emission line gas in mrc1138 - 262 , we wish to investigate the influence of the agn on the excitation and kinematics of the emission line gas using data obtained with the near - infrared integral - field spectrograph spiffi . we observed mrc1138 - 262 in the near - infrared h and k bands using the integral - field spectrograph spiffi on ut1 of the vlt in april 2003 . spiffi uses an image slicer to dissect the 8@xmath248 field of view on the sky into 32 slices or `` slitlets '' with individual pixel sizes in projection of 0.25 . the spectrograph and image slicer have since become a facility instrument at ut4 as part of sinfoni . under good and stable weather conditions , we obtained a total of 140 min of data in k and 75 min in h with a pixel scale of 0.25and seeing of @xmath25 . individual exposure times were 300s in h and 600s in k , respectively . spectral resolution is r=@xmath26 at 1.6 and r=@xmath27 at 2.2 . one `` off '' frame at a sky position well displaced from the target was taken for each `` on '' frame in an off - on - on - off pattern . to reduce the data , we used the iraf @xcite standard tools for the reduction of longslit - spectra , modified to meet the special requirements of integral - field spectroscopy . data are dark frame subtracted and flat - fielded . for bad pixel correction we use the full three - dimensional information , identifying pixels that are 5@xmath7 deviant in the darks and flats . rectification and wavelength calibration are done before night sky subtraction , to account for some spectral flexure between the frames . curvature is measured and removed using an arc lamp , before shifting the spectra to an absolute ( vacuum ) wavelength scale with reference to the oh lines in the data . the three - dimensional data cubes are then reconstructed , assuming that each slitlet covers exactly 32 pixels . they are spatially aligned by cross - correlating the collapsed cubes , and then combined , clipping deviant pixels . telluric correcton is applied to the combined cube . flux scales are obtained from standard star observations . from the light profile of the standard star , we measure the fwhm spatial resolution to be @xmath28 in right ascension and declination , respectively . in fig . [ fig : specnuke ] we show h and k - band spectra of mrc1138 - 262 . the upper panel shows the spectrum of the spatial pixel covering the agn , the lower panel shows the agn removed extended emission integrated over the central 30 kpc . line profiles are not simple gaussians , but have irregular profiles indicating complex kinematics in the extended emission line gas . the lines are generally broad , and [ nii]6583 and are blended , so that is not a reliable kinematic tracer . therefore we use [ oiii]5007 ( with fwhm@xmath29 in the integrated spectrum ) to trace the gas kinematics . we discuss below that the emission line kinematics does not vary strongly with the specific emission line analyzed ( [ sec : iontempdens ] ) , within the accuracy we can reach given the line blending . thus we do not expect and there is no evidence that this choice has a strong impact on our overall conclusions . the luminous broad emission line is the most prominent spectral feature in the agn spectra , with fwhm@xmath30 , and a rest - frame equivalent width , eqw@xmath31 , indicating that the emission from the broad line region can not be neglected ( especially in the broad band near infrared magnitudes ) . previous rest - frame uv spectroscopy of mrc1138 - 262 did not reveal nuclear broad lines , indicating that mrc1138 - 262 might harbor an obscured quasar , similar to mrc2025 - 218 @xcite . broad emission in the h band is measured at much fainter flux : from the balmer decrement in the broad components , @xmath32 , using a galactic extinction law and @xmath33 . the galactic extinction law might not be strictly adequate for broad agn emission lines , nonetheless , the large extinction underlines the interpretation that mrc1138 - 262 harbors an obscured quasar . to correct the k - band data for the agn contribution , we fit and remove the nuclear spectrum from all other spatial pixels . this may lead to a slight oversubtraction of extended line emission along the line of sight of the agn . in our later analysis , we aim at setting lower limits to the emission line kinematics and fluxes , therefore this procedure is nonetheless appropriate . in the k - band , we can use the luminous broad h@xmath5 emission line as an indicator for how well our algorithm removes the agn light . agn removed spectra are shown in the lower panel of fig . [ fig : specnuke ] to illustrate the success of the method . to investigate the spatially resolved kinematics , we extracted spectra from 3@xmath243 pixel box apertures ( @xmath34 ; slightly larger than the estimated seeing in both reduced data cubes ) . using the iraf task splot , we fitted [ oiii]5007 line emission with up to 3 kinematic components , requiring @xmath35 in each component . systematic errors of these measurements are possibly large , and the agn removal in the h band might be not very precise . we thus use these spectra only to identify zones of rather uniform kinematics , and extract integrated spectra of these zones to further investigate their properties . relative velocities of all [ oiii]5007 components are shown in fig . [ fig : velvsdistplot ] as a function of projected radius . they appear to be segregated by velocity into roughly 4 groups with sizes of @xmath36 ( @xmath37 kpc ) over the total radius of @xmath38 ( 20 - 25 kpc ) . overlap between these groups in velocity is small . in fig . [ fig : velocitymap ] we show the map of relative velocities of the brightest [ oiii]5007 line component in each spatial pixel . at least three distinct kinematic zones , each separated in velocity space by @xmath39 1000 , but with rather uniform velocities internally , are evident . velocities in zone 1 decline with increasing projected distance from the nucleus , gradually approaching the values in zone 2 at the eastern bound . this is in agreement with fig . [ fig : velvsdistplot ] , which also suggests that the velocities in zone 1 at its low - velocity end approach the velocities in zone 2 . zone 3 has largely uniform intrinsic velocities , whereas velocities in zone 2 vary gradually with position angle , reaching a maximum at a position angle @xmath40 . the relatively uniform velocities within each structure , ( fig . [ fig : velocitymap ] ) , mirroring the regularity found in fig . [ fig : velvsdistplot ] , are highly reminiscent of what would be expected for overlapping , edge - brightened `` bubbles '' ( such a scenario is also favored by the low filling factor of the emission line gas , [ sec : timescale ] and [ sec : heatgas ] ) . however , we are using the term `` bubble '' somewhat loosely , referring mainly to the clearly distinct zones in velocity space . their morphologies are nearly circular as if projected on the sky , and are consistent with expanding spheres or projected conical outflows . they might be analogues of buoyantly rising , low - density regions in low - redshift radio galaxies , but our analysis does not rely on this interpretation . by using this term , we simply relate to the geometry and velocity structure of the gas , not necessarily the full astrophysical interpretation . this interpretation is also supported by the spatial distribution of line widths . in fig . [ fig : sigmamap ] we show a map of the fwhm of the brightest [ oiii]5007 component in each pixel . line widths vary between fwhm@xmath41 . most striking is a ring of broad line emission ( fwhm@xmath42 ) around zones 1 and 3 , shown as hatched area in the velocity map of fig . [ fig : velocitymap ] . this ring supports the above interpretation that the emission line regions with well separated velocities are geometrically and physical distinct entities , which originate from apparent broadening due to overlap between adjacent bubbles or from turbulence due to interactions along their boundaries . integrated h and k band spectra of the 3 definable bubbles are shown in fig . [ fig : z13spec ] . emission lines are bright in all bubbles . line profiles are complex , and and [ nii]6583 are blended in all zones . to obtain a consistent fit for all lines in a given zone , we identify kinematic subcomponents from the [ oiii]5007 profile , and restrict all other line fits to have the same relative velocities and widths . we leave the flux as a free parameter , except for the ratios of the [ nii]6548,6583 and [ oiii]4959,5007 emission line doublets , where we fix the line ratios to the ratios of their transition probabilities . our data also include the [ sii]6717,6731 line doublet , which has line ratios sensitive to electron densities in the range of 100 to @xmath43 @xmath44 @xcite . we identify two @xmath45 pixel apertures in bubble 2 where the two components are sufficiently narrow to be deblended robustly . line ratios in these apertures are @xmath46 } \equiv f_{\lambda6716}/f_{\lambda6731}= 1.1\pm 0.1 $ ] . this corresponds to a density range @xmath47 @xmath44 , and @xmath48 @xmath44 at best fit value , assuming a temperature @xmath49 k. ( the results of * ? ? ? * ; * ? ? ? * ; * ? ? ? * indicate a most likely temperature range of @xmath50 k in agn narrow line regions , which is sufficiently accurate for our purpose . ) @xcite find densities of @xmath51 @xmath44 in the precursor regions of fast radiative shocks , constrained by radio observations , in good agreement with our result using an independent estimate . the [ sii ] emission originates mainly in a `` partially ionized zone '' , where the thermal pressure is @xmath52 few @xmath2410@xmath53 n@xmath54 dynes @xmath55 and @xmath56 ( e.g * ? ? ? * ) . we use these correction factors in calculations that make use of the total gas density , e.g. , the total mass estimate based on the recombination line flux of hydrogen . moreover , we only use the line flux with the agn contribution removed , as described in [ sec : spectra ] . the blending of [ nii ] and complicates the flux estimates , therefore we only aim at estimating robust lower limits to the kinetic energy in the extended emission line region , and hence minimize the contribution in our fits to the [ nii]@xmath57 line blend , typically obtaining [ nii]@xmath58 . with the adopted cosmology , the total observed flux in all zones corresponds to a total luminosity @xmath59 ergs s@xmath15 . to estimate the ionized gas mass in the diffuse nebula , we assume case b recombination @xcite , and our previous gas density estimates : @xmath60 with luminosity @xmath61 in units of @xmath62 ergs s@xmath15 , planck constant @xmath63 , frequency @xmath64 , effective recombination coefficient , @xmath65 , proton mass @xmath66 , and electron number density @xmath67 in units of 100 s@xmath15 . with the electron density of @xmath68 @xmath44 ( [ sec : iontempdens ] ) this implies a total ionized gas mass @xmath69 . our analysis can be precise at the order - of - magnitude level only , and is dominated by astrophysical rather than the measurement uncertainties . therefore we will adopt an hii mass of @xmath70 , corresponding to the measured luminosity and electron density at face value and not corrected for extinction . thus this estimate is a robust low limit to the intrinsic luminosity and total emission line gas mass . mrc1138 - 262 is not the first hzrg known to have broad , spatially resolved emission line gas ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but we have not found another such galaxy in the literature with integral - field spectroscopy in the rest - frame optical . for galaxies with spatially resolved slit spectroscopy , the slit is typically aligned with the radio axis , so that the morphology of the high - velocity emission line gas and total energy and mass in the flow are difficult to estimate . gas dynamics dominated by the large - scale gravitational potential of the host galaxy would have smooth velocity gradients and rather uniform line widths with fwhms of a few hundred . @xcite discuss the relatively quiescent kinematics in the ly@xmath5 halo of the z=2.49 hzrg mrc2104 - 242 , with fwhm@xmath71 , and argue that the kinematics might be caused by rotation or infalling gas . the kinematics we observe in mrc1138 - 262 , however , are different from mrc2104 - 242 , quantitatively and qualitatively . our integral - field observations are uniquely suited for a more detailed investigation of what causes the extreme kinematics in mrc1138 - 262 , by relating the local bulk and turbulent velocities in the gas . the right panel of fig . [ fig : z13spec ] shows the fwhms as a function of velocity in each bubble . velocities are relative to the mean velocity in each bubble . fwhms and velocities are uncorrelated in zone 2 , in zone 1 and 3 they correlate tightly with correlation coefficients @xmath72 and @xmath73 , respectively . positive correlations like in zone 3 are also found in starburst - driven winds , and indicate outflowing gas being accelerated and becoming more turbulent as the kinetic energy is being dissipated through shocks @xcite . negative correlations like in zone 1 have been interpreted as deceleration shocks @xcite , when gas heated by the agn collides with the confining , higher density ism . starburst powered outflows do not appear as good candidates to power an outflow with the velocity gradients and line widths observed . typical outflow velocities in superwinds , as similarly estimated from optical emission lines , are a few@xmath74 @xcite . these velocities are a factor of a few lower than in mrc1138 - 262 ( not including any correction for projection effects ) . the velocities seen in starburst driven winds do not seem to be a function of either energy injection power or rate @xcite or perhaps redshift @xcite . @xcite find that the outflow observed in the @xmath75 strongly star - forming ( sfr@xmath76 yr@xmath15 ) submillimeter galaxy smmj14011 + 0252 does not exceed velocities of @xmath77 and fwhms @xmath78 . most importantly for this study , the line cores in smmj14011 + 0252 are not offset from systemic by large velocities . in other words , the overall , large scale kinematics of smmj14011 + 0252 do not appear to be strongly influenced by the wind , similar to the integrated line profiles of local starburst galaxies . moreover , pressures within the wind are similar to winds in local starburst galaxies , suggesting similar physical properties , which make it unlikely that the outflow properties would dramatically change with redshift ( see arguments in , e.g. , * ? ? ? nonetheless , the presence of individual bubbles and large and irregular velocity offsets in mrc1138 - 262 do suggest that some sort of feedback mechanism is driving a massive gas outflow . we will analyze the properties of this outflow in close analogy to @xcite and show that energy and timescale arguments strongly favour a wind driven by the agn as the only plausible explanation . our data indicate that the gas reaches large distances , over 25 kpc for the high surface brightness emission , and might even completely escape the host potential . a way of estimating the time scale of the flow is to calculate the crossing time given the observed velocities . with bulk motions of @xmath79 and an average observed radius of the emission line nebula @xmath80 kpc , the dynamical timescale is @xmath81 yrs , providing a rough timescale necessary to drive the flow over the observed distances . if the outflow is powered by the radio source , then this timescale will be approximately similar to the age of the radio source . with estimated jet - head advance speeds of @xmath390.1c ( e.g. , * ? ? ? * with ranges of about 0.01c to about 0.2c ) , and the `` largest angular size '' of the radio source at @xmath82=1.41 ghz in mrc1138 - 262 ( 11.1 , or 92 kpc ; * ? ? ? * ) , we estimate a dynamical timescale of @xmath393@xmath2410@xmath83 @xmath84 yrs . because of the unknown projection angle , this can only be a lower limit ( hence the factor @xmath84 ) , and the intrinsic timescale is likely a few times higher , up to a few @xmath2410@xmath85 yrs . @xcite suggest that the ages of radio sources decrease with increasing redshift . however , they use a flux limited sample , so that the average radio power will increase with redshift , whereas the size decreases , which might at least in part explain this trend . using their relationship between radio power and redshift for the @xmath86 of mrc1138 - 262 implies an age of @xmath390.4 to 1@xmath2410@xmath85 yrs , similar to our previous estimate . in either approach , the age of the radio source is not significantly different from the crossing time of the emission line nebula , and the radio jet and lobes reach a factor of @xmath392 larger radii than the ( detected ) high surface brightness emission line gas . this alone can not prove that the radio jet is powering the outflow , but it is a necessary prerequisite . to estimate a mass outflow rate , we simply assume that the outflow is being accelerated with an efficiency that is approximately constant during the jet lifetime . this implies that the current estimate is representative for the average rate over the lifetime of the source . given the large velocities and size of the nebula , this assumption is reasonable . this very simple scenario implies @xmath87 yr@xmath15 . of course , we do expect variations in the instantaneous outflow rate due to jet precession , variable jet energy or mass density , or perhaps changing coupling efficiency to the ism as might be expected for a jet interacting with a `` clumpy '' surrounding medium . having measured the electron density and relative ionization ( assuming case b recombination ) , we can also constrain the volume filling factor of the emission line gas . the total observed luminosity corresponds to a `` total emission volume '' @xmath88 , which we compare to the volume of the emission line nebula @xmath89 ( @xmath90 kpc@xmath91 , approximated by the total volume of 3 spherical bubbles with the observed radii of each zone ) . this yields a filling factor @xmath92 . we find a filling factor , @xmath93 , likely signaling that distributed in bright , line - emitting sheets , small clouds , and/or filaments . low filling factors arise naturally when a sufficiently intense , expanding hot wind sweeps up , accelerates , and ionizes dense hydrogen clouds within the ism of the host galaxy . relative to the lifetime of the host , an outflow with lifetime @xmath94 yrs is nearly explosive , implying impulsive energy injection ( but constant over the lifetime of the event ) . following , e.g. , @xcite , expansion speed of the shell and energy content of the bubble are related as : @xmath95 where @xmath96 is the ( constant ) energy injection rate in units of 10@xmath97 ergs s@xmath15 , @xmath98=0.5 @xmath44 is the ambient ism density in @xmath44 and @xmath99 is the injection time in units of 10 myrs . we approximate the velocity @xmath100 through the average of velocities in the line profile fits to the 3 bubbles discussed in [ ssec : kinematics ] . from the analysis , we find an average projected velocity of @xmath101 800 . the required energy is then @xmath394.2@xmath2410@xmath102 ergs s@xmath15 . in a second approach , we model the ( constant ) energy injection rate into an energy conserving bubble expanding into a uniform medium with ambient density @xmath98 , which serves as a plausible upper bound to the energy injection rate , @xmath103 with the radius @xmath104 given in units of 10 kpc and velocity @xmath105 given in units of 1000 km s@xmath15 . we estimated the half - light radius of the h@xmath5 emission , after removing the contribution from the broad line region , and found @xmath5210 kpc . for 10 kpc , a flux weighted mean velocity @xmath106 and ambient density @xmath107@xmath44 , the energy injection rate is @xmath108 ergs s@xmath15 . this is likely a minimum energy required since the velocities are seen in projection . if we use the total extent of the nebula for the radius , we find the energy injection rate is @xmath109 ergs s@xmath15 . alternatively , if the bubble is momentum conserving , we estimate that , @xmath110 or @xmath111 dyn , of force must be injected to power the observed flow . if the jet or radiation field of the agn is to be a viable driver of the observed gas dynamics in mrc1138 - 262 , it must inject at least 10@xmath112 ergs s@xmath15 of energy or alternatively , about 10@xmath113 dynes of force into the ambient interstellar medium to produce the outflow in mrc1138 - 262 . each of these estimates is valid within a given model , although the underlying assumptions are very general , involving basic physical principles . we can give a fully model - independent , strictly observational lower bound to the energy injection rate from the observed kinematics of the emission line regions . to this end , we only account for the observed ionized gas masses , not correcting for projection effects lowering velocity measurements , extinction , and material in other phases , especially x - ray emitting or molecular gas , which will contribute significantly . with the measured gas masses and relative velocities , the kinetic energy is simply @xmath114 for the bulk velocities and @xmath115 for turbulent motion . for mrc1138 - 262 and summing over the velocities and dispersions in the individual bubbles , we estimate @xmath116 ergs , and @xmath117 ( with unknown correction for projection , @xmath118 ) . for our dynamical timescale of @xmath119 yrs , this corresponds to an average injection rate @xmath120 ergs s@xmath15 over a lifetime @xmath94 yrs . the results illustrate that energy injection rates at high redshift are not trivial to estimate , which limits the precision of our analysis to an order - of - magnitude level . it is however encouraging that in spite of the large uncertainties , the values are broadly consistent . we will in the following adopt e@xmath121 erg s@xmath15 . this choice is motivated by the approximate range of values indicated by the simple estimates we have made and to the order - of - magnitude spirit in which they were made . if the agn is to be a viable candidate for powering the observed outflow , then it must be able to at least provide the above energy injection rate . it is difficult to know _ a priori _ what is powering the outflow , the overall bolometric luminosity , @xmath122 , which might couple to the ism through radiation pressure on dust grains , or directly through the kinetic luminosity of the jet @xmath123 . we shall start by estimating @xmath122 . following the scaling of x - ray to bolometric luminosity of @xcite and the x - ray luminosity of mrc1138 - 262 at @xmath124 kev , ( @xmath125=45.6 ergs s@xmath15 ; * ? ? ? * ) , we estimate @xmath126 ergs s@xmath15 . we can also use the relationship between @xmath122 and rest - frame optical flux density of the agn in qsos @xcite , and the 5100 continuum flux in our h band spectrum of mrc1138 - 262 , @xmath127 w @xmath128 m@xmath129 , yielding @xmath130 ergs s@xmath15 . given the approximate nature of either estimate , these values are consistent . we will in the following adopt @xmath131 ergs s@xmath15 since the estimate using the optical continuum is likely to be strongly affected by high extinction to the region of agn continuum emission in radio galaxies compared to qsos . estimating the jet kinetic luminosity as a measure of the power that the jet is capable of injecting in the ambient ism is more difficult and challenging and many authors have tried a variety of methods for estimating or constraining the mechanical jet luminosities . we will therefore use and compare a variety of methods to estimate the mechanical luminosity of the jet in mrc1138 - 262 . @xcite empirically calibrate the radio source characteristics in the 3cr as a function of radio power at 178 mhz . for mrc1138 - 262 the vla measurements at 4.86 ghz and the original selection of the molonglo reference catalog ( mrc ) at 408 mhz ( * ? ? ? * and references therein ) , imply a radio spectral index @xmath5=1.34 , and a 178 mhz radio power , p@xmath132=10@xmath133 ergs s@xmath15 hz@xmath15 sr@xmath15 , in the rest - frame . this corresponds to a jet kinetic luminosity of 10@xmath134 ergs s@xmath15 in the calibration of @xcite . @xcite measured an integrated luminosity of the radio jet in mrc1138 - 262 in the rest - frame 0.1 - 1.0 ghz band , @xmath135 ergs s@xmath15 . based almost solely on theoretical arguments , the ratio of mechanical to radio luminosity is typically argued to be about 10 to 1000 ( e.g. , * ? ? ? * ; * ? ? ? estimating that the radio luminosity is equal to @xmath136 , @xmath137 , of the total jet luminosity sets a lower limit to the jet kinetic luminosity of 10@xmath138@xmath137 ergs s@xmath15 . obviously , it could be substatially higher . this is , however , within the scatter of the @xcite estimate . we can additionally estimate the jet power using analyses of x - ray cavities which are thought to be powered by the mechanical energy of the radio jet interacting with the surrounding hot gas . following the analysis in @xcite , we estimate the mechanical luminosity of the jet in mrc1138 - 262 to be about 10@xmath139 ergs s@xmath15 . we think that this estimate is valuable because it is similar to our approach and provides a robust lower limit to the mechanical luminosity . however , @xcite study radio sources at lower redshift , which have different radio morphologies than mrc1138 - 262 and typically lower radio luminosity . moreover , @xcite point out that their estimates might be systematically too low due to various physical effects . as a last approach , we can estimate the kinetic energy necessary to explain the observed expansion speeds of radio lobes which are of the order of 0.1 to 0.4c ( e.g. , * ? ? ? these advance speeds and a plausible model for the medium into which the lobes are advancing , yields estimated jet kinetic powers of 10@xmath140 ergs s@xmath15 for powerful , high luminosity radio sources @xcite . for the subsequent analysis , we will adopt a finducial jet kinetic luminosity of 10@xmath112 ergs s@xmath15 , but suggest that the true value could be a factor of a few higher . thus , both jet power and bolometric luminosity estimates provide sufficient energy to power the observed outflow . both appear similarly powerful , thus their relative impact will depend on their coupling efficiency with the ambient ism and their ability to accelerate ambient interstellar material to large distance from the nucleus . the importance of radiation pressure in driving an outflow depends on the intensity of the radiation field itself , i.e. , the bolometric luminosity , @xmath122 . powerful agn have radiation fields that are likely intense enough to efficiently heat and accelerate the gas near their center through compton heating @xcite and dust opacity @xcite . through coulomb coupling of charged dust grains to the ionized gas , radiation pressure can in principle trigger an outflow . blue shifted x - ray absorption lines in quasar spectra indeed suggest parsec scale outflows , with mass loss rates of @xmath391 m@xmath141 yr@xmath15 and velocities @xmath390.l c ( e.g. , * ? ? ? * ) . however , e.g. , @xcite find that this gas generally can not drive large scale outflows . we describe a momentum driven agn wind , following , e.g. , @xcite and @xcite . the equation of motion for an optically thick wind is , @xmath142 with acceleration @xmath143 , and agn luminosity @xmath144 . @xmath145 denotes the gas mass as a function of radius . @xcite argue that the velocity as a function of the distance for the black hole , @xmath146 , is set by , @xmath147 with a threshold luminosity @xmath148 above which the agn is able to drive a significant , optically thick outflow , gas fraction @xmath149 , velocity dispersion of the host galaxy @xmath7 , and gravitational constant @xmath150 . @xmath151 is the radius where the acceleration sets on , namely the dust sublimation radius at a few pc from the black hole . @xmath148 can be derived in close analogy to the eddington limit for the central agn , @xmath152 . ( see 5 of * ? ? ? * for the full derivation of this threshold . ) . for a @xmath153 galaxy , @xmath154 ergs s@xmath15 . with @xmath155 ergs s@xmath15 , a reasonable gas fraction of 10 - 50% , and a velocity dispersion of 250 km s@xmath15 ( roughly corresponding to the stellar mass of mrc1138 - 262 ) , this suggests that @xmath156 ergs @xmath157 , well above the bolometric luminosity of mrc1138 - 262 . for a lower gas fraction and smaller velocity dispersion , radiation pressure might play a role , but it unlikely dominates in accelerating the gas to the high velocities we observe . for instance , if @xmath158=@xmath159 ( in good agreement with the @xmath160 implied by * ? ? ? * ) , the acceleration would have to have been maintained for a few @xmath24 10@xmath161 yrs in order to accelerate the emission line gas in mrc1138 - 262 to the observed velocities , likely about 1 - 2 orders of magnitude longer than the agn lifetime . typical line velocities would be about @xmath162 those observed , perhaps less since we can not accurately quantify the effect of projection . hence , radiation pressure does not seem sufficient to power the large - scale gas kinematics in mrc1138 - 262 , making the radio jet a more likely candidate to mediate the agn feedback . we derive a simple estimate for the observed coupling efficiency @xmath163 between the energy input rate @xmath164 ergs s@xmath15 of the radio jet ( [ subsec : mrc.lbol ] ) and the energy injection rates derived in [ sec : injection ] , @xmath165 erg s@xmath15 . given the uncertainties in either estimate , and in the spirit of providing lower limits , we adopt a loosely constrained coupling efficiency @xmath166 , corresponding to the jet kinetic energy estimate common to all methods of [ subsec : mrc.lbol ] . the `` dentist drill model '' @xcite is the simplest scenario explaining how the jet might interact with the ambient ism , which uses a termination shock to couple with the ism as the jet - head ploughs through the ambient medium . small variations of the jet direction cause the jet to jitter across the working surface , and the resulting cavity will be somewhat larger than the jet diameter itself . typical assumed values are @xmath167 @xcite . thus in the `` dentist drill '' scenario , the interaction between radio jet and the surrounding medium is limited to the working surface of the jet , and the coupling efficiency is approximately the area of the working surface , divided by 4@xmath168 . this can be parameterized as @xmath169 , or 0.045% for 15opening angle . to explain the @xmath170 erg s@xmath15 kinetic energy injected into the outflow of mrc1138 - 262 with the small @xmath171 implied by the dentist drill model would hence require an energy input of @xmath172 few @xmath173 erg s@xmath15 . this is above even the most extreme estimates of jet kinetic energies discussed in [ subsec : mrc.lbol ] . hence the observed kinematics in mrc1138 - 262 require a significantly more efficient coupling to the ambient ism than implied by the `` dentist drill '' model . line widths and bulk motions in mrc1138 - 262 indicate shock speeds @xmath174 . optical emission line properties of shocks with such speeds are very difficult to model , because emission arises predominantly through ionizing photons escaping upstream ahead of the shock into the `` precursor '' region . thus , accurately modelling the line emission from the shock depends crucially on the detailed properties of the ambient ism and the radiation field ( both background and that generated by the shock ) . following this argument and the modeling of @xcite , we use , @xmath175 to estimate the amount of shocked material from the measured h@xmath5 luminosity in mrc1138 - 262 ( [ sec : iontempdens ] ) . in this equation , @xmath176 is the number of recombinations per hydrogen atom . at least @xmath177 yr@xmath15 of shock material are necessary to produce the entire h@xmath5 luminosity , not accounting for extinction and possibly a lower efficiency of the emission ( fewer recombinations per hydrogen atom for example ) . over the likely @xmath119 yrs lifetime of the radio jet ( [ sec : timescale ] ) , this implies an unphysically large @xmath178 of shocked gas . on the other hand , the predicted surface brightness of a shock with 500 km s@xmath15 velocity is about 2@xmath2410@xmath179 ergs s@xmath15 kpc@xmath129 @xcite , and increases dramatically with velocity ( @xmath180 ) . shock heating across the projected area of the emission line nebula in mrc1138 - 262 , @xmath181 kpc@xmath182 , would therefore produce a total h@xmath5 luminosity of @xmath392.4@xmath2410@xmath183 , @xmath184% of what we observe . accounting for extinction will make this discrepancy yet more dramatic . since the amount of shocked material necessary to power the nebula is unphysically large and the predicted surface brightness is too low , it seems that shock excitation is probably not viable as the sole excitation mechanism of the emission line gas , and ionization by the agn will play a large role . @xcite reach a similar conclusion from longslit rest - frame uv spectra of 10 hzrgs . to estimate whether photoionization by the agn can heat the gas , we assume a uniformly dense , radiation bounded nebula in ionization equilibrium . with the properties we observe in mrc1138 - 262 , this requires an ionizing photon rate of @xmath185 , or @xmath186 for @xmath187 . the results of @xcite imply that on average each ionizing photon in powerful agn has an energy corresponding to @xmath188 times the ionizing potential of hydrogen ( i.e. , @xmath189 ) . we therefore constrain the ionizing luminosity @xmath190 , corresponding to @xmath191 , or @xmath192 . if @xmath193% of the bolometric luminosity of mrc1138 - 262 is ionizing @xcite , i.e. , @xmath194 ergs s@xmath15 ( [ subsec : mrc.lbol ] ) , the nebulosity of mrc1138 - 262 might well be mainly photoionized by the agn , if the escape fraction is @xmath195% . this estimate also implies that the nebula could easily be matter bounded . however , given the small filling factor ( @xmath196 , [ sec : timescale ] ) , a covering factor @xmath193% implies that the gas is distributed in elongated structures with large surface area and relatively small volume . the gas must either be clumped into large numbers of small , approximately spherical clouds , long and thin `` string - like '' structures , or in thin , broad sheets , for either shock or excitation of the nebula . whatever the exact geometry , our results are only consistent with a small volume filling factor , but a relatively large covering fraction . thus , the mechanical and optical emission line properties of mrc1138 - 262 give direct evidence for an outflow which is driven by the expanding radio source , and heated mainly through photoionization by the agn . our analysis of the kpc scale outflow in mrc1138 - 262 suggests that the agn is the best candidate for driving the outflow : the crossing time of the nebula ( @xmath197 yrs ) approximately equals the lifetime of the radio jet and typical estimates of agn lifetimes ( * ? ? ? * and references therein ) , and the agn also provides the energy input rate necessary to power the observed outflow , given it is efficiently coupled to the ism , with @xmath198 . observed mass loss rates are @xmath199 yr@xmath15 ( [ sec : timescale ] ) . population synthesis fits assuming exponentially declining star - formation histories with a variety of e - folding times ( including infinity for continuous star - formation ) over a wide range of maximum star - formation rates , suggest a stellar mass of @xmath200 . the range of best fit masses for all of the models suggest an uncertainty of about a factor 2 - 3 . the observed outflow rate therefore corresponds to a mass loss of @xmath193% of the stellar mass within @xmath201 years . observations of carbon monoxide in high redshift radio galaxies suggest gas masses of @xmath202 @xcite , corresponding to @xmath203% of the stellar mass of mrc1138 - 262 . given the large uncertainties in these estimates , we assume a gas fraction of @xmath3% of the stellar mass , implying that agn feedback could remove the total ism of the host within @xmath204 yrs . we emphasize that the mass loss rate is a strict lower bound estimated from the observed emission line gas mass . the true outflow might be up to an order of magnitude higher , and might well be larger than the total needed to quench future star - formation . the outflow in mrc1138 - 262 appears not aligned with the radio jets , but has a nearly spherical geometry , so that an adequate assessment of the dynamics in the extended emission line region clearly requires integral - field data , especially in the rest - frame optical , to be less sensitive to the source geometry , extinction , and the complex radiation transport in the rest - frame uv given the complex spatial structure of the emission line gas . we do not believe that mrc1138 - 262 is a `` one - off '' special case . broad forbidden emission lines in the nebulosities of hzrgs with fwhm@xmath16 are commonly found in longslit studies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? simulating a longslit spectrum of mrc1138 - 262 from our spiffi data with the `` slit''aligned with the radio axis , yields similar fluxes as given in , e.g. @xcite . we recently obtained sinfoni data of a small sample of hzrgs at z=2.5 - 3.5 , and find similarly extreme kinematics in the extended sources . in the following we will therefore assume that mrc1138 - 262 is in an evolutionary stage , which is short , @xmath205 yrs , but common for massive galaxies , and in particular for those which go through a phase of powerful nuclear activity at high redshift . a massive , agn driven outflow is a catastrophic event which can be expected to leave measurable imprints on the low - redshift descendents of the agn host galaxy . semi - analytical models of massive gas - rich mergers with a multi - phase ism in the early universe are found to reproduce the local luminosity function of massive , red early - type galaxies if they include a phase of strong agn feedback @xcite . @xcite find that these models can explain the tilt in the fundamental plane assuming dissipative , high redshift merging of galaxies with gas fractions @xmath206 , including star - formation and agn - related feedback , and @xcite are able to reproduce the observed m@xmath207@xmath7 relationship using this mechanism . in spite of these successes , such models can not resolve the black hole accretion , therefore agn feedback is included in a rather schematic way , neglecting , e.g . , radiatively inefficient black hole growth @xcite . they assume isotropic thermal coupling between agn and ism with efficiencies of @xmath208% , and imply that star - formation is suppressed within a few @xmath9 yrs . we note that this efficiency is lower than the coupling efficiency we find in mrc1138 - 262 ( @xmath209 ) , but the time scales agree surprisingly well , given the crudeness of the models and , by necessity , the `` order - of - magnitude '' nature of our analysis . to set observational constraints on the impact of agn driven winds , we now investigate whether the outflow in mrc1138 - 262 can explain the observed properties of the low redshift massive galaxy population . a massive outflow related to major mergers at high redshift is in overall agreement with a population of mostly spheroidal galaxies , stellar populations that have luminosity - weighted ages consistent with formation at redshifts @xmath210 @xcite , and with extremely low gas contents . a more quantitative constraint comes from the relative enhancement of @xmath5 elements with respect to the iron content of these galaxies , @xmath8 $ ] ( e.g. , * ? ? ? * ; * ? ? ? the abundance of @xmath5-elements relative to iron traces the relative numbers of thermonuclear and core - collapse supernovae and effectively traces the time scale of star formation due to the relative delay in the thermonuclear supernova . @xcite propose a top - heavy imf as alternative explanation for the observed @xmath8 $ ] overabundance . however , as @xcite point out , the _ relative _ @xmath8 $ ] overabundance in fact is an _ absolute _ fe underabundance , which indicates a relative lack of contribution of the thermonuclear supernovae to the overall metal abundances rather than an enhancement of core - collapse supernovae . this argument does not favor explaining the relative @xmath8 $ ] enhancement compared to the solar value with a top - heavy imf . moreover , @xcite find from semi - analytical modeling that the hypothesis of a top - heavy imf can not explain the observed tight correlation between @xmath8 $ ] enhancement and velocity dispersion in early - type galaxies , but agrees with a more pronounced early truncation of star - formation with increasing galaxy mass . a mechanism which efficiently removes nearly all gas from the host in @xmath211 gyrs is therefore more promising , and was suggested by @xcite and @xcite for massive early - type galaxies . outflows with the observed mass loss rate in mrc1138 - 262 , @xmath212 yr@xmath15 , fulfill this constraint , given that sufficient kinetic energy is injected into the ism during the agn luminous phase . in [ subsec : simpleenergy ] we estimate that @xmath213 ergs or more are injected into the ism of mrc1138 - 262 during an assumed agn lifetime of @xmath119 yrs . the mechanical energy of the agn is able to entrain ambient material , propelling it outwards at the high velocities observed . the maximum amount of ambient material ( if accelerated to escape velocity ) can be approximated by an energy conservation argument as : @xmath214 where @xmath215 is the energy of the outflow in units of @xmath216 ergs and @xmath217 is the escape velocity in units of 500 . @xmath218 is difficult to constrain . for galaxies with nfw profiles @xcite , there is a simple relationship between @xmath218 , virial radius and mean mass density of the halo over the background dark matter density ( see * ? ? ? * ) . for a baryonic mass fraction of 10% , the escape velocity in mrc1138 - 262 is about 700 at the virial radius of about 120 kpc @xcite . equation [ eqn : mejmax ] implies that mrc1138 - 262 may eject , @xmath219 9 @xmath24 10@xmath220 of material ( assuming @xmath215=9 . , which is the maximum of our estimated values , and @xmath217=1.4 ) , or @xmath219 3 @xmath24 10@xmath220 assuming our fiducial value of @xmath215=3 .. this is of course a very crude estimate , but it does suggest that , plausibly , much of the baryonic mass could escape the potential of the galaxy and thus will not be available for future star - formation . however , this represents a strict upper limit to the total mass ejection . we have not assumed a coupling factor between the energy and the ambient gas , and this factor is undoubtably less than one ( we estimate @xmath136 ) . subsequently , we will adopt this maximum mass ejection estimate and a coupling efficiency of 10% as the upper boundary , and the simple observed mass loss rate from as a lower boundary to the most likely range of mass loss rates and to constrain the amount of mass and metals ejected by agn winds . in addition to suppressing star formation within the short timescale set by the [ @xmath5/fe ] `` clock '' , the agn must also be able to inhibit the accretion of gas from the surrounding igm or icm to effectively halt future star formation . @xcite find that it is necessary to incorporate such a mechanism into their simulation , to inhibit galaxy growth in a population of massive red galaxies , and to produce a mass function with the appropriate exponential cutoff at the high end . in a very interesting analysis of a large sample of radio loud agn based on data from the sloan digital sky survey , @xcite find that the dependence of the size of the radio - loud agn fraction on black hole mass mirrors that of the cooling rate within the hot atmospheres of elliptical galaxies . @xcite further argue that the agn could be fueled by such cooling gas and that agn heating could plausibly balance the cooling of the gas over time . could the energy output of mrc1138 - 262 balance gas accretion from the igm with a similar mechanism ? following the arguments in @xcite , we can estimate whether or not it is even plausible for the radio jets in mrc1138 - 262 to inhibit the accretion of gas onto the dark matter halo . @xcite estimate the rate at which gas is accreted onto dark matter halos as a function of halo virial mass and redshift . for a halo with @xmath221 at @xmath222 , they give an accretion rate of @xmath223 yr@xmath15 . the gas is assumed to form into a hot , cooling halo of gas filling the dark matter halo , thus its energy can be estimated using simple kinetic arguments . making this assumption , the rate at which energy is deposited in the halo with a virial velocity , @xmath224 , is approximately : @xmath225 over the lifetime of the agn , this will be , @xmath226 . thus both the instantaneous ( @xmath22710@xmath112 ergs s@xmath15 ) and the total energy ejection rate ( @xmath2273 @xmath24 10@xmath228 ergs ) of the agn appear sufficient to stop the collapse of gas at the virial radius of the dark matter halo . if there is a connection between mass accretion and igniting agn activity in galaxies , then the output of a single radio loud agn cycle appears sufficiently energetic to suppress subsequent gas flow at the virial radius . radio lobes at high redshift reach distances of @xmath229 kpc . therefore the range and amount of energy injection seem sufficient to plausibly resist the gas accretion flow , and to influence the final baryonic mass of galaxies hosting powerful radio sources . of course the energy output from the agn must couple to the infalling gas for it to be effective . certainly the high efficiencies we have estimated for the gas within 10 - 20 kpc of the nucleus would be sufficient . this is a fundamental difference to starburst - driven winds which will heat the gas on similar time scales , but lack the energy and power necessary to unbind significant amounts of gas from the most massive halos , and are very unlikely to stop the accretion flow . if the ejected gas rains back onto the galaxy , and this material supports further star - formation , the enhanced @xmath8 $ ] enhancement will not be preserved . similar arguments disfavor models where star - formation has consumed most of the gas prior to the agn bright phase , e.g. , @xcite , and approaches where ultraluminous starbursts are triggered through _ positive _ agn feedback @xcite , provided that most of the stellar mass observed at @xmath230 are formed in such a burst . the short lifetimes of powerful agn ( @xmath197 yrs ; * ? ? ? * ) compared to a hubble time indicate that nuclear activity is likely episodic . to provide an estimate of the cosmic energy density released through agn feedback , we will therefore assume that agn hosts have only one cycle of activity that lasts @xmath231 yrs with coupling efficiencies to the host s ism and outflow characteristics similar to mrc1138 - 262 . the energy output during one such cycle follows from the previous analysis , @xmath232 ergs ( with @xmath233=10@xmath112 ergs s@xmath15 and @xmath234=10@xmath235 years ) . to estimate the agn duty cycle directly from the number of observed luminous agn , we use the co - moving number density of luminous agn calculated by , e.g. , @xcite , @xmath236 mpc@xmath237 for optically selected high - redshift quasars brighter than @xmath238 . if the `` unified model '' of agn is correct , then qsos are a subsample of all agn , and the observed number density of qsos will be a few times lower than the intrinsic number density of the full agn population . we also assume that a `` radio loud '' episode is an intrisic part of the lifecycle of all powerful agn . figure 9 of @xcite shows that 95% of all quasars are at @xmath239 , within an epoch of @xmath240 gyrs . for @xmath94 yrs , this yields this is again a lower limit , and neglects radiatively inefficient black hole growth . the overall energy budget of powerful agn can then be gauged by , @xmath242 as a result , powerful agn provide a cosmic energy density due to outflows of @xmath243 . we note that this _ global _ estimate is relatively robust against uncertain agn lifetimes , because increased durations leads to lower duty - cycles and vice versa . we can also use the local co - moving mass density of supermassive black holes ( smbh ) as a measure for the integral accreted black hole mass to constrain the cosmological significance of agn feedback . using the co - moving number density of spheroids and the m@xmath244 relation , @xcite estimate the local mass density of nuclear black holes , @xmath245 @xmath43 h@xmath246 mpc@xmath237 . to estimate the total kinetic luminosity corresponding to this black hole mass density , we use the simple scaling , @xmath247 with the efficiency of the kinetic energy output relative to the radiative ( bolometric ) luminosity , @xmath248 , and the radiative efficiency relative to the accreted rest mass energy , @xmath249 . the radiative efficiency of massive black holes has been estimated to be roughly 10% to 30% ( e.g. , * ? ? ? * and references therein ) , whereas there is a broad range of relative kinetic energy efficiency of 0.05 to 1 @xcite . adopting this range of values suggests that the kinetic energy density is approximately 2 to 100 @xmath2410@xmath250 ergs mpc@xmath237 . for a fiducial @xmath248=1 , the estimated kinetic energy density ejected is about 5 @xmath24 10@xmath250 ergs mpc@xmath237 , @xmath251 higher than what we estimated from the qso co - moving space density and duty cycle . this however implies that all black hole growth is related to gas accretion , and occurs during agn luminous phases . both assumptions are unrealistic , as indicated , e.g. , by the large population of obscured quasars and the possibility of bh - bh mergers . this result therefore sets an upper constraint to the energy density ejected by agn . in the spirit of setting lower constraints we adopt the first value , and note that our calculation does not violate any low - redshift constraints from the low - redshift number densities of supermassive black holes . simple energetic arguments alone show the potentially large significance of agn feedback on the evolution of @xmath252 local early type galaxies . @xcite estimate a local co - moving space density of @xmath253 galaxies , 0.0020 h@xmath254 mpc@xmath237 , @xcite find a local co - moving space density of 0.0022 h@xmath255 mpc@xmath237 for m@xmath256 early type galaxies . using the co - moving density of m@xmath256 early type galaxies as a normalization factor , and the upper limit to the kinetic energy density in [ subsubsec : simpleenergy ] , @xmath257 ergs mpc@xmath237 , we find that _ on average _ , each m@xmath256 early - type galaxy ejects @xmath258 ergs , roughly the binding energy of an elliptical galaxy with total mass @xmath259 ( or a galaxy about one magnitude brighter than @xmath260 ; * ? ? ? this is of course a statistical estimate to relate the impact of such outflows to the population of massive galaxies . it does not require or imply that each m@xmath256 early - type galaxy indeed went through a mrc1138 - 262-like epoch . given the large amount of ambient ism that the agn likely drives out from the galaxy and perhaps from the surrounding dark matter halo , it is pertinent to ask : if mrc1138 - 262 is typical , how much mass ( and mass of metals ) could such outflows eject ? using either the observed outflow rate for @xmath119 yrs or the maximum total mass ejection , we have , @xmath261 ( [ subsec : alphafe ] ) , with the co - moving qso number density and the inverse of their duty cycle , we find that on average , @xmath262 of material will be swept out by agn feedback , where @xmath263 is the co - moving space density of qsos in units of 5@xmath2410@xmath264 mpc@xmath237 and @xmath265 is the inverse duty cycle in units of 300 . this implies that approximately @xmath266 of material is ejected per m@xmath256 galaxy . within these loose constraints , this is about the mass of ism in a low - redshift gas rich galaxy and also roughly the molecular gas mass in high - redshift submm and radio galaxies . lacking robust constraints on the gas - phase metallicity in hzrgs , we use the metallicities of low - redshift massive galaxies to constrain the metal loss induced by agn feedback . the local mass - metallicity relationship indicates that local @xmath267 galaxies have solar and greater metallicities @xcite . with our above analysis , and scaled to solar metallicity @xmath268 , this implies that powerful agn feedback expels up to @xmath269 of metals per m@xmath256 galaxy . early - type nearby galaxies about a magnitude brighter than @xmath253 have total baryonic masses of @xmath270 , and super - solar metallicities by @xmath271 dex @xcite , which implies a total metal content of @xmath272 . this indicates that mrc1138 - 262-like feedback could eject a substantial fraction of the metals contained in the stars of an @xmath273 elliptical in the local universe . this metal - rich outflow may contribute significantly to the overall metallicity density at high redshift . @xcite estimate the co - moving metallicity density at z@xmath274 from integrating the star - formation rate density , @xmath275 mpc@xmath237 . if outflow rates in powerful agn are generally similar to mrc1138 - 262 , then our above estimate implies a total metal ejection of @xmath276 mpc@xmath237 . @xcite were able to account for @xmath27750% of the total metallicity density at z@xmath278 in the known population of galaxies . thus , agn feedback from massive galaxies may account for of - order a few percent to about 20% of all metals in the igm at z@xmath392 . although these estimates are highly uncertain , it appears very likely that agn - driven outflows are a significant contributor to the overall metal distribution in the igm in the early universe . we presented a study of the spatially - resolved dynamics in the optical emission line gas around the [email protected] powerful radio galaxy mrc1138 - 262 , using the near - infrared integral - field spectrograph spiffi on ut1 of the vlt . the large - scale kinematics in the emission line nebula of this galaxy are not consistent with the signatures of large - scale gravitational motion or starburst - driven winds . velocities and fwhms of @xmath279 indicate a vigorous outflow with an estimated total energy which is orders of magnitude higher than suggested for starburst - driven winds . based on timescale and energy arguments , we conclude that the observed outflow , which is a strict lower limit to the intrinsic amount of outflowing gas , is most plausibly driven by the agn . this implies a relatively efficient interaction between agn and interstellar medium , with efficiency @xmath280 , mostly mediated by mechanical energy injection through the radio jet . radiation - driven winds do not appear powerful enough to explain the observed velocities over the size of the optical line emitting gas . for mrc1138 - 262 we estimate from simple energetic arguments that the total energy needed to drive the observed outflow of optical emission line gas is of - order few @xmath24 @xmath216 ergs with a mass outflow rate , @xmath281 yr@xmath15 . if we assume that every luminous qso near the peak of their co - moving density at z@xmath392 has such an outflow phase with similar characteristics to mrc1138 - 262 we find that powerful agn eject @xmath282 ergs mpc@xmath237 of energy into the igm and @xmath283 mpc@xmath237 of gas , including @xmath284 mpc@xmath237 in metals . these energy and masses are significant compared to the the total binding energy and mass budget of @xmath253 early - type galaxies in the local universe . if mrc1138 - 262 is indeed archetypal , then agn winds have the potential to be of similar , perhaps even larger , cosmological signficance than starburst - driven winds , especially for the most massive galaxies . they accelerate the outflowing gas to much higher velocities than starburst - driven winds , and much of this material and energy are likely to escape the potential of even the most massive individual galaxy halo . outflows like in mrc1138 - 262 have the potential to explain the observed properties of low - redshift massive galaxies , namely their old , and narrow range of ages and low gas fractions . moreover , the large mass loss rates and short timescales of their outflows can naturally explain the observed enhancement of [ @xmath5/fe ] relative metal abundances observed in the most massive galaxies at low redshift . agn - driven outflows appear to be a plausible mechanism that efficiently suppresses star - formation within a few @xmath9 yrs . they also have high enough energies to perhaps stop the accretion flow of matter from the igm into their dark matter halos . thus , if accretion is followed by agn activity , then agn feedback may be an effective mechanism for regulating galaxy growth @xcite . our study comprises only one source , but we have indirect evidence for mrc1138 - 262 being far from unique . velocities of @xmath285 are not uncommon in the extended emission line gas of powerful radio galaxies , but are generally extracted from longslit spectra aligned with the radio jet axis only . simulating a longslit spectrum from the spiffi data cube , we find good agreement between mrc1138 - 262 and these samples . the bolometric and radio jet luminosities of mrc1138 - 262 are not unusual , and the source follows the @xmath286 relationship of high - redshift radio galaxies , if correcting for the luminous broad line emission . we are currently acquiring rest - frame optical , integral - field data for a larger sample of hzrgs , to make statistically more robust predictions . preliminary results strongly support the present analysis , giving further evidence that agn feedback plays a major role in the evolution of the most massive galaxies in the universe . martini , p. 2004 , invited review to appear in `` carnegie observatories astrophysics series , vol . 1 : coevolution of black holes and galaxies , '' ed . l. c. ho ( cambridge : cambridge univ . press ) , astro - ph/0304009

statistical properties of order - driven double - auction markets with bid - ask spread are investigated through the dynamical quantities such as response function . we first attempt to utilize the so - called _ madhavan - richardson - roomans model _ ( mrr for short ) to simulate the stochastic process of the price - change in empirical data sets ( say , eur / jpy or usd / jpy exchange rates ) in which the bid - ask spread fluctuates in time . we find that the mrr theory apparently fails to simulate so much as the qualitative behaviour ( ` non - monotonic ' behaviour ) of the response function ( denotes the difference of times at which the response function is evaluated ) calculated from the data . especially , we confirm that the stochastic nature of the bid - ask spread causes apparent deviations from a linear relationship between the and the auto - correlation function , namely , . to make the microscopic model of double - auction markets having stochastic bid - ask spread , we use the minority game with a finite market history length and find numerically that appropriate extension of the game shows quite similar behaviour of the response function to the empirical evidence . we also reveal that the minority game modeling with the adaptive ( ` annealed ' ) look - up table reproduces the non - linear relationship ( stands for a non - linear function leading to `-shapes ' ) more effectively than the fixed ( ` quenched ' ) look - up table does .

1011.0748

how a specific trading mechanism effects on the price formation is one of the essential queries to understand the process and outcomes of exchanging assets under a given concrete rule . to investigate the issue , a lot of studies concerning the micro - structure of markets have been done in various research fields @xcite . recently , lots of on - line trading services on the internet were constructed by several major banks such as the sony bank @xcite . as the result , one can gather a lot of trading data sets to investigate the statistical properties extensively . as such studies , several authors focused on the fact that the sony bank uses a trading system in which foreign currency exchange rates change according to a first - passage process ( fpp ) @xcite . automatic forex trading systems such as the sony bank are now popular in japan where many investors use a scheme called _ carry trade _ by borrowing money in a currency with low interest rate and lending it in a currency offering higher interest rates . with these demands in mind , several studies have been done to investigate the stochastic process and made a model of it to reproduce the fpp in order to provide useful information for customers @xcite . the data sets of the sony bank rate @xcite are composed of time index and trading rate at that time . as we explained , a huge number of market data are reduced to a small amount of it , namely , the number of the sony bank rate is reduced by the first - passage process and unfortunately , the market rates behind the sony bank rate are not available for us . as well - known , there are several data sets whose price are determined by the so - called _ double - auction _ system . in the double - auction market , each trader ( investor ) posts his ( or her ) selling price or buying price in market order of a specific commodity with its volume to the market . then , the market maker determines the minimum price of buying orders , what we call _ ask _ , and the maximum price of selling orders , the so - called _ bid _ at each trading time and discloses these prices to the public . then , the difference between the bid and the ask is referred to as _ spread _ or _ bid - ask spread_. in market rates available for traders ( on the web for instance ) , there are two types of bid - ask spread , that is , ` constant ' or ` distributed ' , and which type of spread is disclosed depends on the market makers ( securities companies ) . results of market making , especially , statistical properties of the bid - ask spread might have an impact on the market and several studies have been done to reveal the relationship between the properties of bid - ask spread and behaviour of the market @xcite . for instance , madhavan , richardson and roomans @xcite proposed a phenomenological model to explain the price dynamics of double - auction market in market order , however , their model is apparently limited to the case in which the bid - ask spread remains constant during the price dynamics . therefore , much more extensive studies including empirical data analysis seem to be needed to investigate to what extent the model proposed by madhavan , richardson and roomans can explain the behaviour of market with _ stochastic _ bid - ask spread through some relevant quantity . in this paper , we investigate statistical properties of double - auction markets with bid - ask spread through the dynamical quantities such as response function . we first attempt to examine the so - called _ madhavan - richardson - roomans model _ ( mrr for short ) to simulate the stochastic process of the price - change in empirical data sets ( say , eur / jpy or usd / jpy exchange rates ) in which the bid - ask spread fluctuates in time . we find that the mrr theory apparently does not simulate so much as the qualitative behaviour ( ` non - monotonic ' behaviour ) of the response function calculated from the data sets . it is possible for us to show that a linear relationship @xmath3 between auto - correlation function @xmath2 and response function @xmath0 holds for the mrr model . namely , these two macroscopic quantities are related each other and the relationship should be explained from the microscopic view point as statistical physics provides the microscopic foundation of thermodynamics . moreover , we find that the linear relationship @xmath3 is apparently broken down in order - driven double - auction markets with fluctuating bid - ask spread . this fact tells us that on the analogy of physics , the phenomenological mrr model for the constant spread might be regarded as ` thermodynamics ' which usually deals with the macroscopic quantities such as price , auto - correlation , response functions and the relationship between them . it does not need to consider the detail behaviour of microscopic ingredients such as traders . in this paper , we show that the phenomenological model is apparently limited and fails to reproduce the dynamical quantities @xmath7 efficiently . hence , here we attempt to construct a kind of ` statistical mechanics ' in finance , which provides a microscopic foundation of phenomenological theory such as the mrr model . for this end , we utilize the minority game with a finite market history length having the distributed bid - ask spread to reproduce similar behaviour of macroscopic dynamical quantities as the empirical evidence shows . in our minority game modeling , we first fix each decision component ( buying : @xmath8 , selling : @xmath9 ) in their look - up tables before playing the game ( in this sense , the decision components are ` quenched variables ' in the literature of disordered spin systems such as spin glasses @xcite ) . we also consider the case in which a certain amount of traders update their decision components according to the macroscopic market history ( they ` learn ' from the behaviour of markets ) so as to be categorized into two groups with a finite probability ( in this sense , the components are now regarded as ` annealed variables ' ) . namely , at each round of the game , if the number of sellers is smaller / greater than that of buyers , a fraction of traders , what we call _ optimistic group_/_pessimistic group _ , is more likely to rewrite their own decision components from @xmath9/@xmath8 to @xmath8/@xmath9 . we find that the minority game modeling with the adaptive look - up table reproduces the non - linear relationship @xmath4 ( @xmath5 stands for a non - linear function leading to ` @xmath6 -shapes ' ) more effectively than fixed ( frozen ) look - up table does . this paper is organized as follows . in the next section [ sec : data ] , we explain our data sets and their format . we investigate their statistical properties . in the next section [ sec : data_analysis ] , we evaluate two relevant quantities , namely , the auto - correlation function @xmath2 and the response function @xmath0 , which are our key quantities to discuss the double - auction markets , for our data sets . we find that a linear relationship @xmath3 holds for the data sets having a constant bid - ask spread , however , the relation is broken for the data with a stochastic spread . in section [ sec : mrr ] , we introduce the so - called mrr model as a phenomenological model and derive the @xmath2 and @xmath0 . the difference between the mrr theory and the empirical evidence , the origin of the difference is discussed . in section [ sec : mg ] , we introduce and modify the minority game with a finite market history length and apply to explain the typical behaviour of the response function for the data with stochastic bid - ask spread . in section [ sec : adapt ] , we reveal that the minority game modeling with the adaptive ( ` annealed ' ) look - up table reproduces the non - linear relationship @xmath4 ( @xmath5 stands for a non - linear function leading to ` @xmath6-shapes ' ) more effectively than fixed ( ` quenched ' ) look - up table does . in section [ sec : discuss ] , we comment on the possible extensions of our approach . the last section is summary . in order to check the validity of our modeling of markets , we gathered data sets of double - auction markets from the web site http://www.metaquotes.net/ @xcite by using the free software _ metatrader4_. we shall explain the data format of the _ matatrader 4_. we used the script which is available on the web @xcite . by using the script , the data sets are stored as the following format : + .... 2009/12/24,17:17:40,131.053,131.092 2009/12/24,17:17:41,131.053,131.088 2009/12/24,17:17:41,131.052,131.088 2009/12/24,17:17:43,131.048,131.071 2009/12/24,17:17:44,131.043,131.076 .................................. .................................. .... + from the far left column to the far right , transaction time ( year / month / day , hour : min : sec ) , bid , ask are shown . for instance , the first line denotes the bid is 131.053 and ask is 131.092 on 24th december 2009 at 17:17:40 . in this paper , we treat the data set written by the above format . among data sets concerning various different financial assets , we shall use here specific four data sets , namely , usd / jpy exchange rates ( 23rd-28th october 2009 ) , eur / jpy exchange rates ( 22nd-28th november 2009 ) , nasdaq100 ( 22nd-31st october 2009 ) and price of gold ( 28th-30th october 2009 ) . each data set contains @xmath10-data points . in the conventional ( standard ) data for continuous - time double - auction markets , we usually use the data having transactions ( buying or selling price and the transaction time ) including the quote ( bid and ask prices posted to the market with the time ) . however , unfortunately , the data set provided by the _ metatrader4 _ does not contain any information about the transaction . namely , the ` bid and ask values ' we mentioned above are the best selling price and the best buying price , and the time at which the transaction takes place . therefore , the price itself is not available for these data sets . hence , here we assume that the mid point @xmath11 of the bid @xmath12 and ask @xmath13 at time @xmath14 , that is , @xmath15 is a sort of buying or selling price when the transaction is approved . then , we shall define the sign of the ` return ' of the mid points ( the difference between successive mid points ) as a _ selling - buying signal _ @xmath16 , namely , @xmath17 . of course , these definitions of ` prices ' and the ` selling - buying signals ' are different from the conventional one , however , we shall try to investigate the behaviour of the system having such a slightly different definition of the prices and signals in limited data sets . in this section , we first calculate the histogram of the return of the mid point @xmath18 . the results are shown in fig . [ fig : fg1 ] . from these panels , we clearly find that the return of the mid point is distributed with ` heavy tails ' as the conventional return of the price has @xcite . to compare the results with gaussians , we calculate the empirical mean @xmath19 and the empirical variance @xmath20 for each data and plot the corresponding gaussian @xmath21 in the same panel . for our four kinds of data sets . from the upper left to the lower right , @xmath22 for eur / jpy exchange rates , usd / jpy exchange rates , nasdaq100 and price of gold are plotted . to compare the results with normal gaussians , we calculate the empirical mean @xmath19 and the empirical variance @xmath20 for each data and plot the gaussian @xmath21 in the same panel . from the upper left to the lower right , these normal gaussians are @xmath23 and @xmath24 , respectively . , title="fig:",width=226 ] for our four kinds of data sets . from the upper left to the lower right , @xmath22 for eur / jpy exchange rates , usd / jpy exchange rates , nasdaq100 and price of gold are plotted . to compare the results with normal gaussians , we calculate the empirical mean @xmath19 and the empirical variance @xmath20 for each data and plot the gaussian @xmath21 in the same panel . from the upper left to the lower right , these normal gaussians are @xmath23 and @xmath24 , respectively . , title="fig:",width=226 ] + for our four kinds of data sets . from the upper left to the lower right , @xmath22 for eur / jpy exchange rates , usd / jpy exchange rates , nasdaq100 and price of gold are plotted . to compare the results with normal gaussians , we calculate the empirical mean @xmath19 and the empirical variance @xmath20 for each data and plot the gaussian @xmath21 in the same panel . from the upper left to the lower right , these normal gaussians are @xmath23 and @xmath24 , respectively . , title="fig:",width=226 ] for our four kinds of data sets . from the upper left to the lower right , @xmath22 for eur / jpy exchange rates , usd / jpy exchange rates , nasdaq100 and price of gold are plotted . to compare the results with normal gaussians , we calculate the empirical mean @xmath19 and the empirical variance @xmath20 for each data and plot the gaussian @xmath21 in the same panel . from the upper left to the lower right , these normal gaussians are @xmath23 and @xmath24 , respectively . , title="fig:",width=226 ] we next focus on the bid - ask spread @xmath25 which is one of the key values in this study . we are confirmed that the above four data sets are classified into two types according to each statistical property of the spread . namely , the spread of usd / jpy or eur / jpy exchange rates is time - dependent and fluctuates , whereas , the spread of nasdaq100 or price of gold is a time - independent constant . , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] + , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] + , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] , the return of the mid point @xmath18 as a function of @xmath14 , and the distribution of the spread @xmath26 are plotted . , title="fig:",width=151 ] in fig . [ fig : fg_usd1 ] for eur / jpy and usd / jpy exchange rates , and in fig . [ fig : fg_nas1 ] for nasdaq100 and price of gold , we plot the mid point @xmath11 , the return of the mid point @xmath18 as a function of @xmath14 and the distribution of the bid - ask spread @xmath26 . from these figures , we clearly find that the bid - ask spread for the exchange rates apparently fluctuates , whereas the spread for nasdaq100 or price of gold is a constant leading up to a single delta peak in the empirical distribution . from now on , the data in which the spread fluctuates is refereed to as _ data with stochastic bid - ask spread _ , whereas , the data in which the spread is constant is called as _ data with constant bid - ask spread_. one of the main goals of this paper is to reveal the relationship between the statistical properties of bid - ask spread and the behaviour of auto - correlation and response functions for double - auction markets . in this section , we evaluate two macroscopic dynamical quantities , namely , auto - correlation and response functions by making use of empirical data analysis . these two relevant quantities are explicitly defined by @xmath27 in order to evaluate these functions , we need the information about selling - buying signal @xmath16 . however , as we mentioned in the previous section , the data gathered through the _ metatrader4 _ @xcite does not contain any information about it explicitly . to overcome this problem , we here assume that @xmath16 is given in terms of ` return ' of the mid point : @xmath28 namely , we assume when the mid point increases at the instant @xmath29 , the number of traders who posted their own buying signal to the market also increases . as the result , the selling - buying signal @xmath30 is more likely to take buying @xmath8 at that instant @xmath14 . under the above assumption , for our four data sets , namely , eur / jpy , usd / jpy exchange rates , nasdaq100 and price of gold , we calculate the auto - correlation function @xmath2 and the response function @xmath0 via ( [ eq : empiricalc ] ) and ( [ eq : empiricalr ] ) , respectively . we first plot the auto - correlation function for the above four data sets in fig . [ fig : rho_eur ] . . from the upper left to the lower right , we plot eur / jpy exchange rates ( the estimated @xmath31 ) , usd / jpy exchange rates ( the estimated @xmath32 ) , nasdaq100 ( the estimated @xmath33 ) , price of gold ( the estimated @xmath34 ) . , title="fig:",width=226 ] . from the upper left to the lower right , we plot eur / jpy exchange rates ( the estimated @xmath31 ) , usd / jpy exchange rates ( the estimated @xmath32 ) , nasdaq100 ( the estimated @xmath33 ) , price of gold ( the estimated @xmath34 ) . , title="fig:",width=226 ] + . from the upper left to the lower right , we plot eur / jpy exchange rates ( the estimated @xmath31 ) , usd / jpy exchange rates ( the estimated @xmath32 ) , nasdaq100 ( the estimated @xmath33 ) , price of gold ( the estimated @xmath34 ) . , title="fig:",width=226 ] . from the upper left to the lower right , we plot eur / jpy exchange rates ( the estimated @xmath31 ) , usd / jpy exchange rates ( the estimated @xmath32 ) , nasdaq100 ( the estimated @xmath33 ) , price of gold ( the estimated @xmath34 ) . , title="fig:",width=226 ] from the upper left to the lower right , we plot eur / jpy exchange rates ( under the assumption on the asymptotic form : @xmath35 , the estimated @xmath31 ) , usd / jpy exchange rates ( the estimated @xmath32 ) , nasdaq100 ( the estimated @xmath33 ) , price of gold ( the estimated @xmath34 ) . from these panels , we find that the correlation in the selling - buying signals decreases in the time difference @xmath1 although the result for usd / jpy exchange rate possesses the negative correlation in @xmath36 and converges to zero with a slight oscillation . we next evaluate the response function for our four data sets . the results are shown in fig . [ fig : fg_empirical ] . + in this figure , we plot the response function for the data with stochastic bid - ask spread ( the lower panels ) and for the data with a constant bid - ask spread ( the upper panels ) . from these panels , we find that some ` non - monotonic ' behaviour in @xmath0 appears for the data set with stochastic bid - ask spread . from the definitions , both auto - correlation and response functions are functions of the time - difference @xmath1 . in the previous subsections , we investigated their behaviour independently . however , it might be assumed that these two quantities are related each other . therefore , it is useful for us to make ` scatter plots ' to reveal the dynamical relationship underlying these two quantities . and @xmath2 . the upper two panels are results for the gold and nasdaq100 with constant spreads , whereas the lower panel denotes the result for the eur / jpy exchange rate having fluctuating spreads . we find that the linear relationship @xmath3 holds for the data having a constant spread , whereas the linear relationship is apparently broken down in the eur / jpy exchange rate which possesses a fluctuating bid - ask spread . , title="fig:",width=226 ] and @xmath2 . the upper two panels are results for the gold and nasdaq100 with constant spreads , whereas the lower panel denotes the result for the eur / jpy exchange rate having fluctuating spreads . we find that the linear relationship @xmath3 holds for the data having a constant spread , whereas the linear relationship is apparently broken down in the eur / jpy exchange rate which possesses a fluctuating bid - ask spread . , title="fig:",width=226 ] + and @xmath2 . the upper two panels are results for the gold and nasdaq100 with constant spreads , whereas the lower panel denotes the result for the eur / jpy exchange rate having fluctuating spreads . we find that the linear relationship @xmath3 holds for the data having a constant spread , whereas the linear relationship is apparently broken down in the eur / jpy exchange rate which possesses a fluctuating bid - ask spread . , title="fig:",width=226 ] in fig . [ fig : cr_real ] , we plot the relationship between @xmath0 and @xmath2 by scatter plots . the upper two panels are results for the gold and nasdaq100 with constant spreads , whereas the lower panel denotes the result for the eur / jpy exchange rate having fluctuating spread . we find that the linear relationship @xmath3 is apparently broken down in the eur / jpy exchange rate which possesses a fluctuating bid - ask spread . in the next section , we examine a phenomenological model to explain the non - linear relationship @xmath4 ( @xmath5 denotes a non - linear function ) theoretically . in order to explain the behaviour of the auto - correlation and response functions , we examine a phenomenological approach based on the so - called _ madhavan - richardson - roomans model _ ( mrr for short ) @xcite to simulate the stochastic process of price - change in empirical data sets . in the mrr model , the price @xmath37 updates according to the following rule . @xmath38 where @xmath39 denotes a noise in the market satisfying @xmath40 and @xmath41 . @xmath42 is a constant value to control the slope of instantaneous price change . the label @xmath16 means a _ selling - buying signal _ to represent @xmath43 for a buying signal and @xmath44 vice versa . behaviour of the above update rule is dependent on the statistical properties of selling - buying signals @xmath16 . @xmath45 is a correlation factor and in the mrr theory , we assume that @xmath16 follows a simple markovian process , namely , @xmath46 the price value of @xmath47 provided that the selling - buying signal in the previous time step is @xmath43 should be the ask @xmath13 and the price @xmath47 provided that the signal is @xmath44 should be the bid @xmath12 . hence , we naturally define the time - dependence of ask and bid as follows . @xmath48 where @xmath49 denotes a kind of transaction cost and the value itself is set to a constant in the mrr model . from these rules , we easily find the bid - ask spread at time @xmath14 as @xmath50 namely , in the mrr model , the spread is a time - independent constant during the dynamics . on the other hand , the mid point of the bid and ask is given by @xmath51 therefore , for the parameter choice @xmath52 or @xmath53 , the mid point @xmath11 is identical to the price @xmath37 . for the above update rules of price , bid , ask , spread and mid point , we investigate the macroscopic properties of double - auction markets through the auto - correlation function and the response function . from the definition of markovian process ( [ eq : markov ] ) , the auto - correlation function is given by @xmath54 we should keep in mind that the auto - correlation function is originally defined by ( [ eq : empiricalc ] ) . however , in the limit of @xmath55 , one can replace the time - average by the average over the joint probability of the stochastic variables @xmath56 as @xmath57 according to the law of large number . the correlation factor @xmath45 should be @xmath58 . for a positive @xmath45 , the correlation function decays exponentially as @xmath59 . we next consider the response function of the market , that is defined by @xmath60 where the bracket @xmath61 has the same meaning as that in ( [ eq : corr ] ) has . from the above response function , one obtains some information about the response of the market at time @xmath62 to the selling - buying signal at arbitrary time @xmath14 . namely , the response function measures to what extent the mid point increases ( decreases ) on average for interval @xmath1 when a buying ( selling ) signal is posted to the market @xmath1 steps before we observe the mid point . after simple algebra , we easily obtain the relationship between the response function @xmath0 and the correlation function @xmath2 for the mrr model as follows . @xmath63 as we saw before , for a positive correlation factor @xmath64 , the @xmath2 monotonically decreases as @xmath59 . hence , the response function also behaves monotonically and converges to @xmath42 as @xmath65 ( @xmath66 ) . it should be noticed that the linear relationship between @xmath2 and @xmath0 holds from ( [ eq : rc ] ) . from equation ( [ eq : rc ] ) , we also find @xmath67 and this fact tells us that @xmath68 holds . the volatility defined by @xmath69 also reads @xmath70 and @xmath71 , @xmath72 . we may use the above rigorous equations to check the validity of our computer simulations . in fig . [ fig : fg3 ] , we show the typical behaviour of response function for several choices of @xmath45 . from these panels , we find that for a positive correlation factor , the response function monotonically converges to the value @xmath42 . . the solid lines are theoretical predictions by @xmath73 . the boxes are obtained by numerical simulations for a finite @xmath74 . , title="fig:",width=226 ] . the solid lines are theoretical predictions by @xmath73 . the boxes are obtained by numerical simulations for a finite @xmath74 . , title="fig:",width=226 ] + . the solid lines are theoretical predictions by @xmath73 . the boxes are obtained by numerical simulations for a finite @xmath74 . , title="fig:",width=226 ] . the solid lines are theoretical predictions by @xmath73 . the boxes are obtained by numerical simulations for a finite @xmath74 . , title="fig:",width=226 ] in this figure , we also show the results obtained by simulating the update equation for the price ( [ eq : updatep ] ) and the mid point ( [ eq : pm ] ) , and calculating the response function numerically by making use of ( [ eq : empiricalr ] ) . we find that the both theoretical prediction ( solid lines ) and the simulation ( boxes ) are in good agreement . moreover , for the choice of positive correlation factors , say , @xmath75 and @xmath76 , the response function increases monotonically and converges to @xmath77 as the mrr theory predicted . in fig . [ fig : cr_mrr ] , we show the relationship between @xmath0 and @xmath2 for the mrr model . we clearly find that the linear relationship @xmath78 actually holds . we should notice that this result is completely different from the relationship shown in the lower panel of fig . [ fig : cr_real ] which is result for the data having fluctuating bid - ask spread . and @xmath2 for the mrr model . we find that the linear relationship @xmath78 holds . , width=302 ] this failure of the mrr model to simulate the ` @xmath6-shape ' of the @xmath2-@xmath0 scatter plot for the data with stochastic bid - ask spread is obviously due to the assumption of constant bid - ask spread , namely , @xmath79 on the mrr theory . we also conclude that the breaking of the linear relationship @xmath3 is macroscopically due to the fact that the response function evaluated on the basis of the mrr theory behaves ` monotonically ' and converges to a finite value @xmath42 . this fact is one of the limitations of the mrr theory to explain the empirical data for double - auction markets . to make the efficient model for the double - auction market with stochastic bid - ask spread , we use a game theoretical approach based on the so - called _ minority game_. in order to make a model to simulate the ` non - monotonic ' behaviour of the response function for the financial data with stochastic bid - ask spread , we start our argument from standard minority game @xcite with a finite market history length . in our computer simulations for the minority game , at each round @xmath14 ( time step ) of the game , each trader @xmath80 ( @xmath81 : @xmath82 should be an odd number to determine the ` minority group ' ) decides his ( or her ) decision : @xmath83 ( _ buy _ ) or @xmath84 ( _ sell _ ) to choose the minority group . then , we evaluate the total decision of the traders : @xmath85 for each round @xmath14 . it should be noted that the factor @xmath86 is needed to make the @xmath87 of order @xmath88 object ( independent of the size @xmath82 ) . from the definition ( [ eq : total_bid ] ) , the market is _ seller s market _ for @xmath89 , whereas the market behaves as _ buyer s market _ for @xmath90 . the @xmath87 follows complicated stochastic process and one might consider that the price @xmath91 is updated in terms of the @xmath87 as follows . @xmath92 \ } \label{eq : mgp}\end{aligned}\ ] ] where @xmath93 and @xmath94 are positive constants . the above update rule means that the price increases if the ` buying group ' is majority and decreases vice versa . this setting of the game seems to be naturally accepted . a bias term @xmath94 appearing in ( [ eq : mgp ] ) plays an important role to simulate the auto - correlation function as we will see later on . to decide ` buy ' or ` sell ' , each trader uses the following information vector defined by @xmath95 \\ { \rm sgn}[(1-\zeta)a(t-2)+\zeta z(t,2 ) ] \\ \cdots \\ { \rm sgn}[(1-\zeta)a(t - m)+\zeta z(t , m ) ] \end{array } \right)\end{aligned}\ ] ] where @xmath96 denotes a sign function and @xmath97 is a white noise defined by @xmath98 therefore , each trader uses the information of market through the up - down configuration of the return @xmath99 ( with some additive noise @xmath100 ) back to the previous @xmath101-steps . if the @xmath102 is close to @xmath88 , the ` real ' market history through the @xmath99 is hided by the ` fake ' market history through the noise @xmath100 . for a given information vector @xmath103 chosen from all possible @xmath104 candidates , each trader @xmath80 decides her ( or his ) action at round @xmath1 by the following strategy vector : @xmath105 where we defined @xmath106 as the index ( entry ) of the selected information vector @xmath107 and @xmath108 stands for the number of the possible strategies for each trader . each component of the above strategy vector @xmath109 takes @xmath8 ( buy ) or @xmath9 ( sell ) . therefore , each trader has her ( his ) own look - up table which is defined by a matrix with size @xmath110 as @xmath111 each component of the above look - up table @xmath112 is fixed ( ` quenched ' ) before playing the game . however , in the next section , we consider the case in which the components of look - up tables are rewritten during the game . in this paper , we concentrate ourselves to the simplest case of two strategies @xmath113 and @xmath114 ( ` real ' market history ) . then , the trader @xmath80 changes her / his own pay - off value @xmath115 by the following update rule : @xmath116 where @xmath117 denotes the kronecker s delta and @xmath118 means the _ optimal strategy _ in the sense that @xmath118 is given by @xmath119.\end{aligned}\ ] ] the meaning of the update rule ( [ eq : pa ] ) is given as follows . if @xmath89 and the majority group consists of traders who post their decisions @xmath8 to the market , the trader @xmath80 attempts to post her / his decision as an opposite sign of @xmath87 , namely , @xmath84 . thus , the trader @xmath80 acts so as to satisfy the condition @xmath120 which leads to increase of her / his pay - off value @xmath121 . by taking into account the fact that we are dealing with the case of @xmath113 ( @xmath122 ) , we rewrite the equation ( [ eq : pa ] ) as @xmath123 by means of @xmath124 substituting ( [ eq : bit2 ] ) into the definition of the total bit @xmath87 , we have @xmath125 we should notice that the above equation can be written by using the following relation @xmath126 \eta_{\lambda}^{i } , \,\,\ , w_{\lambda}^{i } = \frac{1}{2 } ( r_{\lambda}^{i1 } + r_{\lambda}^{i2 } ) .\end{aligned}\ ] ] then , we obtain the following coupled non - linear equations with respect to the total decision @xmath87 , the difference of pay - off values for two strategies @xmath127 and the update equation of the price @xmath128 : @xmath129 \eta_{\lambda}^{i } \right\ } \label{eq : al } \\ % % % % % % % % % % % % % % q_{i}(t+1 ) & = & q_{i}(t ) - \frac{1}{\sqrt{n } } \sum_{\lambda=1}^{2^{m } } \delta ( \lambda , \mbox{\boldmath $ \lambda$ } ( a , z : t ) ) \eta_{\lambda}^{i } a(t ) \label{eq : qi } \\ % % % % % p(t+1 ) & = & p(t ) + \beta \ { a(t ) + \psi \ , { \rm sgn}[a(t-1 ) ] \}. \label{eq : xl}\end{aligned}\ ] ] the above rules ( [ eq : qi])([eq : al ] ) and ( [ eq : xl ] ) are our basic equations to discuss the response of double - auction markets having stochastic bid - ask spread to instantaneous selling - buying signals . to make the bid - ask spread in our minority game , we assume that the buying price @xmath130 and the selling price @xmath131 which are posted to the market by each trader @xmath80 at round ( time ) @xmath14 are updated according to the following rules . @xmath132 where @xmath133 and @xmath134 are constants to be set so as to satisfy @xmath135 . @xmath136 is an uncorrelated gaussian variable with mean @xmath137 and covariance @xmath138 ( _ additive white gaussian noise : awgn _ ) . in our simulations , we set @xmath139 . then , the bid - ask spread at round @xmath14 is given by @xmath140 where the groups taking ` buying ' and ` selling ' decisions are refereed to as @xmath141 and @xmath142 , respectively ( @xmath143 ) . in fig . [ fig : fg_mgba ] , we plot the resulting distribution of the spread @xmath144 . generated by the minority game . , width=302 ] from fig . [ fig : fg_mgba ] , we find that the stochastic bid - ask spread generated from the above modeling based on the minority game actually fluctuates and possesses a non - trivial shape of the distribution . for the above set - up of the minority game , we evaluate two relevant statistics , namely , correlation function @xmath145 ( by ( [ eq : empiricalc ] ) ) and the response function @xmath146 ( by ( [ eq : empiricalr ] ) ) to compare the results with the empirical evidence for the data with stochastic bid - ask spread . we first examine the effect of the bias term @xmath94 on the correlation function . the results for @xmath147 are shown in fig . [ fig : fg_c_l2 ] ( left ) . in the minority game for @xmath147 ( left ) . we set @xmath148 and iterated the game @xmath149 rounds . the @xmath2 is zero for @xmath150 . the right panel is auto - correlation function of the mrr model with @xmath75 . the error - bars were calculated by 10-independent trials . , title="fig:",width=226 ] in the minority game for @xmath147 ( left ) . we set @xmath148 and iterated the game @xmath149 rounds . the @xmath2 is zero for @xmath150 . the right panel is auto - correlation function of the mrr model with @xmath75 . the error - bars were calculated by 10-independent trials . , title="fig:",width=226 ] from this panel , we find that @xmath151 for @xmath150 and the result is apparently different from the result for the empirical data . here we set @xmath148 and iterated the game @xmath152 rounds . the right panel of fig . [ fig : fg_c_l2 ] shows the correlation function of the mrr model for @xmath75 . the selling - buying signal @xmath153\end{aligned}\ ] ] is actually correlated automatically through the market history with length @xmath154 , however , the correlation strength is very weak . therefore , we need some other explicit correlation through the bias term @xmath94 which enhances the correlation by two - round back @xmath155 from the present @xmath156 . hence , we here choose the non - zero bias term @xmath94 to reproduce the auto - correlation function as observed in the empirical data . we checked that the results are robust against the slight differences in the parameters appearing in the game such as @xmath157 _ etc_. however , for only parameter @xmath94 , we should be careful to choose the value . this is because from the definition of update rule of the price ( [ eq : xl ] ) , the effect of the @xmath87 on the price change is relatively depressed by the large value of the @xmath94 . therefore , we should choose @xmath94 so as to make the value smaller than the standard deviation of the @xmath87 , namely , square root of the volatility as @xmath158 in our simulation , the square root of the volatility is estimated as @xmath159 . in fig . [ fig : fg_c_l ] , we plot the correlation function for the case of @xmath160 ( left panel ) and @xmath161 ( right ) . we should notice that these two choices of the bias term @xmath94 satisfy the condition ( [ eq : cond_psi ] ) . in the minority game for @xmath160 ( left ) and @xmath161 ( right ) . we set @xmath148 and iterated the game for @xmath149 rounds . the error - bars were calculated by 10-independent trials . , title="fig:",width=226 ] in the minority game for @xmath160 ( left ) and @xmath161 ( right ) . we set @xmath148 and iterated the game for @xmath149 rounds . the error - bars were calculated by 10-independent trials . , title="fig:",width=226 ] from these panels , we find that the correlation function decreases as we observed in the same function for the empirical data sets . we next plot the response function in fig . [ fig : fg_r_l ] for @xmath162(upper panel ) , @xmath160 ( lower left ) and @xmath161 ( lower right ) . in the minority game for @xmath147 ( upper ) , @xmath160 ( lower left ) and @xmath161 ( lower right ) . we set @xmath148 and iterated the game for @xmath149 rounds . the error - bars were calculated by 10-independent trials . , title="fig:",width=226 ] + in the minority game for @xmath147 ( upper ) , @xmath160 ( lower left ) and @xmath161 ( lower right ) . we set @xmath148 and iterated the game for @xmath149 rounds . the error - bars were calculated by 10-independent trials . , title="fig:",width=226 ] in the minority game for @xmath147 ( upper ) , @xmath160 ( lower left ) and @xmath161 ( lower right ) . we set @xmath148 and iterated the game for @xmath149 rounds . the error - bars were calculated by 10-independent trials . , title="fig:",width=226 ] and @xmath0 for minority game with @xmath147 ( upper ) , @xmath160 ( lower left ) and @xmath161 ( lower right ) . we set @xmath148 and iterated the game for @xmath149 rounds and we calculate @xmath7 up to @xmath163 . , title="fig:",width=226 ] + and @xmath0 for minority game with @xmath147 ( upper ) , @xmath160 ( lower left ) and @xmath161 ( lower right ) . we set @xmath148 and iterated the game for @xmath149 rounds and we calculate @xmath7 up to @xmath163 . , title="fig:",width=226 ] and @xmath0 for minority game with @xmath147 ( upper ) , @xmath160 ( lower left ) and @xmath161 ( lower right ) . we set @xmath148 and iterated the game for @xmath149 rounds and we calculate @xmath7 up to @xmath163 . , title="fig:",width=226 ] from these panels , we find that the behaviour of the response function is not monotonically increasing function leading up to the convergence to some constant value but ` non - monotonic ' as the response function of data sets having stochastic bid - ask spread ( eur / jpy , usd / jpy exchange rates ) shows ( see the lower panels of fig . [ fig : fg_empirical ] ) . in fig . [ fig : fg_c_r ] , we plot the relationship between the auto - correlation and response functions . in the upper panel , we show the result for zero bias term @xmath147 . we find that the curve is deviated from the linear relation @xmath3 . however , the shape is not ` @xmath6 ' as observed in the empirical evidence but ` @xmath164-shape ' . in the lower two panels are results for the non - zero bias term @xmath165 . we clearly find that the ` @xmath6-shapes ' appears and the results are qualitatively similar to those of the empirical data . in the previous section , we fixed each decision component ( buying : @xmath8 , selling : @xmath9 ) in their look - up tables before playing the game ( in this sense , the decision components are ` quenched variables ' in the literature of disordered spin systems such as spin glasses ) . however , in this paper , a certain amount of traders update their decision components according to the macroscopic market history ( they ` learn ' from the behaviour of markets ) so as to be categorized into two - groups with a finite probability ( in this sense , the components are now regarded as ` annealed variables ' ) . namely , at each round of the game , if the number of sellers is smaller / greater than that of buyers , a fraction of traders , what we call _ optimistic group_/_pessimistic group _ , is more likely to rewrite their own decision components from @xmath9/@xmath8 to @xmath8/@xmath9 . in realistic trading , we might change our mind and rewrite the components of the look - up table according to the market history . therefore , in this section we consider the case in which some amount of traders can rewrite their own table adaptively . we first consider the case in which each trader updates her / his own look - up table according to the latest information of the market . some of the traders make their decisions as ` buy ' when the market is _ seller s market _ , namely , the signal of the latest market is ` buying ' ( what we call _ optimistic group _ ) . on the other hand , they decide ` sell ' vice versa ( they are referred to as _ pessimistic group _ ) . namely , each trader rewrites the table according to the following algorithm . + + * adaptation algorithm using the latest market information * 1 . fix ( ` quench ' ) each component of the look - up table at the beginning of the game @xmath166 . 2 . at each game round @xmath14 for @xmath167 , each trader rewrites her / his component @xmath168 , where @xmath106 denotes the entry of market history for the information vector @xmath169 , with probability @xmath170 as @xmath171,\,\,\ , r_{\lambda}^{i2 } = { \rm sgn}[a(t-1)].\ ] ] 3 . each trader recovers her / his original ( at the beginning of the game ) look - up table with a probability @xmath172 at the next game round @xmath156 . 4 . repeat ( ii ) and ( iii ) until the game is over . namely , a fraction @xmath173 of the traders is categorized into the ` optimistic group ' if @xmath174 ( _ seller s market _ ) and into the ` pessimistic group ' if @xmath175 ( _ buyer s market _ ) . ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables . we set @xmath176 . @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath179 ( left ) and @xmath180 ( right ) . , title="fig:",width=226 ] ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables . we set @xmath176 . @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath179 ( left ) and @xmath180 ( right ) . , title="fig:",width=226 ] + ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables . we set @xmath176 . @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath179 ( left ) and @xmath180 ( right ) . , title="fig:",width=226 ] ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables . we set @xmath176 . @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath179 ( left ) and @xmath180 ( right ) . , title="fig:",width=226 ] + ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables . we set @xmath176 . @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath179 ( left ) and @xmath180 ( right ) . , title="fig:",width=226 ] ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables . we set @xmath176 . @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath179 ( left ) and @xmath180 ( right ) . , title="fig:",width=226 ] in fig . [ fig : r_mgame_simu3 ] , we show dynamical quantities @xmath2 ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables . we set @xmath176 and @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath179 ( left ) and @xmath180 ( right ) . from this figure , we find that the positive correlation @xmath181 for @xmath36 appears even if we set @xmath147 and non - monotonic behaviour of the response function is reproduced . as the result , ` @xmath6-shape ' in the @xmath2-@xmath0 scatter plots is generated . from these results , we conclude that the adaptive modification of the look - up table by using the latest market information for each trader works well to explain the empirical evidence . in the previous subsection , we succeeded in generating a positive finite auto - correlation by making use of the adaptive look - up table even if we set @xmath147 . however , in this look - up table , each trader changes her / his decision from the latest information about the market . as the result , the auto - correlation function decays to zero for @xmath150 . to modify the weak correlation , we construct the adaptive look - up table by using the information about the market history with length @xmath101 . as we mentioned , the information vector @xmath182 contains the useful information on the market . hence , we shall assume that each trader rewrites the component of her / his table as @xmath183,\,\,\ , r_{\lambda}^{i2 } = { \rm sgn}[\omega_{\lambda}(a , z : t)]\end{aligned}\ ] ] with probability @xmath184 where we defined @xmath185 \\ % % % % \mbox{\boldmath $ w$ } & \equiv & ( 2^{m-1},2^{m-2},\cdots,2^{0})\end{aligned}\ ] ] at each game round @xmath14 . we should keep in mind that each trader recovers her / his original look - up table with a probability @xmath172 at the next game round . the @xmath186 denotes cumulative weighted market status , namely , we assume that the importance of the market information decays as @xmath187 in the history length @xmath188 . for instance , for the information vector ( let us define the entry by @xmath106 ) having @xmath8 for all components : @xmath189,\cdots , { \rm sgn}[a(t - m)])= ( + 1,\cdots,+1)$ ] , that is , the market remains as _ seller s market _ up to @xmath101-times back , we obtain @xmath190 and @xmath191-traders rewrite their components as @xmath192=1 $ ] . as another example , when _ seller s market _ and _ buyer s market _ appears periodically as @xmath193 , we have @xmath194 and @xmath195-traders rewrite their component as @xmath196=-1 $ ] . let us summarize the above procedure as the following algorithm . + + * adaptation algorithm using the market history * 1 . fix ( ` quench ' ) each component of the look - up table at the beginning of the game @xmath166 . 2 . at each game round @xmath14 for @xmath167 , each trader rewrites her / his component @xmath168 , where @xmath106 denotes the entry of market history for the information vector @xmath169 , with probability as @xmath197,\,\,\ , r_{\lambda}^{i2 } = { \rm sgn}[\omega_{\lambda}(a , z , t)]\ ] ] with probability @xmath198 with @xmath199 $ ] . 3 . each trader recovers her / his original ( at the beginning of the game ) look - up table with a probability @xmath172 at the next game round @xmath156 . 4 . repeat ( ii ) and ( iii ) until the game is over . of @xmath200 ( left ) and the typical time - evolution of the probability @xmath201 for the first @xmath202-steps ( right ) . , title="fig:",width=226 ] of @xmath200 ( left ) and the typical time - evolution of the probability @xmath201 for the first @xmath202-steps ( right ) . , title="fig:",width=226 ] we set @xmath203 and select the same values as those of the previous section for the other parameters . in fig . [ fig : fg_omega_f3 ] , we first plot the generated distribution @xmath204 of @xmath200 ( left ) and the typical time - evolution of the probability @xmath201 for the first @xmath202 steps ( right ) . it should be noted that @xmath205 in the left panel is ranged from @xmath206 to @xmath207 because we set the history length @xmath154 in the definition of the @xmath186 . we find from the right panel that the typical time - evolution of the probability @xmath201 obeys complicated dynamics . we next show the results for the macroscopic quantities in fig . [ fig : r_mgame_simu4 ] . ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables for @xmath101 market history length . we set @xmath147 , @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath177 ( left ) and @xmath178 ( right ) . , title="fig:",width=226 ] ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables for @xmath101 market history length . we set @xmath147 , @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath177 ( left ) and @xmath178 ( right ) . , title="fig:",width=226 ] + ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables for @xmath101 market history length . we set @xmath147 , @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath177 ( left ) and @xmath178 ( right ) . , title="fig:",width=226 ] ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables for @xmath101 market history length . we set @xmath147 , @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath177 ( left ) and @xmath178 ( right ) . , title="fig:",width=226 ] + ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables for @xmath101 market history length . we set @xmath147 , @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath177 ( left ) and @xmath178 ( right ) . , title="fig:",width=226 ] ( upper left , middle left ) and @xmath0 ( upper right , middle right ) evaluated for the minority game with adaptive look - up tables for @xmath101 market history length . we set @xmath147 , @xmath177 ( upper panels ) , @xmath178 ( middle panels ) . the lower two panels show the relationship between the auto - correlation and response functions for @xmath177 ( left ) and @xmath178 ( right ) . , title="fig:",width=226 ] from this figure , we confirm that the ` @xmath6-shape ' in @xmath0-@xmath2 scatter plots are much similar to the empirical evidence than the results in the previous subsection . our general set - up presented in this paper is applicable to the analysis for the other quantities or for the other stochastic models . here we shall mention them briefly . as we mentioned in section [ sec : intro ] , we can generate the duration between the price changes within the framework of our minority game . let us introduce the maximum value of the spread @xmath208 which is determined by market makers . then , we might define a set of time points at which the price is updated as @xmath209 for these time points @xmath210 , the duration between successive price changes is given by @xmath211 we can evaluate the distribution @xmath212 and compare the results with well - known distributions , for instance , the mittag - leffler type @xcite . recently , vikram and sinha @xcite proposed a mean - field model to describe the collective behaviour of financial markets . in their model , each trader @xmath80 decides his ( her ) decision : @xmath213 ( buy ) , @xmath9 ( sell ) and @xmath214 ( no action ) at time @xmath14 according to the following probability . @xmath215 = 1-p[s_{i}(t)=0 ] & = & { \exp } \left ( -\mu { \biggr | } \log \frac{p ( t)}{\langle p ( t ) \rangle_{\tau } } { \biggr | } \right ) \label{eq : sinha_s}\end{aligned}\ ] ] where @xmath128 stands for the price at time @xmath14 and the bracket @xmath216 means the moving average over the past @xmath188-time steps . the parameter @xmath217 is a parameter which controls the sensitivity of an agent to the magnitude of deviation of the price from its moving average @xmath218 . for @xmath219 , the system reduces to a binary decision model where every agent trades at all time instants . namely , @xmath220=0 $ ] for @xmath219 means that the trader decides his ( her ) action , buy or sell , certainly . the traders who decide to trade at @xmath14 make his ( her ) action randomly , that is , @xmath221=p[s_{i}=-1]=1/2 $ ] . the price at time @xmath156 is decided by the following recursion relation . @xmath222 where @xmath223 . in fig.[fig : sinha ] , we plot the typical time - evolution of the price @xmath128 and the moving average @xmath224 evaluated by ( [ eq : sinha_s ] ) and ( [ eq : sinha_p ] ) . ( the solid line ) and the moving average @xmath224 ( the broken line ) evaluated by ( [ eq : sinha_s ] ) and ( [ eq : sinha_p ] ) . we set @xmath225.,width=302 ] to construct the double - auction market for the above price change , we shall use the same definitions of buying @xmath131 and selling @xmath130 signals for each trader @xmath80 at time @xmath14 as ( [ eq : ait])([eq : bit ] ) in our minority game modeling . then , we calculate the response function and the auto - correlation function by using the above set - up and plot them in fig . [ fig : sinha2 ] . . the error - bars were calculated by 10-independent trials . , title="fig:",width=151 ] . the error - bars were calculated by 10-independent trials . , title="fig:",width=151 ] . the error - bars were calculated by 10-independent trials . , title="fig:",width=151 ] + . the error - bars were calculated by 10-independent trials . , title="fig:",width=151 ] . the error - bars were calculated by 10-independent trials . , title="fig:",width=151 ] . the error - bars were calculated by 10-independent trials . , title="fig:",width=151 ] we set the number of agents @xmath226 , the number of iterations for the price change @xmath227 . we also choose @xmath228 . the initial condition on the price @xmath128 is chosen from the eur / jpy exchange rate in the empirical data . from this figure , we find that the behaviour is different from the empirical evidence . especially , the auto - correlation function fluctuates with very large amplitudes . this result might come from the fact that in the vikam - sinha model , the buying - selling signal is chosen randomly . therefore , we should modify the vikram - sinha model in order to explain the non - monotonic behaviour of the response function in double - auction markets with stochastic bid - ask spread . in this paper , statistical properties of double - auction markets with bid - ask spread were investigated through the response function . we first attempted to utilize the so - called _ madhavan - richardson - roomans model _ ( mrr for short ) to simulate the stochastic process of the price - change in empirical data sets ( say , eur / jpy or usd / jpy exchange rates ) in which the bid - ask spread fluctuates in time . we found that the mrr theory apparently does not simulate so much as the qualitative behaviour ( ` non - monotonic ' behaviour ) of the response function calculated from the data . especially , we were confirmed that the stochastic nature of the bid - ask spread causes apparent deviations from a linear relationship between the @xmath0 and the auto - correlation function @xmath2 , namely , @xmath3 . to make the microscopic model of double - auction markets having stochastic bid - ask spread , we utilized the minority game with a finite market history length and found numerically that appropriate extension of the game shows quite similar behaviour of the response function to the empirical evidence . we also revealed that the minority game modeling with the adaptive ( ` annealed ' ) look - up table reproduces the non - linear relationship @xmath4 ( @xmath5 stands for a non - linear function leading to ` @xmath6-shapes ' ) more effectively than fixed ( ` quenched ' ) look - up table does . of course , there are still gaps between the theoretical prediction and the empirical evidence . we should modify our modeling of the double - auction market and figure out the micro - macro relationship in the market much more quantitatively . we thank enrico scalas and naoya sazuka for useful comments on auction theory , especially for calling our attention to the reference @xcite . we also thank hikaru hino for giving us useful information about the _ metatrader4 _ @xcite . our special thank goes to fabrizio lillo for valuable comments and his obliging advice on the double - auction markets . we would like to give an address of thanks to anonymous referees for critical reading of the manuscript and giving us useful comments . we were financially supported by grant - in - aid for scientific research ( c ) of japan society for the promotion of science , no .

we introduce the notion of resultant of two planar curves in the tropical geometry framework . we prove that the tropicalization of the algebraic resultant can be used to compute the stable intersection of two tropical plane curves . it is shown that , for two generic preimages of the curves to an algebraic framework , their intersection projects exactly onto the stable intersection of the curves . it is also given sufficient conditions for such a generality in terms of the residual coefficients of the algebraic coefficients of defining equations of the curves . _ keywords _ : tropical geometry , resultants , plane curves msc2000 : 14m25 , 14h50 , 52b20

0805.1305

in the context of tropical geometry , it is well known that two tropical curves may share an infinite number of intersection points without sharing a common component . this problem is avoided with the notion of stable intersection , @xcite . given two curves , there is a well defined set of intersection points that varies continuously under perturbations of the curves . this stable intersection has very nice properties . for example , it verifies a tropical version of bernstein - koushnirenko theorem ( cf . an alternative way of defining a finite intersection set is the following : given two tropical curves @xmath0 and @xmath1 , take two algebraic curves @xmath2 and @xmath3 projecting onto the tropical curves . then , the intersection of the two algebraic curves @xmath4 will project into the intersection of the tropical curves , @xmath5 . in general , the set @xmath6 depends on the election of the curves @xmath2 and @xmath3 . we are proving that , if the coefficients of @xmath2 , @xmath3 are generic , then the algebraic intersection will project exactly onto the stable intersection and there is a correspondence among the multiplicities of the intersection points . moreover , given two curves , we compute residually dense sufficient conditions defining these genericity conditions . the method works in any characteristic and it is essential in the generalization of the geometric construction method of @xcite to the non linear case . this also provides a particular case of bernstein - koushnirenko theorem for fields of positive characteristic . in order to prove this relationship , we introduce the notion of tropical resultant , as the tropicalization of the algebraic resultant . the paper is structured as follows . in section [ sec_notions ] , we give a brief description of the algebraic context we will work in . next , in section [ intersec_estable ] we recall the notion of stable intersection for plane curves and provide a brief discussion about its properties . then , it is introduced the notion of tropical resultant for univariate polynomials ( section [ sec_univariate ] ) and plane curves ( section [ sec_bivariate ] ) . in section [ sec_stable_int ] , we relate the stable intersection of tropical curves with the resultant of the curves and the generic preimage under the tropicalization map . we will provide conditions for the lifts ( preimages ) to be compatible with the stable intersection and the correspondence of the multiplicities . finally , in section [ sec_remarks ] we present some comments and remarks about the results . the algebraic context where the theory is developed is the following : let @xmath7 be an algebraically closed field provided with a non trivial rank one valuation @xmath8 . without loss of generality , we may suppose that @xmath9 and that @xmath10 is onto @xmath11 . we denote by @xmath12 the residual field of @xmath7 under the valuation . we will distinguish two main cases along the paper : the case whether @xmath13 ( hence @xmath14 ) and the case wether @xmath15 . in this case , either @xmath16 ( equicharacteristic @xmath17 ) or @xmath14 ( @xmath17-adic case ) . it is also assumed that we have fixed a multiplicative subgroup @xmath18 that is isomorphic to @xmath11 by the valuation map . the element @xmath19 represents the element of @xmath20 whose valuation is @xmath21 . any element @xmath22 of @xmath23 can be uniquely written as @xmath24 , where @xmath25 . we will denote the principal coefficient of an element @xmath22 of @xmath7 by @xmath26 and @xmath27 . we denote the principal term of an element by @xmath28 . the principal term of an element is only a notation , it is not , in general , an element of @xmath7 . if @xmath29 is an element of @xmath23 , @xmath30 if and only if @xmath31 . the tropicalization map is minus the valuation , @xmath32 . the tropical semiring @xmath33 is the group @xmath11 with the operations of tropical addition @xmath34 and tropical product @xmath35 . with these operations , @xmath36 and , if @xmath37 or @xmath38 then @xmath39 . let @xmath40 $ ] be a polynomial of support @xmath41 , where @xmath42 , @xmath43 , @xmath44 . the set @xmath45 of zeroes of @xmath0 is the set of points in @xmath46 such that the maximum of the piecewise affine function @xmath47 is attained for at least two different indices . it is known ( kapranov s theorem , @xcite ) that if @xmath48 is any polynomial in @xmath49 $ ] such that @xmath50 , then @xmath51 is exactly the set of zeroes of @xmath0 . moreover , if @xmath52 is a point , let @xmath53 be the set of indices where the value @xmath54 is attained and let @xmath55 . we define the residual polynomial of @xmath2 over @xmath56 as : @xmath57\ ] ] then , it happens that : [ initialserie ] let @xmath58 $ ] and @xmath59 be any point , then there is a root @xmath60 of @xmath2 such that @xmath61 , @xmath62 , if and only if @xmath63 is a zero of the tropical polynomial @xmath0 and @xmath64 is a root of @xmath65 in @xmath66 . for a constructive proof of this theorem we refer to @xcite or @xcite . let @xmath67 be a tropical plane curve defined as the zero set of a tropical polynomial @xmath68 . if we multiply @xmath0 by a monomial , the curve it defines stays invariant . we define the support of @xmath67 as the support @xmath41 of @xmath0 modulo a translation of an integer vector in @xmath69 . analogously , given an algebraic curve @xmath67 in @xmath70 defined by an algebraic polynomial @xmath71 , multiplying by a monomial does not change the set of zeroes in the algebraic torus @xmath70 , we also define the support of @xmath67 as the set @xmath41 modulo translations by an integer vector . if @xmath67 is a tropical curve , it may happen that there are polynomials with different support defining @xmath67 , even under the identification by translations we have defined . hence , when we define a tropical curve , we will always fix the support of a defining polynomial . let @xmath41 be the support of a tropical polynomial @xmath0 , the convex hull @xmath72 of @xmath41 in @xmath73 is the newton polytope of @xmath0 . this object is strongly connected with the set of zeroes of @xmath0 . every tropical polynomial @xmath0 defines a regular subdivision of its newton polytope @xmath74 . the topological closure of @xmath45 in @xmath73 has naturally a structure of piecewise affine polyhedral complex . this complex is dual to the subdivision induced to @xmath74 . to achieve this duality we have first to define the subdivision of @xmath74 . let @xmath75 be the convex hull of the set @xmath76 . the upper convex hull of @xmath75 , that is , the set of boundary maximal cells whose outgoing normal vector has its last coordinate positive , projects onto @xmath74 by deleting the last coordinate . this projection defines the regular subdivision of @xmath74 associated to @xmath0 ( see @xcite for the details ) . [ newton - dual ] the subdivision of @xmath74 associated to @xmath0 is dual to the set of zeroes of @xmath0 . there is a bijection between the cells of @xmath77 and the cells of @xmath45 such that : * every @xmath12-dimensional cell @xmath78 of @xmath74 corresponds to a cell @xmath79 of @xmath45 of dimension @xmath80 such that the affine linear space generated by @xmath79 is orthogonal to @xmath78 . ( in the case where @xmath81 , the corresponding dual cell is a connected component of @xmath82 ) * if @xmath83 , then @xmath84 * if @xmath85 , then @xmath86 * @xmath87 where the union is disjoint . * @xmath79 is not bounded if and only if @xmath88 . from this , we deduce that , given a fixed support @xmath41 , there are finitely many combinatorial types of tropical curves with support @xmath41 . these different types are in bijection with the different regular subdivisions of @xmath74 . finally , let @xmath67 be a tropical planar curve of support @xmath41 and newton polygon @xmath74 , let @xmath78 be a one - dimensional cell of the subdivision of @xmath74 dual to @xmath67 , then , the weight of the dual cell @xmath79 is defined as @xmath89 , the integer length of the segment @xmath78 . one of the first problems encountered in tropical geometry is that the projective geometry intuition is no longer valid . if we define a tropical line as the set of zeroes of an affine polynomial @xmath90 , then two different lines always intersect in at least one point . the problem is that sometimes they intersect in more than one point . the usual answer to deal with this problem is using the notion of stable intersection . let @xmath91 , @xmath92 be the set of zeroes of two tropical polynomials @xmath0 and @xmath1 respectively . let @xmath93 be the intersection of the curves , @xmath94 . it is possible that @xmath93 is not the image of an algebraic variety @xmath95 by the map @xmath96 . we want to associate , to each @xmath97 an intersection multiplicity . we will follow the notions of @xcite and we will compare them with the subdivisions of the associated newton polygons of the curves in terms of mixed volumes . see @xcite to precise the comparison between mixed volumes and intersection of algebraic curves . let @xmath98 . it is easy to check that the union of the two tropical curves is the set of zeroes of the product @xmath99 . the newton polygon @xmath100 of @xmath101 is the minkowski sum of @xmath102 and @xmath103 . that is : @xmath104 the subdivision of @xmath100 dual to @xmath105 is a subdivision induced by the subdivisions of @xmath102 , @xmath103 . more concretely , let @xmath56 be a point in @xmath105 , let @xmath106 be the monomials of @xmath0 where @xmath54 is attained and let @xmath107 be the monomials of @xmath1 where @xmath108 is attained . then @xmath109 or @xmath110 . the monomials where @xmath99 attains it maximum are @xmath111 . the newton polygon of these monomials is the minkowski sum of the newton polygons of @xmath112 and @xmath107 , each one of these newton polygons is the cell dual to the cell containing @xmath56 in @xmath100 , @xmath102 and @xmath113 respectively . this process covers every cell of dimension @xmath114 and @xmath115 of @xmath100 . the zero dimensional cells correspond to points @xmath56 belonging neither to @xmath91 nor to @xmath92 . let @xmath116 be the monomials of @xmath0 and @xmath1 where the value at @xmath56 is attained . then the monomial of @xmath99 where @xmath117 is attained is @xmath118 . to sum up , every cell of @xmath100 is naturally the minkowski sum of a cell @xmath119 of @xmath0 and a cell @xmath10 of @xmath1 . the possible combination of dimensions @xmath120 are : * @xmath121 , these cells do not correspond to points of @xmath105 . * @xmath122 , these are edges of @xmath105 that correspond to a maximal segment contained in an edge of @xmath91 that does not intersect @xmath92 . * @xmath123 , correspond to the vertices of @xmath105 that are vertices of @xmath91 that do not belong to @xmath92 . * @xmath124 , this combination defines a vertex of @xmath105 which is the unique intersection point of an edge of @xmath91 with an edge of @xmath92 . * @xmath125 are the edges of @xmath105 that are the infinite intersection of an edge of @xmath91 and an edge of @xmath92 . * @xmath126 corresponds with the vertices of @xmath105 that are a vertex of @xmath92 belonging to an edge of @xmath91 . * @xmath127 this is a vertex of @xmath105 which is a common vertex of @xmath91 and @xmath92 . and the obvious symmetric cases @xmath128 , @xmath129 and @xmath130 . if the relative position of @xmath91 , @xmath92 is generic , then @xmath105 can not contain any cell of type @xmath125 , @xmath126 and @xmath127 . that is , the intersection points @xmath56 of @xmath91 and @xmath92 are always the unique intersection point of an edge of @xmath91 and an edge of @xmath92 . this is the transversal case . the definition of intersection multiplicity , as presented in @xcite for these cells @xmath131 is the following : [ mult_rgst ] let @xmath56 be an intersection point of two tropical curves @xmath91 and @xmath92 . suppose that @xmath56 is the unique intersection line of an edge @xmath132 of @xmath91 and an edge @xmath133 of @xmath92 . let @xmath134 be the primitive vector in @xmath69 of the support line of @xmath132 . let @xmath135 be the corresponding primitive vector of @xmath133 . let @xmath119 be the dual edge of @xmath132 in @xmath102 and let @xmath10 be the dual edge of @xmath133 in @xmath103 , we call @xmath136 and @xmath137 the weight of the edges @xmath132 and @xmath133 respectively . the intersection multiplicity is @xmath138 the absolute value of the determinant of the primitive vectors times the weight of the edges . if the curves are not in a generic relative position , consider the curve @xmath139 obtained by translation of @xmath91 by a vector @xmath10 . if the length of @xmath10 is sufficiently small , @xmath140 ( that is , it is an infinitesimal translation ) , then every cell of @xmath141 of type @xmath121 , @xmath122 , @xmath123 and @xmath124 stays invariant . furthermore , if the translation is generic ( for all but finitely many directions of @xmath10 ) , the cells of type @xmath125 are subdivided into cells of type @xmath121 and @xmath128 . that is , if two edges intersect in infinitely many points , after the translation , every intersection point will disappear . if @xmath56 is an intersection point of @xmath91 and @xmath92 corresponding to a cell of type @xmath130 or @xmath127 and the direction of @xmath10 is generic , this cell is subdivided , after the perturbation , into cells of type @xmath121 , @xmath122 , @xmath124 , @xmath123 . that is , no intersection point is a vertex of @xmath0 or @xmath1 . however , some transversal intersection points appear instead ( of type ( 1,1,2 ) ) in a neighborhood of @xmath56 . the intersection multiplicity of @xmath56 is , in this case , the sum of the intersection multiplicities of the transversal intersection points . now we recall the notion of stable intersection of curves ( see @xcite ) . [ interseccion_estable_def ] let @xmath91 , @xmath92 be two tropical curves . let @xmath142 , @xmath143 be two small generic translations of @xmath144 such that their intersection is finite . the stable intersection @xmath145 of @xmath91 and @xmath92 is the limit set of intersection points of the translated curves @xmath146 . from the previous comments it is clear that let @xmath91 , @xmath92 be two tropical curves , then the stable intersection of @xmath91 and @xmath92 is the set of intersection points with positive multiplicity . this stable intersection has very nice properties . from the definition , it follows that it is continuous under small perturbations on the curves . moreover , it verifies a berstein - koushnirenko theorem for tropical curves . [ bernstein ] let @xmath91 , @xmath92 be two tropical curves of newton polygons @xmath102 , @xmath103 . then the number of stable intersection points , counted with multiplicity is the mixed volumes of the newton polygons of the curves @xmath147 see @xcite in particular , we have the following alternative definition of intersection multiplicity for plane curves : [ def_multiplicidad_tropical ] let @xmath0 , @xmath1 be two tropical polynomials of newton polygons @xmath102 , @xmath103 respectively . let @xmath148 be an intersection point . let @xmath149 , @xmath150 be the cells of @xmath151 , @xmath152 dual to the cells in the curve containing @xmath56 respectively , then , the tropical intersection multiplicity of @xmath56 is : @xmath153 from the classification of intersection points , @xmath56 is an intersection point of multiplicity zero if and only if it belongs to a cell of type @xmath125 in @xmath105 . in this case @xmath154 , because an edge has no area . if @xmath56 is a stable intersection point , let @xmath155 , let @xmath156 be truncated polynomials . it follows from the definition that the intersection multiplicity of @xmath56 only depends in the behaviour of the mixed cell @xmath157 in the dual subdivision of @xmath100 . that is , the intersection multiplicity of @xmath56 as intersection of @xmath91 and @xmath92 equals the intersection multiplicity of @xmath56 as an intersection point of @xmath158 and @xmath159 . but , by construction , the unique stable intersection point of @xmath158 and @xmath159 is @xmath56 itself . hence , by theorem [ bernstein ] , the intersection multiplicity of @xmath56 is @xmath160 let us start with the notion of tropical resultant of two univariate polynomials . in algebraic geometry , the resultant of two univariate polynomials is a polynomial that solves the decision problem of determining if both polynomials have a common root . let @xmath161 , @xmath162 $ ] , where @xmath7 is an algebraically closed field . for simplicity , we assume that @xmath163 . let @xmath17 be the characteristic of @xmath7 . then , there is a unique polynomial in @xmath164 $ ] , up to a constant factor , called the resultant , such that it vanishes if and only if @xmath2 and @xmath3 have a common root . in the definition , it is asked the polynomials to be of effective degree @xmath165 and @xmath166 , this is in order to avoid the specialization problems that usually appear when using resultants . but the polynomials are also asked to have order zero . this restriction is demanded for convenience with tropicalization . recall that the intersection of the varieties with the coordinate hyperplanes is always neglected . hence , the definition of resultant will take this into account . moreover , as the polynomials are described by its support , the resultant will not be defined by the degree of the polynomials , but by their support . this approach will be convenient in the next section , when there will be provided a notion of resultant for bivariate polynomials . let @xmath41 , @xmath167 be two finite subsets of @xmath168 of cardinality at least 2 such that @xmath169 . that is , the support of two polynomials that do not have zero as a root . let @xmath170 be the resultant of two polynomials with indeterminate coefficients , @xmath171 , @xmath172 over the field @xmath7 . @xmath173,\ ] ] ( where @xmath17 is the characteristic of the field @xmath174 . let @xmath175 be the tropicalization of @xmath170 . this is a polynomial in @xmath176 $ ] , which is called the tropical resultant of supports @xmath41 and @xmath167 over @xmath7 . so , our approach is to define the tropical resultant polynomial as the projection of the algebraic polynomial . in this point , one may obtain , for the same support sets @xmath41 and @xmath167 , different tropical resultants , one for each possible characteristic of @xmath7 . this is not good , in the sense that tropical geometry should not be determined by the characteristic of the field we have used to define the projection . hence , one has to take care of what is the common information of these polynomials . the answer is complete : the tropical variety they define is always the same . this variety is the image of any resultant variety over a field @xmath7 , so it will code the pairs of polynomials with fixed support that have a common root . the tropical variety @xmath177 does not depend on the field @xmath7 , but only on the sets @xmath41 and @xmath167 . let @xmath178 be the newton polytope of the resultant defined over a field @xmath179 of characteristic zero , @xmath180 . it is known that the monomials of @xmath181 corresponding to vertices of @xmath178 ( extreme monomials ) have always as coefficient @xmath182 ( see , for example , @xcite or @xcite ) . hence , the extreme monomials in @xmath170 are independent of the characteristic of the field @xmath7 and so is @xmath178 . if @xmath183 is a monomial of @xmath170 that does not correspond to a vertex of @xmath178 , then @xmath184 , @xmath185 , where @xmath186 are vertices of @xmath178 . @xmath187 and , as coeff(@xmath22 ) is an integer ( or an integer mod @xmath17 ) , it is contained in the valuation ring , that is , @xmath188 . @xmath189 is finite and not zero if and only if we are dealing with a @xmath17-adic valuation and @xmath17 divides @xmath190 . it is @xmath191 if and only if the characteristic of @xmath7 divides the coefficient . hence , for any evaluation @xmath192 of the indeterminates , we have that @xmath193 , where @xmath194 . it follows that the maximum of the piecewise affine function @xmath195 is never attained in the monomial @xmath22 alone and that @xmath22 does not induce any subdivision in the cell it is contained . thus , this monomial does not add anything to the tropical variety defined by @xmath195 . the tropical hypersurface @xmath177 is , as a polyhedral complex , dual to the subdivision of @xmath178 induced by @xmath195 ( cf . @xcite ) . in this case , the subdivision of @xmath178 induced by the tropical polynomial is @xmath178 itself . so @xmath196 is always the polyhedral complex dual to @xmath178 centered at the origin . this complex is independent of @xmath7 . hence , fixed two supports @xmath41 , @xmath167 , there may be different tropical polynomials that can be called the resultant of polynomials of support @xmath41 and @xmath167 . however , the variety all of them define is always the same , so there is a good notion of resultant variety . now we prove that the resultant variety @xmath177 has the same geometric meaning than the algebraic resultant variety . let @xmath197 be two support subsets as before . let @xmath198 , @xmath199 be two univariate tropical polynomials of support @xmath41 and @xmath167 . then , @xmath0 and @xmath1 have a common tropical root if and only if the point @xmath200 belongs to the variety defined by @xmath195 . suppose that @xmath200 belongs to @xmath195 . by theorem [ initialserie ] , we can compute an element @xmath201 in the variety defined by @xmath170 . in this case , @xmath202 and @xmath203 are lifts of @xmath0 and @xmath1 . that is , @xmath204 , @xmath205 . moreover , their coefficients belong to the algebraic resultant , so the algebraic polynomials have a common root @xmath206 that is non zero by construction ( @xmath169 ) . projecting to the tropical space , @xmath0 and @xmath1 have a common root @xmath207 . conversely , if @xmath0 and @xmath1 have a common root @xmath56 , we may take any lift @xmath208 of @xmath1 . then , by theorem [ initialserie ] , we may lift @xmath56 to a root @xmath206 of @xmath3 . finally , note that the coefficients of @xmath0 belong to the hyperplane defined by the equation @xmath209 , so it can be lifted to an algebraic solution @xmath210 of the affine equation @xmath211 , the polynomial @xmath212 projects onto @xmath0 and has @xmath206 as a root . by construction , @xmath2 , @xmath3 share a common root @xmath206 , hence , their coefficients @xmath213 belong to the algebraic resultant variety . projecting again , the coefficient vector ( @xmath214 , @xmath215 ) of @xmath0 and @xmath1 belong to the tropical resultant . this lemma about the geometric meaning of the resultant also shows that the variety defined by @xmath195 does not depend on the field @xmath7 . at least as a set of points , because the tropical characterization of two tropical polynomials having a common root does not depend on the field @xmath7 . consider the easiest nonlinear case , @xmath216 , the resultant of two quadratic polynomials . if @xmath217 , @xmath218 , the algebraic resultant in characteristic zero is @xmath219 and , over a characteristic 2 field it is @xmath220 . if @xmath221 , the tropical polynomial is @xmath222 . if @xmath14 and @xmath223 , the tropical polynomial is @xmath224 . finally , if @xmath225 then the tropical polynomial is @xmath226 . the unique difference among these polynomials is the term @xmath227 . this monomial lies in the convex hull of the monomials @xmath228 and @xmath229 and it does not define a subdivision because its tropical coefficient is always @xmath230 . the piecewise affine functions @xmath231 , @xmath232 and @xmath233 are the same . so the three polynomials define the same tropical variety . in this section , the notion of univariate resultant is extended to the case where the polynomials are bivariate . let @xmath2 and @xmath3 be two bivariate polynomials . in order to compute the algebraic resultant with respect to @xmath22 , we can rewrite them as polynomials in @xmath22 . @xmath234 where @xmath235 and @xmath236 , @xmath237 are elements of valuation zero . let @xmath238 $ ] be the algebraic univariate resultant of supports @xmath41 , @xmath167 . the algebraic resultant of @xmath2 and @xmath3 is the polynomial @xmath239 $ ] . analogously , let @xmath240 , @xmath241 , @xmath242 , @xmath243 , where @xmath244 let @xmath245 $ ] be the tropical resultant of supports @xmath41 and @xmath167 . then , the polynomial @xmath246 $ ] is the tropical resultant of @xmath0 and @xmath1 . again , we have different tropical resultant polynomials , one for each possible characteristic of the fields @xmath7 and @xmath12 . we want to check that this notion of tropical resultant also has a geometric meaning . in the algebraic setting , the roots of the resultant @xmath247 are the possible @xmath29-th values of the intersection points of the curves defined by @xmath2 and @xmath3 . this is not the case of the tropical resultant , because @xmath248 only has finitely many tropical roots , while the intersection @xmath249 may have infinitely many points and there may be infinitely many possible values of the @xmath29-th coordinates . again , this indetermination is avoided with the notion of stable intersection . we will prove that the roots of @xmath250 are the possible @xmath29-th values of the stable intersection @xmath251 . this will be made in several steps , the first one is to check that @xmath252 , provided that @xmath253 are residually generic . sometimes , for technical reasons , it is better to work with an affine representation of the polynomials . the set of polynomials @xmath254 of fixed support @xmath41 is an open subspace ( a torus ) of a projective space . the projective coordinates of @xmath2 are its coefficients @xmath255 $ ] . we may fix an index @xmath256 . then , the affine representation of @xmath2 with respect to this index is obtained by setting @xmath257 and @xmath258 . we prove that this dehomogenization process is also compatible with tropicalization . that is , if we divide each algebraic coefficient @xmath259 and @xmath260 by @xmath261 and @xmath262 respectively and substitute each coefficient @xmath263 , @xmath264 of the tropical polynomials @xmath0 and @xmath1 by @xmath265 and @xmath266 respectively , still we have that @xmath267 . [ tropicalizar_resultante_no_homogenea_char0p ] let @xmath268 , @xmath269 $ ] , where the coefficients are @xmath270 and let @xmath271 , @xmath272 , @xmath273 be the corresponding tropical polynomials . suppose that @xmath236 , @xmath237 are residually generic . then @xmath267 . first , we suppose that @xmath13 . in general , the composition of polynomials does not commute with tropicalization , because , in the algebraic case , there can be a cancellation of terms when performing the substitution that does not occur in the tropical case . recall that , by the nature of tropical operations , a cancellation of terms in the tropical development of the polynomial never happens . so , we have to check that there is never a cancellation of terms in the algebraic setting . first , it is proved that there is no cancellation of monomials when substituting the variables by polynomials without dehomogenizing . @xmath274 is homogeneous in the set of variables @xmath275 and in the set of variables @xmath215 . as the substitution is linear in the variables @xmath236 and @xmath237 , @xmath276 is homogeneous in @xmath277 and @xmath237 . if we have two different terms @xmath278 , @xmath279 of @xmath280 , then there is a variable with different exponent in both terms . assume for simplicity that this variable is @xmath281 with degrees @xmath282 and @xmath283 respectively . after the substitution , the monomials obtained by expansion of @xmath278 are homogeneous of degree @xmath282 in the set of variables @xmath284 and the monomials coming from @xmath279 are homogeneous of degree @xmath283 in the variables @xmath284 . thus , it is not possible to have a cancellation of terms and we can conclude that the homogeneous polynomial projects onto the tropical homogeneous polynomial . in the case we dehomogenize @xmath2 and @xmath3 with respect to the indices @xmath285 , @xmath286 respectively . by the homogeneous case , we can suppose that all the variables @xmath287 and @xmath288 in @xmath280 have already been substituted by the polynomials @xmath289 and @xmath290 respectively . the only possibility to have a cancellation of terms is if there are two monomials of the form @xmath291 , @xmath292 with @xmath293 and @xmath294 is a monomial in the variables @xmath236 , @xmath237 . but , as the polynomial is multihomogeneous in @xmath295 and @xmath296 , it must happen that @xmath297 and @xmath298 . that is , the original monomials were the same . so , a cancellation of terms is not possible and the dehomogenized polynomial projects into the dehomogenized tropical polynomial . in particular , @xmath299 . now suppose that @xmath15 . in this case , it is not necessarily true that the tropicalization of the algebraic resultant is the tropical resultant . but we are going to check that the monomials where these two tropical polynomials differ do not add anything to the tropical variety @xmath300 . so , we are going to compare the monomials in @xmath301 and @xmath302 . the support of both polynomials is contained in the support of @xmath303 , where @xmath179 is an equicharacteristic zero field . the first potential difference in the monomials are those obtained by expansion of a monomial @xmath166 of the univariate resultant @xmath304 whose coefficient has valuation in @xmath305 . that is , @xmath17 divides @xmath306 . it happens that @xmath166 is never a extreme monomial . that is , @xmath307 , @xmath308 and @xmath309 are extreme monomials . so , for every @xmath132 , @xmath310 . hence , the monomials of @xmath311 never add anything to the tropical variety defined by @xmath247 , because they are never greater than the monomials that appear by the extreme monomials . the other source of potential differences in the monomials is the decreasing of the tropicalization of some terms of the power @xmath312 due to some combinatorial coefficient @xmath313 divisible by @xmath17 . but , in the tropical context , it happens that @xmath314 as piecewise affine functions . the rest of the terms in the expansion do not contribute anything to the tropical variety . the only terms that may play a role are @xmath315 , @xmath316 . so , even if the tropicalization of the polynomials @xmath317 depends on the algebraic field @xmath7 , the tropical variety they define is always the same and it is the tropical variety defined by @xmath302 , including the weight of the cells . so , the previous lemma provides a notion of tropical resultant for bivariate polynomials with respect to one variable . they also prove that this polynomials define the same variety as the projection of the algebraic resultant in the generic case . our next goal is to provide a geometric meaning to the roots of the tropical resultant in terms of the stable intersection of the curves . let @xmath0 be a tropical polynomial of support @xmath41 defining a curve , let @xmath102 be the convex hull of @xmath41 . by proposition [ newton - dual ] , the coefficients of @xmath0 induce a regular subdivision in @xmath102 dual to @xmath0 . this subdivision is essential in the definition of tropical multiplicity and stable intersection for the case of curves . next , it is proved that , for sufficiently generic lifts @xmath2 and @xmath3 , their intersection points correspond with stable intersection points of @xmath0 and @xmath1 . [ generic - lift_stable - int ] let @xmath0 and @xmath1 be two tropical polynomials in two variables . let @xmath318 be its stable intersection . then , for any two lifts @xmath2 , @xmath3 such that their coefficients are residually generic , the intersection of the algebraic curves projects into the stable intersection . @xmath319 if every intersection point of @xmath0 and @xmath1 is stable , then there is nothing to prove . let @xmath56 be a non stable intersection point . this means that @xmath56 belongs to the relative interior of two parallel edges of @xmath45 and @xmath320 . the residual polynomials @xmath321 and @xmath322 can be written ( after multiplication by a suitable monomial ) as @xmath323 , @xmath324 . if @xmath2 , @xmath3 have a common point projecting into @xmath56 then there is an algebraic relation among their residual coefficients . namely , the resultant of the polynomials @xmath325 , @xmath326 with respect to @xmath327 must vanish . if the residual coefficients of @xmath2 , @xmath3 do not belong to the resultant defined by each non stable intersection cell , the intersection in the torus of @xmath2 , @xmath3 projects into the stable intersection of @xmath0 and @xmath1 . so , there is a natural relation between the stable intersection of two tropical curves @xmath0 and @xmath1 and the intersection of two generic lifts @xmath2 and @xmath3 of the curves . on the other hand , the intersection of two generic lifts can be determined by the algebraic resultant of the defining polynomials . applying tropicalization , this relationship links the notion of stable intersection with the resultants . to achieve a true bijection between the roots of the resultant and the intersection points of the curves , it is used the relationship between the tropical and algebraic resultants . so , one needs to concrete the generality conditions for the values values @xmath236 , @xmath237 that makes lemma [ tropicalizar_resultante_no_homogenea_char0p ] and proposition [ tropicalizar_resultante_no_homogenea_char0p ] hold . next proposition shows how to compute the residually conditions for the compatibility of the resultant . [ resultantes ] let @xmath2 , @xmath328 $ ] . then , there is a finite set of nonzero polynomials in the principal coefficients of the coefficients of @xmath2 , @xmath3 , that depends only on the tropicalization @xmath0 and @xmath1 such that , if no one of them vanishes , then @xmath329 where @xmath330 is the evaluation of the tropical resultant of supports @xmath41 and @xmath167 in the coefficients of @xmath0 and @xmath1 as polynomials over @xmath22 . write @xmath331 , @xmath332 , and take @xmath333 , @xmath334 , @xmath335 , @xmath336 , @xmath337 , @xmath338 . let @xmath41 , @xmath167 be the support of @xmath0 and @xmath1 with respect to @xmath22 . consider both resultants @xmath339 it happens that @xmath340 and the equality holds if and only if the term @xmath341 of @xmath342 is different from @xmath343 . as in the generic case the resultant projects correctly by lemma [ tropicalizar_resultante_no_homogenea_char0p ] , the polynomials @xmath344 corresponding to vertices of the subdivision of the newton polytope of the resultant polynomial ( that in this case is a segment ) are nonzero polynomials in @xmath345 $ ] . if no one of them vanish , the resultant tropicalizes correctly . with all these results we are ready to prove our main result , we can provide a bijection between the stable intersection of two tropical curves and the intersection of two generic lifts of the curves . moreover , sufficient residual conditions for the genericity can be explicitly computed . [ interestable ] let @xmath2 , @xmath328 $ ] . then , it can be computed a finite set of polynomials in the principal coefficients of @xmath2 , @xmath3 depending only on their tropicalization @xmath0 , @xmath1 such that , if no one of them vanish , the tropicalization of the intersection of @xmath2 , @xmath3 is exactly the stable intersection of @xmath0 and @xmath1 . moreover , the multiplicities are conserved . proposition [ resultantes ] provides a set @xmath347 of polynomials in the principal coefficients of @xmath2 and @xmath3 such that , if no one vanishes , the algebraic resultants @xmath348 and @xmath349 define the same tropical varieties as @xmath350 and @xmath351 . these two resultants define a finite set @xmath93 that contains the stable intersection . the problem is that , in the tropical case , it is possible that the intersection of @xmath93 with both curves may be strictly larger than the stable intersection of the curves , see example [ resultex ] . so , we need another polynomial in order to discriminate the points in this intersection that are not stable points . take @xmath352 , any natural number such that the affine function @xmath353 is injective in the finite set @xmath93 . make the monomial change of coordinates @xmath354 . the polynomial @xmath355 encodes the values @xmath356 of the common roots of @xmath2 and @xmath3 . we add to the set @xmath347 the restrictions in the principal coefficients of this resultant to be compatible with tropicalization according to proposition [ resultantes ] . these values @xmath356 of the algebraic intersection points correspond with the possible values @xmath353 of the tropicalization of the roots . as the linear function is injective in @xmath93 , then @xmath357 is exactly the tropicalization of the intersection points of any system @xmath358 verifying the restrictions of @xmath347 . by , lemma [ generic - lift_stable - int ] , this set is contained in the stable intersection of @xmath0 and @xmath1 . to prove that the multiplicities are conserved , consider the field @xmath359 of generalized puiseux series , in this case @xmath360 because the sum on the left is bounded by the mixed volume of the residual polynomials @xmath321 , @xmath322 over @xmath56 by bernstein - koushnirenko theorem ( c.f . @xcite @xcite @xcite ) . this mixed volume is , by definition , the tropical multiplicity of @xmath56 on the right . on the other hand , the sum on the left is , over any field , the sum of the multiplicities of the algebraic roots of @xmath361 projecting onto @xmath56 . by the previous results on the correct projection of the resultant , this multiplicity does not depend on @xmath7 , because it is the degree minus the order of the residual polynomial @xmath362 , or , equivalently , the multiplicity of @xmath56 as a root of @xmath363 . moreover , this multiplicity is the mixed volume of the residual polynomials over @xmath56 . that is , the inequality @xmath364 holds for any field . the total number of roots of @xmath2 and @xmath3 counted with multiplicities in the torus equals the sum of multiplicities of the stable roots of @xmath0 and @xmath1 , because , in both cases , this is the degree minus the order of @xmath365 . from this , we conclude that @xmath366 hence , the projection of the intersection of @xmath2 and @xmath3 is exactly the stable intersection . along the proof of the theorem we have proved the following result , that asserts that the tropical resultant of two tropical curves has a geometric meaning analogous to the algebraic resultant . let @xmath0 , @xmath367 $ ] be two tropical polynomials . let @xmath368 $ ] be a tropical resultant of @xmath0 and @xmath1 with respect to the @xmath22 variable . then , the tropical roots of @xmath369 are exactly the @xmath29-th coordinates of the stable intersection of @xmath0 and @xmath1 . [ resultex ] consider @xmath370 , two conics . their stable intersection is the set @xmath371 , @xmath372 , @xmath373 , @xmath374 . compute the resultants : @xmath375 by symmetry @xmath376 . their roots are the lines @xmath377 , @xmath378 , @xmath379 and @xmath380 , @xmath381 , @xmath382 respectively . in both cases the multiplicity of the roots @xmath383 and @xmath114 is 1 , while the multiplicity of @xmath343 is @xmath115 . the intersection of this lines and the two curves gives the four stable points plus @xmath384 and @xmath385 . we need another resultant that discriminates the points . see figure [ fig : discrimine ] . take @xmath386 , the first affine function @xmath353 that is injective over these points . its roots are @xmath389 , all with multiplicity 1 . it is easy to check now that the intersection of the two curves and the three resultants is exactly the stable intersection . the two extra points take the values -4 , 4 in the monomial @xmath390 , moreover , every point has intersection multiplicity equal to one . two generic lifts of the cubics are of the form : @xmath391 @xmath392 the residual conditions for the compatibility of the algebraic and tropical resultant with respect to @xmath22 are : + @xmath393 @xmath394 @xmath395 @xmath396 @xmath393 @xmath395 @xmath397 @xmath398 @xmath399 @xmath397 @xmath396 @xmath400 @xmath394 @xmath401 , @xmath402 @xmath394 @xmath403 @xmath404 @xmath402 @xmath403 @xmath397 @xmath405 @xmath406 @xmath394 @xmath407 @xmath408 @xmath409 @xmath410 , @xmath411 @xmath394 @xmath407 @xmath412 @xmath394 @xmath413 @xmath414 @xmath408 @xmath409 @xmath414 @xmath412 @xmath413 @xmath397 @xmath411 , @xmath393 @xmath395 @xmath397 @xmath411 @xmath415 @xmath394 @xmath401 @xmath393 @xmath394 @xmath395 @xmath414 @xmath399 @xmath416 @xmath414 + for the resultant with respect to @xmath29 , the compatibility conditions are : + @xmath417 @xmath398 @xmath414 @xmath410 @xmath417 @xmath414 @xmath396 @xmath418 @xmath419 @xmath398 @xmath420 @xmath421 @xmath396 @xmath410 , @xmath422 @xmath398 @xmath420 @xmath417 @xmath414 @xmath396 @xmath422 @xmath421 @xmath396 @xmath413 @xmath417 @xmath398 @xmath414 @xmath413 , @xmath423 @xmath396 @xmath413 @xmath424 @xmath398 @xmath401 @xmath393 @xmath398 @xmath395 @xmath413 @xmath393 @xmath395 @xmath396 @xmath422 , @xmath393 @xmath394 @xmath395 @xmath396 @xmath393 @xmath395 @xmath397 @xmath398 @xmath399 @xmath397 @xmath396 @xmath425 @xmath394 @xmath401 . + finally , the third resultant is a degree twelve polynomial in the variable @xmath327 . the residual conditions for its compatibility with the tropical resultant are : + @xmath426 @xmath394 @xmath427 @xmath396 @xmath411 @xmath414 @xmath405 @xmath397 @xmath428 @xmath429 @xmath430 @xmath431 @xmath396 @xmath411 @xmath414 @xmath405 @xmath432 @xmath395 @xmath433 @xmath434 @xmath420 @xmath405 @xmath396 @xmath435 @xmath434 @xmath398 @xmath436 @xmath420 @xmath405 @xmath437 @xmath434 @xmath398 @xmath436 @xmath414 @xmath411 @xmath410 @xmath438 @xmath401 @xmath396 @xmath439 @xmath434 @xmath423 @xmath410 @xmath440 @xmath430 @xmath401 @xmath441 @xmath411 @xmath414 @xmath405 @xmath442 @xmath395 @xmath433 @xmath434 @xmath436 @xmath439 @xmath410 @xmath443 @xmath430 @xmath427 @xmath428 @xmath414 @xmath411 @xmath396 @xmath410 @xmath432 @xmath430 @xmath427 @xmath428 @xmath420 @xmath405 @xmath396 @xmath440 @xmath430 @xmath401 @xmath398 @xmath436 @xmath414 @xmath411 @xmath410 @xmath399 @xmath430 @xmath401 @xmath398 @xmath436 @xmath420 @xmath405 @xmath444 @xmath394 @xmath401 @xmath423 @xmath397 @xmath414 @xmath411 @xmath396 @xmath410 @xmath442 @xmath430 @xmath427 @xmath428 @xmath436 @xmath439 @xmath410 @xmath443 @xmath395 @xmath433 @xmath434 @xmath414 @xmath411 @xmath396 @xmath410 @xmath445 @xmath430 @xmath427 @xmath428 @xmath436 @xmath411 @xmath414 @xmath405 @xmath446 @xmath394 @xmath401 @xmath423 @xmath436 @xmath439 @xmath397 @xmath410 @xmath447 @xmath430 @xmath431 @xmath420 @xmath405 @xmath448 @xmath394 @xmath401 @xmath423 @xmath436 @xmath411 @xmath397 @xmath414 @xmath405 @xmath449 @xmath430 @xmath431 @xmath414 @xmath411 @xmath410 @xmath450 @xmath394 @xmath427 @xmath414 @xmath411 @xmath397 @xmath428 @xmath410 @xmath451 @xmath394 @xmath427 @xmath420 @xmath405 @xmath397 @xmath428 @xmath449 @xmath401 @xmath434 @xmath423 @xmath414 @xmath411 @xmath410 @xmath447 @xmath401 @xmath434 @xmath423 @xmath420 @xmath405 @xmath399 @xmath430 @xmath401 @xmath441 @xmath439 @xmath410 @xmath429 @xmath401 @xmath396 @xmath411 @xmath434 @xmath423 @xmath414 @xmath405 @xmath432 @xmath394 @xmath427 @xmath397 @xmath428 @xmath396 @xmath439 @xmath410 @xmath438 @xmath430 @xmath431 @xmath396 @xmath439 @xmath410 @xmath437 @xmath434 @xmath441 @xmath411 @xmath414 @xmath405 @xmath452 @xmath394 @xmath401 @xmath423 @xmath397 @xmath420 @xmath405 @xmath396 @xmath435 @xmath434 @xmath441 @xmath439 @xmath410 @xmath453 @xmath397 @xmath394 @xmath395 @xmath398 @xmath436 @xmath414 @xmath411 @xmath410 @xmath454 @xmath397 @xmath394 @xmath395 @xmath398 @xmath436 @xmath420 @xmath405 @xmath454 @xmath394 @xmath395 @xmath397 @xmath441 @xmath439 @xmath410 @xmath453 @xmath394 @xmath395 @xmath397 @xmath441 @xmath411 @xmath414 @xmath405 @xmath445 @xmath395 @xmath433 @xmath434 @xmath436 @xmath411 @xmath414 @xmath455 + @xmath456 @xmath430 @xmath431 @xmath439 @xmath420 @xmath405 @xmath457 @xmath430 @xmath431 @xmath458 @xmath414 @xmath410 @xmath459 @xmath460 @xmath458 @xmath414 @xmath405 @xmath461 @xmath434 @xmath439 @xmath401 @xmath420 @xmath405 @xmath462 @xmath434 @xmath463 @xmath411 @xmath410 @xmath464 @xmath430 @xmath401 @xmath465 @xmath405 @xmath466 @xmath397 @xmath411 @xmath394 @xmath401 @xmath463 @xmath405 @xmath467 @xmath434 @xmath411 @xmath395 @xmath463 @xmath405 @xmath468 @xmath397 @xmath439 @xmath394 @xmath401 @xmath420 @xmath410 @xmath469 @xmath434 @xmath439 @xmath395 @xmath420 @xmath410 @xmath470 @xmath434 @xmath465 @xmath405 @xmath471 @xmath434 @xmath458 @xmath401 @xmath414 @xmath410 @xmath472 @xmath394 @xmath395 @xmath397 @xmath465 @xmath405 @xmath473 @xmath394 @xmath431 @xmath458 @xmath397 @xmath414 @xmath405 @xmath440 @xmath434 @xmath458 @xmath427 @xmath414 @xmath405 @xmath474 @xmath394 @xmath431 @xmath475 @xmath397 @xmath410 @xmath476 @xmath394 @xmath395 @xmath397 @xmath463 @xmath411 @xmath410 @xmath477 @xmath430 @xmath401 @xmath463 @xmath411 @xmath410 @xmath478 @xmath460 @xmath475 @xmath410 @xmath479 @xmath430 @xmath427 @xmath439 @xmath420 @xmath410 @xmath480 @xmath397 @xmath439 @xmath394 @xmath427 @xmath420 @xmath405 @xmath481 @xmath397 @xmath458 @xmath394 @xmath427 @xmath414 @xmath410 @xmath482 @xmath430 @xmath427 @xmath411 @xmath463 @xmath405 @xmath399 @xmath434 @xmath475 @xmath427 @xmath483 + @xmath433 @xmath434 @xmath409 @xmath413 @xmath463 @xmath405 @xmath424 @xmath430 @xmath427 @xmath407 @xmath458 @xmath410 @xmath484 @xmath430 @xmath427 @xmath420 @xmath411 @xmath410 @xmath464 @xmath434 @xmath485 @xmath439 @xmath414 @xmath405 @xmath486 @xmath395 @xmath434 @xmath423 @xmath463 @xmath405 @xmath473 @xmath430 @xmath401 @xmath422 @xmath407 @xmath439 @xmath414 @xmath410 @xmath487 @xmath397 @xmath394 @xmath485 @xmath458 @xmath395 @xmath410 @xmath488 @xmath397 @xmath422 @xmath394 @xmath407 @xmath411 @xmath395 @xmath420 @xmath405 @xmath489 @xmath397 @xmath422 @xmath394 @xmath407 @xmath439 @xmath395 @xmath414 @xmath410 @xmath454 @xmath434 @xmath422 @xmath407 @xmath420 @xmath411 @xmath405 @xmath453 @xmath434 @xmath422 @xmath407 @xmath414 @xmath439 @xmath410 @xmath477 @xmath434 @xmath409 @xmath413 @xmath420 @xmath411 @xmath410 @xmath412 @xmath430 @xmath427 @xmath407 @xmath439 @xmath414 @xmath405 @xmath490 @xmath430 @xmath401 @xmath409 @xmath413 @xmath463 @xmath405 @xmath393 @xmath430 @xmath401 @xmath409 @xmath413 @xmath420 @xmath411 @xmath410 @xmath491 @xmath430 @xmath427 @xmath411 @xmath420 @xmath405 @xmath413 @xmath492 @xmath430 @xmath427 @xmath439 @xmath414 @xmath413 @xmath410 @xmath493 @xmath394 @xmath401 @xmath428 @xmath463 @xmath405 @xmath397 @xmath494 @xmath394 @xmath401 @xmath428 @xmath420 @xmath411 @xmath397 @xmath410 @xmath490 @xmath430 @xmath401 @xmath485 @xmath439 @xmath414 @xmath405 @xmath484 @xmath395 @xmath434 @xmath423 @xmath420 @xmath411 @xmath410 @xmath495 @xmath394 @xmath401 @xmath428 @xmath407 @xmath439 @xmath397 @xmath414 @xmath405 @xmath496 @xmath394 @xmath401 @xmath428 @xmath407 @xmath458 @xmath397 @xmath410 @xmath497 @xmath394 @xmath401 @xmath428 @xmath413 @xmath411 @xmath420 @xmath405 @xmath397 @xmath498 @xmath394 @xmath401 @xmath428 @xmath413 @xmath439 @xmath414 @xmath397 @xmath410 @xmath491 @xmath395 @xmath423 @xmath413 @xmath420 @xmath434 @xmath411 @xmath405 @xmath492 @xmath395 @xmath423 @xmath413 @xmath414 @xmath434 @xmath439 @xmath410 @xmath499 @xmath397 @xmath409 @xmath394 @xmath413 @xmath395 @xmath463 @xmath405 @xmath487 @xmath397 @xmath409 @xmath394 @xmath413 @xmath395 @xmath420 @xmath411 @xmath410 @xmath486 @xmath430 @xmath427 @xmath463 @xmath405 @xmath477 @xmath434 @xmath485 @xmath458 @xmath410 @xmath499 @xmath397 @xmath394 @xmath485 @xmath439 @xmath414 @xmath405 @xmath395 @xmath393 @xmath430 @xmath401 @xmath485 @xmath458 @xmath410 @xmath474 @xmath430 @xmath401 @xmath422 @xmath407 @xmath411 @xmath420 @xmath405 @xmath412 @xmath434 @xmath423 @xmath407 @xmath439 @xmath395 @xmath414 @xmath405 @xmath424 @xmath434 @xmath423 @xmath407 @xmath458 @xmath395 @xmath483 + @xmath500 @xmath430 @xmath397 @xmath409 @xmath485 @xmath411 @xmath420 @xmath405 @xmath501 @xmath485 @xmath409 @xmath420 @xmath411 @xmath410 @xmath501 @xmath502 @xmath458 @xmath410 @xmath503 @xmath397 @xmath409 @xmath485 @xmath439 @xmath414 @xmath410 @xmath504 @xmath434 @xmath505 @xmath439 @xmath407 @xmath414 @xmath410 @xmath506 @xmath507 @xmath463 @xmath405 @xmath508 @xmath502 @xmath439 @xmath414 @xmath405 @xmath509 @xmath397 @xmath505 @xmath407 @xmath420 @xmath411 @xmath410 @xmath508 @xmath485 @xmath409 @xmath463 @xmath405 @xmath510 @xmath507 @xmath420 @xmath411 @xmath410 @xmath506 @xmath505 @xmath407 @xmath439 @xmath414 @xmath405 @xmath509 @xmath397 @xmath422 @xmath511 @xmath458 @xmath410 @xmath512 @xmath434 @xmath505 @xmath411 @xmath407 @xmath420 @xmath405 @xmath513 @xmath511 @xmath422 @xmath439 @xmath414 @xmath410 @xmath514 @xmath511 @xmath422 @xmath411 @xmath420 @xmath405 @xmath510 @xmath505 @xmath407 @xmath458 @xmath410 @xmath515 @xmath397 @xmath505 @xmath407 @xmath463 @xmath405 @xmath516 @xmath434 @xmath517 @xmath413 @xmath463 @xmath405 @xmath518 @xmath434 @xmath517 @xmath413 @xmath420 @xmath411 @xmath410 @xmath515 @xmath397,@xmath422 @xmath511 @xmath439 @xmath414 @xmath405 @xmath518 @xmath434 @xmath409 @xmath485 @xmath458 @xmath410 @xmath519 @xmath517 @xmath413 @xmath439 @xmath414 @xmath410 @xmath520 @xmath517 @xmath413 @xmath411 @xmath420 @xmath405 @xmath516 @xmath434 @xmath409 @xmath485 @xmath439 @xmath414 @xmath455 + @xmath521 @xmath517 @xmath407 @xmath410 @xmath522 @xmath523 @xmath511 @xmath409 @xmath405 @xmath524 @xmath508 @xmath525 @xmath526 @xmath510 @xmath527 @xmath528 @xmath529 @xmath502 @xmath422 @xmath522 @xmath410 @xmath530 @xmath434 @xmath517 @xmath407 @xmath524 @xmath405 @xmath516 @xmath434 @xmath409 @xmath511 @xmath526 @xmath509 @xmath397 @xmath517 @xmath407 @xmath528 @xmath531 @xmath507 @xmath413 @xmath524 @xmath405 @xmath532 @xmath397 @xmath505 @xmath485 @xmath405 @xmath524 @xmath515 @xmath397 @xmath422 @xmath502 @xmath526 @xmath518 @xmath434 @xmath507 @xmath413 @xmath528 @xmath533 @xmath397 @xmath409 @xmath511 @xmath522 @xmath410 @xmath506 @xmath505 @xmath485 @xmath526 @xmath501 @xmath485 @xmath505 @xmath528 @xmath534 @xmath434 @xmath505 @xmath485 @xmath522 @xmath410 as a consequence of theorem [ interestable ] , a new proof of bernstein - koushnirenko theorem for plane curves over an arbitrary algebraically closed field can be derived from the classic theorem over @xmath535 ( @xcite , @xcite ) . we refer to @xcite for a direct proof in positive characteristic . let @xmath2 , @xmath3 be two polynomials over @xmath7 , an algebraically closed . let @xmath102 , @xmath103 be the newton polytope of the polynomials @xmath2 and @xmath3 respectively . then , if the coefficients of @xmath2 and @xmath3 are generic , then the number of common roots of the curves in @xmath70 counted with multiplicities is the mixed volume of the newton polygons @xmath536 if the coefficients of the polynomials are generic , the number of roots in the torus counted with multiplicities is the degree minus the order of the resultant of the two polynomials with respect to one of the variables . this number does only depend on the support of the polynomials , and it is equal to the mixed volume of the newton polygons , because this is the number of stable intersection points of two tropical curves of newton polygons @xmath102 , @xmath103 . another application of the techniques developed in this article is the computation of tropical bases . theorem [ initialserie ] proves that for a hypersurface @xmath2 , the projection @xmath537 . this is not true for general ideals . if @xmath538 $ ] and @xmath539 is the variety it defines in @xmath540 , @xmath541 but it is possible that both sets are different . a set of generators @xmath542 of @xmath543 such that @xmath544 is called a tropical basis of @xmath543 . in @xcite , it is proved that every ideal has a tropical basis and it is provided an algorithm for the case of a prime ideal @xmath543 . an alternative for the computation of a tropical basis of a zero dimensional ideal in two variables is the following . let @xmath545 be a zero dimensional ideal in two variables . let @xmath546 , @xmath547 be the resultants with respect to @xmath22 and @xmath29 of the curves . let @xmath93 be the intersection of the projections @xmath548 and @xmath549 . this is always a finite set that contains the projection of the intersection of @xmath2 , @xmath3 . it may happen that @xmath93 is not contained in the stable intersection of the corresponding tropical curves @xmath0 and @xmath1 , though . let @xmath352 be a natural number such that @xmath353 is injective in @xmath93 . let @xmath550 be another resultant . then , it follows that @xmath551 is a tropical basis of the ideal @xmath552 . this alternative approach is very similar to the regular projection method that has been developed by hept and theobald @xcite . along the article , the notion of tropical resultant has been defined as the projection of the algebraic resultant . it is needed a precomputation of the algebraic resultant in order to tropicalize it . for the case of plane curves , it would be preferable to have a determinantal formula . that is , to prove that the determinant of the sylvester matrix of two polynomials define the resultant variety . but the proof of the properties is achieved by a careful look to the polynomials involved , paying special attention to the cancellation of terms . in the case of the determinant of the sylvester matrix , the tropical determinant of the sylvester matrix is the projection of the permanent of the algebraic determinant . there are cancellation of terms even in the equicharacteristic zero case . it is conjectured that still the determinant of the sylvester matrix is a tropical polynomial that defines the same tropical variety as the resultant does . the author has checked that it is the case for polynomials up to degree four with full support . j. richter - gebert , b. sturmfels , and t. theobald . first steps in tropical geometry . in _ idempotent mathematics and mathematical physics _ , volume 377 of _ contemp . _ , pages 289317 . soc . , providence , ri , 2005 .

the effect of noise on various protocols of secure quantum communication has been studied . specifically , we have investigated the effect of amplitude damping , phase damping , squeezed generalized amplitude damping , pauli type as well as various collective noise models on the protocols of quantum key distribution , quantum key agreement , quantum secure direct quantum communication and quantum dialogue . from each type of protocol of secure quantum communication , we have chosen two protocols for our comparative study ; one based on single qubit states and the other one on entangled states . the comparative study reported here has revealed that single - qubit - based schemes are generally found to perform better in the presence of amplitude damping , phase damping , squeezed generalized amplitude damping noises , while entanglement - based protocols turn out to be preferable in the presence of collective noises . it is also observed that the effect of noise entirely depends upon the number of rounds of quantum communication involved in a scheme of quantum communication . further , it is observed that squeezing , a completely quantum mechanical resource present in the squeezed generalized amplitude channel , can be used in a beneficial way as it may yield higher fidelity compared to the corresponding zero squeezing case .

1603.00178

in 1984 , bennett and brassard proposed the first protocol for quantum key distribution ( qkd ) which is now known as bb84 protocol @xcite . this pioneering work drew a considerable amount of attention from the scientific community , as it was shown to be able to provide unconditional security , a desired feature for all key distribution schemes , but unachievable in the domain of classical cryptography . the fact that unconditional security can be provided if we use quantum resources for key distribution led to extensive studies on the protocols of secure quantum communication ( see ref . @xcite for further details ) . initial studies were limited to qkd @xcite . these initial studies on qkd brought out a number of facts which were further established later . here , in the context of the present work , we wish to specially stress on a specific aspect . in 1991 , ekert proposed a protocol for qkd using entangled state which can be reduced to bb84 protocol ( which uses single photon ( qubit ) states ) under certain conditions @xcite . later , bennett introduced a single - photon - based scheme for qkd which requires only 2 states , now known as the b92 protocol @xcite . soon bbm protocol @xcite was introduced , and it was found that bbm protocol may be viewed as an entangled - state - based analogue of the single - photon - based b92 protocol . thus , these studies indicated that the security achieved by a single - photon - based scheme can also be achieved by a corresponding entangled - state - based scheme . as we have already mentioned , initial studies on quantum cryptography were limited to qkd @xcite . later on , several other aspects of secure quantum communication were investigated . for example , protocols were proposed for quantum secret sharing @xcite , quantum key agreement ( qka ) @xcite , quantum dialogue ( qd ) @xcite , quantum secure direct communication ( qsdc ) @xcite , and deterministic secure quantum communication ( dsqc ) @xcite . for the purpose of the present study , all these schemes of secure quantum communication can be broadly divided in two classes : class a : single - qubit - based schemes which do not use entangled states to implement the protocol , like bb84 protocol @xcite , b92 protocol @xcite , lm05 protocol @xcite , and class b : entangled - state - based protocols , which uses one or more entangled states to implement the protocol . ekert protocol @xcite , bbm protocol @xcite , ping - pong ( pp ) protocol @xcite , are some of the protocols belonging to class b. in fact , there exists a one to one map between the protocols of class a and class b. in principle , any task that can be implemented using single qubit states can also be implemented using an entangled - state - based scheme . of course , device independent schemes can be realized only using the protocols of class b. however , we do not wish to stress on that feature ( device independence ) here . excluding ideas of device independence , it can be shown that the security provided by a scheme of class a and the corresponding scheme of class b is equivalent in the ideal situation , where noise is not present . to illustrate this point in table [ tab : class1 and 2 ] , we have listed protocols of class a and class b for various tasks related to secure quantum communication . as we have already mentioned in an ideal situation , these schemes ( i.e. , any two schemes shown in the same row of table [ tab : class1 and 2 ] ) are equivalent as far as the ability to perform the cryptographic task in a secure manner is concerned . however , to the best of our knowledge this equivalence is not investigated in the realistic situation ( i.e. , in the presence of noise ) . keeping this fact in mind , this paper aims to perform a comparative study of the protocols for secure quantum communication under various noise models . specifically , we wish to compare single - qubit - based protocols ( protocols of class a ) with entangled - state - based protocols ( protocol of class b ) under various noise models . here , it may be noted that although , such comparative study has not yet been performed for protocols of class a and class b mentioned above , a similar comparative study has been performed on conjugate - coding - based protocols of secure quantum communication and orthogonal - state - based protocols of secure quantum communication , which are equivalent in the ideal situation ( @xcite and references therein ) , but not in noisy environment ( @xcite and references therein ) . further , there are various equivalent but different decoy - qubit - based strategies ( such as the bb84 subroutine , gv subroutine ) for eavesdropping checking that are used in standard protocols of secure quantum communication . these subroutines are also known to be equivalent in a noise free environment , but a recent study has established that they are not equivalent in a noisy environment @xcite . this recent observation has further motivated us to perform the present investigation and to systematically investigate the effect of different type of noises on various type of schemes of secure quantum communication . from class a & protocol from class b + 1 & qkd & b92 protocol @xcite & bbm protocol @xcite + 2 & qka & chong et al . protocol @xcite & shukla et al . protocol @xcite + 3 & qsdc & lm05 protocol @xcite & pp protocol @xcite + 4 & qd & shi et al . protocol @xcite & ba an protocol @xcite + there are several noise models @xcite . here , we will restrict ourselves to the study of the effects of amplitude damping ( ad ) channel , phase damping ( pd ) channel @xcite , collective noise and pauli noise . finally , we will also discuss squeezed generalized amplitude damping ( sgad ) channel @xcite and note that results for the generalized amplitude damping ( gad ) channel as well as that for the ad channel can be obtained from the results computed for the sgad channel . the motivation to study these noise models is that the ad noise model deals with an interaction of the quantum system with a zero temperature ( vacuum ) bath . an energy dissipation is involved in this noise model while not in pd . these two noise models can bring about the phenomena of entanglement decay and entanglement sudden - death @xcite . here , as we wish to analyze the equivalence between a single - qubit - based scheme with an entanglement - based one , these two noise models become relevant . collective noise is a coherent effect on all the qubits , viz . , all the polarization encoded photons traveling through an optical fiber undergo the same birefringence @xcite . the pauli noise channels include various physically relevant cases , such as bit flip , phase flip , and depolarizing channels @xcite . sgad channels are a generalization of the ad family of channels , which includes the gad and involves the dissipative interaction with a non - zero temperature bath with non - vanishing squeezing @xcite . the squeezing , being a quantum resource , provides an edge over gad channels , which study a dissipative interaction with a finite temperature bath without squeezing @xcite . hence , the choice of sgad channel enables investigations into both non - zero as well as vanishing regimes of squeezing . the wide applicability of all these noise models sets our motivation to systematically study various schemes for secure quantum communication under noisy environment and to analyze their equivalence . the remaining part of the paper is organized as follows . in section [ sec : noise - models ] , we briefly discuss the noise models we are going to apply on the schemes mentioned in table [ tab : class1 and 2 ] . the next section is dedicated to the method adopted to study the effect of noise models described in section [ sec : noise - models ] . in section [ sec : various - protocols ] , we briefly describe the protocols listed in table [ tab : class1 and 2 ] , and report the effect of various type of noises on these protocols with a clear aim to compare single - qubit - based scheme for a specific cryptographic task with the corresponding entangled - state - based scheme . finally , we conclude in section [ sec : conclusion ] . the most important and widely studied noise models are the ad , pd , collective and pauli noise models . apart from these , generalization of ad considering a dissipative interaction with a thermal and squeezed thermal bath have been studied as gad and sgad , respectively . here , we describe only the sgad channels as the effect of the gad channel can be obtained as its limiting case for zero bath squeezing . further , as the ad noise is a limiting case of gad , it provides a consistency check of the obtained results under sgad noise . in what follows , we will study the effect of all these noise models on the protocols of secure quantum communication that are listed in table [ tab : class1 and 2 ] . the noise models we have opted to study in the present paper are briefly described below . the ad noise simulates the dissipative interaction of a quantum system with a vacuum bath . a perception about the importance of this noise model can be obtained easily if we consider the large number of theoretical and experimental works on this noise model reported in the recent past ( @xcite and references therein ) . the kraus operators of an ad channel are given by @xcite @xmath0,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,e_{1}=\left[\begin{array}{cc } 0 & \sqrt{\eta}\\ 0 & 0 \end{array}\right],\label{eq : krauss - amp - damping}\ ] ] where @xmath1 ( @xmath2 ) is the probability of error or decoherence rate . similarly , kraus operators for phase - damping noise model are @xcite @xmath3,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,e_{1}=\sqrt{\eta}\left[\begin{array}{cc } 1 & 0\\ 0 & 0 \end{array}\right],\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,e_{2}=\sqrt{\eta}\left[\begin{array}{cc } 0 & 0\\ 0 & 1 \end{array}\right],\label{eq : krauss - phase - damping}\ ] ] where @xmath1 ( @xmath2 ) is the decoherence rate . this is another widely studied noise model . for instance , pd noise is discussed in refs . ( @xcite and references therein ) . this noise model is also experimentally simulated in refs . @xcite . a coherent effect of environment on all the travel qubits passing through a channel @xcite can be studied using collective rotation ( cr ) and dephasing ( cd ) noise models . it is known that the singlet states are resistant to an arbitrary collective noise @xcite . recently , the effect of collective noise on various schemes of quantum communication has been studied @xcite . interestingly , these studies provided protocols for quantum communication , which use logical qubits to avoid the effect of collective noise ( cf . before we proceed further let us briefly introduce cr and cd noise models . cr noise transforms @xmath4 and @xmath5 . here , @xmath6 is the noise parameter @xcite . mathematically , a rotation operator acts on the quantum state of travel qubits corresponding to this transformation . cd noise leaves @xmath7 unchanged while transforms @xmath8 as @xmath9 , where @xmath10 is the noise parameter @xcite . this is equivalent to a phase gate . the set of all pauli channels is a tetrahedron . the phase flip and phase damping channels correspond to a proper subset of the pauli channels . depolarizing channels forms a 1-simplex embedded within the convex polytope representing the pauli channels @xcite . pauli noise @xcite is studied using operators @xmath11 where @xmath12 , @xmath13 , @xmath14 , and @xmath15 . here , @xmath16 corresponds to the probability with which a particular pauli operation is applied @xcite . corresponding expression for the depolarizing channel can be obtained with @xmath17 for @xmath18 and @xmath19 . specifically , it would mean that with a certain probability the state remains unchanged while with the remaining probability , it becomes completely mixed . further , information regarding bit flip , phase flip and bit - phase flip channels can be obtained with @xmath19 and @xmath20 for @xmath21 and 2 , respectively . this kind of noise channel is studied for noise estimation @xcite , channel characterization @xcite and error correction @xcite . sgad channel is a generalization of the ad and gad channels and is characterized by the following kraus operators @xcite @xmath22,\end{array}\ ] ] @xmath23,\end{array}\ ] ] @xmath24,\end{array}\ ] ] @xmath25,\end{array}\label{eq : sgad - kraussoperators2 - 1}\ ] ] where @xmath26-\exp\left(-\gamma_{0}\left(2n+1\right)t\right)\right\ } , $ ] @xmath27 and @xmath28 . here , @xmath29 is the spontaneous emission rate , @xmath30 and @xmath31 where @xmath32 and @xmath33 is equal to the bath squeezing angle . the analytic expression for the parameter @xmath34 is quite involved , and can be obtained from ref . the beauty of sgad channel is that for zero bath squeezing ( @xmath35 ) , it reduces to gad channel , which can further be reduced to zero temperature bath ( ad channel ) , where @xmath34 becomes 1 . hereafter , we will avoid the time @xmath36 in the argument of all the expressions under sgad noise for simplicity of notations . quasiprobability distributions and tomogram of the single and two qubit spin states under the sgad channels have been studied recently in @xcite . the influence of sgad noise on a quantum cryptographic switch was analyzed in @xcite . the effect of noise can be studied by using a distance - based measure , fidelity , between the final quantum state expected in the absence of noise and the final state obtained when one of the noise models discussed above is considered . to be precise , the strategy adopted in ref . @xcite will be used here . before we discuss various protocols of secure quantum communication and the effect of noise on them , we will briefly summarize the strategy adopted for the task . consider an initial pure state @xmath37 which is to be evolved under a noisy environment . the evolution of the state after applying the kraus operators characterizing a particular noise is @xmath38 , where @xmath39s are the kraus operators for the chosen noise model under consideration . specifically , the kraus operators of ad , pd , sgad and pauli channels are given in section [ sec : noise - models ] . further , in case of the coherent effect of noise on all the qubits , i.e. , collective noise , the transformed state is obtained as @xmath40 , where @xmath41 is the unitary operation due to corresponding noise . the unitary operations for both collective noises are given in the previous section . finally , fidelity , defined as @xmath42 between the final state after the effect of noise @xmath43 and pure initial state @xmath44 is used as a measure of the effect of noise . it would be worth mentioning here that the fidelity expression used here has been used in refs . ( @xcite and references therein ) . however , conventionally , an equivalent , but a slightly different definition of fidelity is used , and fidelity for two quantum states @xmath45 and @xmath46 is defined as @xmath47 . in the current study , we have assumed that one of the noise models is studied at a time . further , we have also considered that only the travel qubits are affected by the environment , while the qubits not traveling through the channel , i.e. , home qubits remain unaffected . we briefly review two protocols for each type of secure quantum communication task ( namely , qkd , qka , qsdc and qd ) , and study the effect of the above described noise models on them . for this we chose one protocol from class a and another one from class b. specifically , for a cryptographic task listed in the second column of table [ tab : class1 and 2 ] , a protocol from class a ( b ) is mentioned in the third ( fourth ) column . here , we aim to compare the protocol mentioned in the third column of table [ tab : class1 and 2 ] with the protocol mentioned in the fourth column of the same row under different type of noise models . the purpose , is to investigate their equivalence when subjected to different noise models discussed in section [ sec : noise - models ] . specifically , the strategy mentioned in the previous section is used here to perform the comparison by comparing fidelity . we obtain expressions of fidelity for the quantum states to be recovered at the end of each protocol . further , we would like to mention that all the fidelity expressions reported here are obtained as an average fidelity for all possible choices of initial states and encoding on them . for example , if we consider ba an protocol of qd where a predecided entangled state is used as initial state , then there will be 16 possible cases as alice and bob each can encode messages using 4 different operations . similarly , for a single - qubit - based qd scheme there are 16 possible cases with 4 initial states and 2 possible encodings by each party . this is why for each type of qd average fidelity is obtained by computing fidelity for all cases and then averaging . a similar approach is adopted in the rest of the paper to obtain average fidelity for various protocols . here , as we compare a single - qubit - based scheme for qkd with a qkd scheme which requires an entangled state . specifically , we opt for b92 protocol @xcite as an example of single - qubit - based qkd scheme and bbm protocol @xcite as its entangled state counterpart . a modified version of bb84 with less resources was proposed by bennett in 1992 @xcite . hence , the protocol is referred to as b92 protocol . the b92 scheme can be summarized in the following steps : b92 1 : : : alice sends a random string of @xmath48 and @xmath49 to bob , where it is assumed that @xmath48 and @xmath49 correspond to bit values 0 and 1 , respectively . + we can easily observe the modification from bb84 as , in bb84 a random string of @xmath50 was prepared . b92 2 : : : bob measures the received qubits in either computational @xmath51 or diagonal @xmath52 basis randomly . + here , bob does not announce his choice of basis , which is in contrast to bb84 . b92 3 : : : from his measurement outcome bob keeps only the qubits with measurement outcome @xmath53 or @xmath54 and announces the same . subsequently , alice also discards the rest of the qubits . the reason behind discarding the measurement outcomes can be understood by noting that a contribution to measurement outputs @xmath48 or @xmath49 can be from both the initial states @xmath48 and @xmath49 , due to which the measurement outcomes @xmath48 or @xmath49 can lead to a non - conclusive result . therefore , only @xmath53 or @xmath54 outcomes are considered which correspond to alice s bit values 1 and 0 . hence , these qubits can be used to generate a random symmetric key . b92 4 : : : bob announces the measurement outcomes of a part of the generated string with the positions of the qubits for verification of eavesdropping . for the corresponding qubits alice checks the measurement outcome with the initial state as @xmath55 and @xmath56 . for the errors above a tolerable limit the protocol is discarded . otherwise a secure and symmetric key can be generated between the two users . the protocol described above can be studied under various noise models . when the qubit prepared by alice travels to bob under the effect of ad noise , the obtained fidelity is @xmath57 here , and in the remaining part of the paper , required expressions of fidelity are provided using a notation of the form @xmath58 , where @xmath59 is the type of noise model ; @xmath60 is 1 and 2 for single - qubit - based and entanglement - based schemes , respectively ; and @xmath61 denotes the type of secure quantum communication , i.e. , @xmath62 . now , considering that travel qubits have propagated via a pd channel , we obtain @xmath63 in the collective noisy environment , the obtained fidelity expressions are @xmath64 and @xmath65 for cd and cr noise channels , respectively . the analytic expressions of fidelity under the effect of pauli and sgad channels are @xmath66 and @xmath67 respectively . the bbm protocol @xcite is a variant of the ekert protocol @xcite , with reduced resources . specifically , ekert protocol uses three mutually unbiased bases ( mubs ) to calculate the correlation function for detecting eavesdropping when the entanglement source was kept in between the two authenticated users alice and bob @xcite . in contrast , in bbm protocol , the source of entangled photon is given to alice and the requirement of three mubs are reduced to two @xcite . the protocol can be summarized as follows . bbm 1 : : : alice prepares a string of the singlet state @xmath68 and sends the second qubit to bob keeping the first qubit with herself . bbm 2 : : : both alice and bob measure their qubits of shared quantum state in either computational @xmath51 or diagonal @xmath52 basis randomly . both the users announce their choices of measurement basis , but not the measurement outcomes . bbm 3 : : : both users decide to discard the measurement outcomes where their choices of measurement basis were different , as in all the remaining cases their measurement outcomes are supposed to be correlated . bbm 4 : : : finally , both the users choose around half of the string of the undiscarded instances and announce corresponding measurement outcomes . if the error in the measurement outcomes is below certain tolerable limit both alice and bob can obtain a symmetric key using the outcomes of the measurements performed on the remaining qubits which are not used for eavesdropping check . in other words , a lack of correlation in the measurement outcomes is a signature of the presence of an adversary . the entanglement - based protocol of qkd considered here , i.e. , bbm scheme , under the ad , pd , cd and cr noises lead to the following fidelities expressions @xmath69 @xmath70 @xmath71 and @xmath72 respectively . in case of pauli channels , the fidelity only depends on the probability with which the state remains unchanged @xmath73 hence , a linear plot is expected . when the qubits travel under the dissipative sgad channel , the compact form of fidelity is @xmath74 now , we will try to make a comparative analysis of the obtained fidelities expressions in the qkd protocols from the two classes . when the travel qubits are subjected to ad noise we observe that b92 protocol performs better than bbm protocol for all values of decoherence rate @xmath1 . quite similar nature is observed with pd noise as well . the gradual decrease in the fidelity with the increasing decoherence rate for all these cases can be observed in fig . [ fig : qkd-2d ] a and b. the fact observed here is consistent with some of our recent observations that single qubits perform better while traveling through ad and pd channels @xcite . interestingly , fidelity obtained for both the qkd schemes considered here is the same when the travel qubits are subjected to cr noise , as can be seen in fig . [ fig : qkd-2d ] c. in the presence of another type of collective noise ( namely cd noise ) , which is dephasing in nature , b92 is again seen to perform better ( cf . [ fig : qkd-2d ] d ) . here , it is worth noting that the singlet state is decoherence free in an arbitrary collective noise when both the qubits of the single state travel through the noisy channel . however , when one of the qubits travels through the channel having collective noise , it gets affected by the noise and as a consequence singlet state also gets affected by the noise . this is what we have observed here . expressions of fidelity under pauli noise reveal that for bbm protocol equally affected states will be obtained for bit - flip , phase - flip , bit - phase flip errors with certain probability . for the depolarizing channel the fidelity expression shows a similar nature . the same expression is obtained for b92 protocol with bit - phase flip error . further , bit - flip and phase flip errors with equal probabilities affect the state in a similar manner as the expression becomes same in both the cases . the values are always higher than all others for the same amount of error as shown in fig . [ fig : qkd-2d ] e. the obtained fidelity in depolarizing channels can be seen to be intermediate between the last two . from the fidelity expressions of sgad noise , the corresponding fidelities under ad and gad channels can be obtained as limiting cases . in fig . [ fig : qkd-2d ] f we can see the advantage obtained due to squeezing . specifically , for non - zero squeezing we can obtain higher fidelity than that with a gad noisy environment for a longer time period . further , it can also be observed that a state is more affected while traveling through finite temperature bath than in the vacuum bath ( ad ) . qkd in ad , pd , cr and cd noises is shown in ( a ) , ( b ) , ( c ) and ( d ) , respectively . the smooth ( blue ) and dashed ( red ) lines in all these plots correspond to b92 and bbm protocols , respectively . for cr noise both b92 and bbm protocols have the same fidelity ( in ( c ) ) . in ( e ) , the smooth ( blue ) line corresponds to the fidelity variation with probability of bit - phase flip error for b92 protocol . this same curve also illustrates the dependence of fidelity on probability in all four possible cases discussed in the text for bbm protocol . the dashed ( red ) and dotted dashed ( cyan ) lines show fidelity variation in b92 scheme with bit / phase flip error and depolarizing channel , respectively . ( f ) demonstrates the effect of ad ( in smooth ( blue ) and dashed ( red ) lines ) ; gad ( in dotted dashed ( cyan ) and dotted ( magenta ) lines ) with temperature @xmath75 ; and sgad ( in large dashed ( orange ) and large dotted dashed ( purple ) lines ) with @xmath75 and squeezing parameters @xmath76 and @xmath77 for b92 and bbm protocols , respectively . ] in realistic scenarios , it may be preferable that a single party does not control the whole key . in such scenarios qkd can be circumvented by a key agreement protocol , where all the parties can equally contribute in the final key . to be precise , qka schemes are studied under two notions : weaker and stronger . in the weaker notion of qka protocols , the final key is generated after negotiation between both the parties . if we follow this notion , then many of the qkd schemes can be viewed as qka schemes , such as bb84 , b92 and bbm discussed in the previous subsection . however , in the strong notion all the parties contribute equally to the final shared key . many qka schemes have been proposed in the past ( @xcite and references therein ) . a single - qubit - based quantum key agreement protocol given by chong et al . , in 2010 @xcite can be described in the following steps : qka1 - 1 : : alice randomly prepares an @xmath78 bit raw key @xmath79 and a random string of 0 and 1 . qka1 - 2 : : alice prepares @xmath78 qubits in such a way that for every 0 ( 1 ) in the key she prepares either @xmath48 or @xmath49 ( @xmath53 or @xmath54 ) depending upon the corresponding bit value in the random string 0 or 1 , respectively . finally , she sends all the qubits to bob . qka1 - 3 : : bob also prepares an @xmath78 bit raw key @xmath80 . now , to encode this key he applies @xmath81 ( @xmath82 ) on the received qubits for 0 ( 1 ) . qka1 - 4 : : bob selects a random sequence from the qubits as verification string and announces the positions of the corresponding qubits . he also announces his raw key . qka1 - 5 : : alice can extract a final key as @xmath83 from her and bob s keys . subsequently , she broadcasts the obtained values corresponding to the qubits bob had chosen as verification string along with the information of basis chosen for each qubit in qka1 - 2 . qka1 - 6 : : using the information of the basis chosen bob can also extract the final key @xmath84 . if the obtained values for alice and bob have errors below a tolerable limit they share an unconditionally secure quantum key . if the single - qubit - based qka scheme described above is implemented using a quantum channel having ad noise then we obtain @xmath85 whereas under pd noise we have @xmath86 on the effect of cd noise the fidelity becomes @xmath87 while under the influence of cr noise it is @xmath88 in case the travel particles go through a pauli channel the obtained fidelity is @xmath89 for an interaction with a squeezed thermal bath , the fidelity depends on various parameters as @xmath90 there are various protocols of quantum key agreement that exploit entanglement . here , we wish to summarize a protocol proposed by shukla et al . in 2014 @xcite . qka2 - 1 : : alice prepares @xmath91 , where @xmath92 . she also prepares a raw key @xmath79 of @xmath78 bits . she prepares a string of all the first particles to be sent to bob keeping all the second qubits with herself . qka2 - 2 : : alice prepares @xmath93 bell states @xmath94 as decoy qubits and concatenates them with the string of the first particles of the bell states and sends the @xmath95 qubits to bob after applying a permutation operator @xmath96 . qka2 - 3 : : after an authentic acknowledgment of the receipt of all the qubits alice announces the positions of the decoy qubits , i.e. , information of @xmath97 . using this information bob performs a bell state measurement on partner pairs and calculates error rate . it would be relevant to mention that the decoy - qubit - based security achieved here with gv subroutine can be equivalently done by bb84 subroutine where single qubit decoy qubits are used . they decide to proceed if error rates are below a certain value . qka2 - 4 : : bob also prepares a raw key @xmath80 . further , on the remaining qubits bob encodes his raw key by applying @xmath81 or @xmath98 operations for 0 and 1 , respectively . subsequently , he prepares @xmath93 bell states as decoy qubits and permutes the string of @xmath95 qubits by permutation operator @xmath99 after concatenating the decoy and encoded qubits . finally , he sends them to alice . qka2 - 5 : : bob informs the coordinates of the decoy qubits using which alice computes the error rate . from this they choose whether to proceed or not . qka2 - 6 : : alice announces her key publicly from which bob can generate the final key @xmath83 . qka2 - 7 : : bob announces the permutation operator to rearrange the particles in the encoded string with alice . using this alice performs a bell state measurement on the partner pairs of home and travel qubits . the measurement outcome would reveal bob s key to alice . qka2 - 8 : : alice can also obtain the final shared , unconditionally secure , quantum key @xmath84 . the fidelity expression for the entanglement - based qka scheme when subjected to ad and pd noise are @xmath100 and @xmath101 respectively . the quantum state evolves under the collective noise such that the obtained fidelity with the expected pure state is @xmath102 and @xmath103 for cd and cr noise , respectively . here , it may be noted that the two noise parameters @xmath104 and @xmath105 correspond to each round of the travel qubit . the pauli channels have a symmetric expression for fidelity , given by @xmath106 the closed form analytic expression of fidelity , under the sgad channel , for the above described qka scheme @xcite is @xmath107 for qka schemes two way quantum communication is involved unlike qkd protocols in the previous section , where only sender to receiver communication is involved . in both single - qubit - based and entangled - state - based qka schemes fidelity falls gradually with an increase in decoherence rate @xmath1 when subjected to ad and pd noisy environments of identical strength ( cf . fig . [ fig : qka-2d ] a and b ) . similar to the qkd scheme , single - qubit - based schemes perform better than the entangled - state - based ones in both these noisy channels . further , this similarity between the single - qubit - based qkd and qka schemes for collective noises , is depicted in the corresponding curves shown in figs . [ fig : qkd-2d ] c and d and figs . [ fig : qka-2d ] c and d. however , the entangled - state - based qka scheme is seen to benefit under collective noise as fidelity for the entangled - state - based qka scheme is more than that of single - qubit - based protocol , under the assumption of the same noise strength in both rounds of the travel qubit . this fact can be attributed to different choices of bell state in entanglement - based qkd and qka protocols . further , as discussed in the previous section , the collective noise parameter remains the same for all the qubits traveling through a channel at a particular time , but can have a different value at any other time . the effect of two different values of noise parameters , of the collective noises , on the fidelity of the obtained state can be studied by showing either 3 dimensional variation or contour plots . [ fig : qka-3d ] a and b ( c and d ) show both these kinds of plots for qka scheme subjected to cr ( cd ) noise . hereafter , we will stick to the contour plots to illustrate the effect of two parameters . interestingly , it can be observed that it is possible to obtain states with unit or null fidelity for some values of noise parameters . for the single - qubit - based scheme , the analytic expressions for fidelities are the same for all three types of pauli channels ( i.e. , for bit flip , phase flip and bit - phase flip channels ) . the expressions of fidelity for bit flip and phase flip channels are also the same for the entangled - state - based protocol , but for bit - phase flip error , we obtain a different expression for fidelity , and it is observed that the obtained value of fidelity is smaller compared to the corresponding values for bit flip and phase flip errors . [ fig : qka-2d ] e shows variations of fidelity in all these error channels , where an increase in fidelity for entanglement - based qka schemes can be attributed to the presence of quadratic terms in the fidelity expression . the variation of fidelity in fig . [ fig : qka-2d ] f considering a dissipative interaction via sgad channel for both kinds of qka schemes reemphasize the facts established by their qkd counterparts ( cf . [ fig : qkd-2d ] f ) . specifically , with increase in temperature , dissipation increases , causing decay in the fidelity of the recovered state . also , squeezing turns out to be a useful resource here , as observed from the increased fidelity of the sgad plots compared to their gad ( without squeezing ) counterparts after a certian evolution period . ( a)-(d ) illustrates the fidelity obtained for qka protocols when subjected to ad , pd , cr and cd noises , respectively . the smooth ( blue ) and dashed ( red ) lines correspond to single - qubit - based and entangled - state - based qka protocols , respectively . for cr and cd noises it is assumed that the noise parameter is same for both the directions of travel of the qubit ( i.e. , alice to bob and bob to alice ) . in ( e ) , the effect of bit flip error on single - qubit - based qka and entangled - state - based qka protocols are shown using smooth ( blue ) and dashed ( red ) lines , respectively . in the same plot , dotted dashed ( cyan ) and large dotted dashed ( purple ) lines correspond to the effect of depolarizing channel on shi et al.s and shukla et al.s qka schemes , respecetively ; and the dotted ( magenta ) line illustrates the effect of bit - phase flip error on the single - qubit - based qka scheme . ( f ) corresponds to the effect of ad in smooth ( blue ) and dashed ( red ) lines ; gad in dotted dashed ( cyan ) and dotted ( magenta ) lines with @xmath75 ; and sgad in large dashed ( orange ) and large dotted dashed ( purple ) lines with @xmath75 and squeezing parameters @xmath76 and @xmath77 for single - qubit - based and entangled - state - based qka protocols , respectively . ] contour and 3d variation of fidelity of entangled - state - based qka protocol is shown in cr and cd noises . specifically , in ( a ) and ( c ) the contour plots under the effect of cr and cd noise are shown . corresponding 3d plots can be seen in ( b ) and ( d ) , respectively . the same plots are obtained for pp protocol as well . a detailed discussion follows in subsection [ sub : qsdc - protocols ] . ] quantum secure communication not necessarily involves a key generation or key agreement . there are direct communication protocols avoiding key generation and such protocols are referred to as the protocols for secure direct quantum communication . these protocols can be broadly categorized as qsdc and dsqc protocols depending upon the requirement of additional classical communication for decoding of the information . qsdc protocols do not require any additional classical communication other than that involved in eavesdropping checking , while dsqc protocols do . here , we wish to discuss two qsdc protocols and compare them in the presence of noise . a qsdc protocol without using entanglement was proposed by lucamarini and mancini in 2005 which is now known as lm05 protocol @xcite . the protocol can be briefly describe in the following steps : lm1 : : bob ( receiver ) prepares a random string of @xmath108 and sends it to alice . lm2 : : alice chooses randomly half of the received qubits as a verification string ( to be used as decoy qubits ) and performs eavesdropping checking on these qubits . specifically , alice measures all the qubits in the verification string in mubs @xmath109 or @xmath110 randomly . then she announces the choice of basis with the position of qubits . based on this , bob announces the qubits where he has chosen the same basis to prepare the initial state . depending on this , the measurement outcomes of alice are expected to be the same with the state prepared by bob in the absence of any attempt of eavesdropping . for errors below a tolerable limit they proceed to the next step , else they start afresh . lm3 : : to encode her message on half of the remaining qubits alice applies operator @xmath81 ( @xmath82 ) for sending 0 ( 1 ) . subsequently , she returns the encoded qubits to bob . here , it would be nice to mention that using such a scheme , for both choices of encoding , a particular initial state will transform into orthogonal states . consequently , at bobs end , a message can be easily decoded by measuring the state in the basis it was prepared . lm4 : : alice announces the coordinates of the qubits she had not encoded on ( as she wished to use them as decoy qubits for alice to bob communication ) . bob measures corresponding qubits in the basis he had prepared them initially to check the presence of eve for alice to bob travel of the encoded particles . the same task can also be achieved by alice encoding on all the remaining qubits after eavesdropping in lm2 , while she prepares additional string of equal number of qubits in @xmath108 randomly for eavesdropping checking in this step . lm5 : : only if bob is convinced of the absence of eve , he decodes the message sent by alice by measuring the qubits in the same basis he had prepared them in lm1 , otherwise they abort the protocol . the effect of the ad noise on the single - qubit - based qsdc protocol ( lm05 protocol ) opted here , lm05 , can be deduced from the fidelity expression @xmath111 the corresponding expression under the effect of pd noise is @xmath112 similar to the entanglement - based qka scheme , two rounds of quantum communication is involved here , due to which the expressions of fidelity under cd noise @xmath113 and that for cr noise @xmath114 involve two noise parameters ( @xmath115 or @xmath116 ) each . as usual , the fidelity expression for pauli channels with four parameters is @xmath117 the presence of quadratic terms is signature of two rounds of quantum communication . when the travel qubits undergo a dissipative interaction characterized by the sgad channel , the fidelity is obtained as @xmath118 an entangled - state - based qsdc protocol was proposed by @xmath119 and felbinger in 2002 @xcite . precisely , lm05 protocol is a single - qubit - based counterpart of pp protocol . the pp protocol works as follows : pp1 : : bob prepares @xmath91 , where @xmath92 . then he sends all the first particles to alice keeping all the second qubits with himself . pp2 : : alice forms a verification string by randomly choosing a set of @xmath93 qubits to perform bb84 subroutine as was done in lm2 . specifically , alice measures the qubits randomly in @xmath109 or @xmath110 basis and announces the choice of basis . bob also measures his qubits in the same basis . in the absence of eve , their measurement outcomes are expected to be correlated . in the absence of such a correlation they discard the protocol and return to pp1 , otherwise they proceed . pp3 : : out of half of the remaining qubits alice randomly makes two sets of equal number of qubits . one set for encoding her message and another set for eavesdropping check for alice to bob communication . to encode 1 alice applies @xmath98 gate before sending the qubit to bob , and for sending 0 she returns the qubit unchanged . pp4 : : alice informs the coordinates of verification string and bob performs bb84 subroutine to compute the error rate . pp5 : : for low error rates , bob performs bell state - measurement on the partner pairs to decode the message sent by alice . the analytical expressions of fidelity in the case of the pp protocol exactly match those for entanglement - based qka scheme @xcite . therefore , we avoid repetition of the expressions and carry on with the discussion regarding the comparison between lm05 and pp protocols under noisy environments . both the qsdc protocols when subjected to noise are affected to different extent . precisely , as observed in the protocols discussed so far , the single - qubit - based schemes have been found to be more efficient as compared to entangled - state - based schemes in ad and pd noisy channels . this is also observed here in figs . [ fig : qsdc-2d ] a and b. under the assumption of the same noise parameter for cd and cr noise for alice to bob and bob to alice travel of the qubits , pp protocol is affected by the cr and cd noise in a manner similar to the entangled - state - based qka protocol . in fact , in the entangled - state - based qka scheme one of the parties sends the raw key by pp type qsdc while the other party announces it . therefore , the effect of noise is the same as in pp protocol . the single - qubit - based scheme has different nature in fig . [ fig : qsdc-2d ] c and d as compared to the corresponding qkd and qka protocols . this can be attributed to the two way quantum communication associated in this scheme , unlike the last two cases where it was unidirectional . further , in the presence of cd noise , the benefit of bidirectional communication can be easily observed as the observed fidelity is more than the previous cases . the single qubits ( in lm05 ) perform better when subjected to cd noise , but suffer more under the influence of cr noise . in fig . [ fig : qsdc-3d ] , we have not shown the contour plots for the fidelity for pp protocol under collective noise as the expressions are exactly the same as that illustrated through fig . [ fig : qka-3d ] for shukla et al s qka scheme . the contour plots also show that very low fidelity is also possible for some particular values of noise parameters during the two directions of transmission . further , under the effect of cd noise , a similar nature of the fidelity variation under lm05 and pp protocols can be observed in fig . [ fig : qsdc-3d ] . however , a closer look reveals that under cd noise fidelity obtained for lm05 protocol is more than that obtained for pp protocol , indicating that for cd noise , single - qubit - based lm05 protocol performs better than the corresponding entangled - state - based pp protocol . the expressions of fidelity under pauli noise reveal that the fidelity for pp protocol in bit , phase and bit - phase flip is the same as the fidelity for bit - phase flip errors for equal probability of error in lm05 protocol . in this case the fidelity resurrects to 1 for maximum probability of error . quite a similar nature is observed for fidelity under bit flip and phase flip errors in lm05 scheme though it remains less than that of the corresponding values in the pp protocol . similarly , under the influence of the depolarizing channel the fidelity fails to revive but remains always more for lm05 protocol . the advantage of squeezing , a purely quantum resource , can be observed in fig . [ fig : qsdc-2d ] f , where in the presence of squeezing after an appreciable amount of time , fidelity higher than the corresponding case of zero squeezing can be observed . specifically , higher fidelity under sgad channel relative to ad channel shows that coherence can be sustained using squeezing that would have been lost due to the presence of non - zero temperature . qsdc under ad , pd , cr and cd noises are depicted in ( a ) , ( b ) , ( c ) and ( d ) , respectively . the smooth ( blue ) and dashed ( red ) lines correspond to lm05 and pp protocols , respectively . for cr and cd noises it is assumed that the noise parameter is same for both the directions of travel of the qubit ( i.e. , alice to bob and bob to alice ) . in ( e ) , the fidelity under the depolarizing channel for lm05 and pp protocols are shown in dotted ( magenta ) and dotted dashed ( cyan ) lines , respectively . in all the remaining cases of pp and bit phase flip for lm05 the red line illustrates the fall and revival in fidelity . lastly , the blue line corresponds to the fidelity in bit flip and phase flip errors in lm05 scheme . ( f ) illustrates the effect of ad ( i.e. , an interaction with a zero temperature and squeezing bath ) in smooth ( blue ) and dashed ( red ) lines ; gad ( i.e. , an interaction with a non - zero temperature and zero squeezing bath ) in dotted dashed ( cyan ) and dotted ( magenta ) lines with @xmath75 ; and sgad ( finite temperature and squeezing bath ) in large dashed ( orange ) and large dotted dashed ( purple ) lines with @xmath75 and squeezing parameters @xmath76 and @xmath77 for lm05 and pp protocols , respectively . ] contour plots of fidelity of lm05 protocol when subjected to cr and cd noises in ( a ) and ( b ) , respectively . ] one of the most efficient secure quantum communication schemes is the quantum dialogue ( qd ) . in this scheme both the legitimate parties encode their information on the same qubits and at the end of the protocol each party can deduce the others message . the first qd scheme was proposed by ba an using bell states in 2004 @xcite . recently , yang and hwang proposed a qd scheme immune to the collective noise using logical qubits @xcite . here , we consider two qd protocols for analyzing their performance when subjected to noisy environments . a modified qd protocol using only single qubit states and mubs was proposed by shi et al . in 2010 @xcite . shi et al . protocol can be described in the following steps : qd 1 : : bob prepares a sequence of @xmath95 single qubits randomly in @xmath120 or @xmath121 basis . he intentionally prepares two copies of @xmath78 qubits , i.e. , two adjacent photons are in the same quantum state . for example , bob prepares a string of single qubits as @xmath122 , and out of each pair , one qubit will be used for encoding while the other one will be used to send the initial state information . he also prepares some additional decoy qubits to be used for eavesdropping check in each round of communication . finally , he sends all the @xmath95 qubits after inserting decoy qubits randomly in them to alice . qd 2 : : bob and alice perform security checking for the received qubits . specifically , bob will announce positions of the decoy qubits he chooses for this round of communication and alice announces her choice of measuring basis and corresponding outcome . using that bob determines the error rate and decides whether to proceed or call off the protocol . qd 3 : : with the help of bob , alice can separate three sequences : the first one of decoy qubits , and two sequences of one copy of initial states each . out of these three sequences she encodes her message on the second sequence using @xmath81 ( @xmath82 ) operation for sending bit value 0 ( 1 ) . subsequently , she encodes a checking message on the decoy qubits using the same scheme and concatenates these two encoded sequences . finally , she sends this concatenated sequence after randomizing to bob while keeping the last sequence with herself . qd 4 : : after receiving authenticated receipt of all the qubits from bob , alice will announce the positions of the decoy qubits and being aware of the preparation basis he decodes the message and announces it publicly . with this checking message alice decides whether to go to the next step or start afresh . qd 5 : : if they decide to proceed , bob also encodes on the received qubits after rearranging them using the same encoding scheme as alice . subsequently , he performs measurement on all these qubits in the basis in which they were prepared and announces the measurement outcomes . from the measurement outcomes , bob gains the knowledge of alice s encoding as he knows his encoding apart from the initial and final state . further , the measurement outcomes also reveal the choice of the basis used for preparation of the states . using this information alice measures the third sequence , she had kept with herself in qd 3 , in a suitable basis and learns the initial state of bob . from the information of the initial and final states , alice can extract the message of bob by using her knowledge of her encoding . the fidelity expression of single - qubit - based qd scheme under ad channel contains cubic terms @xmath123 which signify bidirectional quantum communication apart from a qsdc to inform bob about the initial state . this fact can also be observed in pd noise @xmath124 for the two rounds of communication under collective noisy channels , characterized by two noise parameters , the fidelity for cd is @xmath125 and for cr noise is @xmath126 similarly , all the qubits traveling through a pauli channel give rise to the fidelity as a function of various parameters , as @xmath127 finally , the obtained fidelity for this qd scheme under the action of the sgad channel is @xmath128 in the originally proposed ba an s qd scheme , both parties can communicate simultaneously using bell states @xcite . the protocol can be summarized in the following steps : qd : ba 1 : : bob prepares @xmath129 . he encodes his message on the first qubit ( travel qubit ) and keeps the second qubit with himself as home qubit . to encode his message he uses dense coding , i.e. , he applies unitary operations @xmath130 and @xmath131 to encode @xmath132 and @xmath133 , respectively . qd : ba 2 : : bob sends all the first qubits to alice and confirms their receipt . qd : ba 3 : : alice also encodes on the travel qubit using the same rule as was used by bob and sends them back to bob . bob performs a bell measurement on the partner particles ( bell measurement is done on a qubit from the sequence of home qubits and another qubit from the sequence of travel qubits , which was initially entangled with the chosen home qubit ) . qd : ba 4 : : after alice s disclosure bob comes to know whether it was message mode ( mm ) or control mode ( cm ) . bob announces his measurement outcome in the mm using which both alice and bob can learn each others message . while in cm alice announces her encoding which bob uses for eavesdropping checking . in the original ba an s qd scheme , when subjected to ad and pd noise , the fidelity can be seen to be @xmath134 and @xmath135 respectively . the presence of quadratic terms is a signature of bidirectional quantum communication involved . under the coherent effect of cd noise on the travel qubits , we obtain @xmath136 and for cr noise the fidelity is found to be @xmath137 when the travel qubit is transmitted through a pauli channel the fidelity is the same as that obtained in case of pp protocol . the analytic expression of fidelity for ba an protocol of qd , when subjected to sgad noise is @xmath138 when both the protocols of qd are subjected to ad noise the fidelity obtained for the entangled - state - based protocol is comparable of that of the single - qubit - based one . this is in contrast with the earlier observations reported in the present work , where single - qubit - based schemes were found to be preferable in cases of ad and pd noise models . though , in the large decoherence limits the single - qubit - based qd turns out to be a suitable candidate ( cf . [ fig : qd-2d ] a ) . it is worth commenting here that the decay in the fidelity of single - qubit - based qd scheme when compared with the corresponding qsdc protocol ( as they are of the same order in entanglement - based schemes ) can be attributed to an extra single qubit traveling through the noisy channel in step qd 1 . however , under the effect of pd channels , the observation established from the previous three secure quantum communication schemes ( namely , qkd , qka and qsdc protocols ) remains valid ( cf . [ fig : qd-2d ] b ) , in other words , it is observed that the single - qubit - based schemes perform better in pd channels . when ba an protocol of qd is subjected to collective noise the same nature of fidelity variation as was observed in pp protocol is observed if the same noise parameters are used in to and fro travel of the qubits . however , in the single - qubit - based qd scheme a different nature from lm05 protocol is observed . interestingly , a close look at figs . [ fig : qsdc-2d ] c ( d ) , and fig . [ fig : qd-2d ] c ( d ) reveals that compared to lm05 protocol , an extra dip is observed at @xmath139 ( where for lm05 fidelity was obtained to be unity ) . this dip was observed in single - qubit - based qka scheme , too ( cf . fig . [ fig : qka-2d ] c ) . this point further establishes the fact that the fidelity of single - qubit - based qd schemes decays , when subjected to ad noise . the contour plots shown in fig . [ fig : qd-3d ] also demonstrate that the single - qubit - based qd scheme has a different nature of fidelity variation compared to that in lm05 protcol , while in entangled - state - based protocol it remains similar to that observed in pp . the expression of fidelity under pauli noise shows that for ba an protocol of qd it is the same as in pp protocol . however , the presence of cubic terms in the expressions of fidelity for single - qubit - based qd is a signature of the nature observed in fig . [ fig : qd-2d ] e , i.e. , the descent for very low and high error probabilities . for the bit / phase flip error the single - qubit - based scheme remains the preferred choice , but for very high probability of errors it should be avoided . a similar nature is also observed under the influence of a depolarizing channel . for bit - phase flip error , fidelity indicates better performance of entanglement - based scheme as compared to their single - particle counterparts . under a dissipative interaction with a non - zero temperature bath , a behavior similar to that observed under an ad channel however , due to the non - zero squeezing , single - qubit - based qd scheme turns out to be a better candidate . [ fig : qd-2d ] f further reiterates the facts observed in fig . [ fig : qd-2d ] a - d , i.e. , the obtained nature in the case of ba an protocol of qd is the same as in pp protocol ; and in single - qubit - based scheme , it can be explained as a compound effect of single - qubit - based qka and qsdc protocols . qd protocols are analyzed under ad , pd , cr and cd noises in ( a)-(d ) , respectively . the smooth ( blue ) and dashed ( red ) lines correspond to the single - qubit - based and ba an s qd protocols , respectively . for cr and cd noises it is assumed that the noise parameter is same for both the directions of travel of the qubit ( i.e. , alice to bob and bob to alice ) . in ( e ) , bit / phase flip is shown together for the single - qubit - based qd and ba an protocol of qd in smooth ( blue ) and dotted ( magenta ) lines , respectively . for ba an protocol of qd bit phase flip errors matches exactly with the previous case . however , for single - qubit - based scheme , it is shown in dashed ( red ) line . under the depolarizing channel the fidelity variation for the single - qubit - based scheme and ba an protocol of qd is demonstrated by dotted dashed ( cyan ) and large dotted dashed ( purple ) lines , respectively . ( f ) corresponds to the effect of ad ( i.e. , an interaction with a zero temperature and squeezing bath ) in smooth ( blue ) and dashed ( red ) lines ; gad ( i.e. , an interaction with a non - zero temperature and zero squeezing bath ) in dotted dashed ( cyan ) and dotted ( magenta ) lines with @xmath75 ; and sgad ( finite temperature and squeezing bath ) in large dashed ( orange ) and large dotted dashed ( purple ) lines with @xmath75 and squeezing parameters @xmath76 and @xmath77 for the single - qubit - based and ba an s qd protocols , respectively . ] contour plots illustrating the variation of fidelity of the single - qubit - based shi et al . protocol and entangled - state - based ba an protocol of qd in cr and cd noises , respectively . ] the comparative study of single - qubit - based and entangled - state - based schemes of secure quantum communication performed in the present work has lead to a number of interesting conclusions . firstly , the equivalence observed in the ideal noiseless scenario is lost in more practical scenarios where noise is present . next , it is observed that it is not possible to say unambiguously that in a noisy environment entangled - state - based protocols perform better than the single - qubit - based protocols or vice versa . in fact , it depends on the nature of the noise present in the channel . specifically , single - qubit - based schemes are generally found to be the suitable choice in the presence of ad and/or pd noises , while entanglement - based protocols turns out to be preferable in the presence of collective noises . as sgad and gad channels are generalizations of the ad channel , conclusions similar to that for the ad channel are also applicable to them . however , with an increase in temperature , fidelity is seen to decay . squeezing is seen to play the role of a beneficial quantum resource , in a host of scenarios , in practical quantum communication . also , it is observed that the effect of noise depends upon the number of rounds ( how many times a travel qubit travels through the noisy channel ) of quantum communication involved . for instance , qkd protocols are least affected by noise , while qd protocols are most affected as in qkd protocols a travel qubit travels only once through the noisy channel , whereas in ba an protocol of qd , it travels twice through the noisy channel . further , the single - qubit - based qd scheme involves three rounds of communication as it requires alice to bob and bob to alice transmission of qubits and an additional alice to bob travel of equal number of qubits . as a consequence , single - qubit - based qd scheme is found to be the most affected among the four different single - qubit - based schemes for secure quantum communication discussed in this paper . 10 bennett , c. h. , brassard , g. : quantum cryptography : public key distribution and coin tossing . in : proceedings of the ieee international conference on computers , systems , and signal processing , bangalore , india , p. 175 ( 1984 ) yuan , h. , song , j. , zhou , j. , zhang , g. , wei , x. : high - capacity deterministic secure four - qubit w state protocol for quantum communication based on order rearrangement of particle pairs . int . * 50 * , 2403 ( 2011 ) sharma , r. d. , thapliyal , k. , pathak , a. , pan , a. k. , de , a. : which verification qubits perform best for secure communication in noisy channel ? quantum inf . process . ( 2015 ) doi 10.1007/s11128 - 015 - 1207 - 6 bourennane , m. , eibl , m. , gaertner , s. , kurtsiefer , c. , cabello , a. , weinfurter , h. : decoherence - free quantum information processing with four - photon entangled states . lett . * 92 * , 107901 ( 2004 ) chiuri , a. , rosati , v. , vallone , g. , p@xmath140dua , s. , imai , h. , giacomini , s. , macchiavello , c. , mataloni , p. : experimental realization of optimal noise estimation for a general pauli channel . * 107 * , 253602 ( 2011 ) turchette , q. a. , myatt , c. j. , king , b. e. , sackett , c. a. , kielpinski , d. , itano , w. m. , monroe , c. , wineland , d. j. : decoherence and decay of motional quantum states of a trapped atom coupled to engineered reservoirs . a * 62 * , 053807 ( 2000 ) ; myatt , c. j. , king , b. e. , turchette , q. a. , sackett , c. a. , kielpinski , d. , itano , w. m. , monroe , c. , wineland , d. j. : decoherence of quantum superpositions through coupling to engineered reservoirs . nature * 403 * , 269 ( 2000 ) thapliyal , k. , pathak , a. : applications of quantum cryptographic switch : various tasks related to controlled quantum communication can be performed using bell states and permutation of particles . quantum inf . process . * 14 * , 2599 ( 2014 )

by performing n - body simulations of chemodynamical evolution of galaxies with dusty starbursts , we investigate photometric evolution of gas - rich major mergers in order to explore the nature of ultraluminous infrared galaxies ( ulirgs ) with the total infrared luminosity ( for m ) of . main results are the following five . \(1 ) global colors and absolute magnitudes the during dusty starburst of a major merger do not change with time significantly , because interstellar dust heavily obscures young starburst populations that could cause rapid evolution of photometric properties of the merger . \(2 ) dust extinction of stellar populations in a galaxy merger with large infrared luminosity ( ) is selective in the sense that younger stellar populations are preferentially obscured by dust than old ones . this is because younger populations are located in the central region where a larger amount of dusty interstellar gas can be transferred from the outer gas - rich regions of the merger . \(3 ) both and the ratio of to band luminosity ) increases as the star formation rate increase during the starburst of the present merger model , resulting in the positive correlation between and . the dust temperature and the flux ratio of also increase with the increase of the star formation rate , because a larger number of young stars formed by massive starburst can heat the dusty interstellar gas as the star formation becomes larger . \(4 ) the star formation efficiency , total gas mass , the degree of dust extinction ( ) , , , , and depend strongly on the separation of two cores of the merger , which clearly reflects the fact that dynamical processes of galaxy merging play an important role in determining the photometric evolution of dusty starbursts in the merger . \(5 ) the two - dimensional distribution of global colors ( e.g. , ) shows a negative color gradient during starburst , mainly because central young starburst populations are preferentially and very heavily obscured by dust . furthermore , the peak of the two dimensional distribution of coincides with that of . these results clearly suggest that the spatial distribution of gas and stars is one important determinant of the two - dimensional photometric properties of major mergers . our numerical results clearly demonstrate that dynamical processes of major galaxy merging , which can control the time evolution of the relative spatial distribution of dusty interstellar gas and young stars , play a vital role in determining photometric properties of ulirgs .

astro-ph0012365

ultraluminous infrared galaxies ( ulirgs ) with @xmath16 @xmath3 @xmath4 @xmath5 discovered by the @xmath17 @xmath18 @xmath19 ( iras ) are considered to be ideal laboratories for studying the formation of dusty starburst galaxies , elliptical galaxies , and qsos . detailed morphological studies of these ulirgs have revealed that virtually all show evidence of galaxy merging or strong tidal interaction ( e.g. , tidal tails and double or multiple nucleus ) and thus that most of ulirgs are formed by galaxy interaction / merging ( e.g. , sanders et al . 1988 ; clements et al . 1996 ; murphy et al . 1996 ; scoville et al . 1999 ; surace et al . 1999 ) . these ulirgs are observed to show mixtures of two types of activity ; starburst and active galactic nuclei ( agn ) ( sanders et al 1988 ; sanders & mirabel 1996 ; veilleux , kim , & sanders 1999 ; zheng et al . it is still controversial whether dust - enshrouded agn or a nuclear dusty starburst is the primary energy source for the huge luminosity of ulirgs ( sanders & mirabel 1996 for a review ; genzel et al . 1998 ; lutz et al . 1998 ; smith , lonsdale , & lonsdale 1998 ) . all of these of ulirgs are rich in molecular gas with the total mass of @xmath20 ranging from @xmath21 @xmath22 to 3 @xmath23 @xmath24 @xmath22 ( sanders et al . 1988 ) , and a large fraction of this gas is in very dense regions near nuclei ( solomon et al . ulirgs appear to be about 2 times more numerous than optically selected qsos and the luminosity function of ulirgs is nearly similar to that of qsos ( soifer et al . these results give rise to the argument that quasar - like activity is also being initiated in major mergers and ulirgs can finally evolve into qsos ( sanders et al . 1988 ) . recent high - resolution optical / infrared imaging of ulirgs have revealed a number of very bright star - forming knots probably formed by merging processes and furthermore clarified diverse morphology reflecting their varying stages of galactic dynamical evolution ( scoville et al . 1999 ; surace et al . 1999 ) . these peculiar phenomena seen in ulirgs ( i.e. , ongoing major mergers ) have raised theoretical interests in the interstellar gas and star formation process in gas - rich major mergers with strong dusty starbursts . numerical simulations of galaxy mergers including gas dynamics indicate that a large amount of gas , as much as @xmath21 @xmath22 , can be funneled to the central kpc region of the remnant ( barnes & hernquist 1991 , 1992 ; olson & kwan 1990 ; noguchi 1991 ) . mihos et al . ( 1992 ) suggested that the gas accumulated in the central kpc region of galaxy mergers can be converted into massive starburst , though the magnitude of the starburst is not strong enough to explain that observed in ulirgs . by using a more sophisticated model of hydrodynamical evolution of galaxy mergers , mihos & hernquist ( 1996 ) demonstrated that the strength and the epoch of massive starburst depend on initial internal structure of merger progenitor disks rather than initial orbital configurations of the disks . bekki & noguchi ( 1994 ) and bekki ( 1995 ) demonstrated that the dynamical heating by two sinking cores and subsequent dissipative cloud - cloud collisions in a merger can drive a large fraction of disk interstellar gas into the central tens pc where nuclear starburst and agn activity are observed for ulirgs . gerritsen & icke ( 1999 ) investigated how thermal energy from supernovae ( i.e. , feedback of supernovae ) affects the dynamical evolution of interstellar gas and consequently controls the star formation rate in a major merger and found that the supernovae energy from the nuclear high density gas region ( @xmath3 1 kpc ) can prevent the star formation in the outer region of the merger . although these theoretical studies have been quite successful in clarifying the detailed processes of gaseous inflow to the central 10 @xmath3 1000 pc and in describing star formation history for gas - rich major galaxy mergers , they have not yet investigated at all how photometric and spectroscopic properties of merger evolve with time . therefore it remains highly uncertain how the photometric properties of ulirgs is determined by the detailed dynamical processes of major galaxy merging . the purpose of the present paper is to investigate numerically morphological , structural , and photometric evolution of gas - rich major mergers with dusty starburst in an explicitly self - consistent manner . by using a new numerical code for calculating the spectral energy distribution ( sed ) of a galaxy with dusty interstellar gas ( bekki & shioya 2000a ) , we particularly investigate the following five points : ( 1 ) the fundamental properties of seds of dusty starburst mergers with the infrared luminosity larger than @xmath7 @xmath5 , ( 2 ) time evolution of the relative spatial distribution of dusty gas and old and new stars , ( 3 ) the dependence of @xmath13 on stellar ages and radius from the center of mass of a merger , ( 4 ) time evolution of colors , absolute magnitude , surface brightness , dust temperature , and the degree of dust extinction ( @xmath13 ) , and ( 5 ) physical correlations between @xmath0 , central gas density , luminosity ratio ( e.g. , @xmath25 , where @xmath26 is the @xmath8 band luminosity ) , star formation efficiency , and morphology during starburst . the above points have not been investigated extensively by previous theoretical studies . accordingly , the present numerical results can provide new clues to the nature and the origin of ulirgs . the layout of this paper is as follows . in 2 , we summarize numerical models used in the present study and describe briefly the methods for deriving the seds corrected by internal dust extinction . in 3 , we present numerical results on the time evolution of morphology , sed , and photometric properties in a gas - rich major merger . in 4 , we discuss the nature of ulirgs , in particular , the importance of the relative distribution of gas and stars in determining photometric and spectroscopic properties of ulirgs . the conclusions of the preset study are given in 5 . since numerical methods and technique for solving chemodynamical and photometric evolution of major galaxy mergers with dusty starburst have been already given in detail by bekki & shioya ( 2000a ) , we briefly describe the present merger model . we construct models of galaxy mergers between gas - rich disk galaxies with equal mass by using the fall - efstathiou ( 1980 ) model . the total mass and the size of a progenitor disk are @xmath27 and @xmath28 , respectively . from now on , all the mass and length are measured in units of @xmath27 and @xmath28 , respectively , unless specified . velocity and time are measured in units of @xmath29 = @xmath30 and @xmath31 = @xmath32 , respectively , where @xmath33 is the gravitational constant and assumed to be 1.0 in the present study . if we adopt @xmath27 = 6.0 @xmath23 @xmath24 @xmath34 and @xmath28 = 17.5 kpc as a fiducial value , then @xmath29 = 1.21 @xmath23 @xmath35 km / s and @xmath31 = 1.41 @xmath23 @xmath36 yr , respectively . in the present model , the rotation curve becomes nearly flat at 0.35 radius with the maximum rotational velocity @xmath37 = 1.8 in our units . the corresponding total mass @xmath38 and halo mass @xmath39 are 5.0 and 4.0 in our units , respectively . the radial ( @xmath40 ) and vertical ( @xmath41 ) density profile of a disk are assumed to be proportional to @xmath42 with scale length @xmath43 = 0.2 and to @xmath44 with scale length @xmath45 = 0.04 in our units , respectively . the central bulge component is represented by a plummer sphere with the mass of 0.25 and the scale length of 0.04 in our units , which means that the mass ratio of bulge to disk is 0.25 and the scale length ratio of bulge to disk is 0.2 . the collisional and dissipative nature of the interstellar medium is modeled by the sticky particle method ( @xcite ) and the initial fraction of gas mass ( @xmath46 ) in a disk is set to be 0.2 . the number of particles for an above isolated galaxy is 10000 for dark halo , 3165 for stellar bulge , 10000 for stellar disk components , and 10000 for gaseous ones . in all of the simulations of mergers , the orbit of the two disks is set to be initially in the @xmath47 plane and the distance between the center of mass of the two disks ( @xmath48 ) is 140 kpc . the pericenter distance ( @xmath49 ) and the orbital eccentricity ( @xmath50 ) are assumed to be 17.5 kpc and 1 ( i.e. , parabolic ) , respectively . the spin of each galaxy in a merger is specified by two angle @xmath51 and @xmath52 , where suffix @xmath53 is used to identify each galaxy . @xmath51 is the angle between the @xmath54 axis and the vector of the angular momentum of a disk . @xmath52 is the azimuthal angle measured from @xmath55 axis to the projection of the angular momentum vector of a disk on to @xmath47 plane . in the present study , we show the results of only one model with @xmath56 = 30.0 , @xmath57 = @xmath58 , @xmath59 = 0.0 , @xmath60 = 0.0 : this model describes a nearly prograde - retrograde merger . the results of the models with variously different @xmath51 , @xmath52 , @xmath49 , @xmath50 , and @xmath48 will be described in our future papers ( bekki & shioya 2000c ) . the time when the progenitor disks merge completely and reach the dynamical equilibrium is less than 15.0 in our units for the present model ( @xmath3 2.12 gyr ) . star formation is modeled by converting the collisional gas particles into collisionless new stellar particles according to the schmidt law ( schmidt 1959 ) with exponent @xmath61 = 2.0 ( 1.0 @xmath62 @xmath61 @xmath62 2.0 , @xcite ) . chemical enrichment through star formation during galaxy merging is assumed to proceed both locally and instantaneously in the present study the fraction of gas returned to interstellar medium in each stellar particle and the chemical yield are 0.3 and 0.02 , respectively . initial metallicity @xmath63 for each stellar and gaseous particle in a given galactic radius @xmath40 ( kpc ) from the center of a disk is given according to the observed relation @xmath64 of typical late - type disk galaxies ( e.g. , zaritsky , kennicutt , & huchra 1994 ) . for calculating the sed that is not modified by dust extinction for a merger , we use the spectral library gissel96 which is the latest version of bruzual & charlot ( 1993 ) . for deriving the sed modified by dust extinction , we use our original ray tracing method that can be applied to n - body simulations . the details of this method and advantages and disadvantages in using this method are given in bekki & shioya ( 2000a ) . all the calculations related to the above dynamical evolution including the dissipative dynamics , star formation , and gravitational interaction between collisionless and collisional component have been carried out on the grape board ( @xcite ) at astronomical institute of tohoku university . the parameter of gravitational softening is set to be fixed at 0.03 in all the simulations . the time integration of the equation of motion is performed by using 2-order leap - flog method . energy and angular momentum are conserved within 1 percent accuracy in a test collisionless merger simulation . based on the above numerical model , we investigate mainly the sed and the photometric evolution of major mergers and thereby provide some implications on the origin of ulirgs . before describing our numerical results , we should point out that total infrared luminosity ( @xmath0 ) of a major merger depends strongly on the initial disk mass of the merger . accordingly , the @xmath0 of a major merger during starburst can exceed @xmath3 @xmath4 @xmath5 that is a defined value for ulirgs ( sanders & mirabel 1996 ) , only if the disk mass of the merger exceeds a certain value ( for a given orbital configuration and gas mass fraction in major merging ) . a possible dependence of @xmath0 on the initial disk mass of a major merger is described in detail in appendix a. in the present model with the initial disk mass of 6.0 @xmath23 @xmath24 @xmath34 , the maximum @xmath0 is @xmath3 5.9 @xmath23 @xmath7 @xmath5 , which is appreciably smaller than the defined value of ulirgs . accordingly , the present merger model with structural and kinematical properties nearly the same as those of the milkey way does not show the evolution of ulirg in a strict sense . we however consider that it is not unreasonable for us to discuss extensively the origin and the nature of ulirgs by using the present model . this is because some changes of merger parameters ( e.g. , gas mass fraction , initial disk mass , and orbital configurations of mergers ) can increase the @xmath0 a factor of 2 and because the defined value of @xmath3 @xmath4 @xmath5 itself does not probably have essentially important physical meanings . bekki & shioya ( 2000a ) have already demonstrated that if the gas mass fraction of initial disks of a merger is rather large ( 0.5 , which is an unrealistic value for the present disk galaxies but not so an unrealistic one for the high redshift dusty starburst galaxies such as faint submillimeter sources discovered by smail et al . 1997 ) , the maximum infrared luminosity exceeds @xmath4 @xmath5 . in the present study , we described the model with @xmath65 = 0.2 and a nearly prograde - retrograde orbital configuration as a fiducial one showing a typical behavior of dusty major mergers that can form ulirgs . the essentially important results on photometric evolution of dusty starburst mergers do not depend so strongly on orbital configurations and internal structure in mergers , though the epoch and the strength of starbursts depend on these parameters . the dependence of photometric evolution on merger parameters will be described in our future papers ( bekki & shioya 2000c ) . figures 1 , 2 , 3 , and 4 describe the time evolution of global morphology of the present merger model for each of three components , bulge , old disk , and gas disk ( and new stars ) . > from now on , for convenience , the time @xmath66 represents the time that has elapsed since the two disks begin to merge . seven fundamental properties of morphological evolution of the less inclined and nearly prograde - retrograde merger with bulges are described as follows . firstly , a strong stellar bar is not developed even in the disk orbiting in a prograde sense for the first encounter of the merger ( 0.56 @xmath67 @xmath66 @xmath67 0.85 gyr ) . this is because the central massive bulges of the two disks can prevent the growth of non - axisymmetric perturbation ( triggered by tidal interaction of the merger ) . secondly , as galaxy merging proceeds , the two disks are strongly disturbed to form a long tidal tail in the disk orbiting in a prograde sense at 0.56 @xmath67 @xmath66 @xmath67 0.71 gyr . a strong tidal tail is not developed in the disk orbiting in a retrograde sense . this one long tidal arm is characteristics of prograde - retrograde mergers . thirdly , the size of one disk orbiting in a prograde sense becomes smaller owing to the efficient tidal stripping of disk components whereas that of the other disk orbiting in a retrograde sense does not change so greatly for 1.12 @xmath67 @xmath66 @xmath67 1.48 gyr . as a natural result of this , the major merger @xmath68 be observationally identified as a minor merger with the mass ratio of the two disks rather small ( @xmath69 @xmath67 0.1 ) or as unequal - mass one ( @xmath69 @xmath3 0.3 ) during violent interaction of the major merger ( corresponding to 1.12 @xmath67 @xmath66 @xmath67 1.48 gyr in the present study ) . fourthly , the tidal tail formed in the first encounter becomes less discernible when the two disks finally sink into the center of massive dark halos owing to dynamical friction during merging and consequently are completely destroyed by violent relaxation of galaxy merging ( 1.41 @xmath67 @xmath66 @xmath67 1.83 gyr ) . fifthly , an elliptical galaxy formed in the present very dissipative and less inclined merger model has a rather flattened shape . sixthly , old stellar components initially located within disks and bulges are more diffusely distributed than new stars in the merger remnant , primarily because new stars that were previously gaseous components experience much more amount of gaseous dissipation during merging . seventhly , these results are broadly consistent with those of previous dissipative merger models such as mihos & hernquist ( 1996 ) . during violent major merging , interstellar gas is very efficiently transferred to the central region of the two disks owing to gaseous dissipation of colliding gas clouds and gravitational torque . gas accumulated in the central region of the merger is then consumed by massive starburst and consequently converted into new stars . figure 5 describes the total gas mass accumulated within the central 100 , 200 , and 500 pc for each of the merging disks . during the first tidal encounter of the merger ( 0.56 @xmath67 @xmath66 @xmath67 1.12 gyr ) , tidal force transforms the disk orbiting in a prograde sense more dramatically and furthermore changes the mass distribution of the disk . consequently angular momentum redistribution of gas and stars proceeds in the disk and some fraction of gas can be accumulated into the central region of the disk ( and some fraction of gas is tidally stripped away from the disk as a natural result of angular momentum redistribution ) . on the other hand , the disk orbiting in a retrograde sense does not suffer so severely from such tidal effects , and accordingly the mass distribution is not greatly changed for 0.56 @xmath67 @xmath66 @xmath67 1.12 gyr . thus , as is shown in figure 5 , total gas mass accumulated in the central region is larger for the disk orbiting in a prograde sense for 0.56 @xmath67 @xmath66 @xmath67 1.12 gyr . for the final violent relaxation phase ( 1.41 @xmath67 @xmath66 @xmath67 1.69 gyr ) , the total gas mass is not so largely different and the gas mass is appreciably larger for the disk orbiting in a retrograde sense ( in particular , for the central 100 pc region ) . these results imply that the difference in the central gas mass between two disks in a merger can be more clearly observed in the merger early phase when two cores are still well separated ( i.e. , the epoch of early tidal encounter ) . as is shown in figure 6 , owing to the large central bulge , strong starbursts can occur only in the late phases of the merger when two disks are completely destroyed by violent relaxation to form the central high density core of an elliptical galaxy . the star formation rate becomes the maximum value of @xmath3 120 @xmath22 @xmath70 at @xmath66 = 1.53 gyr . the time scale within which the star formation rate exceeds 100 @xmath22 @xmath70 is rather short ( less than 0.1 gyr ) , which implies that the time scale for a gas - rich merger to become an ulirg is an order of 0.1 gyr . absorption of stellar light for stars in a galaxy can be modeled according to the following reddening formulation ( black 1987 ; mazzei et al . 1992 ) ; @xmath71 where @xmath72 and @xmath73 are gaseous column density and gaseous metallicity , respectively . if we follow this reddening formulation , stellar light from a star is more strongly affected by dusty gas for the higher density and the larger metallicity of gas surrounding the star . accordingly estimating gas density and metallicity around each of stars in a galaxy is very important for deriving the total amount of dust extinction and thus the sed modified by dust extinction . thus we here show the time evolution of gaseous density and metallicity around stars before describing the time evolution of seds of the present merger model . figure 7 gives the time evolution of the mean value for gaseous density , metallicity , and the degree of dust extinction around new stars in the present merger model . in deriving the mean value for each time @xmath66 in this figure , we first find gaseous particles located within 0.02 ( in our units ; corresponding to 350 pc ) of each stellar particle and thereby estimate gaseous density and metallicity around each stellar particle . next we calculate the product of gaseous density and metallicity for each stellar particle ( this product is proportional to e(b - v ) in the above reddening formulation ) . for convenience we call this product as the degree of extinction in explaining figure 7 , 8 , 9 , and 10 . based on the above three values of each particle , we calculate the mean of these values at each time @xmath66 both for old stars initially located in disks and bulges and for new ones formed from gas . as is shown in figure 7 , both the mean gaseous density and the mean gaseous metallicity around stars are larger in new stars than in old ones . as a natural result of this , the mean of the degree of extinction is considerably larger in new stars than in old ones during galaxy merging . the reason for these dependences is essentially that new stars are formed preferentially in the merger central region with very high density gas and are more compactly distributed within the central cores of the merger compared with old stars . these results accordingly suggest that seds from new stars are more dramatically modified by dust than those from old stars during dynamical evolution of the merger . furthermore the difference in the mean of the degree of dust extinction between new stars and old ones becomes larger during massive starburst ( 1.48 @xmath67 @xmath66 @xmath67 1.62 gyr ) , mainly because a large amount of gas is the most efficiently transferred to the nuclear regions where young stars are located . this result implies that since most of stellar light comes from young stellar components during starburst , the merger shows the largest degree of dust extinction during its starburst . figure 8 shows the age of each star and the gaseous metallicity around the star for all new stars at @xmath66 = 1.53 gyr corresponding to the epoch when the merger show the maximum star formation rate of @xmath74 @xmath22 @xmath75 . for a given stellar age , gaseous density , gaseous metallicity , the degree of extinction around stars are all very widely distributed , which reflect the fact that both chemical evolution and local density enhancement of gas proceed in a very inhomogeneous way during starburst of the merger . the number excess of stars around the age of 0 gyr in figure 8 is due mainly to the massive starburst around @xmath66 = 1.53 gyr . furthermore , a larger number of new stars are located in the region with larger dust extinction ( the upper parts in figure 8) for the age of @xmath3 0 gyr . figure 9 shows the radial dependence of gaseous density , gaseous metallicity , mean age of new stars ( estimated for each radial bin ) , and the mass fraction of new stars to all stars ( old stars and new ones ) at @xmath66 = 1.53 gyr ( the maximum starburst epoch ) . as is shown in this figure , both gaseous density and gaseous metallicity are likely to be higher and mean age of new stars is likely to be younger in the inner region of the merger . this suggests that stars with younger ages can be more heavily obscured by dust than those with older ages : the dust extinction is very selective in the sense that only very young components can be heavily obscured by dust . the selective dust extinction processes are described in detail in later sections . the fraction of new stars is likely to be larger in the inner region of the merger , which implies that the sed of the inner region of the merger is more strongly affected by dust extinction . figure 10 and 11 show clearly that the dust extinction can be very selective during starburst in the merger . figure 10 describes the dependence of mean gaseous density , mean gaseous metallicity , and mean of the degree of extinction around new stars on the mean age of new stars ( with ages smaller than 1.2 gyr ) at @xmath66 = 1.48 ( the beginning of starburst ) , 1.53 ( the maximum starburst ) , 1.55 , 1.62 , and 1.83 gyr ( the end of starburst ) . the top panel of figure 10 demonstrates that irrespectively of time @xmath66 , the mean gaseous density around new stars is likely to be larger for stars with younger ages and furthermore that the age dependence is more clearly seen for the epoch of stronger starburst ( @xmath66 = 1.53 and 1.55 gyr ) . for example , mean gas density for new stars with the ages ranging from 0 gyr to 0.06 gyr is about a factor of 5.5 times larger than that for new stars with the ages ranging from 1 gyr to 1.06 gyr at @xmath66 = 1.53 gyr . furthermore the middle panel of figure 10 indicates that although the mean gaseous metallicity around new stars is likely to be larger for new stars with younger ages , the age dependence is more clearly seen for the end of starburst ( @xmath66 = 1.83 gyr ) . the reason for the clearer age dependence of gaseous metallicity around new stars at @xmath66 = 1.83 gyr is that chemical evolution in the central region where young stellar components are located proceeds to a larger degree in the later phase of galaxy merging . as a result of the above age dependences , the mean of the degree of dust extinction is likely to be larger for new stars with younger ages at all time @xmath66 ( 1.48 , 1.53 , 1.55 , 1.62 , and 1.83 gyr ) and the age dependence of the degree of extinction is clearer for the epoch of stronger starburst ( see the bottom panel in figure 10 ) . for example , the mean degree of dust extinction for new stars with the ages ranging from 0 gyr to 0.06 gyr is about a factor of 4.8 times larger than that for new stars with the ages ranging from 1 gyr to 1.06 gyr at @xmath66 = 1.53 gyr . the age - dependent dust extinction derived in figure 10 can be seen also in figure 11 for very young stars with the ages less than 0.12 gyr . these results in figure 10 and 11 thus demonstrate that dust extinction for young starburst components in major gas - rich mergers is very selective in the sense that stars with younger ages are more heavily obscured by dusty gas . this selective extinction is suggested to be crucially important for explaining spectroscopic properties of dusty starburst galaxies ( poggianti & wu 1999 ; shioya & bekki 2000 ) . we discuss this importance of selective dust extinction in the section of discussion ( 4.2 ) . photometric evolution of dusty starburst galaxies is basically determined by the seds , and the seds furthermore depend on the age - metallicity distribution of stellar populations , the dust temperature ( represented by @xmath11 from now on ) , and the degree of dust extinction ( @xmath13 for extinction in @xmath76 band ) . therefore the time evolution of @xmath11 and @xmath13 is important for photometric evolution of dusty starburst galaxies , in particular , in infrared bands . in the present model , @xmath11 is calculated for each of gaseous particles and @xmath13 can be estimated for each of the stellar ones ( bekki & shioya 2000a ) . the sed of a galaxy thus depends strongly on the detailed distribution of @xmath11 and @xmath13 . figure 12 shows the location of each of stellar particle ( with different ages ) on an age-@xmath13 plane at @xmath66 = 1.53 gyr . as is shown in this figure , the value of @xmath13 is rather diverse ; ranging from @xmath3 0.0 mag to @xmath3 300 mag for stars with ages larger than 5.0 @xmath23 @xmath77 yr . this result reflects the fact that gaseous density and metallicity around stars depend on the location of each star ( i.e. , whether a star is in the central high density region or in the outer low density one ) and are thus variously different between stars . for stars with the ages less than 5.0 @xmath23 @xmath77 yr , there is a tendency that younger stars have larger @xmath13 . this result thus confirms the selective dust extinction that is already described before in this paper . these young starburst populations are mostly located in the central high density gaseous region and @xmath13 of a new star is controlled greatly by the surrounding gaseous density . a new star with younger age ( @xmath67 5.0 @xmath23 @xmath77 yr ) , which is located in the rather inner part of the central starburst region , is more heavily obscured by high density gas in the central starburst region and consequently the @xmath13 of the star is larger . figure 13 shows the location of each of gaseous particles on @xmath78 plane , where @xmath41 is gaseous metallicity . we can discernibly observe a weak tendency that a more metal - enriched gas has a larger temperature ( in particular , for the metallicity range of @xmath41 @xmath67 0.02 ) , though the dispersion in @xmath11 distribution is rather large for a given @xmath41 . the main reason for this weak dependence is that a more metal - enriched gaseous particle is likely to be located in the central region where young massive stars emit a larger amount of stellar light that can be absorbed by dusty gas . the age dependence of @xmath13 and the metallicity dependence of @xmath11 derived in figure 12 and 13 can be seen at other @xmath66 during starburst ( bekki & shioya 2000c ) , which implies that such dependences are essential ingredients of dusty starburst major mergers . these dependences are furthermore important for ( mean ) @xmath13 and @xmath11 of the merger at each time @xmath66 . figure 14 shows the time evolution of @xmath13 and @xmath11 in the merger . both @xmath13 and @xmath11 gradually increase with time and become maximum at the epoch of maximum starburst ( @xmath66 = 1.53 gyr ) . after starburst , @xmath13 rapidly decreases with time , mainly because gaseous density around new stars becomes considerably smaller owing to rapid consumption of gas by starburst ( see also figure 7 ) . @xmath11 also becomes low after starburst because most of young massive stars emitting a large amount of stellar light ( that can be absorbed by dusty gas ) died out . the time evolution of the sed and the shape of the sed in the present merger model depends strongly on the time evolution of @xmath13 and @xmath11 . figure 15 describes the seds of the merger which we derive based on the mass distribution of stellar and gaseous component and the age and metallicity distribution of stellar populations of the merger at each time @xmath66 . the infrared and submillimeter luminosities become larger during 1.41 @xmath67 @xmath66 @xmath67 1.53 gyr in the present merger . this is firstly because star formation rate , which is closely associated with the total amount of stellar light absorbed by interstellar dust , becomes considerable higher owing to the efficient gas transfer to the central region of the merger and secondly because the density of dusty gas becomes also very high in the later phase of the merging ( 1.41 @xmath67 @xmath66 @xmath67 1.53 gyr ) so that the gas can heavily obscure the strong starburst . after the maximum starburst at @xmath66 = 1.53 gyr , both infrared and submillimeter luminosities rapidly decline ( 1.53 @xmath67 @xmath66 @xmath67 1.83 gyr ) this is principally because most of interstellar gas indispensable for strong starburst and dust obscuration is rapidly consumed by star formation in the merger till @xmath66 @xmath3 1.83 gyr . thus the time evolution of sed of a merger depends strongly on that of the star formation rate , which is basically controlled by the dynamical evolution of the merger . figure 16 describes the sed of the merger for old stars , new ones , and gas at the maximum starburst ( @xmath66 = 1.53 gyr ) in models with and without dust extinction . one of advantages of the present numerical model is that we can investigate the sed of each of stellar particles and the dust reemission from each of gaseous one . although stellar light from new stars dominates the sed in @xmath79 band ( @xmath80 0.3 @xmath81 ) for the model without dust extinction , the difference in the @xmath79 flux between the two stellar populations is appreciably smaller for the model with dust extinction . this is mainly because new stars are located in the central region , where there is a large amount of high density gas , and consequently are more heavily obscured by dust than old ones ( see also figure 8) . on the other hand , the contribution of old stars to the sed in optical and near - infrared bands is appreciably larger compared with that of new stars in the model with dust extinction , though the shape of the sed at optical and near - infrared bands is not significantly different between new stars and old ones in the model without dust extinction . this result suggests that photometric properties of dusty starburst mergers ( i.e. , luminous infrared galaxies , lirgs and ulirgs ) in optical and near - infrared band are determined largely by old stellar populations ( old stars ) owing to selective dust extinction of new stars . figure 17 describes the time evolution of absolute magnitude from ultraviolet to near - infrared wavelength ( in @xmath82 , @xmath8 , @xmath76 , @xmath40 , @xmath83 , and @xmath84 band ) for the models with and without dust extinction . firstly we summarize the evolution of the model without dust extinction . owing to the aging of old stellar populations , the absolute magnitude of the merger in most of bands becomes fainter as galaxy merging proceeds . the absolute magnitude in @xmath82 band is very sensitive to star formation rate and therefore appreciably brightens at @xmath66 @xmath3 0.85 gyr when the two disks in the merger first encounter and consequently weak starburst occurs in the disk rotating in a prograde sense . during strong starburst ( 1.4 @xmath67 @xmath66 @xmath67 1.6 gyr ) , the absolute magnitude of the merger rapidly brightens with time because of increased stellar light from young massive stars . furthermore , the increase of luminosity depends on wavelength such that the magnitude difference between the epoch of prestarburst ( @xmath66 @xmath67 1.4 gyr ) and that of starburst is larger for shorter wavelength ( e.g. , @xmath3 1.7 mag for @xmath82 band and @xmath3 0.5 mag for @xmath84 one ) . the @xmath84 band absolute magnitude during starburst becomes brighter than @xmath851 that is the mean value of ulirgs ( sanders & mirabel 1996 ) . the absolute magnitude of the merger rapidly becomes faint with time after starburst because of the death of young massive stars . the difference in absolute magnitude between starburst and poststarburst is also larger for shorter wavelength . next we summarize the evolution of the model with dust extinction . as is shown in figure 17 , the time evolution of the absolute magnitude of the merger ( 0 @xmath80 @xmath66 @xmath80 2.3 gyr ) becomes rather moderate compared with the model without dust extinction for all bands , though the time evolution is qualitatively similar to that of the model without dust extinction . this is principally because young massive stars , which are important factor for the rapid increase and decrease in photometric properties of major mergers , are heavily obscured by dusty gas . the difference in absolute magnitude between the two models with and without dust extinction is larger for shorter wavelength during starburst , which reflect the fact that stellar light with shorter wavelength can be more greatly absorbed by dust . furthermore , irrespectively of wavelength , the difference in absolute magnitude between the two models is likely to be the largest at the epoch of maximum starburst ( @xmath66 = 1.53 gyr ) . this is because young and very luminous components formed by starburst are the most heavily obscured by dust at the maximum starburst epoch when a large amount of dusty gas is transferred to the central region to form very high density region around the compact starburst . the @xmath84 band magnitude at @xmath66 = 1.53 gyr is about @xmath86 mag that is about 0.9 mag fainter ( less luminous ) than the mean @xmath84 band magnitude of ulirgs ( sanders & mirabel 1996 ) . this is either because the initial total mass of the host disk of a typical ulirg is about 2 times larger than that of the galaxy adopted as a merger progenitor disk in the preset study or because the present model underestimates the star formation rate of the merger ( e.g. , owing to the smaller initial gas mass fraction ) . figure 18 describes the time evolution of global colors from optical to near - infrared wavelength ( in @xmath87 , @xmath88 , @xmath89 and @xmath14 ) for the models with and without dust extinction . by comparing the results of the model with dust extinction with those of the model without dust extinction , we can clearly observe that the effect of dust on color evolution is basically similar to that on the above evolution of absolute magnitude : the time variation of colors during galaxy merging becomes rather moderate owing to dust extinction . to be more specific , although the @xmath87 color becomes @xmath3 0.4 mag bluer in the maximum starburst ( @xmath66 = 1.53 gyr ) than in the prestarburst ( @xmath66 = 1.4 gyr ) for the model without dust extinction , the color difference between the two epochs is only 0.04 mag for the model with dust extinction . this result implies that a major merger does not necessarily show very blue colors when it becomes a lirg ( or an ulirg ) owing to very strong dust extinction . we here point out that the color evolution derived above is based only on the present model with a certain orbital configuration . we accordingly suggest that although the derived essential dust effects on color evolution could be true for lirgs and ulirgs , the @xmath90 @xmath91 of the colors of lirgs and ulirgs can depend on orbital configurations and internal structure of galaxy mergers . bekki & shioya ( 2000b ) found that a nearly retrograde - retrograde merger suffers the more remarkable dust extinction of stellar light than a prograde - prograde merger . since the colors of mergers depend also on the degree of dust extinction , lirgs and ulirgs formed by mergers with variously different orbital configurations and internal structure probably show different colors . figure 19 shows the time evolution of mid - infrared and far - infrared fluxes at 25 @xmath2 m , 60 @xmath2 m , and 100 @xmath2 m of the merger , that of the ratio of 25 @xmath2 m flux to 60 @xmath2 m one ( @xmath92 ) , and that of the ratio of 60 @xmath2 m flux to 100 @xmath2 m one ( @xmath12 ) . the present model does not include the effects of small grains on seds and accordingly can not so precisely estimate the infrared flux around the order of 10 @xmath2 m . here we thus point out that although the following evolution of the fluxes could be qualitatively similar to the evolution observed in lirgs and ulirgs , the absolute values of the fluxes can be largely different from observational ones . the 25 @xmath2 m , 60 @xmath2 m , and 100 @xmath2 m fluxes are rather sensitive to star formation rate and dust temperature and therefore depend strongly on star formation histories and the degree of central gas concentration in major mergers . the first remarkable increase of the fluxes ( in particular , for 25 @xmath2 m and 60 @xmath2 m ) corresponds to the epoch of the first encounter of two disks of the merger ( @xmath66 = 0.85 gyr ) . the second and the largest increase at @xmath66 = 1.53 gyr is due to the maximum starburst heavily obscured by dusty gas . the increase of these fluxes is associated closely with the increase of dust temperature and the fraction of young stars obscured by dust . as is shown in figure 19 , the flux ratio of @xmath12 becomes considerably larger ( @xmath3 0.4 ) during starburst compared with the initial value ( @xmath3 0.04 ) . this result implies that mergers in the merger late phase when two cores become very close ( less than a few kpc ) are more likely to show larger @xmath12 . the time evolution of the flux ratio of @xmath92 is rather moderate and furthermore the values of @xmath92 is less than 0.02 during merging . the derived value is much smaller than that of the so - called warm ulirgs which are observationally suggested to have agns ( e.g. , sanders & mirabel 1996 ) . thus our present study , which does not include the effects of small grains though , confirms that dusty starburst alone can not contribute to the large flux ratio of @xmath92 observed in warm ulirgs . figure 20 gives the time evolution of @xmath0 and @xmath10 in the present merger model . both @xmath0 and @xmath10 become considerably larger during strong starburst , because a larger number of young stars are created and then obscured by dusty gas during starburst . although the change of @xmath8 band magnitude is at most 0.3 mag during starburst ( see figure 17 ) , that of @xmath0 is @xmath3 an order of magnitude . this result indicates that major mergers with similar luminosities can show very different @xmath0 and furthermore that @xmath0 reflects the dynamical stages of major galaxy merging . furthermore the @xmath10 during starburst becomes larger than 10 that is similar to the order of @xmath10 of lirgs and ulirgs . this result confirms that the rather large luminosity ratio of @xmath10 observed in lirgs and ulirgs are associated closely with major galaxy merging . considering the results shown in figure 19 and 20 , we can say that a merger with higher @xmath10 shows warmer @xmath12 and also that @xmath12 increases with the increase of @xmath0 . these derived correlations are consistent reasonably well with the observational ones ( e.g. , sandered & mirabel 1996 ) . figure 21 shows the two - dimensional distribution ( projected onto the @xmath55-@xmath93 plane ) of @xmath84 band surface brightness , @xmath14 color , 60 @xmath2 m flux , and @xmath13 at the epoch of maximum starburst ( @xmath66 = 1.53 gyr ) . the @xmath84 band surface brightness is higher in the inner region of the merger and shows peculiar double - peak morphology in the central region . the main reason for this central peculiar morphology is that the merger at this epoch has non - axisymmetric structure ( composed both old stars and new ones ) formed by dynamical relaxation of major merging . the difference in @xmath84 band surface brightness between the central region ( @xmath3 2 kpc ) and the outer one ( @xmath3 8kpc ) is about 7 mag @xmath94 at this epoch . the distribution of @xmath14 color is more complicated and more strongly disturbed than that of @xmath84 band surface brightness , which implies that the signature of major galaxy merging can be more clearly observed in the @xmath14 color distribution than the @xmath84 band surface brightness distribution . the reason for this peculiar morphology of color distribution is that stellar light in @xmath40 band ( rather than in @xmath84 band ) is strongly affected by dusty gas that is distributed in a very irregular way owing to dynamical perturbation of the merger . the merger at the very luminous infrared phase shows the redder color in the central region than in the outer region , which means that ulirgs and lirgs are more likely to have negative color gradients . the reasons for the derived negative gradient are ( 1 ) that the central starburst components are rather heavily obscured by dust and ( 2 ) that initially the central stellar components are more metal - rich owing to the bulge components . the absolute magnitude of the negative @xmath14 color gradient is estimated to be 0.08 mag / kpc for the central 8.5 kpc of the merger at this epoch . as is shown in figure 21 , the infrared flux at 60 @xmath2 m is distributed rather preferentially to the central region of the merger . the peak of the 60 @xmath2 m flux coincides both with the peak of the @xmath13 distribution and with the ( lower ) peak of the @xmath84 band surface brightness one . this result means that the infrared emission from dusty gas comes exclusively from the central compact high density gaseous regions where very young starburst components are located . the coincidence of the two peaks ( in @xmath84 band surface brightness and @xmath13 ) is furthermore associated with the origin of peculiar morphology of @xmath84 band surface brightness distribution . the distribution of @xmath13 is also very compact and shows the rather large negative radial gradient ( @xmath3 3.0 mag / kpc for the central 2 kpc ) . figure 21 furthermore shows that the @xmath14 color distribution is more similar to the @xmath13 one than the @xmath84 band surface brightness one . this implies that the two - dimensional distributions of dust extinction in luminous infrared galaxies formed by major mergers are important not only for the infrared luminosity distribution but also for the color distribution . lirgs and ulirgs are observed to show a number of fundamental correlations between physical properties such as @xmath0 , @xmath10 , infrared colors ( e.g. , @xmath95 ) , star formation efficiency estimated from the ratio of @xmath0 to @xmath96 molecular gas , and core separation of two merging disks ( sanders & mirabel 1996 ; gao & solomon 1999 ; young 1999 ) . for example , gao & solomon ( 1999 ) found a correlation between the co ( 1 - 0 ) luminosity and the projected separation of merger nuclei in a sample of 50 lirg mergers . such observed correlations are considered to contain valuable information on the formation and the evolution of luminous infrared galaxies ( sanders & mirabel 1996 ) and thus should be explained by any theories of galaxy evolution and formation . figure 22 gives correlations between the separation of two merging disks ( cores ) and five physical properties of the merger at 15 time steps ( @xmath66 = 0 , 0.28 , 0.56 , 0.71 , 0.85 , 0.99 , 1.12 , 1.41 , 1.48 , 1.53 , 1.55 , 1.62 , 1.69 , 1.83 , and 2.26 gyr ) . in order to estimate of the core separation at each time step , we initially place a stellar ( collisionless ) particle in the center of mass of each disk and thereby investigate the orbit of each particle . by measuring the separation of the two particles at each time step , we estimate the core separation ( @xmath40 ) of the merger . as is shown in figure 22 , total gas mass within the central 500 pc is more likely to be smaller for the smaller core separation . the decrease of gas mass with the increase of the core separation for 0.0 @xmath67 @xmath97 @xmath67 0.5 in our units ( corresponding to 0.56 @xmath67 @xmath66 @xmath67 0.99 gyr ) reflects the fact that the two disks separate from with each other after the first encounter of the disks and before final coalescence . the reason of the decrease of gas mass with the decrease of core separation ( in particular for 0.5 @xmath67 @xmath97 @xmath67 @xmath98 in figure 22 ) is simply that as the major merging proceeds , a larger amount of gas within the central 500 pc is consumed by star formation . gao & solomon ( 1999 ) found a correlation between the co ( 1 - 0 ) luminosity corresponding to a measure of molecular mass @xmath99 and the projected separation of merger nuclei ( the indicator of merging stages ) in a sample of 50 lirg mergers . since most of interstellar gas with rather high gaseous density is observationally demonstrated to become @xmath96 gas , we consider that it is reasonable to compare the present results on gaseous distribution with the observational ones on @xmath96 gaseous distribution . thus , although the present study does not investigate the time evolution of @xmath99 gas mass , it is not unreasonable for us to say the present numerical model at least quantitatively reproduces the observed correlation . this result in figure 22 implies that the observed correlation can be understood in terms of dynamical evolution of major merging . the star formation efficiency estimated from the ratio of star formation rate ( @xmath100 ) to total gas mass within the central 500 pc is larger for the smaller core separation for @xmath101 @xmath6 @xmath102 ( corresponding to @xmath66 @xmath67 1.53 gyr , i.e. , before the epoch of maximum starburst ) . the reason of this derived dependence is that owing to the adopted density - dependent star formation rate ( i.e. , the schmidt law ) , the star formation efficiency becomes larger when the two cores approach with each other and the degree of the central gas accumulation becomes larger . after the maximum starburst , the star formation efficiency rapidly declines with time and thus with the core separation ( @xmath66 @xmath6 1.53 gyr ) , which means that there is no remarkable positive correlation between core separation and star formation efficiency for mergers after massive starburst . gao & solomon ( 1999 ) also found an anti - correlation between @xmath103 corresponding to a global measure of the star formation rate per unit gas mass and the projected separation of merging two cores in lirgs . our results successfully reproduce this observed dependence , in particular , for the strong starburst epoch that can be interpreted as a lirg phase . we suggest that the observed correlation can be also understood in terms of gas accumulation processes of major galaxy mergers . here , we again point out that the gas mass in our simulations is not for @xmath96 molecular mass estimated from co luminosity but for gas mass ( i.e. , mass for all gas components ) . thus more quantitative comparison of our theoretical results with observations ( e.g. , those by gao & solomon 1999 ) requires high - resolution numerical simulations that can derive the time evolution of @xmath96 mass ( e.g. , transformation from hi gas to @xmath96 one ) and that of spatial distribution of @xmath96 gas . we plan a future study to include the formation of molecular gas ( @xmath96 ) from atomic one ( hi ) in numerical simulations and thereby to investigate time evolution of @xmath96 mass and spatial distribution of @xmath96 . the luminosity ratio of @xmath10 , the flux ratio of @xmath12 , and the degree of dust extinction ( @xmath13 ) are likely to be larger for the smaller core separation . the essential reason for the dependence of @xmath10 is that stellar light in @xmath8 band of young stellar populations is more heavily absorbed by dusty gas and reemitted in the infrared band and consequently becomes smaller for the smaller core separation . the main reason for the dependence of @xmath12 is that the dust temperature is likely to increase with the decrease of core separation . the reason for the dependence of @xmath13 is that the degree of the central gaseous accumulation becomes larger as major merging ( and dynamical relaxation ) proceeds . we accordingly suggest that @xmath10 , @xmath12 , and @xmath13 observed in lirgs and ulirgs clearly depend on to what degree galaxy mergers are dynamically relaxed . we also suggest that the diversity of @xmath10 observed in lirgs and ulirgs ( e.g. , sanders & mirabel 1996 ) can be understood in terms of the difference in merging stages ( or in the degree of dynamical relaxation ) . figure 22 furthermore indicates that although the total mass of stellar components in the merger does not increase significantly during galaxy merging ( only a factor of 0.15 increase ) , there is an order of magnitude difference in @xmath10 between the merger at different dynamical stages . such large difference can be seen also in @xmath12 , though the @xmath104 @xmath105 does not change so greatly with time and core separation . these results are only true for the present merger model with a certain orbital configuration of galaxy merging . accordingly , the diversity in initial physical conditions of galaxy merging ( e.g. , gas mass fraction and orbital configurations ) could furthermore cause considerably greater difference in @xmath10 and @xmath12 . thus we suggest that @xmath10 and @xmath95 are rather diverse ( more than an order of magnitude difference ) in galaxy mergers with variously different merger stages . figure 23 shows the location of the present merger model on the @xmath0 @xmath106 @xmath10 plane , the @xmath0 @xmath106 @xmath12 one , and @xmath12 @xmath106 @xmath107 one for @xmath66 = 1.48 , 1.53 , 1.55 , 1.62 , and 1.69 gyr when the merger shows the infrared luminosity larger than @xmath7 @xmath5 . as is shown in this figure , @xmath10 and @xmath12 are likely to be larger for larger @xmath0 . the main reason for the @xmath0 dependence of @xmath10 is that both @xmath0 and the @xmath10 increase with the increase of star formation rate . the main reason for the @xmath0 dependence of @xmath12 is that both @xmath0 and the dust temperature ( an important determinant the evolution of @xmath12 ) increase with the increase of star formation rate . the @xmath0 dependences derived in figure 23 are qualitatively consistent with observational results ( e.g. , sanders & mirabel 1996 ) . figure 23 furthermore demonstrates that the star formation efficiency is likely to be larger for the larger @xmath95 and thus that the present merger model can reproduce qualitatively the observed correlation between @xmath95 and star formation efficiency ( e.g. , young 1999 ) . it is generally considered that investigating theoretically the time evolution of dusty starburst galaxies is important for better understanding the nature and the origin of ulirgs . it is particularly important to investigate spectral energy distributions ( seds ) of dusty starburst galaxies , primarily because galactic seds are considered to be one of essentially important factors that can determine the nature of dusty starburst galaxies . the following previous theoretical studies have succeeded in clarifying important dependences of seds on physical parameters such as spatial distribution of dust , geometries of stellar and gaseous components , and dust properties in dusty starburst galaxies . witt et al . ( 1991 ) investigated transfer processes of stellar light for models with variously different spherical geometries of dust with special emphasis on the effects of scattered light on photometric properties of galaxies and seds . witt & gordon ( 1996 ) investigated the radiative transfer processes of a central stellar source surrounded by a spherical , statistically homogeneous but clumpy two - phase scattering medium and found that the structure of dusty medium can greatly affect the conversion of uv , optical , and near - infrared radiation into thermal far - ir dust radiation in a dusty system . by comparing the observed seds of 30 starburst galaxies with theoretical radiative transfer models of dusty systems , gordon , calzetti , & witt ( 1997 ) discussed the importance of geometry of stellar and gaseous components in determining the sed of a dusty starburst galaxy . we here suggest that among the important factors for galactic seds , the relative spatial distribution of gas and stars is a particularly important determinant for the evolution of seds and thus for the photometric evolution in dusty starburst galaxies . in particular , we stress that the relative spatial distribution of dusty interstellar gas and young stars ( i.e. , starburst populations ) is critically important , because most of stellar light during starburst comes from young starburst components in galaxies . as is demonstrated in the present study , photometric evolution such as the time evolution of @xmath14 color , @xmath0 , @xmath108 and @xmath10 , clearly reflects the time evolution of the spatial distribution of dusty gas and young stars . the difference in the relative distribution of gas and young stars is also suggested to explain the diversity in seds and @xmath10 observed in ulirgs ( bekki & shioya 2000b , c ) . then can we observationally confirm the suggested importance of the relative distribution of gas and young stars in photometric evolution of ulirgs ? recent high - resolution observational studies on hi and @xmath96 gaseous distribution have already revealed the detailed spatial distribution of interstellar gas ( e.g. , hibbard & yun 1999 ) . accordingly it will be possible for us to answer the above question if ongoing and future observational studies can derive the detailed spatial distribution of star - forming regions and thus that of young stellar populations for a starburst galaxy consisting both of old and young stars . based on the keck spectra for each of the projected region of 3c48 ( which is a qso at @xmath54 = 0.38 and also identified as an ulirg with dusty starburst ) , canalizo & stockton ( 1999 ) examined the nature of stellar populations for each of the regions and thereby derived spatial distribution of young stellar population for 3c48 . however , there are only a few observational studies which investigate in detail the two - dimensional distribution of spectroscopic properties and address the projected distribution of young stellar populations for dusty galaxies . young starburst populations are suggested to be relatively correctly traced by the spatial distribution of h@xmath109 regions without being affected seriously by dust extinction ( kennicutt 1998 ) . furthermore we can derive the detailed two dimensional distribution of h@xmath109 regions by using the technique of integral - field spectroscopy ( e.g. , chatzichristou 1998 ) . accordingly we expect that future ( and ongoing ) systematic spectroscopic studies with integral - field spectroscopy ( for spatial distribution of star - forming regions ) combined with very high - resolution mapping of dusty interstellar gas will assess the validity of the suggested importance of the relative distribution of gas and young stars . recently poggianti & wu ( 1999 ) have investigated in detail spectroscopic properties of ulirgs and found that some ulirgs show both strong balmer absorption and relatively modest [ oii ] emission ( the so - called e(a ) spectra ) . this peculiar spectroscopic properties of ulirgs are also found in galaxies in distant clusters of galaxies ( dressler et al . 1999 ; poggianti et al . 1999 ; smail et al . 1999 ) , which implies that this spectra is not characteristic only for ulirgs . poggianti et al . ( 1999 ) suggested that this e(a ) spectra comes from the starburst regions obscured by dust . although a growing number of observational results on e(a ) galaxies with possible dusty starburst have been accumulated , there are only a few theoretical studies addressing the formation and the evolution of these e(a ) populations . poggianti & wu ( 1999 ) first discussed that if the dust extinction for a stellar population decreases with the stellar age , e(a ) spectra can be obtained for a starburst galaxy . they furthermore suggested that if the location and thickness of dust patches depend on the age of the embedded stellar populations in a starburst galaxy , the effects of the above selective dust extinction become remarkable for the galaxy . based on a one - zone chemophotometric evolution model of dusty starburst galaxies , shioya & bekki ( 2000 ) confirmed that if a young starburst population is preferentially obscured by dust than old one in a dusty starburst galaxy , the galaxy shows e(a ) spectra . then , when and how the proposed difference in the degree of dust extinction between stellar populations with different ages is possible during galaxy merging ? we here suggest that the difference of dust extinction between the central region and the outer one in a galaxy is a main cause for the above age - dependent extinction . to be more specific , since younger stellar populations formed by secondary starburst in the central region with a larger amount of dust interstellar gas are more heavily obscured by dust than the outer old components , the age - dependent extinction can be achieved in mergers . this radial dependence of dust extinction is considerably reasonable and realistic , considering that secondary dusty starburst occurs preferentially in the central part of a galaxy owing to efficient inward gas transfer driven by non - axisymmetric structure ( such as stellar bars ) during galaxy merging . the present numerical simulations on gaseous and stellar distribution in merging disk galaxies with dusty starburst furthermore have demonstrated that the central young stellar component formed by nuclear starburst is more heavily obscured by dusty gas than the outer old component initially located in merger progenitor disks . we accordingly suggest that galaxies with the central dusty starburst ( e.g. , ulirgs ) are more likely to show the selective dust extinction and thus have spectroscopic properties characteristic for e(a ) galaxies . although we have not yet investigated spectroscopic properties of ulirgs formed by major merging in detail , it is not unreasonable to claim that the e(a ) spectral appearance is due essentially to the selective dust extinction ( i.e. , age - dependent extinction ) in the central regions of gas - rich major mergers with strong dusty starburst . in our preliminary studies ( bekki & shioya 2000c ) , a dusty starburst merger shows k+a / a+k spectra ( i.e. , the classical `` e+a '' galaxy ) at the epoch of post - starburst and e(a ) spectra at the epoch of maximum starburst . we have investigated in detail morphological , structural , and photometric evolution of gas - rich major mergers with strong dusty starburst in an explicitly self - consistent manner in order to explore the nature of ulirgs . main conclusions are summarized as follows . \(1 ) a massive starburst triggered by dynamical processes of major galaxy merging is found to be very heavily obscured by high density , dusty gas in a gas - rich merger . as a result of this , the merger can become a very luminous infrared galaxy with @xmath0 of @xmath3 5.9 @xmath23 @xmath7 @xmath5 at the maximum starburst epoch which corresponds to the epoch of the formation of central high density core of an elliptical galaxy . \(2 ) dust extinction of stellar populations in a galaxy merger with large infrared luminosity ( @xmath0 @xmath6 @xmath7 @xmath5 ) is selective in the sense that younger stellar populations are preferentially obscured by dust than old ones . this is because younger populations are located in the central region where a larger amount of dusty interstellar gas can be transferred from the outer gas - rich regions in the merger . this selective extinction provides physical basis for the origin of spectroscopic properties of ulirgs , such as e(a ) spectra recently revealed by poggianti & wu ( 1999 ) . \(3 ) both @xmath0 and @xmath10 increase as the star formation rate increases during starburst of the present merger model , resulting in the positive correlation between @xmath0 and @xmath10 in the merger model . the time evolution of the relative distribution of dusty gas and young starburst components is crucially important for the time evolution of these properties . \(4 ) the dust temperature @xmath110 and the flux ratio of @xmath12 increase with the increase of the star formation rate . this is because a larger number of young stars formed by massive starburst can heat the dusty interstellar gas as the star formation rate becomes higher . \(5 ) global colors ( e.g. , @xmath14 ) are redder in the central region than in the outer one at the epoch of strong starburst in a dusty major merger , which implies that most ulirgs show negative color gradients . furthermore the central surface brightness ( e.g. , in @xmath84 band ) becomes higher as the star formation rate becomes larger during starburst primarily because both old and young stellar populations are more strongly concentrated in the nuclear region of the merger . two - dimensional color distributions seem to be more strongly disturbed than surface brightness maps . the peak of two dimensional distribution of @xmath15 flux coincides with that of @xmath13 . these results clearly suggest that spatial distribution of gas and stars is one of important determinants for the two - dimensional photometric properties of major mergers . \(6 ) the star formation efficiency in a major merger becomes higher as the separation of the two cores becomes smaller , which clearly reflects the fact that dynamical processes of galaxy merging play an important role in determining the time evolution of dusty starburst in the merger ( e.g. , the time evolution of chemical components and local gaseous density in the starburst regions ) . total gas mass , the degree of dust extinction ( @xmath13 ) , @xmath11 , @xmath0 , @xmath10 , and @xmath12 also depend strongly on the separation of two cores of the merger . \(7 ) the star formation efficiency during starburst in a merger is found to correlate positively with the flux ratio of @xmath12 , which is consistent with recent observational results by young ( 1999 ) . furthermore @xmath10 and @xmath12 are found to be proportional to @xmath0 and to the central gaseous density of the merger , which is also consistent with fundamental properties observed in ulirgs ( e.g. , sanders & mirabel 1996 ) . we lastly stress that better understanding of dissipative and dissipationless dynamics of galaxy merging furthermore leads us to clarify more thoroughly the photometric properties of ulirgs . we are grateful to the referee guy worthey for valuable comments , which contribute to improve the present paper thanks the japan society for promotion of science ( jsps ) research fellowships for young scientists . the expected mass ( luminosity ) dependence of @xmath0 is described as follows . the @xmath0 is observationally suggested to be proportional to star formation rate ( kennicutt 1998 ) . therefore , the mass dependence of @xmath0 depends on the mass dependence of ( maximum ) star formation rate . here we define the maximum value of @xmath0 derived in the present study as @xmath111 . based on @xmath111 for the model with the mass of the merger progenitor disk equal to 6.0 @xmath23 @xmath24 @xmath34 , we predict the @xmath112 for mergers with different @xmath113 , where @xmath113 is the initial total mass of a disk in a merger . we also define the total mass of luminous and dark matter and size of the progenitor as @xmath114 and @xmath115 , respectively . we consider here that gas mass in a disk is proportional to @xmath116 for simplicity . the dynamical time scale represented by @xmath117 is given as @xmath118 assuming the freeman s law and the constant ratio of @xmath115 to the scale length of exponential disk , we derive @xmath119 we assume that the degree of self - gravity of a galactic disk is described as @xmath120 where @xmath121 is a parameter that controls the degree of self - gravitating of the disk and should be determined by observational studies . @xmath112 is proportional to star formation rate and thus can be scaled to @xmath122 ( in our units ) . then we can derive @xmath123 if we assume that @xmath121 is 0 ( 0 @xmath80 @xmath121 @xmath80 0.6 , saglia 1996 ) , this relationship predicts that @xmath112 becomes larger as @xmath116 increases . by using the derived value of 5.9 @xmath23 @xmath7 @xmath124 for @xmath125 , we can expect the following @xmath116 dependence of @xmath112 , @xmath126 the expected mass dependences of @xmath0 for the cases of @xmath121 = 0.0 and 0.6 are shown in figure 24 . in this paper , we show the results of only one merger model with a prograde - retrograde configuration . figure 25 shows clearly this orbital configuration of merging two disk galaxies . stellar and gaseous particles in the galaxy labeled as galaxy 1 orbit the center of the galaxy in a counter - clockwise whereas the galaxy itself orbits the center of mass of the merger in a counter - clockwise . this galaxy is considered to be rotating in a prograde sense for the merger , because intrinsic spin vector of the galaxy is nearly the same as that of the orbital one . in the galaxy labeled as galaxy 2 , on the other hand , stellar and gaseous components orbit the center of the galaxy in a clockwise whereas the galaxy itself orbits the center of mass of the merger in a counter - clockwise . this galaxy is considered to be rotating in a retrograde sense because intrinsic spin of the galaxy goes in the opposite direction of the orbital one . the merger with this orbital configuration is referred to as a nearly prograde - retrograde merger in the present paper .

recent experiments have demonstrated a remarkable progress in implementing and use of the particle image velocimetry ( piv ) and particle tracking techniques for the study of turbulence inhe . however , an interpretation of the experimental data in the superfluid phase requires understanding how the motion of tracer particles is affected by the two components , the viscous normal fluid and the inviscid superfluid . of a particular importance is the problem of particle interactions with quantized vortex lines which may not only strongly affect the particle motion , but , under certain conditions , may even trap particles on quantized vortex cores . the article reviews recent theoretical , numerical , and experimental results in this rapidly developing area of research , putting critically together recent results , and solving apparent inconsistencies . also discussed is a closely related technique of detection of quantized vortices negative ion bubbles inhe . pacs numbers : 67.40.vs,47.37.+q,47.27.-i

0909.4451

in this review we will be mostly concerned with the motion of small solid particles in turbulent @xmath0he . this new and rapidly developing area of research has been initiated by the recent success of implementation of the particle image velocimetry ( piv ) and the particle tracking techniques in superfluid helium , see donnelly _ _ et al.__@xcite , van sciver and co - workers@xcite , and publications of the group involving paoletti , bewley , lathrop , sreenivasan and their co - workers@xcite . ( the only difference between these two , otherwise identical , techniques is that the results of piv yield the local average particle velocities obtained by calculating cross - correlations of particle ensembles and , therefore , result in the fluid - like , smooth velocity field , while the particle tracking aims at investigation of individual particle trajectories . ) these are perhaps the only two well developed techniques which can identify the flow patterns in turbulent superfluid helium . the piv has been a standard technique in classical fluid dynamics for several decades ( see , for example , the book by raffel _ _ et al.__@xcite ) . in experimental studies of classical turbulence , one can be confident that the motion of sufficiently small particles will reveal the details of turbulent motion of the viscous fluid . in superfluids , the two - fluid nature of @xmath0he makes an interpretation of piv measurements much more difficult : a solid particle interacts with both the normal fluid and the superfluid ; moreover , the particle may interact strongly with quantized vortex lines in the superfluid component , and even become trapped on them . the mechanism of trapping of solid particles is essentially the same as that of trapping the charge carriers on quantized vortices in @xmath0he , primarily because e.g. the negative ion ( electron ) forms around itself an almost macroscopic bubble of radius @xmath1 from which helium atoms are excluded . the ion trapping technique was used for detection of quantized vortices since late 50s , and , although not being suitable for studying the normal fluid patterns in @xmath0he , can be considered as the technique closely related to the piv and the particle tracking methods . we will review the development of the ion trapping technique in section 7 of this article . the aims of experimental physicists working in the area of superfluid turbulence are to understand the vortex tangle ( quantum turbulence ) , to measure velocity fluctuations in both the normal fluid and the superfluid , and possibly to make comparisons between quantum turbulence and ordinary turbulence . until recently the major difficulty was lack of direct flow visualization technique near absolute zero . existing methods , such as measurements of temperature differences to detect extra dissipation , ion trapping , measurements of second sound attenuation by quantized vortices , etc . probed only the vortex line density @xmath2 ( total vortex length per unit volume ) averaged over an experimental cell and did not reveal turbulent velocity fluctuations ( although a remarkable resolution has been recently achieved in second sound measurements by roche _ _ et al.__@xcite of the local tangle density ) . experimental breakthrough was made in 2002 when the piv technique was successfully introduced in @xmath0he ( see references in the first paragraph of this section ) . this technique , which has been standard in classical fluid dynamics for many years , is based on injecting many small particles into the liquid . two images are produced using short laser pulses of different frequencies ( corresponding e.g. to the green and red colors ) focused into a narrow sheet and separated in time by a few milliseconds . software then analyzes the images and identifies green and red dots corresponding to the same particle at the two different times . in this way the observed distance between the corresponding dots yields the component of the local velocity in the plane of the light sheet . in classical fluids , provided the particle size is sufficiently small ( in turbulence studies , `` small '' means smaller than the kolmogorov length ) , the dominating force acting on the particle is the viscous drag force , so that a researcher can be confident that small particles trace the fluid motion ( in particular , turbulent velocity fluctuations ) rather well . because the viscosity of liquid helium is very low the kolmogorov scale in the turbulent normal fluid can be very small , so that it is essential that the particles used in visualization experiments are as small as possible . in the cited experimental works the typical particle diameter was of the order of @xmath3 . what do tracer particles trace in superfluid helium ? one may expect that , although the viscosity of @xmath0he is low , the dominating force exerted on the particle by the fluid will still be the viscous drag , so that solid particles should trace the normal fluid . however , this is not always true : due to the two - fluid nature of superfluid helium , the particles interact not only with the normal fluid , but also with the superfluid component through inertial and added mass forces ; moreover , the particles interact strongly with quantized vortices in superfluid and may even be trapped on superfluid vortex lines . therefore , if we want to interpret results of piv and particle tracking measurements correctly , we must answer first the question asked in the beginning of this paragraph . this article is divided in sections where theoretical and numerical models of increasing complexity are compared to each other and to experimental results . the first model , described in section 2 , is the one - way coupling model . in sections 2.1 and 2.2 we derive the governing equations of motion of particle tracers in the presence of two imposed fluids , the viscous normal fluid and the inviscid superfluid , under the assumptions that the particles do not disturb the flow , are smaller than any flow scale of interest , and do not become trapped in vortices . the one - way coupling model allows us to discuss the problem of the stability of particles trajectories , which is relevant to the visualization a pure superflow . the one - way coupling model is powerful enough to derive the general principles which lead to particles being trapped on vortices . in section 2 we present the experimental evidence for this trapping ( section 2.3 ) , and show how the mutual friction affects the motion of particles near vortex cores ( section 2.4 ) . three case studies are discussed of particle trajectories near vortices : vortex ring , thermal counterflow tangles , and vortex tangles at low temperatures ( section 2.5 ) . section 3 introduces the more sophisticated ( and computationally more expensive ) two - way coupling model ; in this model the back reaction of the flow on the particle is taken into account , and the dynamics of the particle - vortex interaction and the trapping can be studied in great detail , including what happens at the surface of the particle ( which we assume to be spherical for simplicity ) . section 3.1 contains the mathematical formulation of the two - way coupling model ; sections 3.2 and 3.3 are devoted to the numerical calculations of typical vortex - particle interactions . section 4 makes use of the results of the numerical simulations described in section 3 to derive a simpler analytical model which explains in a quantitive way experiments performed in florida@xcite in which heavy particles fell through a tangle of vortices . the two - way coupling model is also applied to particles moving in tangles generated by a thermal counterflow : these numerical results are applied to particle - tracking experiments performed in maryland@xcite . by considering numerical calculations at small and high values of the vortex line density , we solve the apparent disagreement between the florida and maryland experiments : the discussion will reveal that the two experiments refer to two different regimes , which are both explained by the two - way coupling model . in section 4 we also discuss the experimental observation of flow structures observed behind and in front a cylindrical obstacle set in the middle of a counterflow channel , and propose a simple analytical two - dimensional model which accounts , at least in principle , for the qualitative existence of these flow structures . section 5 describes the most recent experimental discoveries obtained using tracer particles : the observation of turbulent boundary layer flows , the visualization of individual vortex reconnections , and the measurement of velocity statistics . section 7 reviews other techniques based on trapping ions and imaging @xmath4 molecules ; these techniques share important principles ( but not the size of the trapped object ) with piv and particle tracking techniques . section 8 contains the final discussion . we begin answering this question with a relatively simple , `` one - way coupling '' model which follows the approach standard in classical two - phase turbulence studies . we will generalize to the two - fluid model of superfluid helium the equations of motion of a solid spherical particle of radius @xmath5 in a nonuniform flow of classical fluid . we assume that 1 ) the presence of particles does not modify the turbulence , 2 ) the flow velocities vary little over a distance of the order of particle size , and 3 ) particles do not interact strongly with quantized vortex lines and are certainly not trapped on these lines . these assumptions require that the particle size be much smaller than both the kolmogorov lengthscale , @xmath6 in the normal fluid , and the mean intervortex distance , @xmath7 in the superfluid . below we formulate , under these assumptions , the equations of particle motion . we start with the fluid - particle interaction in classical liquids . we consider a spherical solid particle of radius @xmath5 in the nonuniform flow . let the ambient ( that is , in the absence of the particle ) fluid velocity field be @xmath8 . according to the assumptions formulated above , the size of the particle is much smaller than the flow lengthscale , @xmath9 , i.e. @xmath10 , so that we can introduce a small parameter @xmath11 where @xmath12 is the particle velocity . as shown by auton _ _ et al.__@xcite , the total force acting on the particle in the nonuniform , inviscid flow can , under assumption ( [ varepsilon ] ) , be represented in the form @xmath13 where @xmath14 are the inertial and the added mass force , respectively , @xmath15 is the particle volume , @xmath16 and @xmath17 is the added mass coefficient ( for spherical particle @xmath18 ) ; @xmath19 where @xmath20 is the vorticity , represents the lift force arising due to stretching of vortex lines in the vicinity of the sphere s surface . we will consider the motion of a solid particle in the viscous fluid assuming that the particle reynolds number is small : @xmath21 where @xmath22 is the kinematic viscosity . note that small particle reynolds numbers are typical of piv and particle tracking experiments , both in classical fluids and in superfluid helium . we will consider the particle motion under the assumptions formulated above ( in particular that the parameter @xmath23 introduced by formula ( [ varepsilon ] ) is small ) . detailed analysis of the forces acting on the particle in the nonuniform viscous flow can be found in works by maxey and riley@xcite , mei@xcite , and kim , elghobashi , and sirignano@xcite . the total force acting on the particle can be represented as a sum of several contributions , i.e. the gravity , viscous drag , the inertial and added mass force , faxn correction arising due to the local non - uniformity of the ambient flow , the saffman lift force arising due to the local shear , and the magnus lift force arising due to rotation of the particle . the main contribution , dominating in most particulate flows , is the viscous stokes drag force : @xmath24 surprisingly , in the viscous flow the inertial and added mass forces , @xmath25 and @xmath26 are determined by the same formulae ( [ inert_and_added])-([f ] ) as for the inviscid flow , with the same added mass coefficient , @xmath18 for the spherical particle . to determine the force acting on the particle in @xmath0he , we simply add together all the forces exerted by the normal fluid and the superfluid . since the superflow is potential , @xmath27 is identically zero , provided the particle does not become trapped on quantized vortex lines . as shown by poole _ _ et al.__@xcite , for the flow properties and physical parameters typical of the normal component of @xmath0he , the faxn correction , the history and lift forces can be neglected provided the particle reynolds number and the parameter @xmath28 are small and the particle size is significantly smaller than the kolmogorov length . therefore , the total force acting on the particle can be approximated as @xmath29 where the subscripts @xmath30 and @xmath31 refer to the normal fluid and the superfluid , respectively . here @xmath32 is the combination of gravity and buoyancy , and @xmath33 is the viscous drag force exerted by the normal fluid : @xmath34 where @xmath35 is the viscosity of @xmath0he . substantial derivatives required for determining the inertial and added mass forces in the normal and superfluid components are now defined by formula ( [ ddt ] ) with @xmath36 replaced by @xmath37 and @xmath38 , respectively . we arrive at the following equation of particle motion@xcite : @xmath39 this equation must be considered together with the kinematic equation @xmath40 where @xmath41 ( @xmath42 ) should be regarded as a lagrangian trajectory of the solid particle . ( [ motion1 ] ) and ( [ kinematic ] ) constitute the closed system for the unknown particle position and velocity , @xmath43 and @xmath44 , respectively . often ( although not always ) in the piv and particle tracking experiments neutrally buoyant particles ( with @xmath45 ) are used in order to eliminate unwanted effects of gravity on the particle motion . for neutrally buoyant particles , eq . ( [ motion1 ] ) can be written in a more concise form @xmath46 where @xmath47 the parameter @xmath48 , which shows how quickly the particle adjusts its motion to the viscous flow , plays an important rle in the study of particle motion in fluids , and is commonly known as either the particle response time , or viscous relaxation time . analyzing eqs . ( [ kinematic ] ) and ( [ motion2 ] ) , poole _ _ et al.__@xcite showed that , provided @xmath49 , where @xmath50 is the timescale of the fluid motion ( e.g. the kolmogorov time ) , the neutrally buoyant particle tracks the motion of the normal fluid . if @xmath51 , the particle moves with a velocity corresponding to the total current density , @xmath52 . however , two important issues may invalidate these conclusions : 1 ) instability of particle trajectories , and 2 ) trapping of particles on superfluid vortex lines . the analysis of particle trapping will require more elaborate , self - consistent , two - way coupling model which would account for deformation of the vortex filament by the approaching particle , including possible reconnection of the vortex with the particle surface ( see below section [ self - consistent ] ) . instability of lagrangian trajectories of the neutrally buoyant particle in the classical viscous fluid was discovered and studied relatively recently by babiano _ _ et al.__@xcite . to illustrate such an instability in turbulent superfluid at finite temperature such that both the normal fluid and the superfluid component are present , we consider the simplest case assuming that @xmath53 down to the length scales comparable to the vortex line spacing . the lagrangian equations of particle motion , which become @xmath54 have , provided @xmath55 and @xmath56 , a formal solution @xmath57 , @xmath58 , where @xmath59 is a trajectory of the fluid point . therefore , it would seem natural to conclude that neutrally buoyant particles follow the fluid exactly . however , let us consider now the particle motion in the so - called abc ( arnold - beltrami - childress ) flow , frequently used as the simplest model of turbulence , whose velocity field is @xmath60 the time sequence@xcite illustrating the position of large number of neutrally buoyant tracer particles in the abc flow , starting from their uniform distribution , is shown in fig . [ segregation ] . it can be seen that particles do not follow the fluid , but instead , due to instabilities of their trajectories , travel from the regions of high vorticity to the regions of high rate of strain . ( in classical multiphase fluid dynamics similar phenomenon for particles heavier than the fluid has been studied in detail . however , the mechanism of segregation of heavy particles is quite different from that of neutrally buoyant particles . ) note , though , that this instability develops rather slowly : the last frame of fig . [ segregation ] , recovering the lines of minimum vorticity and maximum rate of strain , corresponds to 3 times of turnover of the abc flow . instability of particle trajectories is more pronounced in the case @xmath61 when the normal fluid is absent . the equations of motion of neutrally buoyant particle become @xmath62 and have a formal solution @xmath63 , @xmath64 , where @xmath65 , so that , in the case of very low temperature when the normal fluid is absent , it can be expected that neutrally buoyant particles trace the superfluid . however , this is not the case either . consider the simplest case of the neutrally buoyant particle moving around a single , stationary , straight vortex line . in cylindrical polar coordinates @xmath66 the equations of particle motion are : @xmath67 where @xmath68 . if at the initial moment the particle velocity does not coincide exactly with the fluid velocity , the particle will spiral either outwards , or inwards . we arrive at the conclusion which remains valid in the general case of particle motion at temperature @xmath61 ( sergeev _ _ et al.__@xcite ) : unless the initial velocity of neutrally buoyant particle matches exactly the velocity of the fluid point , the trajectory of the solid particle deviates significantly from the trajectory of the fluid point . such an instability is amplified by any macroscopic mismatch between the velocity of the superfluid and the velocity of the particle at the beginning of the experiment . this is further illustrated by fig . [ three_vortices ] showing the trajectories of solid and fluid particles around three vortices . why can the superfluid vortex trap the particle ? to answer this question , we have to take into account a possibility of reconnection of the quantized vortex to the surface of moving particle ( a more detailed analysis of the mechanism of this process will be given below in section [ self - consistent ] ) . below three different versions , or , rather , three different ways of explaining the reason for particle trapping are suggested . imagine that the quantized vortex reconnects symmetrically to the surface of spherical particle , as shown in the right part of fig . [ asymmetric ] . the kinetic energy of the flow field created by the straight quantized vortex can be easily calculated , and in the symmetric configuration is reduced by @xmath70 where @xmath71 is the vortex core radius . note that assuming @xmath72 this result follows from the substitution energy calculated by parks and donnelly@xcite for the ion bubble trapped by the quantized vortex line , see below sec . [ related_techniques ] and formula ( [ substitution ] ) therein . formula ( [ reduced_e ] ) determines the kinetic energy which the particle would require to break free from the symmetric vortex configuration shown in fig . [ asymmetric ] ( right ) . the flow field of the vortex creates a pressure gradient @xmath74 attracting the particle to the vortex . if the particle - vortex configuration is symmetric , as in the right part of fig . [ asymmetric ] , then , obviously , the force acting on the particle is zero . now imagine that this symmetry is perturbed as shown in the left part of this figure . the superfluid component is inviscid , so that the following arguments apply based on bernoulli s integral : the fluid velocity on the right side of the particle surface , where the two vortex strands are closer , is greater than that on the left side , and , therefore , the pressure is bigger on the left side of the sphere . this provides a net force restoring the symmetric particle - vortex configuration . experimental evidence comes from the publication of bewley , lathrop and sreenivasan@xcite ( see also more recent papers@xcite ) . [ sreeni_nature ] , published in the cited paper@xcite , shows that the researchers , using the piv technique , actually `` painted '' quantized vortices by tracer particles . another evidence comes not from experiments , but from the quantum calculation based on the gross - pitaevskii equation . berloff and roberts@xcite studied an interaction between the negative ion and the quantized vortex . in superfluid helium , the negative ion forms around itself a bubble of diameter approximately @xmath76 ( an order of magnitude higher than the size of the vortex core , @xmath77 ) and , therefore , can be treated , for our purpose , as a particle ( albeit very small ) . results of berloff and roberts calculation are illustrated in fig . [ ion ] : the ion bubble approaches the vortex , which deforms ( ( b ) and ( c ) ) and then reconnects to the particle surface ( d ) ; the reconnection excites kelvin waves ( ( f ) , ( g ) , and ( h ) ) which carry away the energy , and , eventually , the bubble - vortex configuration relaxes and the ion becomes trapped on the vortex core ( i ) . at this point , the question can be asked whether it is worth or not further exploiting the one - way coupling model which neglects any influence of the particle on the vortex evolution . the answer is `` yes '' : there are cases where trapping events are not very frequent , so that a useful information about the behaviour of tracer particles can be obtained by ignoring their trapping on vortex cores . we will also show that the mutual friction between quantized vortices and the normal fluid can prevent trapping . in the vicinity of the vortex core , the mutual friction induces , in the normal fluid , the vortex dipole whose typical lengthscale is expected to be about @xmath78 . the results of numerical calculation by idowu _ _ et al.__@xcite of the dipole - like normal fluid disturbance are shown in fig . [ normal_disturbances ] . this normal flow disturbance can deflect the tracer particle which otherwise would have collided with , and possibly trapped by the vortex . typical trajectory , calculated by sergeev _ _ et al.__@xcite , of the particle moving from right to left and interacting with the superfluid vortex and normal fluid disturbance is shown in fig . [ deflection ] by the solid line . for comparison , the trajectory calculated without taking into account the normal fluid disturbance is shown by the dashed line . this trajectory leads to the collision with the vortex core located at the origin . normal fluid vortical structures induced by the mutual friction were predicted by hall and vinen@xcite and kivotides , barenghi , and samuels@xcite , but so far , because of low resolution of experimental techniques , there was no direct experimental proof of existence of the normal fluid disturbances . owing to the much higher resolution , the piv and particle tracking techniques can provide such an evidence . perhaps the first experimental confirmation of existence of normal flow structures induced by the mutual friction has already been found . fig . [ jostling ] shows a typical particle trajectory observed by bewley , lathrop , sreenivasan , and paoletti@xcite , by means of the particle tracking technique , in the thermal counterflow . the particle trajectory has characteristic ` kinks ' resembling the ` deflected ' trajectory shown by the solid line in fig . [ deflection ] . in the remaining part of this section we will analyze , neglecting trapping , three examples of particle motion in @xmath0he . this study@xcite led to a proposal of a simple experiment ( not performed yet ) which could , in principle , justify the use of piv technique for measuring instantaneous normal fluid velocity patterns . we consider a single vortex ring propagating normally to a plane sheet of neutrally buoyant particles , see fig . [ ring_sheet ] ( left ) . in the considered model , the mutual friction between the normal fluid and the superfluid vortex is taken into account leading to emergence of the normal fluid disturbances described earlier . the motion of the superfluid vortex whose core is defined parametrically by @xmath79 is governed by the following equation derived by idowu _ _ et al.__@xcite : @xmath80 where @xmath81 , @xmath82 , @xmath83 and @xmath84 are the known mutual friction coefficients , and the vortex - induced superfluid velocity @xmath85 is given by the biot - savart integral @xmath86 the motion of incompressible ( @xmath87 ) normal fluid is governed by the equation @xmath88 where @xmath89 is the mutual friction force per unit volume . this force is determined as the sum of the drag force and the iordanskii force , @xmath90 , per unit length : @xmath91\ , . \label{iordanskii}\ ] ] here the new mutual friction coefficients , @xmath92 and @xmath93 are known and can be expressed explicitly through @xmath82 , @xmath83 and @xmath84 . trajectories of neutrally buoyant particles were found by numerical integration of eqs . ( [ kinematic ] ) and ( [ motion2 ] ) . configurations of solid particles before and after the ring has passed the particulate sheet are shown in fig . [ ring_sheet ] ( right ) . [ histograms_sheet ] shows the histograms of the angle and of the relative difference between @xmath37 and @xmath12 ; it can be seen that @xmath37 and @xmath12 , to a very good degree of accuracy , are identical both in magnitude and direction . moreover , the calculation showed that trapping events are relatively rare only 42 out of 900 particles approached the ring to a distance smaller than three particle diameters . this enables us to conclude that , in the proposed experiment , the measurement of particle velocities can provide direct information about instantaneous normal flow patterns . later , based on the self - consistent , two - way coupling model , kivotides and wilkin@xcite performed more elaborate study of interactions between solid particles and vortex rings , see below sec . [ kivotides_wilkin ] . in this example@xcite we will be concerned with the t - i state of @xmath0he turbulence such that the vortex tangle in the superfluid component is present but the normal flow is laminar . for simplicity , the normal flow is assumed uniform , @xmath94 . the superfluid velocity can be represented as @xmath95 , where the mean ( counterflow ) superfluid velocity , @xmath96 is linked with the normal fluid velocity by the relation @xmath97 , and @xmath85 is the fluctuating superfluid velocity induced by the vortex tangle . the dynamic vortex tangle is modeled , in the periodic box , taking into account the mutual friction between the normal fluid and quantized vortices , as well as the biot - savart interaction between vortex filaments . an influence of superfluid vortices on the motion of normal fluid is neglected . based on eqs . ( [ kinematic ] ) and ( [ motion2 ] ) , the motion of neutrally buoyant tracer particles of diameter @xmath98 was calculated after the vortex tangle has reached the statistically steady state . calculations , performed at temperature @xmath99 for @xmath100 and @xmath101 ( corresponding values of the counterflow heat flux are @xmath102 and @xmath103 ) , and at temperature @xmath104 for @xmath105 ( @xmath106 ) , showed that the particle velocity is very narrowly peaked around the constant normal velocity . three histograms of fig . [ histograms_cos_counterflow ] , corresponding to the three cases considered above , show the absence of alignment ( left ) between particle velocities and the superfluid velocity , @xmath85 induced by the vortex tangle , and illustrate anisotropy in @xmath85 ( center and right ) . the above two examples enable us to expect that , in the range of parameters typical of piv measurements , neutrally buoyant particles should trace the normal fluid well , and , provided strong interactions of particles with quantized vortices can be neglected , particle velocity fluctuations induced by interactions between particles and quantized vortices should be relatively small . however , these conclusion can be invalidated by trapping of solid particles on quantized vortex cores . this phenomenon will be addressed below in the sec . [ self - consistent ] . below the results discussed in section [ temperature ] will suggest that at temperature @xmath107 , when the normal fluid is practically absent and the damping force on the particle can be neglected , trapping of neutrally buoyant solid particles on quantized vortices can , most likely , be ignored and the motion of solid particle can be modeled by a simpler , one - way coupling model . however , at these temperatures the presence and motion of the particles still affects the vortex filaments , so that a certain modification should be necessary of the one - way coupling model . in order to understand some features of the particle motion in the vortex tangle at such a low temperature , we start with a simple , two - dimensional model of the tangle@xcite . in such a model the vortex lines become vortex points , and the biot - savart law reduces to a simple statement that each vortex point moves as a fluid point in a flow field created by all other vortices . ( such a two - dimensional system of vortex points is known as the onsager s point vortex gas@xcite . ) we consider , in the periodic box , the motion of neutrally buoyant particles in the system of vortex points of random polarity . we neglect trapping of particles on quantized vortices as well as any influence of particles on the motion of vortex points . in this approximation the particle motion is governed by eqs . ( [ motiontto0 ] ) , and the following problem can be immediately identified : in the case where the position of the particle coincides with that of the vortex point , the pressure gradient force , @xmath108 becomes unphysically singular . based on the mechanism , described below in sec . [ self - consistent ] of the particle - vortex collision at very low temperature ( in the absence of damping force ) , we will resolve this manifest difficulty by modifying the model ( [ motiontto0 ] ) as follows . the cause of the problem is that in the real , three - dimensional tangle at temperature @xmath107 , even when the particle breaks through the vortex , it nevertheless reconnects with the vortex filament when the distance between the particle and the vortex core becomes of the order of particle radius . since the vortex line attached to the particle is necessarily orthogonal to the particle surface , the reconnection results in a dramatic decrease of the force exerted on the particle ; this force is zero when the particle - vortex configuration is symmetric . then , the one - way coupling model can be modified by assuming that there exists a force - free region for @xmath109 , where @xmath110 is a cut - off distance . in the two - dimensional calculation we set @xmath111 . trajectories of an inertial particle and a fluid point are illustrated in fig . [ tr2d_tr3d ] ( left ) for a system of 20 vortex points [ cols="^,^ " , ] xu and van sciver concluded@xcite that in the wall region the velocity distribution agrees reasonably well with the classical @xmath30th - power law ( for @xmath30 ranging from 7 to 8.8 ) . they also addressed the following questions which yet to be answered . why the velocity profile is wider and flatter than that in the classical viscous channel flow ? what is the nature of dependence on reynolds number of the results shown in fig . [ boundary_layer ] ( left ) ? why , as seen from fig . [ boundary_layer ] ( right ) , the normal fluid density does not seem to affect the shape of velocity profiles ? in a recent experiment@xcite bewley , paoletti , sreenivasan , and lathrop observed the motion of solid hydrogen particles trapped on quantized vortices . the specific purpose of this work was a direct experimental investigation of vortex reconnections in turbulent @xmath0he . a sequence of images@xcite illustrating the motion of particles trapped on vortex filaments is reproduced in fig . [ recon_images ] . to quantify their results , bewley _ _ et al.__@xcite assumed that the evolution of reconnecting vortices can be characterized by a single scale parameter , @xmath112 . using , as a measure of @xmath112 , the experimentally observed distance between two particles closest to the point of reconnection , they found that the evolution obeys the scaling @xmath113 , where @xmath114 corresponds to the moment of reconnection of two vortices . the particle tracking technique was further developed by paoletti , fisher , sreenivasan , and lathrop @xcite to investigate , by analyzing the trajectories of tracer particles , the velocity statistics in decaying quantum turbulence . the decay of turbulence produced initially by the thermal counterflow was studied after the counterflow has been stopped by switching the heater off . _ found that the pdf of the particle velocity is strongly non - gaussian with a pronounced tail obeying @xmath115 power law , see fig . [ vel_stat ] . the authors attributed such a tail to high velocities produced by reconnections of quantized vortices . the experiment has stimulated calculations @xcite of turbulent velocity statistics in the context of three - dimensional and two - dimensional atomic bose - einstein condensates ; after finding similar non - gaussian velocity statistics in these related quantum systems the authors argued that , in general , non - gaussian statistics arise from the singular nature of the velocity field around quantized vortices . the results of fundamental importance reported in these two reviewed publications @xcite raise , however , the following question : what effect have the particles trapped on vortex filaments on the evolution of quantized vortices and , in particular , on the reconnection of two approaching vortices ? in their work@xcite reviewed earlier in sec . [ tracking ] the authors claimed that , because in their experiments the distance between micron - size particles trapped on the same vortex filament was typically about @xmath78 , the influence of trapped particles on the evolution of vortices can be neglected . this claim is yet to be justified ( especially where the motion and evolution of vortex filaments near the reconnection point is concerned ) . it is unlikely that at present this problem can be treated experimentally ; a more realistic alternative seems to be a development of the mathematical model of a vortex filament loaded by trapped solid particles@xcite . the technique of vortex detection by ion trapping was recently reviewed in a number of regular research publications including some in this journal@xcite , so that we will present here a rather brief overview of theoretical and experimental aspects of this technique . the idea of using charge carriers for detection of quantized vortices in superfluids is much older than that of using solid particles for the flow visualization in @xmath0he and dates back to the works of careri@xcite , and reif and meyer@xcite . the technique is based on the phenomenon that a negative charge ( a single electron ) injected into @xmath0he self - localizes itself in a spherical void ( known as the ion or electron bubble ) of radius @xmath1 ( depending on pressure ) from which helium atoms are excluded . ( a relatively large size of this , almost macroscopic bubble justifies inclusion of this chapter into our review otherwise concerned with the motion of much larger , solid particles in @xmath0he . ) the model of the negative ion bubble was proposed by ferrel and later elaborated by kuper and other authors@xcite and confirmed experimentally by levine and sanders@xcite . consistent with the bubble model of the negative charge carrier is the effective mass of the ion , @xmath116 determined in ref . it appears in the langevin equation of motion for the ion s drift velocity , @xmath117 in the electric field , @xmath118 : @xmath119 , where @xmath120 is the electron charge , and @xmath48 the phenomenological relaxation time . the effective mass of the ion was found@xcite to be of the order 100 masses of @xmath0he atom ( in agreement with the bubble model ) , and @xmath48 appears to be the viscous relaxation time introduced above in sec . [ lagrangian_equations ] . ( the detailed treatment for the stokes problem of the ion bubble motion , albeit in normal @xmath0he , was given by ostermeier and schwarz@xcite . ) the possibility of using negative ions as detectors ( or `` probe particles '' ) of quantized vortices is due to the fact that the ion bubble is attracted to the vortex by the bernoulli force . the combined action of the bernoulli effect and the reduced condensation energy of the core produces a potential well of the depth of the order @xmath121@xcite and hence traps the ion . this mechanism@xcite is practically identical to that described above in sec . [ trapping ] of trapping of solid particles on the vortex cores . ( note that the mechanism of trapping of the ion bubble by the quantized vortex in @xmath0he modeled as the bose condensate was studied by berloff and roberts@xcite and described above in sec . [ evidence ] . ) the model of the positive charge carrier in @xmath0he was developed by atkins@xcite in 1959 and became since then commonly accepted . according to this model the positive ion exerts an electrostrictive attraction on the surrounding fluid thus causing a liquid - solid transition resulting in a core ( `` cluster '' or `` snowball '' ) of solid helium . the radius of this snowball is from 7 to @xmath122 , depending on pressure . like the negative ion bubble , such a snowball will be attracted , by the bernoulli force , to the vortex core so that , in principle , positive ions can also be used for detection of quantized vortices . however , a smaller than that of the ion bubble radius of the snowball and , therefore , the cross - section of the snowball - vortex interaction make the experimental technique using positive ions more difficult and less practical . hence in this review we will discuss only works concerned with negative charge carriers . the realization of the negative ion technique can be illustrated by the following arrangements of the experiment , typical of the early observations of single vortex lines in rotating @xmath0he : quantized vortex lines are charged first by ion bubbles trapped on them using e.g. the electron beam emitted orthogonally to the axis of rotation . the vortex lines are then detected by applying the electric field parallel to the axis of rotation so that trapped electron bubbles slide along vortex filaments to a collector attached to an electrometer . the latter registers the amount of collected charge which is proportional to the number of vortex lines present . in its early , simplest version this technique allowed to determine only the total number of ( almost ) straight quantized vortex lines in a slowly rotating container . its subsequent modifications made possible also detection of individual vortex lines and later the measurement of the vortex line density in turbulent @xmath0he . the theory of ion trapping by the quantized vortex has been developed in classical works of donnelly , roberts , and parks@xcite . to analyze ion - vortex interactions , two competing mechanisms were taken into account : trapping of the ion bubble , moving in the electric field , in the potential well of the vortex , and escape of the trapped ion due to its brownian motion , which was considered to be in equilibrium with thermal ( quasiparticle ) excitations in @xmath0he . two parts of this problem were calculations of 1 ) the trapping ( or capture ) cross - section , @xmath123 , and 2 ) the escape probability , @xmath124 ; the second part was analyzed using smoluchowski or fokker - planck equations in the framework of the kramers - chandrasekhar method for calculation the probability of escape of a particle by diffusion from a potential well . the capture cross - section , which was found to be of the order @xmath125 to @xmath126 and to decrease with the electric field and increase with temperature to about @xmath127 , at which temperature it drops sharply . the escape probability ( or , equivalently , the mean trapping lifetime , @xmath128 ) was found in a good agreement with experimental data obtained by springett , tanner , and donnelly@xcite as well as with experimental results of douglass and cade@xcite . the effective cross - section can then be calculated as @xmath129 , where @xmath130 is some characteristic time . parks and donnelly@xcite further developed the theory in order to make possible calculation of the ion bubble radius from the experimentally measured mean trapping lifetime . this required , in particular , a calculation of the so - called substitution energy , i.e. a kinetic energy of rotating superfluid excluded by the trapped bubble . for the ion bubble of radius @xmath131 trapped symmetrically on the vortex core this was found as @xmath132 \label{substitution}\ ] ] ( note that formula ( [ reduced_e ] ) of sec . [ trapping_mechanism ] follows from eq . ( [ substitution ] ) assuming @xmath133 ) . in the context of further experimental studies discussed below , it is worth noticing that the substitution energy ( equal to the depth of the potential well ) is about @xmath121 . parks and donnelly s study of bubble radii was further developed by springett and donnelly@xcite who used the measurements of trapping cross - sections in the rotating container to deduce that the radius of the ion bubble decreases with pressure . the theory developed by donnelly _ et al . _ also predicted that the mean trapping lifetime should increase with the superfluid density and with the radius of ion bubble . these predictions were soon confirmed experimentally by springett@xcite who measured , at various pressures and temperatures , the cross - section of the ion capture and ion mobilities in order to derive from these data the pressure and temperature dependence of the ion bubble radius and the mean trapping lifetime . later pratt and zimmermann@xcite also measured the mean trapping lifetime in the wide range of temperatures and pressures ( from vaporization and solidification ) and showed that at constant pressure the lifetime rapidly decreases with @xmath134 . in particular , it was found that at saturated vapor pressure trapping becomes negligible at temperature above @xmath135 , the value which is now often referred to as the `` abrupt lifetime edge '' . it was also shown that the temperature , below which trapping becomes significant increases with pressure . glaberson@xcite measured the mobility , as a function of temperature and pressure , of negative ions trapped on quantized vortex lines , and arrived at the important conclusion that the negative ion bubbles do not deform as they are trapped on vortex cores . he also developed a new model , which appeared to account satisfactory for available experimental results , for the drag exerted on the trapped ion bubble , based on the assumption that the vortex line has several thick central core surrounded by a tail of excess roton density with momenta opposite to the direction of circulation . among the first applications of the ion trapping technique for an investigation of vortex structures in @xmath0he was an observation by northby and donnelly@xcite of a nonlinear dependence of the number of quantized vortex lines on the angular velocity of the rotating container . this nonlinearity was found to be associated with the existence of a near - wall region , observed in the cited work , free of quantized vortices . the early experiments@xcite have indicated that ions interact strongly with turbulence in @xmath0he , but no quantitative data have been obtained yet . the first experimental study of the turbulent vortex tangle by the ion trapping technique was undertaken by sitton and moss@xcite in 1969 ( in fact it was the first direct experimental confirmation that the turbulence in the superfluid component of @xmath0he is composed of individual , quantized vortex lines indistinguishable from those produced in the ( slowly ) rotating container except for their configurations ) . the vortex tangle was produced by the supercritical heat current at temperature above @xmath127 . since most of the vortex lines are no longer straight but have a configuration of loops and kinks , the charge can no longer slide along the filaments to a collector but is trapped inside the tangle . the fraction @xmath136 of the trapped charge was measured and then linked with the vortex line density by the relation @xmath137 , where @xmath138 is the ion velocity , and @xmath2 can be linked with the heat flux and properties of @xmath0he by the well known vinen equation@xcite . a series of experimental studies and their interpretation , beginning 1972 and spanning the period of nearly 30 years , was undertaken by the group led by packard and williams . in 1972 , packard and sanders@xcite showed the possibility of detecting individual vortex lines in rotating @xmath0he . the experiments were performed in a cylindrical container whose rotation slowly accelerated so that the number of vortex lines present increased with time . the experimental arrangements were similar to those described above,.i.e . the charge trapped on vortex lines was measured by applying the axial electric field thus transferring the charge to a collector attached to an electrometer . the appearance of each new vortex line was detected by a steplike increase in the electrometer s reading . bringing the container slowly to rest , the authors also discovered the existence of remanent vorticity studied later by awschalom and schwarz ( see below ) . later williams and packard@xcite developed the first photographic technique which directly visualized spatial positions of individual quantized vortex lines in rotating @xmath0he . the experimental arrangements were similar to those of ref . @xcite , but in addition the magnetic focusing was used to stabilize beams of electrons emitted from the charged vortex lines ; these beams impinged on a phosphor screen so that the positions of quantized vortices could be actually photographed . experiments were performed at temperatures lower than @xmath139 , and up to 0.8% of @xmath140he ( which acted as a fixed amount of normal fluid ) was added to stabilize the vortices . among other results , it was found , in agreement with the classical feynman s prediction , that the average intervortex spacing is @xmath141 , where @xmath142 is the angular speed of rotation , but in this work an expected stable triangular lattice of vortices was never observed . the photographs of regular , symmetric arrays of vortices were obtained when this experiment was repeated some years later@xcite . ( these photographs should be very familiar to a reader through numerous reproductions in other publications and hence are not shown here . ) the authors ascribed to mechanical disturbances their earlier failure to observe the regular structures . they also argued that the symmetric state of the system is determined not by the absolute minimum of the free energy , as has been believed earlier , but by the combination of the past history of the system and the local minimum of the free energy , concluding that the symmetric state is highly metastable . the detailed description of the photographic technique for visualizing the positions of quantized vortices in rotating @xmath0he was given in work@xcite . much later one of the authors claimed@xcite that usual helium ions are not really suitable for detection of thermally excited vortices , and that multielectron bubbles should be employed for this purpose . also such an experiment was outlined in the cited paper , the authors of this review are not aware of any practical development in this direction . using the experimental technique of ref . @xcite , this research group also revisited@xcite the problem , analyzed earlier theoretically by donnelly , roberts , and parks@xcite , of the lifetime of ions trapped on the vortex lines . theoretical analysis predicted that in the temperature interval @xmath143 the mean lifetime should be longer than @xmath144 and that it should increase with decreasing temperature . however , the experiments revealed that the observed lifetime was many orders of magnitude smaller and was actually decreasing as temperature decreased . to resolve the contradiction with the theoretical predictions , the authors suggested that at temperatures lower than @xmath145 the trapping lifetime is no longer determined by the intrinsic properties of bubble - vortex interaction but by a time scale of the vortex motion until it is destroyed by the container s wall . the details of this mechanism were suggested by the authors as follows : as the moving vortex filament encounters the wall , it is destroyed ( so that the charge is collected by the wall ) , but , to maintain the equilibrium value of the vortex line density , another , uncharged filament is created . in the case where the time scale of migration of the vortex filament to the wall is smaller than the intrinsic trapping lifetime of the ion bubble , the vortex motion will be a limiting factor of the charge loss . the authors concluded that at low temperatures such that the normal fluid density and hence viscous damping ( by the mutual friction ) become negligible , the observed lifetime is actually the measure of the timescale of the vortex migration to the wall and , therefore , should be independent of temperature , in agreement with the authors observations . furthermore , the authors argued that their conclusions can also be related to the earlier observations by cheng , cromar , and donnelly@xcite that the ion trapping reduces in the presence of the axial heat current . the explanation suggested in ref . @xcite is that in the counterflow turbulence the mutual friction increases the intensity of the vortex motion and hence the rate of destruction of vortices by the wall . employing the ion trapping technique , one of the works of key importance for understanding the mechanism of the onset of quantum turbulence is that of awschalom and schwarz@xcite . this experiment used two parallel plates , immersed in @xmath0he , one of which serves as a charge collector detecting the vortices pinned to both plates . ( the experiment aimed , in particular , to support the earlier schwarz s idea@xcite , based on his analysys of vortex reconnections , that under some conditions vortex singularities can multiply . ) the experiment showed that quiescent @xmath0he contains a rather large number of quantized vortex filaments which are pinned metastably to the parallel plates . the line density of these remanent vortices seems to be history independent , and its existence was established upon going , in temperature , down through the @xmath9 transition . ( note that in this work the vortex line density was not measured directly but estimated , considering the rather complicated geometry of electric field lines , from the measurement of the collected charge . ) awschalom and schwarz found that the line density of remanent vortices is very close to the critical line density , introduced by tough@xcite , below which vortices disappear . the authors interpretation of the critical line density was that the vortex filament do not disappear but become immobilized by pinning to the walls and hence can not be observed by any conventional experimental technique . the authors also justified their ideas by analyzing the evolution of the vortex line density in decaying turbulence produced by ultrasound . they observed that , after the ultrasound has been switched off , @xmath2 decays not to zero but to the value of the remanent vortex line density . ( earlier experiments of milliken schwarz , and smith@xcite produced qualitatively similar results . ) the authors concluded that any volume `` will be penetrated _ ab initio _ by quantized vortices stabilized by surface pinning '' , and hence all experiments on the evolution of the vortex line density should be interpreted as if @xmath2 starts from the value corresponding to the remanent vortex line density . the first attempt of systematic experimental study , by means of the ion trapping technique , of the decay of quantum turbulence was made by davis , hendry , and mcclintock@xcite . in turbulence , generated by the oscillating grid , some ions emitted from the tip get trapped on vortices within the tangle and hence reduce the current arriving at the collector . the evolution of the vortex line density can be estimated from the evolution of this current , and the experiment clearly demonstrated the production and decay of quantum turbulence as well as its spatial distribution . however , this experiment had certain disadvantages . in particular , the value of the vortex line density in the low temperature limit ( from 200 down to @xmath146 ) remained unknown , and the data on trapping cross - sections at these temperature had not yet been available . nevertheless , thies experiment yielded a very useful information on the time scale of decay , and also showed that in the temperature range from 22 to @xmath147 the process becomes temperature independent . clearly , a modification of the classical ion trapping experimental arrangements was required that would make possible , in the low temperature limit @xmath148 , to measure directly the dynamics of the vortex tangle as well as the trapping cross - sections . such a new technique was recently developed by walmsley , golov and their co - workers@xcite ; the detailed description of the experimental cell and methods can be found in the first two of the cited works . the experimental cell , mounted on a rotating cryostat , is a cube whose sides are electrodes serving as the charge collectors . in the earlier experiment@xcite just one ion emission tip was fitted at the bottom side of the cell , while later experiment@xcite used two emission tips fitted at the bottom and one of the side plates . in order to provide the means of measuring the spatial properties of the vortex tangle , a difference of electric potential was kept between some of the electrodes to ensure depletion of the ion current emitted by the tips . an analysis of ion trapping by quantized vortices suggests the exponential decay of the ion current , collected by the electrodes , with the vortex line density , so that the latter can be recovered from the measurements of the current @xmath149 from the relation @xmath150 $ ] , where @xmath151 is the size of the experimental cell . the turbulence was produced by spin up or / and spin down of the rotating experimental cell . in ref . @xcite the authors argued in favor of using , as probe particles , the charged vortex rings rather than bare ions . the reason is that the trapping cross - section of the bare ion bubble is small , @xmath152@xcite , and was never measured at temperature below @xmath153 . on the other hand , at low temperatures , in the case where the vortex nucleation velocity , @xmath154 is lower than landau critical velocity , @xmath155 , the ion , upon reaching the velocity @xmath154 will nucleate the vortex ring that captures the ion thus forming the stable ion - ring complex . the details of this mechanism can be found in the classical paper by rayfield and reif@xcite ( see also the work by nancolas , bowley , and mcclintock@xcite and donnely s monographs@xcite ) . while attempting , initially , to repeat the experimental conditions of the work by davis , hendry , and mcclintock@xcite , who tried to eliminate nucleation of vortex rings by ions , walmsley , golov and their co - authors eventually came to a conclusion that at low temperatures charged rings are preferable as proble particles . the trapping cross - section of the charged ring is of the order of its diameter , @xmath156@xcite , which is much larger than the trapping cross - section of the bare ion bubble , and hence it is considerably easier to detect quantized vortices by charged rings ( as was stated much earlier by guenin and hess@xcite ) . some disadvantage of using charged vortex rings is their dynamics which is more complicated than that of bare ion bubbles ( we refer the reader to the theoretical studies@xcite ) . the experimental studies@xcite of walmsley _ _ showed that using charged rings rather than bare ions it is possible to detect the change in the vortex line density upon starting and stopping rotation at temperatures from 0.5 down to @xmath157 . monitoring in ref . @xcite the trapping of charged rings by rectilinear vortices generated by slow rotation the authors were able to measure the ring - vortex trapping cross - section at these temperatures . furthermore , at low temperatures injection of charged rings can be used to create a vortex tangle@xcite . an application of this technique to the study of quantum turbulence in the limit @xmath61 has already brought some non - trivial results . thus , walmsley _ _ et al.__@xcite studied , at temperatures from 1.6 down to @xmath158 , a decay of the homogeneous vortex tangle produced by a sudden spin down of the rotating cell . they found that the decay of the energy flux , @xmath28 is the same as that of the classical turbulence at high reynolds numbers , i.e. @xmath159 , and that the vortex line density decays as @xmath160 . most importantly , it was found that at @xmath161 the effective kinematic viscosity , @xmath162 drops sharply from the value @xmath163 and approaches @xmath164 as @xmath61 . the authors linked this drop to the transition to the new form of turbulence , the so - called `` ultraquantum '' ( or `` vinen '' as opposed to the classical , `` kolmogorov '' ) turbulence . this , less structured than ( quasi)classical regime of turbulence is characterized by the cascade in which the energy is transferred to smaller scales by kelvin waves on individual vortex lines ( and , therefore , by the absence of any large scale motion ) . the mechanism of energy dissipation is though to be acoustic : smallest perturbations are emitted as phonons . clearly , the ultraquantum regime should be characterized by an effective kinematic viscosity , @xmath165 different from the quasiclassical @xmath162 . it should be emphasized , though , that the regime of turbulence in the cited work@xcite was essentially quasiclassical due to forcing at large scales ( rotation ) . the existence and properties of ultraquantum regime at temperatures below @xmath166 were addressed a year later in the recent work of walmsley and golov@xcite . the vortex tangle with was produced not by spin down of the rotating scale but by charged vortex rings so that no large - scale flow was generated . it was found that in the ultraquantum regime the vortex line density decays as @xmath167 , and that the effective kinematic viscosity of the vinen turbulence is @xmath168 . in conclusion of this section we would like to emphasize that the considered so far two experimental techniques of visualization ( or , at least , detection ) of flow properties in turbulent @xmath0he , i.e piv / particle tracking and the ion trapping technique do not compete but rather complement each other . indeed , the ion trapping technique detects quantized vortices without disturbing the flow , and it is efficient in the low temperature regime ; at temperatures above @xmath135 ion bubbles can no longer be trapped on quantized vortices . as strongly emphasized by charalambous _ _ et al.__@xcite , the obvious disadvantage of this technique is that ion bubbles `` can not provide the kind of detailed information about turbulent flow patterns that is available in the case of classical turbulence '' . we fully endorse this statement . actual visualization of quantized vortex filaments is also hardly possible ( but see some new developments discussed below ) . on the contrary , solid , micron - size particles being used in the piv and particle tracking techniques , albeit disturbing the flow , can visualize the normal flow and can even be used to `` paint '' quantized vortices . however , as was demonstrated in secs . [ instability ] and [ tto0 ] , in the low temperature regime , as @xmath61 solid particles do not follow the flow , and , as was shown in sec . [ temperature ] , most likely can not be trapped on quantized vortices . in the remainder of this chapter we will briefly discuss two new promising techniques that potentially can be used for direct flow visualization . the first of these methods is being developed at brown university by maris and his colleagues@xcite . the idea of the method comes from the fact that an application of a negative pressure to @xmath0he leads to an increase of the radius of the ion bubble which explodes , and its size grows substantially when the negative pressure has reached the critical value , @xmath169 . furthermore , maris _ et al . _ argue that since the pressure around the quantized vortex is reduced due to the bernoulli effect , the ion bubbles trapped on the vortex have the size larger than those in the bulk of @xmath0he and explode at the critical pressure of smaller magnitude ( hence the possibility of direct distinguishing the bubbles trapped on vortices from those in the bulk of helium ) . in these experiments the pressure variations were produced by an acoustic , ultrasonic transducer generating either focused or planar ( as in the last of cited works ) sound wave . by choosing a suitable frequency and amplitude of the emitted sound , bubbles can be make so large that they can be visualized by the conventional optical methods ( e.g. by a laser beam and a photomultiplier ) . the other advantage of this method is that it uses neither emission of the electron beam nor the electric field to accelerate the ions . the charged particles are those that already present in the bulk of @xmath0he , either as a result of direct ionization by e.g. cosmic rays , or through the more complicated process involving ionization , production of uv photons , and ejection of electrons from the walls into helium by the photoelectric effect . the absence of the applied electric field means that electron bubbles , unless they are trapped on quantized vortices , will faithfully follow the normal flow . indeed , the results of visualization showed that in the weak , laminar counterflow most of the bubbles moved along the streamlines of the normal fluid , but alongside those about 10% of the observed electron bubbles followed snakelike paths ; these could be assumed the electron bubbles trapped by vortices and sliding along vortex filaments . so far no new information on quantum turbulence has been obtained using this technique , but its relative simplicity has a potential to make it a tool competing with the piv and particle tracking techniques . another , completely new visualization technique is currently being developed by mckinsey and his co - workers@xcite at yale university . this technique employs metastable @xmath4 triplet molecules , produced e.g. by the radioactive source , which can be excited by two infrared photons from the ground @xmath170 state to the @xmath171 . about 90% of the excited molecules will decay into @xmath172 state emitting photons at @xmath173 well separated in the wavelength from the excitation photons and hence can be detected by standard techniques . all excited molecules decay back to the ground state within about @xmath174 , so that the process can be repeated many times to make possible a detection of a single molecule . the authors claim that @xmath4 molecules should be unaffected by quantized vortices at temperatures above @xmath175 and so allow the resolution of the normal flow even at the kolmogorov scale . however , vinen commented@xcite that at sufficiently low temperatures triplet molecules may as well become trapped on quantized vortex core . in this case three - dimensional images of vortex lines can be produced by e.g. stereoscopic imaging . if successful , the further development of this method may be useful for recovering the velocity field and probability distribution function of the normal velocity fluctuations , and can also provide a new , nonintrusive tool for visualization of quantized vortices . this article reviewed the recent progress in understanding the motion of solid particles and mechanisms of their interactions with both the normal fluid and quantized vortices in turbulent @xmath0he . this problem is addressed in the context of the piv and the particle tracking techniques recently implemented in @xmath0he , although we were certainly biased ( perhaps , not surprisingly , considered our own research expertise ) towards theoretical and numerical works and interpretation of experimental results . the problem of the interaction of particle tracers with the normal fluid and with quantized vortices is difficult , at least in principle : it is three - dimensional , time dependent and strongly nonlinear . this is why the best model available ( the two - way coupling model described in section 3 ) is computationally very expensive . in this review we have shown that , fortunately , some aspects of this particle - vortex interaction can be understood using simpler models , either numerical or analytical , particularly given the insight gained from the two - way coupling model . although implemented recently ( the first publications date 2002 ) , the piv and the particle tracking techniques applied to the problem of quantum turbulence in @xmath0he had already produced some new results . among them : the discovery of large - scale eddies , which do not have an analogue in the classical fluid dynamics , in the thermal counterflow past a cylinder ; the wider and flatter , compared to classical , velocity profile in the turbulent boundary layer ; scaling for the time evolution of the parameter characterizing reconnections of quantized vortices ; non - gaussian probability distribution function , with the tail scaled as @xmath115 , of the velocity i decaying quantum turbulence . these results seem to suggest that both the piv and the particle tracking techniques have great potential for studying quantum turbulence . besides , the motion of solid particles in turbulent @xmath0he seems to be itself an interesting and non - trivial phenomenon worth the detailed theoretical and experimental study . although the review of this rapidly development area of research can not be comprehensive , we attempted to resolve , where possible , contradictions in theoretical interpretations of recent experimental observations and to address yet unresolved issues . we also reviewed the related experimental methods , first of all the ion trapping technique which is perhaps the main tool for studying quantized vortices in the low temperature limit . we argue that these two experimental methods , i.e. the piv / particle tracking and the ion trapping techniques do not compete but complement each other . indeed , each of these techniques is most efficient in its own temperature and flow regime . we are grateful to our collaborators , w.f . vinen and d. kivotides who contributed to some of the original research reviewed in this article , and to k.r . sreenivasan for permission to use , in sec . [ mutual_friction ] , the results of yet unpublished experimental results of bewley , lathrop , sreenivasan , and paoletti . g. careri , f. scaramuzzi , and j.o . thomson , _ nuovo cimento _ , * 13 * , 186 ( 1959 ) , and * 18 * , 957 ( 1960 ) ; g. careri , in c.j . gorter , editor . _ progress in low temperature physics_. interscience publishers , new york , vol . 3 , p. 58 , 1961 ; g. careri , w.d . mccormick , and f. scaramuzzi , _ phys . _ , * 1 * , 61 , ( 1962 ) . r.a . ferrel , _ phys . _ , * 108 * , 167 , ( 1957 ) ; c.g . kuper , _ phys . _ , * 122 * , 1007 , ( 1961 ) ; j. jortner , n.r . kestner , s.a . rice , and m.h . cohen , _ j. chem _ , * 43 * , 2614 , ( 1965 ) ; k. hiroike , n.r . kestner , s.a . rice , and j. jortner , _ _ , * 43 * , 2625 , ( 1965 ) . kuper , in g. careri , editor . _ liquid helium _ , proc . enrico fermi int . school of physics . academic press , new york , 1963 , course xxi ; a.j . dahm and t.m . sanders , _ phys . _ , * 17 * , 126 , ( 1966 ) and _ j. low temp _ , * 2 * , 199 , ( 1970 ) . donnelly , _ _ , * 14 * , 39 , ( 1965 ) ; r.j . donnelly and p.h . roberts , in j.f . allen , editor . _ superfluid helium_. academic press , new york , 1966 , and proc . ( london ) a , * 312 * , 519 ( 1969 ) . p.m. walmsley , a.i . golov , h.e . hall , a.a . levchenko , and w.f . vinen , _ phys . _ , * 99 * , 265302 , ( 2007 ) ; p.m. walmsley , a.i . golov , h.e . hall , w.f . vinen , and a.a . levchenko , _ j. low temp . _ , * 153 * , 127 , ( 2008 ) . barenghi , r.j . donnelly , and w.f . vinen , _ j. low temp . _ , * 52 * , 189 , ( 1983 ) ; d.c . samuels and r.j . donnelly , _ phys . rev . _ , * 67 * , 2505 , ( 1991 ) ; m. tsubota and h. adachi , _ j. low temp . phys . _ , in print ( 2009 ) . j. classen , c .- k . su , m. mohazzab , and h.j . maris , _ phys . b _ , * 57 * , 3000 , ( 1998 ) and _ j. low temp . _ , * 110 * , 431 , ( 1998 ) ; a. ghosh and h.j . maris , _ phys . _ , * 95 * , 265301 , ( 2005 ) ; w. guo and h.j . maris , _ j. low temp . phys . _ , * 148 * , 199 , ( 2007 ) .

nonradial pulsations in the primary white dwarfs of cataclysmic variables can now potentially allow us to explore the stellar interior of these accretors using stellar seismology . in this context , we conducted a multi - site campaign on the accreting pulsator sdssj161033.64 - 010223.3 ( v386 ser ) using seven observatories located around the world in may 2007 over a duration of 11 days . we report the best fit periodicities here , which were also previously observed in 2004 , suggesting their underlying stability . although we did not uncover a sufficient number of independent pulsation modes for a unique seismological fit , our campaign revealed that the dominant pulsation mode at 609s is an evenly spaced triplet . the even nature of the triplet is suggestive of rotational splitting , implying an enigmatic rotation period of about 4.8 days . there are two viable alternatives assuming the triplet is real : either the period of 4.8 days is representative of the rotation period of the entire star with implications for the angular momentum evolution of these systems , or it is perhaps an indication of differential rotation with a fast rotating exterior and slow rotation deeper in the star . investigating the possibility that a changing period could mimic a triplet suggests that this scenario is improbable , but not impossible . using time - series spectra acquired in may 2009 , we determine the orbital period of sdssj161033.64 - 010223.3 to be 83.8.9min . three of the observed photometric frequencies from our may 2007 campaign appear to be linear combinations of the 609s pulsation mode with the first harmonic of the orbital period at 41.5min . this is the first discovery of a linear combination between nonradial pulsation and orbital motion for a variable white dwarf .

1003.5703

cataclysmic variables are interacting binary systems in which a late - type star loses mass to an accreting white dwarf . photometric variations consistent with nonradial g - mode pulsations were first discovered in the cataclysmic variable gw librae in 1998 @xcite ; such pulsations had previously been observed only among the non - interacting white dwarf stars . this discovery has opened a new venue of opportunity to learn about the stellar parameters of accreting variable white dwarfs using asteroseismic techniques ( e.g. * ? ? ? * ) . a unique model fit to the observed periods of the variable white dwarf can reveal information about the stellar mass , core composition , age , rotation rate , magnetic field strength , and distance ( see the review papers * ? ? ? * ; * ? ? ? * ; * ? ? ? there are now thirteen accreting pulsating white dwarfs known ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite are pioneering the effort to empirically establish the pulsational instability strip for accretors and to test the theoretical framework laid down by @xcite . the instability strip(s ) for these pulsators has to be established separately from the zz ceti strip @xcite , and are also known as the zz ceti stars . ] because accretion enriches their envelopes with he and metals . this is distinct from the pure h envelope of the non - interacting da white dwarfs , where h ionization causes them to pulsate as zz ceti stars . @xcite find a h / hei instability strip for accreting model white dwarfs with a blue edge near 12000k for a 0.6@xmath1star , similar to the zz ceti instability strip . they also find an additional hotter instability strip at @xmath215000k due to heii ionization for accreting model white dwarfs with a high he abundance ( @xmath30.38 ) . the spectrum of an accreting pulsator includes prominent broad absorption lines from the white dwarf as well as the central emission features from the accretion disk . when the orbital period of a cataclysmic variable is @xmath480 - 90min , it is near the evolutionary orbital period minimum , where the rate of mass transfer is theoretically expected to be the smallest @xmath5@xmath1/yr @xcite . due to the low rates of mass transfer , the white dwarf is expected to be the source of 90% of the optical light observed from these systems @xcite . this makes it possible to detect white dwarf pulsations in these cataclysmic variables . accreting pulsators have probably undergone a few billion years of accretion and thousands of thermonuclear runaways . studying these systems will allow us to address the following questions : to what extent does accretion affect the white dwarf mass , temperature , and composition and how efficiently is angular momentum transferred to the core of the white dwarf . these systems are also crucial in understanding the above effects of accretion on pulsations . asteroseismology can allow us to obtain meaningful mass constraints for the pulsating primary white dwarfs of cataclysmic variables . previously any such constraints on the mass of the accreting white dwarf could only be established for eclipsing cataclysmic variables ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? constraining the population , mass distribution , and evolution of accreting white dwarfs is also important for studying supernovae type ia systematics . for example , @xcite show empirically that the maximum mass of white dwarf progenitors has to be at least 7.1@xmath1 thus constraining the lower mass limit for supernovae progenitors . @xcite deduced that sdssj161033.64 - 010223.3 ( v386 ser ; hereafter sdss1610 - 0102 ) is a cataclysmic variable from early sdss spectroscopic observations @xcite ; sdss has single - handedly led to a substantial increase in the number of known cataclysmic variables . subsequently @xcite discovered photometric variations in the light curve of sdss1610 - 0102 consistent with non - radial pulsations . they determined two independent pulsation modes with periods near 607s and 345s , also finding their harmonics and linear combinations in the data . noting the amplitude modulation in the light curves , @xcite concluded that the dominant mode was a multiplet , but were unable to resolve it . they also determined the orbital period of 80.52min from the observed double - humped modulation in their light curves . we chose to target sdss1610 - 0102for a multi - site asteroseismic campaign from eleven possibilities known then for the following reasons . previous observations of the system had revealed two independent pulsation frequencies , while several similar systems show no more than one pulsation frequency . each independent pulsation frequency serves as a constraint on the stellar structure ; detecting a larger number of frequencies is essential in obtaining a unique seismological fit . sdss1610 - 0102 has an equatorial declination , making it easily accessible to observatories in both northern and southern hemispheres . the objective of the multi - site participation is to keep the sun from rising on the target star @xcite ; reducing the gaps in the stellar data due to daytime increases the contrast between the true frequency and its aliases , thus making multi - site observations more effective than single - site observations . we acquired optical time - series photometry on the accreting white dwarf pulsator sdss1610 - 0102 over a duration of 11 days in may 2007 using multiple telescopes . our multi - site campaign involved using the prime focus time - series photometer argos @xcite on the 2.1 m otto struve telescope at mcdonald observatory ( mo ) , and the time - series photometer agile @xcite on the 3.5 m telescope at apache point observatory ( apo ) . both instruments are frame transfer ccd cameras devoid of mechanical shutters , where the end of an exposure and the beginning of a new exposure is triggered directly by the negative edges of gps - synchronized pulses without any intervention from the data acquisition software . there is also no dead time between consecutive exposures from ccd read times , making argos and agile ideal instrumentation for the study of variable phenomena with millisecond timing accuracy . argos and agile consist of back - illuminated ccds with an enhanced back - thinning process for higher blue quantum efficiency . additionally , the e2v ccd47 - 20 in agile has an ultra - violet coating to enhance the wavelength efficiency of the region 200370 nm to 35% . as apache point observatory is located at an altitude of 2788 m , we expect to detect at least some of the blue photons in the range of 320370 nm . we utilized the calar alto faint object spectrograph ( cafos ) in imaging mode on the 2.2 m telescope at calar alto observatory ( cao ) during the campaign . the duration between the fits time stamp and the actual opening of the shutter is expected to be a fraction of a second . network time protocol ( ntp ) is used to discipline the data acquisition computer , and we expect that the uncertainty in timing for a cafos image is of the order of 0.5s . the site ccd in cafos has a quantum efficiency greater than 80% in the wavelength range of 370780 nm . data was acquired on the robotic 2.0 m faulkes telescope north ( ftn ) at haleakala , hawaii , using the instrument hawkcam1 ; this telescope belongs to the las cumbres observatory global telescope ( lcogt ) network . the data acquisition computer for hawkcam1 is synchronized to a gps time server . the delay between the utstart timestamp in the fits images and the actual opening of the mechanical shutter is expected to be of the order of a few milliseconds . this instrument contains an e2v ccd42 - 40 , which is thinned and back - illuminated for blue sensitivity . the camera salticam @xcite was used on the effective 10 m south african large telescope ( salt ) at the south african astronomical observatory ( saao ) to acquire data for our campaign . the frame transfer time for this large format 2k@xmath64k ccd is 0.1s ; the uncertainty in timing should be significantly smaller than a tenth of a second . the ccds used in salticam are thinned , back - illuminated , and made from deep depletion silicon which provides less fringing and additional sensitivity in the near infrared without photon loss in the blue and ultra - violet . salticam has an efficiency greater than 80% in the wavelength range of 320940 nm . including atmospheric extinction and reflectivity losses at the different mirrors , salticam has an efficiency greater than 60% from 380 to 840 nm . we observed on the 1.5 m telescope of the observatorio astrnomico nacional in san pedro martir ( oan - spm ) using the instrument ruca @xcite . ruca comprises of the detector site1 si003 ccd with a quantum efficiency greater than 55% in the wavelength range 500800 nm . on the 2.5 m irne du pont telescope at las campanas observatory ( lco ) , we used the direct ccd camera ( _ ccd _ ) , which includes a 2k@xmath62k tek ccd with a quantum efficiency greater than 70% in the wavelength range of 400700 nm . table [ jour - obs ] gives the journal of observations , and figure [ xcov ] shows the extent of our coverage during the campaign . llllllll apo 3.5 m & agile & asm & 19 may 2007 08:17:03 & 08:55:33 & 77 & 30 & bg40 + apo 3.5 m & agile & asm & 21 may 2007 08:19:36.1 & 11:09:36.1 & 340 & 30 & bg40 + cao 2.2 m & cafos & aa & 15 may 2007 00:05:49.0 & 01:29:47.5 & 56 & 75 & roeser bv + cao 2.2 m & cafos & aa & 15 may 2007 21:05:20.4 & 03:37:49.7 & 576 & 25 & none + cao 2.2 m & cafos & aa & 16 may 2007 03:38:06.1 & 03:58:22.7 & 27 & 30 & none + cao 2.2 m & cafos & aa & 16 may 2007 20:43:01.2 & 21:01:40.1 & 22 & 35 & none + cao 2.2 m & cafos & aa & 16 may 2007 21:01:56.2 & 21:19:09.4 & 23 & 30 & none + cao 2.2 m & cafos & aa & 16 may 2007 21:19:24.7 & 04:02:19.0 & 598 & 25 & none + cao 2.2 m & cafos & aa & 17 may 2007 22:18:01.6 & 03:57:36.6 & 502 & 25 & none + cao 2.2 m & cafos & aa & 18 may 2007 22:36:00.1 & 03:56:11.4 & 441 & 25 & none + cao 2.2 m & cafos & aa & 19 may 2007 22:15:41.6 & 22:28:50.6 & 20 & 25 & none + cao 2.2 m & cafos & aa & 19 may 2007 22:29:06.5 & 22:37:54.4 & 11 & 30 & none + cao 2.2 m & cafos & aa & 19 may 2007 22:40:37.8 & 00:12:00.4 & 62 & 35 & none + cao 2.2 m & cafos & aa & 20 may 2007 00:13:11.5 & 00:44:15.4 & 28 & 40 & none + cao 2.2 m & cafos & aa & 20 may 2007 00:44:30.9 & 02:06:20.6 & 43 & 35 & none + cao 2.2 m & cafos & aa & 20 may 2007 21:45:26.1 & 23:34:01.6 & 122 & 35 & none + cao 2.2 m & cafos & aa & 20 may 2007 23:34:17.7 & 00:54:14.4 & 77 & 30 & none + cao 2.2 m & cafos & aa & 21 may 2007 00:54:29.8 & 01:02:21.7 & 11 & 25 & none + cao 2.2 m & cafos & aa & 21 may 2007 01:02:39.7 & 01:06:57.9 & 6 & 30 & none + cao 2.2 m & cafos & aa & 21 may 2007 01:07:16.4 & 02:37:28.3 & 88 & 35 & none + cao 2.2 m & cafos & aa & 21 may 2007 02:37:44.6 & 03:29:59.7 & 63 & 30 & none + cao 2.2 m & cafos & aa & 21 may 2007 03:30:15.0 & 03:40:49.2 & 16 & 25 & none + cao 2.2 m & cafos & aa & 21 may 2007 03:41:12.2 & 03:51:36.1 & 14 & 30 & none + lco 2.5 m & _ ccd _ & mrs , as & 17 may 2007 01:39:26.4 & 03:14:38.1 & 101 & 40 & v + lco 2.5 m & _ ccd _ & mrs , as & 17 may 2007 03:15:51.3 & 04:01:04.7 & 61 & 30 & v + lco 2.5 m & _ ccd _ & mrs , as & 17 may 2007 04:01:45.4 & 04:31:19.0 & 46 & 25 & v + lco 2.5 m & _ ccd _ & mrs , as & 17 may 2007 04:34:20.4 & 05:42:11.8 & 86 & 35 & v + lco 2.5 m & _ ccd _ & mrs , as & 17 may 2007 05:46:22.7 & 06:10:28.1 & 25 & 45 & v + lco 2.5 m & _ ccd _ & mrs , as & 18 may 2007 01:05:50.7 & 02:20:54.2 & 101 & 30 & v + lco 2.5 m & _ ccd _ & mrs , as & 18 may 2007 02:21:43.3 & 05:32:14.0 & 293 & 25 & v + lco 2.5 m & _ ccd _ & mrs , as & 18 may 2007 05:32:50.8 & 09:05:45.5 & 362 & 20 & v + lco 2.5 m & _ ccd _ & mrs , as & 18 may 2007 09:06:06.7 & 09:25:37.6 & 31 & 25 & v + lco 2.5 m & _ ccd _ & mrs , as & 18 may 2007 09:26:11.3 & 09:51:10.2 & 35 & 30 & v + lco 2.5 m & _ ccd _ & mrs , as & 19 may 2007 01:16:31.5 & 08:32:53.0 & 650 & 25 & v + lco 2.5 m & _ ccd _ & mrs , as & 20 may 2007 01:02:52.3 & 10:11:13.3 & 843 & 25 & v + lco 2.5 m & _ ccd _ & mrs , as & 21 may 2007 01:59:23.6 & 02:42:51.1 & 62 & 28 & v + lco 2.5 m & _ ccd _ & mrs , as & 21 may 2007 07:30:19.0 & 09:48:57.5 & 190 & 25 & v + lcogt 2.0 m & hawkcam1 & robotic & 15 may 2007 09:28:33.5 & 14:30:07.2 & 261 & 60 & v + lcogt 2.0 m & hawkcam1 & robotic & 16 may 2007 09:42:41.3 & 14:29:40.8 & 244 & 60 & v + lcogt 2.0 m & hawkcam1 & robotic & 17 may 2007 08:49:59.3 & 14:33:50.9 & 291 & 60 & v + lcogt 2.0 m & hawkcam1 & robotic & 18 may 2007 09:53:09.2 & 12:03:08.0 & 113 & 60 & v + lcogt 2.0 m & hawkcam1 & robotic & 18 may 2007 13:32:21.9 & 14:15:57.6 & 38 & 60 & v + lcogt 2.0 m & hawkcam1 & robotic & 19 may 2007 09:17:22.6 & 14:10:54.9 & 195 & 60 & v + mo 2.1 m & argos & asm , bf & 12 may 2007 04:29:30.0 & 11:17:10.0 & 1223 & 20 & bg40 + mo 2.1 m & argos & asm , bf & 13 may 2007 04:07:39.0 & 11:10:19.0 & 1268 & 20 & bg40 + mo 2.1 m & argos & asm , bf & 14 may 2007 04:12:44.0 & 11:19:04.0 & 1279 & 20 & bg40 + mo 2.1 m & argos & asm , bf & 15 may 2007 08:24:49.0 & 11:12:49.0 & 504 & 20 & bg40 + mo 2.1 m & argos & asm , bf & 21 may 2007 08:58:08.0 & 10:58:08.0 & 240 & 30 & bg40 + mo 2.1 m & argos & asm , bf & 22 may 2007 04:09:51.0 & 11:00:39.0 & 1218 & 20 & bg40 + oan - spm 1.5 m & ruca & gt , svz & 16 may 2007 03:59:53 & 06:15:32 & 200 & 35 & bg40 + oan - spm 1.5 m & ruca & gt , svz & 16 may 2007 06:17:15 & 11:48:51 & 581 & 25 & bg40 + oan - spm 1.5 m & ruca & gt , svz & 17 may 2007 04:09:31 & 11:49:48 & 950 & 20 & bg40 + oan - spm 1.5 m & ruca & gt , svz & 18 may 2007 04:12:23 & 11:47:23 & 799 & 25 & bg40 + oan - spm 1.5 m & ruca & gt , svz & 19 may 2007 04:20:45 & 11:45:23 & 850 & 25 & bg40 + oan - spm 1.5 m & ruca & gt , svz & 20 may 2007 03:56:09 & 11:40:19 & 892 & 25 & bg40 + oan - spm 1.5 m & ruca & gt , svz & 21 may 2007 04:01:27 & 04:47:46 & 50 & 50 & bg40 + oan - spm 1.5 m & ruca & gt , svz & 21 may 2007 04:50:12 & 10:39:22 & 500 & 35 & bg40 + saao 10 m & salticam & rs & 15 may 2007 21:36:18.6 & 00:23:41.3 & 605 & 3 & b - s1 + saao 10 m & salticam & rs & 17 may 2007 23:49:51.7 & 01:15:55.5 & 287 & 10 & b - s1 + we used a standard iraf reduction to extract sky - subtracted light curves from the ccd frames using weighted circular aperture photometry @xcite . after extracting the light curves , we divided the light curve of the target star with a sum of one or more comparison stars using brighter stars for the division whenever available as opposed to faint stars ; this choice leads to comparatively lower noise in the target star light curve after the process of division . after this preliminary reduction , we brought the data to a fractional amplitude scale ( @xmath7 ) and converted the mid - exposure times of the ccd images to barycentric coordinated time ( tcb ; * ? ? ? the first data point of the campaign was used to define a reference zero time of 2454232.693409 tcb , which we applied to all the reduced light curves . we then computed a discrete fourier transform ( dft ) for all the individual observing runs up to the nyquist frequency . all stellar pulsators reveal wavelength - dependent amplitudes whenever their atmospheres suffer from limb - darkening effects ; this is also observed to be true for white dwarf pulsators @xcite . @xcite have measured the broadband amplitudes of the pulsation frequencies in sdss1610 - 0102 , confirming the trend of larger amplitudes at bluer wavelengths . including photons redward of 6500 , which are less modulated by the pulsation process , reduces the measured amplitude by as much as 2040% @xcite . based on the pulsation amplitude of the dominant mode , we divided the data from different sites , acquired using different instruments with distinct wavelength responses and different filters , into three groups . the data from the mo 2.1 m , apo 3.5 m , and the saao 10 m telescopes along with the roeser bv data acquired on the 15th of may using the cao 2.2 m telescope constitute group 1 ; these data reveal an amplitude of [email protected] for the 609s mode . the data obtained from the lco 2.5 m , lcogt 2.0 m , and oan - spm 1.5 m telescopes , that form group 2 , yield an amplitude of [email protected] . the calar alto white light data that forms group 3 amounts to an amplitude of [email protected] . these values of amplitude are not consistent with each other within the uncertainties . the pulsation amplitude obtained for sites in group 2 is approximately 15% lower compared to sites in group 1 , while group 3 shows a pulsation amplitude 31% lower than group 1 . hence to even out the 1531% amplitude differences between the various data segments , we scaled the group 2 data by a factor of 1.15 and group 3 data by a factor of 1.31 . without scaling the data , we would effectively be fitting a constant amplitude sinusoid to a multi - site light curve with different embedded amplitudes . after this scaling , the entire 11-day long light curve could be subjected to least squares analysis . blue bandpass filters are ideal to study these hot white dwarf variables . typically we are forced to acquire white light data whenever the photon count rate proves to be inadequate . scaling these noisy light curves to match the pulsation amplitude of the other higher s / n blue observations also implies scaling the noise with the same factor . the drawback of scaling the data is an overall increase in noise ; we do find that the 3@xmath8 level increases from 1.49mma to 1.71mma as a result of scaling . by comparison , the highest amplitude peak in the 609s triplet increases from 26.10mma to 28.78mma due to scaling . we do not have to worry about scaling the other observed frequencies similarly , except the 41.5min period , as they are all low amplitude modes . their amplitudes in the various data segments are already consistent with each other within the uncertainties . we also checked explicitly that the chosen scaling does not alter the results of the least squares analysis for any of the other frequencies , and no artificial frequencies or frequency splittings arise due to this scaling . note that the amplitudes obtained for the 41.5min period for sites in group 1 and group 2 are [email protected] and [email protected] respectively . sites in group 2 show an amplitude approximately 33% lower compared to sites in group 1 . the wavelength dependence of the 41.5min period differs from the 609s mode ; this is consistent with the findings of @xcite . the calar alto white light data ( group 3 ) are too noisy to determine a reliable amplitude for the 41.5min period . this implies that our chosen scaling of 1.15 improves the amplitude differences for the 41.5min period as well , and evens it out to the point that the data segments have an amplitude consistent with each other within uncertainties . all of the subsequent data analysis and the discrete fourier transform ( dft ) shown in figure [ dft ] is based on this scaled multi - site 11-day long light curve obtained during the campaign . although we computed the dft right up to the nyquist frequency of 0.025hz , there were no significant peaks above the 3@xmath8 limit of 1.71mma between 0.005 and 0.025hz . hence the interesting portion of the dft is shown in figure [ dft ] . we determine the amount of white noise in the light curve empirically by using a shuffling technique ( see * ? ? ? all the best - fit frequencies were initially subtracted to obtain a prewhitened light curve . preserving the time column of this light curve , we shuffled the corresponding intensities . this exercise destroys any coherent signal in the light curve while keeping the time sampling intact , leaving behind a shuffled light curve of pure white noise . the average amplitude of a dft computed for the shuffled light curve is close to the 1@xmath8 limit of the white noise . after shuffling the light curve ten times , we average the corresponding values for white noise to determine a reliable 3@xmath8 limit . table [ bfit - per ] indicates the best fit periodicities along with the least squares uncertainties we obtained using the program period04 @xcite . we cross - checked the periods , amplitudes , and phases obtained with our own linear and non - linear least squares code and find that the values obtained from both programs are almost identical . the program period04 has the added advantage of computing monte - carlo uncertainties , which are shown as italicized numbers in table [ bfit - per ] . previous papers on sdss1610 - 0102 did not resolve the triplet @xcite , and hence their determinations of the dominant period differ slightly in value from ours ( see the second to last paragraph in section 7 ) . however the pulsation spectrum in 2007 is essentially similar to that observed in 2004 and 2005 . lllll f@xmath9 & 1641.796@xmath0 0.028 & 609.0893@xmath0 0.010&28.78@xmath0 0.50&596.0@xmath0 2.2 + & _ @xmath0 0.024 _ & _ @xmath0 0.0089 _ & _ @xmath0 0.63 _ & _ @xmath0 2.8 _ + f@xmath10 & 1643.074 @xmath0 0.057 & 608.615@xmath0 0.021 & 8.22@xmath0 0.51 & 255.0@xmath0 6.1 + & _ @xmath0 0.057 _ & _ @xmath0 0.021 _ & _ @xmath0 0.49 _ & _ @xmath0 8.2 _ + f@xmath11 & 1640.665@xmath0 0.071&609.509 @xmath0 0.026 & 7.95 @xmath0 0.66 & 50.3 @xmath0 6.9 + & _ @xmath0 0.082 _ & _ @xmath0 0.031 _ & _ @xmath0 0.79 _ & _ @xmath0 9.1 _ + 2f@xmath12 & 3283.939 @xmath0 0.040 & 304.5124 @xmath0 0.0037 & 7.46@xmath0 0.40 & 132.0 @xmath0 2.7 + & _ @xmath0 0.053 _ & _ @xmath0 0.0049 _ & _ @xmath0 0.56 _ & _ @xmath0 3.3 _ + 2f@xmath12 & 3282.534 @xmath0 0.073 & 304.6427 @xmath0 0.0068 & 4.03 @xmath0 0.40 & 139.9 @xmath0 5.0 + & _ @xmath0 0.089 _ & _ @xmath0 0.0082 _ & _ @xmath0 0.52 _ & _ @xmath0 6.3 _ + f@xmath13 & 1240.517 @xmath0 0.071 & 806.115 @xmath0 0.046 & 3.62 @xmath0 0.40 & 483 @xmath0 14 + & _ @xmath0 10 _ & _ @xmath0 6.5 _ & _ @xmath0 1.4 _ & _ @xmath0 130 _ + f@xmath14 & 4523.14 @xmath0 0.10 & 221.0854 @xmath0 0.0049 & 2.59 @xmath0 0.40 & 201.0 @xmath0 5.4 + & _ @xmath0 29 _ & _ @xmath0 1.4 _ & _ @xmath0 1.2 _ & _ @xmath0 78 _ + f@xmath15 & 2882.582 @xmath0 0.076 & 346.9112 @xmath0 0.0091 & 3.41 @xmath0 0.40 & 91.6 @xmath0 6.5 + & _ @xmath0 23 _ & _ @xmath0 2.8 _ & _ @xmath0 1.5 _ & _ @xmath0 96 _ + f@xmath16 & 401.310 @xmath0 0.053 & 2491.84 @xmath0 0.33 & 4.90 @xmath0 0.40 & 826 @xmath0 32 + & _ @xmath0 0.075 _ & _ @xmath0 0.46 _ & _ @xmath0 0.49 _ & _ @xmath0 45 _ + f@xmath17 & 34.766 @xmath0 0.015 & 28764 @xmath0 13 & 15.76 @xmath0 0.43 & 5860@xmath0 110 + & _ @xmath0 0.023 _ & _ @xmath0 19 _ & _ @xmath0 0.56 _ & _ @xmath0 130 _ + note that we also concatenated all the light curves from various sites without any scaling and computed the corresponding unscaled dft , to find that it showed the same peaks as the dft obtained after scaling the data . additionally , we computed dfts for the unscaled light curves from the first and second groups as defined earlier . these dfts also revealed the same frequency set as shown in figure [ dft ] . these tests show that scaling the data did not introduce any artificial new frequencies in our data set . in order to compute the best - fit periodicities for the unscaled light curve , we subjected it to period04 similar to the procedure carried out for the scaled data . the best - fit values for the scaled and unscaled data are very similar , lending credence to the process of scaling the multi - site data , especially in its ability to preserve the original frequency set without introducing any new artifacts . although three of the seven sites boast millisecond timing accuracy , others like cafos involve a 0.5s timing uncertainty per image . to test the effects of timing uncertainty on our results , we simulated a light curve including all ten frequencies listed in table [ bfit - per ] along with their respective amplitudes , adding in white noise with an amplitude of 2mma . using the same time sampling as the real data , we added in a gaussian uncertainty of @xmath8=2s to * each * point to account for timing errors such as software delays in time - shared data acquisition computers , jitter in the opening of mechanical shutters , etc . this is worse than the worst case possible for badly synchronized clocks . our results show that we are still able to recover the closely spaced frequencies of the 609s triplet from the simulated light curve ; the output best - fit period values were at most 0.003% different from the values used to generate the simulated light curve . amplitudes and phases were affected at the 2% level or less . this result is not surprising given that the timing uncertainty is significantly smaller than the 609s period , and the s / n ratio improves with each passing cycle . we acquired followup time - series spectroscopy on sdss1610 - 0102 on the 28th of may , 2009 , using the dual imaging spectrograph ( dis ) mounted on the apo 3.5 m telescope with intermediate resolution gratings ( resolution about 2 ) and a 1.5 arcsec slit . our goal was not to get a precise determination of the orbital period , but to measure the orbital frequency well enough to identify it or any of its harmonics present in the campaign dft . the spectroscopic data cover a period of just about 3 hours , adding up to a total of sixteen 10-min exposures . individual exposures had a s / n of about 10 for the lines and 57 for the continuum . standard iraf routines were utilized for data reduction and subsequent flux and wavelength calibration . we used a boxcar smoothing over 3 points prior to implementing the centroid - finding routine in the iraf splot package to obtain the equivalent widths and fluxes for the hydrogen balmer lines . fitting a sine curve to the radial velocity curves for each line , we determined the velocity of the center of mass @xmath18 , the semi - amplitude k , the orbital period @xmath19 , and the time of the red - to - blue crossing of the emission lines t@xmath20 ; these values are shown below in table [ spec - porb ] . note that the exposure times constitute a non - negligible part of the orbital period ; this causes some amplitude reduction . the amplitude values listed in table [ spec - porb ] have been corrected for this effect . the top panel of figure [ spec ] shows the complete combined spectrum . lllll h@xmath21 & -23@xmath02 & 96@xmath014 & 410 & 83 + h@xmath22 & 28@xmath03 & 208@xmath023 & 412 & 86 + emission lines from the accretion disk and the hot spot , where the matter stream hits the accretion disk , affect the spectral lines in cataclysmic variables . as a result , the velocity of the center of mass @xmath18 can be different for different balmer lines , and may even have different signs ( see table [ spec - porb ] ) . however the determination of the orbital period from both lines is consistent with each other within uncertainties . we also applied the double gaussian method outlined by @xcite and developed by @xcite on the h@xmath21 line . this method involved the convolution of two gaussian functions of opposite signs and equal fwhm of 300km / s , chosen to be the same order as the wavelength resolution . a peak separation of 1800km / s was used to reach further into the wings without compromising too much on the signal - to - noise ratio . this ensured that the measured radial velocities traced the inner part of the accretion disc , where azimuthally symmetric structure traces the orbital motion of the white dwarf . this technique produced the radial velocity curve ( bottom panel of figure [ spec ] ) that led to an orbital period determination of [email protected] . this spectroscopic determination is consistent with the photometric measurement of [email protected] ( obtained from 1/2f@xmath16 ) within the uncertainties . this orbital period not only indicates that sdss1610 - 0102 is close to the period minimum , but also includes sdss1610 - 0102 within the 8086min period - minimum spike that is largely made up of white dwarf dominated cataclysmic variables @xcite . @xcite observed a double - humped modulation in their light curves , and determined values of [email protected] and [email protected] for the orbital period . using the first harmonic of the orbital period , they deduced additional measurements of [email protected] and [email protected] ; these photometric determinations have lower uncertainties compared to the direct measurements above inspite of their lower amplitude because they involve twice the number of cycles compared to the orbital period itself . large gaps in the combined light curve prevented @xcite from identifying a unique orbital period , and they present the alternatives of 80.52min and 85.08min . computing a weighted average from their individual measurements above , weighted inversely as the uncertainties , we arrive at an orbital period determination of [email protected] . this value is consistent with the orbital period measurement of [email protected] from the h@xmath21 line . we pre - whitened all the frequencies from our 11-day light curve except for the 41.5min periodicity , and folded it on the orbital period . the folded light curve and its running average are shown in figure [ 2hump ] to indicate its apparent double - humped nature . this pulse shape is different from the one shown by @xcite , but it is consistent with the least squares analysis as expected . table [ bfit - per ] reveals that apart from the first harmonic , no other period related to the orbital motion is observed . hence we should expect to see two pulses with a period of about 41.5min at an amplitude of 4.9mma in the folded light curve . the pulse shape from figure [ 2hump ] fits a sine curve of period 41.24min with an amplitude of 4.3mma . several short period cataclysmic variables show similar double - humped light curves @xcite , and in some cases the orbital period has been independently determined using eclipses to confirm that the photometric and spectroscopic periods are actually the same within uncertainties . @xcite proposed that a double humped variation is observed when the bright spot , where the accretion stream hits the white dwarf , is visible even when it is behind the optically thin disk . however @xcite and @xcite argue that the permanent double - humped light curve is an attribute of bounced - backed systems ; these systems consist of cataclysmic variables that have evolved beyond period minimum with quiescent accretion disk radii at a 2:1 resonance . the accreting pulsating white dwarf constitutes a clock in orbit ; the changing light travel time during an orbit is reflected in the pulse arrival times . the observed phase of the clock ( o ) compared to the phase of a stationary clock at the same period ( c ) should show a modulation at the orbital period @xcite . we used the o - c method ( with a small modification ) to attempt a detection of the orbiting motion of the 609s clock . however our uncertainties of 45s were too large to reveal the expected sub - second change in light travel time during the orbiting motion of the white dwarf . each independent pulsation frequency is a constraint on the structure of the star . harmonics and linear combinations in our data merely arise as a result of non - linearities introduced by relatively thick convection zones @xcite . identifying the linearly independent frequencies in a pulsation spectrum is the first step to undertake in seismology . of the several periods shown in figure [ dft ] , we suspect that the longest observed period of 28764s or 8hr is related to airmass or extinction variations or possibly the gaps in the 11-day light curve . other cataclysmic variables have shown photometric or radial velocity periods that are much longer than the orbital period and bear no relation to it . for example , @xcite discuss long photometric periods such as the 2hr modulation in gwlibrae , a possible 3hr period in fs aur , and a 4.6hr period in v2051 oph . @xcite show that the precession of the magnetically accreting white dwarf can successfully explain such long periods . however in this case , we do not believe that the 8hr period is stellar in origin . the 2491.84s or the 41.5min period is clearly the first harmonic of the orbital period , measured spectroscopically to be [email protected] . this identification is consistent with @xcite . the 609s triplet is clearly an independent pulsation mode since it has the highest amplitude of all observed periods ; typically harmonics and linear combinations have lower amplitudes than the parent modes . we can also readily identify the periods close to 304.5s as the first harmonic of the 609s triplet . a triplet can be expected at 304.5s because the harmonic of a triplet should also be a triplet . however we are unable to resolve it , due most likely to the relatively lower amplitudes compared to the 609s triplet . there is also a small chance that the 609s photometric period represents the rotation of the star , and we are misinterpreting it as nonradial pulsation . sdss1610 - 0102 is neither strongly magnetic nor a source of x - rays , and therefore we do not expect to see a rotating hot spot as in the intermediate polars . it is difficult for a hot spot or belt to match the observed pulse shape of the 609s period . if the 609s period was the rotation period of the star , it would also be difficult to explain the evenly spaced triplet , but the triplet could perhaps have been caused by reprocessing or reflection effects . the ratio of the uv to optical pulsation amplitude of the 609s period is consistent with low order g - mode pulsations @xcite . for all of these reasons collectively , nonradial pulsation is certainly the favourable model to explain the 609s photometric triplet . however please note that we are currently unable to absolutely exclude the possibility that the 609s period is the rotation period of the star . for the rest of this paper , we will adopt the more likely model of nonradial pulsation for the 609s triplet . let f@xmath23 , f@xmath13 , f@xmath12 , f@xmath15 , and f@xmath14 denote the frequencies corresponding to the 41.5min , 806s , 609s , 347s , and 221s periods . the linear combinations in our frequency set are : f@xmath14=3f@xmath12-f@xmath23 , f@xmath15=2f@xmath12-f@xmath23 , and f@xmath13=f@xmath12-f@xmath23 . note that the amplitudes of all the proposed combination modes are smaller than the suggested parent modes as expected . justifying the frequency f@xmath14 involves invoking 3f@xmath12 , which has not been observed directly . @xcite identify f@xmath15 as a second independent pulsation mode ; however given the relation f@xmath15=2f@xmath12-f@xmath23 within uncertainties , we arrive at the inevitable conclusion that it is merely a linear combination mode . this is probably the first instance for pulsating white dwarfs where linear combination frequencies have apparently emerged as a result of an interplay between a non - radial pulsation mode f@xmath12 and a harmonic of the orbital frequency f@xmath23 . the idea of a physical interaction between nonradial pulsation and tides caused by orbital motion may be relevant here . we defer a thorough investigation to a future theoretical paper . a measure of the gross ability of our data to resolve the components of the 609s mode is evident from a comparison of the dft and the window function , shown in the top and bottom panels of figure [ trip ] respectively . the window function is the dft of a single frequency noiseless sinusoid sampled at exactly the same times as the actual light curve . comparing the dft to the window function unambiguously reveals the low frequency component and definite additional width , indicating the presence of more than one frequency . using least squares analysis in conjunction with the technique of prewhitening , we are able to separate these resolved components . prewhitening is the standard procedure for isolating and characterizing closely spaced spectral components ; it involves subtracting out the best fit for the highest amplitude component of the multiplet from the light curve and re - computing the dft . carrying out this procedure for the 609.089s period gives clear residuals with similar amplitude at similar spacing above and below in frequency , as shown in the second panel of figure [ trip ] . continuing to prewhiten with the 608.615s ( third panel ) and then the 609.509s periods ( fourth panel ) leads to marginal residual power near the 3@xmath8 detection limit of 1.71mma . this exercise demonstrates that the 609s mode is made up of three components in the form of a triplet as listed in table [ bfit - per ] . measured pulsation amplitudes are typically lower than the intrinsic pulsation amplitudes due to geometric cancellation . this effect has three independent causes : disk averaging , inclination angle , and limb darkening . the inclination angle dictates the distribution of bright and dark zones in our view for a given mode ; this introduces a large amount of scatter in observed pulsation amplitudes . eigenmodes with different @xmath24 values exhibit different cancellation patterns ( see @xcite ) . hence we did not expect that all members of the 609s triplet should have had the same amplitude . given that the implications of the triplet spacing are most unexpected , we conducted simulations to verify whether the triplet is indeed genuine or merely caused by a singular changing period . although the pulsation period of a non - interacting zz ceti only drifts due to stellar cooling @xcite , the period of an accreting pulsator can drift on a faster timescale . during a dwarf nova outburst , an accreting white dwarf is heated to temperatures well beyond the instability strip and it ceases to pulsate ( * ? ? ? * ; * ? ? ? * ; * ? ? ? once the white dwarf has cooled down close to its quiescent temperature and pulsations have resumed in the star , @xcite explain how mode frequencies can drift a little due to the continued cooling of the outer envelope . they calculate that the longer period modes such as the 609s mode should drift at the rate of d@xmath25/dt@xmath4-@xmath26hz / s . this expected model drift rate from the cooling of the outer envelope is much faster than the observed drift rate of d@xmath25/dt@xmath4-@xmath27hz / s for zz ceti pulsators from the cooling of the entire star @xcite . during our 11 day campaign , we should expect a drift in the pulsation period of the order of 1@xmath28hz , comparable to the observed triplet spacing . hence it becomes necessary to check whether a drifting period is responsible for the observed triplet , a fairly general way of introducing spurious peaks in the dft . as a first step , we simulated constant - amplitude light curves with a single variable period near 609s ; the period was varied at different non - zero constant rates . the goal of the simulation was to check if it is possible to fit a triplet to any of the generated light curves , while processing them in the same manner as the real light curve from the campaign . for each drift rate dp / dt , we simulated a light curve with exactly the same time sampling as the real data , adding in gaussian noise with an amplitude comparable to the observed white noise . we computed a dft from the simulated light curve , and then used the techniques of least squares analysis and prewhitening to determine its frequency components , proceeding in exactly the same way as the real data . figure [ pdrift1 ] shows the dfts corresponding to the drift rates dp / dt=6@xmath29s / s and dp / dt=6@xmath30s / s , also including the dft of a single constant period at 609s for reference . figure [ pdrift1 ] clearly demonstrates that the amplitude of the dft based on the 11-day simulated light curve is inversely proportional to the drift rate ; faster the drift rate of the period , smaller the amplitude of the variable period in the dft . the drift rate of the period has to be fast enough to cause a triplet spacing of 1.2@xmath28hz in 11days , and slow enough that the amplitude of the dft does not fall significantly below the simulation amplitude . we find that the drift rate of 6@xmath29s / s comes close to reproducing the triplet spacing . however we are unable to use non - linear least squares code to fit three closely - spaced frequency components to the light curve simultaneously , indicating that they are not resolved . for the second step , we simulated mono - periodic signals whose period and amplitude vary continuously as low order functions . in order to keep our simulations realistic , we determined possible low - order functions using the observed data . to this end , we identified 19 runs of 4.8hr duration or longer , acquired during the course of the campaign . figure [ dm ] indicates our measurements of period and amplitude for these runs ; the amplitudes from filterless data are scaled by a factor of 1.31 and v filter data by a factor of 1.15 . figure [ dm ] shows a first and third order polynomial fit ; both sets of fits are weighted inversely as the squares of the uncertainties . points with uncertainties in period less than 0.6s and uncertainties in amplitude less than 2mma are isolated in the lower panels of figure [ dm ] . note that apart from two points close to the 17th and 18th of may , there is no strong evidence for any change in period . for each low order function that fit the individual period measurements , we simulated light curves with all possible functions that fit the amplitude measurements , to exhaust all possibilities of how the 609s periodicity could have changed . in most cases , the amplitude of the dft computed from the 11day simulated light curve fell to values in the range of 1520mma due to the varying period , thus making them less plausible . it is crucial to note that the best fit to the amplitudes from the 19 long runs suggests that the amplitude stayed constant at about 29mma , making a significant variation in amplitude unlikely . secondly , the amplitude of the combined dft computed from the real data is 28.8mma ( see table [ bfit - per ] ) , at par with the individual amplitude measurements . this strongly suggests that a substantial change in period during the 11-day campaign is also unlikely , as otherwise we would have observed a lower amplitude in the combined dft shown in figure [ dft ] compared to the nightly amplitude measurements . the low order functions we adopted above were meant to provide a realistic idea of how the period and amplitude could change over the duration of the campaign . however none of our simulations with these functions could produce a triplet that matched the observed triplet with its spacing and high amplitude . for a fraction of the simulations that seemed plausible due to a reasonably high amplitude principal component , we also used the pre - whitening and least squares analysis code to determine the individual frequency components . for all of these cases , a simultaneous 3-frequency fit to the entire light curve using the non - linear least squares program failed to converge . this is encouraging as the lack of convergence apparently implies that the non - linear least squares program has the ability to distinguish between a single variable period and a multiplet . since we are unable to run a successful simulation where a single variable period mimics a triplet while being constrained by the observed criteria , we can only conclude that this possibility is not very likely . however we do recognize that we can only carry out a finite number of simulations and scenarios , and therefore our simulations do not completely rule out the possibility that a variable periodicity could mimic a triplet . we strongly recommend another campaign on this star to see if a triplet with exactly identical components is observed again , and whether the future data stay in phase with the observations presented here . as a last step , we also simulated a light curve for a triplet using the best - fit periods , amplitudes , and phases listed in table [ bfit - per ] . we isolated 19 segments from the light curve corresponding to the 19 long runs discussed earlier . then we determined the value of the period and amplitude of the dominant frequency for those 19 simulated runs ; we indicate these values in figure [ dm2 ] and show the corresponding observed values for comparison . this exercise helps us understand what fluctuations in period and amplitude we can expect to see for the triplet model when we fit a single frequency to runs shorter than the duration needed to resolve the triplet . this also explains why @xcite and @xcite measure different values of the dominant period in sdss1610 - 0102 . any significant departure from the triplet model would pose as evidence in favour of a variable period model . apart from a single point close to the 17th of may , there is no significant deviation in period from the triplet model . the observed amplitudes on 17th and 18th of may also do not match the expected amplitudes due to the triplet model . despite these points , we conclude that the triplet model matches the observed values well within the uncertainties , and is most probably correct . the accreting pulsator gw librae also showed unresolved multiplet structure @xcite with 1@xmath28hz splitting for the 370s and 650s modes . @xcite found similar unresolved multiplet structure for the 336.7s mode of the accreting pulsator v455 and ( hs2331 + 3905 ) . thus the multiplet spacing we observe for sdss1610 - 0102 is neither unique nor unusual for accreting white dwarf pulsators ; however this is the first data set in which a multiplet has been cleanly resolved . each of the pulsation eigenmodes that can be excited in the star is typically described by a set of indices : @xmath31 , @xmath32 , & @xmath24 . the radial quantum number @xmath31 gives the number of nodes in the radial eigenfunction , and @xmath32 and @xmath24 index the spherical harmonic @xmath33 function forming the angular dependence of the mode . spherical harmonics ( @xmath34 ) are used to describe these eigenmodes due to the spherical gravitational potential ; this quantization is similar to the quantum numbers used to describe the state of an electron bound by the spherical electrostatic potential of the nucleus . these non - radial motions penetrate below the surface with reducing amplitude ; the radial eigenfunction , which depends on @xmath31 and mode trapping , dictates the pulsation amplitude at different depths . montgomery & winget ( 1999 ) and montgomery , metcalfe , & winget ( 2003 ) showed that pulsations probe up to the inner 99% of a model white dwarf . multiplets in pulsation modes indicate a loss of spherical symmetry , caused by either stellar rotation or a magnetic field @xcite . in the absence of a prominent magnetic field or rotation , normal modes with different @xmath24 values will have the same frequency , although they represent different fluid motion patterns . according to @xcite , frequency splitting in the presence of a magnetic field is given by the following equation : @xmath35 where the frequency shift @xmath36 depends on @xmath37 for a dipole magnetic field . hence a given mode splits into @xmath38 frequencies with uneven spacing . a triplet caused by a magnetic field would require @xmath39 , and the individual components of the multiplet would have @xmath24 values of 0 , 1 , & 2 . such a triplet could not be evenly spaced because the frequency shift of each component depends on @xmath37 . since the observed 609s triplet is evenly spaced within uncertainties , we conclude that it is not caused by a magnetic field . the even spacing of the 609s triplet is consistent with rotational splitting . in light of the remarkable implications this may have for the rotational properties of the star , a more careful discussion of the significance of this result is necessary . since the ratio of the frequency splitting to the central frequency for this triplet is small , it should be safe to use perturbation analysis to linear order , except possibly in the case of differential rotation . @xcite show that the frequency splitting due to rotation in linear perturbation theory is given by : @xmath40 where @xmath8 denotes the frequency in the rotating star , @xmath41 the frequency of the corresponding mode in a non - rotating star ( with the same @xmath31 , @xmath32 , @xmath24 ) , and @xmath42 denotes the solid body angular rotation frequency . the uniform rotation co - efficient @xmath43 depends on the radial and azimuthal order ( @xmath44 ) , while @xmath45 , relevant for non - uniform or differential rotation , also depends on @xmath46 . initially ignoring the term @xmath45 allows us to arrive at a basic constraint on the rotation of the star , but we will revisit differential rotation below . uniform rotation splits a mode of azimuthal quantum number @xmath32 into @xmath47 evenly spaced modes . @xcite showed that in the limit of high radial order @xmath48 , the co - efficient @xmath43 can be simplified as follows : @xmath49 using the mode structure of gw lib as a guide @xcite , the 609s mode likely has @xmath50 and so qualifies marginally for the above approximation . this formalism makes it possible to obtain a gross estimate for the implied spin period of the white dwarf . the spacing of the components of the 609s triplet are [email protected]@xmath28hz and [email protected]@xmath28hz respectively . this spacing is consistent with being even within uncertainties , and we adopt the spacing of [email protected]@xmath28hz . assuming that we are seeing the @xmath51 components of an @xmath52 mode , the implied spin period is @xmath53days . if even approximately correct , the spin period of 4.8 days is remarkably long for an object accreting matter in a binary with an orbital period of 83.8 minutes . the implied rotational velocity for a period of 4.8days would be @xmath541km / s as opposed to the typical rotational velocities of 300400km / s observed for non - magnetic accreting white dwarfs @xcite . the measured surface velocity ( vsini ) for the cataclysmic variables vw hyi is 600km / s @xcite , wz sge is 200400km / s @xcite , and u gem is 50100km / s @xcite . @xcite find the rotation period of the zz ceti star g226 - 29 to be 8.9hr , while @xcite compute the rotation period of zz ceti itself to be 1.5 days ; all known zz ceti stars are non - magnetic . whether they pulsate or not , non - interacting white dwarfs in general are known to rotate slowly @xcite . alternatively , perhaps the splitting of the 609s mode is only indicative of the rotation period of the region of the star that it samples well . in other words , we examine the possibility of differential rotation with a rapidly rotating exterior and a slowly rotating interior . @xcite find that the accreted angular momentum is shared with the accumulated envelope on short timescales . given the small splitting , any differential rotation must be small or constrained to a very localized layer of the star ( e.g. the surface ) . this again provides strong constraints on any diffusive mechanism for angular momentum transport . evaluating the constraints quantitatively requires defining both the mode eigenfunctions and candidate rotation profiles , as well as how the rotational shear depends on latitude . such an extensive theoretical analysis is beyond the scope of this observational paper , but will be pursued in the near future by members of our collaboration . such a long spin period of 4.8 days derived from the triplet spacing can not be discounted out of hand . while white dwarfs gaining matter are thought to spin up ( e.g. * ? ? ? * ; * ? ? ? * ) , those undergoing classical novae are thought to eject any accreted mass ( see discussion in @xcite ) or lose small quantities of mass based on the abundance in their ejecta ( e.g. @xcite ) . therefore these objects have ample opportunity , depending on how angular momentum is exchanged between the accreted envelope and core , to either gain or shed angular momentum and allow the core to spin up or down over the several gyr accretion history of old cataclysmic variables like sdss1610 - 0102 . @xcite propose a model in which the primary white dwarf loses accreted angular momentum during nova outbursts . the demonstration by @xcite of how internal gravity waves can extract angular momentum from the solar core during its evolution might provide a model for how the core of an accreting white dwarf could be spun down , with the angular momentum carried away with the envelopes over many nova ejections . the transport of angular momentum in stably stratified layers within stars remains poorly understood . diffusive prescriptions , even ones which depend on magnetic effects like that of @xcite used by @xcite , are contradicted by the observed rotational profile of the sun @xcite . despite having been spun down on the main sequence , there is no observed gradient in the rotation profile of the solar core , an essential aspect of angular momentum transport in any diffusive prescription . additionally such prescriptions predict rotation periods much shorter than those observed for isolated non - magnetic white dwarfs @xcite . @xcite showed that a radial magnetic field in a partially crystallized white dwarf would be sheared by differential rotation , leading to an increase in the azimuthal component proportional to the cumulative angle of differential rotation . he also indicated that when the field strength reached about @xmath55 gauss , the star would be locked as a rigid rotator . @xcite also demonstrate the need for a magnetic field for rigid body rotation in a white dwarf . perhaps in the context of sdss1610 - 0102 this implies that its magnetic field is weak . we conducted a multi - site campaign on the accreting pulsator sdss1610 - 0102 for a duration of 11 days using seven observatories around the globe in may 2007 . the photometric periods in our light curve were consistent with previous observations from 2004 and 2005 @xcite , indicating their long - term stability . the most interesting result from our multi - site campaign is the detection of a resolved evenly spaced pulsational triplet at 609s . the even spacing of the triplet suggests that it is induced by rotation , and the rotational period of 4.8 days derived from the spacing has strong implications for the transport of angular momentum and its long - term evolution . we investigated the possibility that variability of period and/or amplitude in a single frequency light curve could produce the observed triplet , and find that this is not a likely possibility . either the period of 4.8 days is a measure of the uniform rotation period for the entire star or it is suggestive of differential rotation in the star . in either case , the prospects of constraining rotation in an accreting white dwarf with asteroseismic techniques is immensely exciting . conducting similar multi - site campaigns on other accreting pulsators could help us form a picture of how angular momentum is exchanged in the interior of the white dwarf , and its significance from the perspective of binary evolution and stability . the rotation of the underlying white dwarf is also important for type ia supernovae ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? our spectroscopic measurement of the orbital period is [email protected] is consistent within uncertainties with the photometrically observed first harmonic of the orbital period at 41.5min . our second striking result is the detection of linear combination frequencies apparently caused by an interplay of the dominant pulsation mode at 609s and the first harmonic of the orbital period . such a physical interaction between nonradial pulsation and orbital motion has never been detected before for variable white dwarfs , and is perhaps suggestive of tides . a thorough investigation is left to a future theoretical paper . szkody and mukadam thank nsf for financially supporting this project through the grant ast-0607840 . dr . bildsten is grateful to nsf for the grant ast-0707633 . tovmassian and zharikov acknowledge the conacyt grant 45847/a for supporting the observations made at oan - spm . professors woudt and warner acknowledge research funding from the university of cape town and from the national research foundation . based on observations obtained with the apache point observatory 3.5-meter telescope , which is owned and operated by the astrophysical research consortium and on observations collected at the centro astronmico hispano alemn ( caha ) at calar alto , operated jointly by the max - planck institut fr astronomie and the instituto de astrofsica de andaluca ( csic ) . some of the observations reported in this paper were obtained with the southern african large telescope ( salt ) . along with added gaussian noise . this figure helps us understand the fluctuations we can expect in the measurements of period and amplitude from runs much shorter than the duration required to resolve the triplet , when fitting a single frequency to the data.[dm2],title="fig : " ] along with added gaussian noise . this figure helps us understand the fluctuations we can expect in the measurements of period and amplitude from runs much shorter than the duration required to resolve the triplet , when fitting a single frequency to the data.[dm2],title="fig : " ]

we consider the kondo - lattice model on the kagome lattice and study its weak - coupling instabilities at band filling fractions for which the fermi surface has singularities . these singularites include dirac points , quadratic fermi points in contact with a flat band , and van hove saddle points . by combining a controlled analytical approach with large - scale numerical simulations , we demonstrate that the weak - coupling instabilities of the kondo - lattice model lead to exotic magnetic orderings . in particular , some of these magnetic orderings produce a spontaneous quantum anomalous hall state .

1407.5369

in a seminal work , haldane demonstrated that a magnetic field is not required to induce integer quantum hall states @xcite . adding a complex hopping to the tight - binding hamiltonian on the honeycomb lattice opens a gap at the dirac points of the electron band structure . this gap leads to a topologically nontrivial electronic state for a half - filled band , i.e. a chern insulator or quantum anomalous hall ( qah ) state . the key characteristic of this state is the appearance of chiral edge channels at the sample boundaries , in which current can flow along one direction only , implying dissipationless charge transport due to the absence of backscattering . this property makes qah states attractive for ultra - low - power consumption applications . one of the current challenges in condensed matter physics is to find ways of stabilizing a qah state at ambient temperature . many paths toward realization of the qah effect have been suggested , most of which are limited to graphenelike dirac systems . proposals include spin - orbit coupled magnetic semi - conductors @xcite , spin - orbit coupled ad - atoms on graphene @xcite , and spin - polarized qah states @xcite . experimental signatures of qah have been detected @xcite and advanced nanostructures such as `` molecular graphene '' @xcite open the possibility of controlled manipulation of dirac fermions . a proposal based on doped magnetic topological insulator materials @xcite has led to the first robust observation of the qah state very recently @xcite . another proposed scenario involves spontaneous chiral symmetry breaking due to electron - electron interactions at dirac points @xcite . however , relatively large electron - electron coulomb interactions are required to induce chiral states due to the vanishing density of electron states . @xcite . an alternative mechanism for robust , high temperature qah states has generated much interest @xcite . in certain correlated multi - orbital compounds , conduction electrons are coupled to localized magnetic moments ( spins ) at each lattice site . the spin may arise , for example , from @xmath0-electrons in transition - metal oxide materials @xcite or @xmath1-electrons in lanthanide based materials @xcite . qah states are possible if the spin ordering is _ noncoplanar _ , in which case the local berry curvature ( scalar spin chirality ) acts as an effective magnetic field on the orbital motion of the conduction electrons . this mechanism does not require band structures with dirac points . qah states may also arise from weak - coupling instabilities at quadratic band crossings @xcite and nested fermi surfaces @xcite . nonzero chirality implies broken time - reversal and spatial parity symmetries , a necessary ingredient for qah states . note that these discrete symmetries can be spontaneously broken at finite temperature in two - dimensions , even though the continuous @xmath2 spin rotational symmetry has to remain intact . in this paper we investigate the kondo - lattice model ( klm ) on the kagome lattice . the klm is the simplest model that captures the interplay between conduction electrons and localized magnetic moments ; see fig . [ fig : klm ] for an illustration . because we are interested in cases where the spins of localized electrons develop a net magnetic moment below a certain temperature ( @xmath3 for @xmath4 ) , the model can be further simplified by assuming that the localized moments are classical variables , i.e. , that there is no kondo effect . in klm , conduction electrons interact with the localized spins through an on - site exchange coupling @xmath5 . this exchange corresponds to hund s coupling for the case of transition metal oxides and kondo coupling for the case of intermetallics . the magnetic ordering induced by this exchange coupling thus depends on the dispersion of the itinerant electrons at the relevant filling fractions . the kagome lattice has a rich band structure containing dirac points , quadratic fermi points , and perfectly nested fermi surfaces with van hove singularities . it is remarkable that a single model contains all of the above features , which have been studied individually in triangular @xcite , honeycomb @xcite , checkerboard @xcite , cubic @xcite , and pyrochlore lattices @xcite . thus , the kagome klm is a good candidate to search for robust topologically non - trivial states . to investigate the complex magnetic orderings of the kagome lattice , our approach includes group - theoretical symmetry analysis at the ordering wavevectors , variational fourier - space calculations , and large - scale unconstrained numerical simulations . all approaches agree in the small @xmath5 limit . our large - scale simulations are based on a recently developed method to simulate classical degrees of freedom interacting with fermions @xcite , which enables the unbiased study of magnetic ordering at any @xmath5 . we have uncovered some surprising spin textures that would be very difficult to stabilize in local moment mott insulating systems . among them , there are several spin orderings appearing at different filling factors that lead to a spontaneous qah effect . this variety of magnetic phases opens a path towards the realization of the qah effect at ambient temperature . our results are also relevant for interacting electron systems without preformed local moments , such as hubbard - type models @xcite . the kagome tight - binding model is expected to exhibit weak - coupling instabilities when the electron filling reaches the qbcp or the van hove singularities . consequently , an infinitesimal @xmath5 will immediately produce magnetic ordering . on the other hand , since the dirac points are stable against weak perturbations , robust magnetic ordering can only be expected at intermediate electron - electron interactions . moreover , the magnetic or spin - density - wave ordering in the interacting systems has to complete with other instabilities , particularly the superconductivity order @xcite . the situation is more complicated for the second van hove singularity due to sublattice interference phenomenon @xcite ; we will discuss its consequences in sec . [ filling2 ] kondo - lattice model ( klm ) on the kagome lattice . the model can be viewed as consisting two subsystems represented here by two fictitious kagome layers . conduction electrons hop on the top layer according to a nearest - neighbor transfer integral @xmath6 . the bottom layer consists of localized spins ( magenta arrows ) at every kagome site . the electron spins ( cyan arrows ) are coupled to the localized moments through an on - site exchange constant @xmath5 . the numbers 1 , 2 , 3 denote the three inequivalent sublattices of kagome . ] the paper is organized as follows . in section [ modelmethod ] we present the klm on the kagome lattice , discuss the spectral features , and explain our numerical techniques . next , we present our magnetic ordering results for electron filling fractions @xmath1 of special interest . in section [ vanhove ] we discuss the case of van hove fillings ( @xmath7 and @xmath8 ) in which the fermi surfaces are nested by three commensurate ordering vectors . in section [ fermipoints ] we discuss the band touching points , including dirac points ( @xmath9 ) and quadratic band crossing ( @xmath10 ) . section [ conclusions ] summarizes our findings . in this work we consider the klm , which is illustrated in fig . [ fig : klm ] . its hamiltonian is given by @xmath11 the first term describes electron hopping between nearest - neighbor ( nn ) sites ; @xmath12 is the transfer integral , @xmath13 creates an electron with spin @xmath14 on site @xmath15 , and @xmath16 denotes a nn pair . the itinerant electrons interact with the localized spins through an on - site exchange coupling @xmath5 as described by the second term , where @xmath17 is a vector of the pauli matrices . note that summation over repeated indices is assumed . we consider the classical limit @xmath18 for the localized spins . in this limit , the electron spectrum is independent of the sign of @xmath5 and the eigenstates corresponding to opposite signs are connected by a global gauge transformation @xcite . ( a ) band structure of the tight - binding model on the kagome lattice . here and in the following figures , the band energy @xmath19 is measured in units of nn hopping @xmath6 . ( b ) brillouin zone ( bz ) of the kagome lattice . at filling fraction @xmath20 the fermi surface shrinks to two inequivalent dirac points @xmath21 at the corners of the bz . these two dirac points are connected by wavevectors @xmath22 . ( c ) the dispersion has three saddle points @xmath23 at the edge of the bz , giving rise to a logarithmically divergent dos at filling fractions @xmath24 and @xmath8 . the corresponding fermi surface is a regular hexagon inscribed within the bz . pairs of parallel edges of this fermi surface are perfectly nested by wavevectors @xmath25 and @xmath26 . ] the energy dispersion for the kagome lattice tight - binding model , i.e. the first term in eq . ( [ eq : h - klm ] ) , can be obtained analytically . we first label the kagome sites as @xmath27 , where @xmath28 denotes the position of site @xmath15 , which is decomposed into the triangular bravais lattice point @xmath29 and the sublattice index @xmath30 ; see fig . [ fig : klm ] . after fourier transformation @xmath31 , we have @xmath32 . the hopping matrix is @xmath33 where @xmath34 , @xmath35 are primitive lattice vectors of the triangular bravais lattice , and @xmath36 . after diagonalizing this matrix , we obtain two dispersive bands @xmath37 and a flat band @xmath38 at the top of the dispersion . as mentioned in the introduction , the kagome lattice electronic band structure [ fig . [ fig : dispersion](a ) ] exhibits several special points , which we now present and discuss in more detail . first , there are two inequivalent dirac cones at the corners @xmath21 of the brillouin zone ( bz ) schematically represented in fig . [ fig : dispersion](b ) . as these dirac points are the isolated crossing points of the two dispersive bands , the fermi surface shrinks to the two @xmath39points at filling fraction @xmath40 . in accordance with the standard theory of electronic berry phases , these dirac fermi points are characterized by nonzero berry flux @xmath41 @xcite . in the case of the kagome lattice , it has been shown that gapping out a pair of dirac points can lead to topological phases such as spontaneous qah and quantum spin hall insulators @xcite , in agreement with the general theory of gapped dirac fermions in two dimensions @xcite . however , the vanishing density of states ( dos ) renders the dirac fermi points robust against weakly repulsive electron - electron interactions . in general , a finite interaction stength is required to open a gap at @xmath40 and stabilize a topologically nontrivial phase @xcite . another topologically nontrivial band - touching occurs at the @xmath42-point between the upper dispersive and flat bands . at filling fraction @xmath43 the fermi surface consists of a single point at @xmath44 with a quadratic dispersion in its vicinity : @xmath45 . this so - called quadratic band - crossing point ( qbcp ) is characterized by a @xmath46 berry phase . in momentum space , the topological qbcp resembles a vortex with a @xmath47 winding number @xcite . contrary to dirac points , quadratic fermi points are unstable against arbitrarily weak short - range interactions due to their finite dos @xcite . the perturbed quadratic fermi point either splits into two fundamental dirac points or it is completely gapped out , giving rise to a topological insulator @xcite . however , because the whole flat band @xmath48 is degenerate with the @xmath42 fermi point , finding the magnetic structure stabilized by @xmath49 in eq . ( [ eq : h - klm ] ) for @xmath50 is still a rather complicated problem . a third distinct spectral feature is related to the two dispersive bands , @xmath51 and @xmath52 . the dos of @xmath51 and @xmath52 contains a van hove singularity at fillings @xmath24 and @xmath8 , respectively . a logarithmically divergent dos at these filling fractions results from the three saddle points @xmath53 at the edges of the bz [ fig . [ fig : dispersion](c ) ] . for just nn hopping , the fermi surface at these filling fractions is a regular hexagon inscribed within the bz , as shown in fig . [ fig : dispersion](c ) . remarkably , pairs of parallel edges of this fermi surface are perfectly nested by wavevectors @xmath25 and @xmath26 . the perfect fermi surface nesting combined with a divergent dos leads to a magnetic susceptibility of the conduction electrons that diverges as @xmath54 ( @xmath55 ) . consequently , the system is unstable against developing a triple-@xmath56 magnetic order even for small @xmath57 . our goal is to find the magnetic ground state of the klm for each of these special filling fractions and to establish the nature of the corresponding electronic states . in the process of uncovering the magnetic ordering at the special fillings mentioned above , we will also discuss specific fillings different from these three . a quadrupling of the unit cell as a consequence of finite @xmath58 ordering allows for commensurate fillings other than the van hove fillings @xmath59 and @xmath60 , which are @xmath61 and @xmath62 . we have investigated the magnetic ordering and the corresponding electronic state at these fillings as well . in a nave approach to monte carlo sampling of the classical spins , every trial update of the spin configuration requires diagonalization of a new single - particle electron matrix to determine the change in energy . direct diagonalization costs order @xmath63 numerical operations , so this simulation approach is limited to systems with @xmath64 spins ( e.g. ref . ) . we require much larger scale systems to study weak - coupling instabilities . small systems obscure susceptibility divergences because of their larger momentum space discretization cut - off . for example , fermi surface nesting yields a divergent susceptibility that scales as @xmath65 , and stabilizing the desired triple-@xmath56 magnetic orders may require @xmath66 spins @xcite . evidently , direct diagonalization at every monte carlo step is impractical . to simulate the klm at large scales , we employ two complementary numerical methods . first , we perform unconstrained , finite - temperature simulations with a recently developed langevin method , which we review below . second , we use a fourier - space variational method to verify and analyze the candidate orderings at zero temperature . in our variational approach , we assume a known magnetic unit cell size and search for the lowest energy ordering among spin configurations within the extended unit cell @xcite . for example , the most general triple-@xmath56 orderings have a quadrupled magnetic unit cell of 12 spins . the electronic part can then be solved by applying a fourier transformation . the total energy obtained by summing over states within the reduced bz is a function of the magnetic configuration in the extended unit cell . we use standard simulated annealing to find the minimum - energy configurations with constrained periodicity @xcite . more details can be found in appendix [ appendix1 ] . our langevin method enables high accuracy , unconstrained simulation of very large systems , and is orders of magnitude faster than previous monte carlo methods @xcite . we approximate langevin forces using an extension of the kernel polynomial method ( kpm ) @xcite . kpm provides fast estimates of the density of electron states ( dos ) , @xmath67)$ ] , where @xmath68 are the electron energy levels , i.e. the eigenvalues of the single - particle electron matrix @xmath69 . kpm approximates the dos using a truncated series in chebyshev polynomials @xmath70 , @xmath71 the numerical factors @xmath72 associated with the jackson kernel @xcite optimally damp gibbs oscillations @xcite . the chebyshev moments are given by @xmath73 . for efficiency , we use the kpm stochastic approximation @xmath74 , where @xmath75 is a random vector with elements that satisfy the ensemble average @xmath76 . we use the chebyshev recursion relation in the form @xmath77 to iteratively build the vectors @xmath78 . thus , the moment estimates @xmath79 may be evaluated at linear cost in the number of spins @xmath80 by using only sparse matrix - vector multiplications . kpm accuracy is controlled by the series truncation order @xmath53 and the number of random vectors @xmath81 over which we average . the kpm dos allows estimation of the ( free ) energy @xmath82 of a spin configuration @xmath83 . after integrating out the electrons at fixed chemical potential @xmath84 and inverse temperature @xmath85 we obtain @xmath86 = -\beta^{-1 } \log { \rm tr}_{c } \,e^{-\beta ( \mathcal{h } - \mu\sum_i c^{\dagger}_i c^{\,}_i ) } = \int \rho(\epsilon ) f(\epsilon ) d\epsilon \;,\ ] ] where @xmath87\}$ ] is an anti - derivative of the fermi function . chebyshev - gauss quadrature allows fast and accurate numerical integration over the kpm dos estimate @xcite . to efficiently sample spin configurations @xmath83 we extend kpm to also estimate forces @xmath88 . one path is the numerically exact technique of automatic differentiation with reverse accumulation @xcite . indeed , _ all _ forces @xmath88 may be simultaneously estimated by a `` reverse '' recursion relation with a cost linear in @xmath80 that is equivalent to the cost of estimating @xmath89 @xcite . we apply our kpm based force estimates to sample spins according to overdamped langevin dynamics , @xmath90 where @xmath91 are uncorrelated gaussian random variables with unit variance , @xmath92 is the langevin time , and we use an implicit lagrange multiplier to constrain @xmath93 . accuracy is again controlled by two tunable parameters : the series truncation order @xmath53 and the fraction @xmath94 of langevin integration time per random vector . the cost to integrate one unit of langevin time scales as @xmath95 . kpm inaccuracies may be viewed as introducing an effective temperature . series truncation effectively smooths the fermi function on the scale @xmath96 , whereas stochastic errors in the force effectively increase the thermal noise an amount @xmath97 . to search for ground states , we randomize the initial spin configuration and then integrate the langevin dynamics with @xmath98 . we perform most simulations with @xmath99 and @xmath100 . the effective accuracy improves with decreasing exchange coupling . at @xmath101 we can often distinguish between spin textures with energies that differ at the fifth significant digit . because the langevin dynamics eq . ( [ eq : langevin ] ) can be viewed as the overdamped limit of the landau - lifshitz - gilbert equation , our simulations also capture the physical emergence of mesoscale topological defects and their dynamics . indeed , we find different domain structures in the same magnetic ordering for different filling fractions at finite temperatures . this subtle difference can be attributed to the different types of effective long - range spin - spin interactions that are mediated by the conduction electrons . we start by considering the magnetic orderings for electron filling fractions right at the van hove singularities of the dos , i.e. the filling fractions @xmath102 and @xmath60 . as mentioned above , the fermi surface of the ideal nn kagome lattice tight - binding model is a regular hexagon inscribed within the hexagonal bz . the corners of the fermi surface hexagon are the three inequivalent @xmath53-points of the bz . these special points are the dominant source of the divergent susceptibility as they are the saddle points of the electron dispersion . the @xmath53-points are nested by three commensurate wavevectors @xmath25 , and @xmath26 , which generically give rise to a quadrupled unit cell since these nesting wavevectors are half of the reciprocal lattice vectors . before we present the results of our numerical simulations , we will provide a symmetry - based perspective of general triple-@xmath56 magnetic orderings , with the purpose of gaining insight into the magnetic and electronic properties of such orderings . in addition , we will demonstrate how the magnetic ground states found in the numerical simulations may be understood analytically on the basis of symmetry constraints @xcite . we then move on to discuss in detail the magnetic order parameters describing the spin textures obtained from our numerical minimizations for @xmath103 and @xmath60 , respectively . triple-@xmath56 magnetic ordering on the kagome lattice can be discussed from the perspective of lattice symmetries . this will shed light on the electronic properties of the conduction electrons in the presence of such ordering . furthermore , we will illustrate how triple-@xmath56 magnetic ground states can be derived systematically by imposing symmetry requirements on the most general form of the spin order parameter @xcite . the numerical simulations then confirm that these are the correct requirements to impose . as a consequence of the commensurability of the @xmath53-point wavevectors we have @xmath104 ( modulo reciprocal lattice vectors ) and hence @xmath105 . the most general spin state modulated by these @xmath106 vectors can be written as @xmath107 where every site on the kagome lattice is labeled by @xmath108 , @xmath109 denoting the sublattice and @xmath110 denoting the bravais lattice unit cell . the set @xmath111 describes nine vector order parameters , one for each sublattice and @xmath56-vector . in the context of the klm with classical spin states , we should constrain the vector order parameters @xmath111 to configurations that satisfy : @xmath112 , i.e. , equal spin length on every site . for the moment we will ignore this constraint and work with the most general set of order parameter components , which is captured by @xmath111 . the constraint will be reinstated in a natural way later . we may write the vector order parameters a product of a scalar and a vector with unit length , i.e. @xmath113 . focusing first exclusively on the scalar part , we can organize them in terms of distinct representations of the lattice symmetry group . for simplicity , we group the scalar order parameters into three vectors @xmath114 which together form a 9-dimensional representation of the lattice symmetry group . its irreducible representations describe possible site orderings or density - wave states on the kagome lattice . ( a)(c ) schematic diagram showing the triple-@xmath56 scalar order ( or collinear spin order ) corresponding to the three irreducible representations @xmath115 , @xmath116 , and @xmath117 . the blue ( red ) circles denote on - site scalar variable @xmath118 ( @xmath119 ) , while empty site corresponds to @xmath120 . panel ( d ) shows the collinear and orthogonal configurations of the triple-@xmath56 vector order parameters . ] the point group of the kagome lattice is @xmath121 . however , each ordering vector @xmath122 is left invariant by the subgroup @xmath123 , called the small group of the ordering vectors . the latter admits only one - dimensional ( @xmath124 ) irreducible representations . taking one of the @xmath58 , for instance @xmath125 , we can evaluate the constraints of the symmetry elements in @xmath126 on @xmath127 . doing so yields three solutions for @xmath127 , which is consistent with the three sublattices of the kagome lattice . the solutions correspond to the irreducible representations @xmath115 , @xmath117 and @xmath116 of @xmath126 . having found the solutions for one @xmath58 , we obtain the solutions for the others by ( rotational ) symmetry . in total we have nine scalars @xmath128 and for each of the three solutions for a single @xmath58 we can form a symmetric combination and the two @xmath129-wave combinations , yielding nine states in total . the symmetric combinations will transform as a @xmath124 irreducible representation of @xmath121 , which are @xmath115 , @xmath117 and @xmath116 . these three states are the relevant building blocks for the spin ordered states . the explicit form of the @xmath128 corresponding to the three @xmath124 representations can be determined by explicitly evaluating the symmetry operations of the @xmath126 group . for instance , the two - fold rotation requires @xmath130 , and @xmath131 . the two @xmath132 representations may be distinguished by their transformation properties under reflection . explicit calculations show that @xmath133 where the @xmath134 or @xmath135 signs correspond to the @xmath117 or @xmath116 modes , respectively . once @xmath136 is determined , the other two vectors can be obtained by applying the three - fold rotations . the pattern of the site - ordered state corresponding to the above three irreducible representations is shown in fig . [ fig : irrep](a)(c ) . these also correspond to the spin configurations in collinear sdw states . we take these three symmetric combinations and focus on the different ways of embedding the obtained @xmath137 in a vector order parameter , i.e. the different ways of combining them with @xmath138 . a fully collinear spin state would be given by @xmath139 for all sublattices @xmath109 , and nesting wavevectors @xmath106 . this fully collinear arrangement of spins does not change the symmetry of the electronic state , which is still @xmath115 , @xmath117 and @xmath116 , respectively . due to the fact that the collinear @xmath115 state uses only a single @xmath122 per kagome sublattice , it fulfills the requirement of uniform spin length , whereas the ( collinear ) @xmath140 states do not . we note that the @xmath115 state would be the kagome lattice version of the uniaxial spin density wave state reported in @xcite . these collinear spin states all manifestly break translational invariance as a consequence of ordering at finite @xmath122 momentum vectors . one can restore an effective translational invariance by choosing @xmath141 , i.e. single unit vector for each @xmath106 , while at the same time demanding @xmath142 for @xmath143 [ see fig . [ fig : irrep](d ) ] . translational invariance is effectively preserved for these spin configurations , as the translations which appear to be broken by finite @xmath106 ordering can be combined with global @xmath144 rotations of the spins to leave the state invariant ( see also ref . @xcite ) . for each of the three scalar states there is such a corresponding translationally invariant spin state . because these spin states are translationally invariant , they clearly satisfy the constraint of uniform spin length . a key result of our numerical work , which we discuss below , is that these three triple-@xmath56 spin states are indeed the ground states ( modulo a global spin rotation ) at the different commensurate filling fractions . we summarize the symmetry constraints giving the three noncoplanar spin configurations as follows : the electronic state i ) transforms as a @xmath124 irreducible representation of @xmath121 , and ii ) is translationally invariant up to global spin rotation . knowing the symmetry properties of the triple-@xmath56 spin configurations puts us in a position to immediately deduce the symmetries of the electronic state . first , the effective translational invariance mandates a full double degeneracy of the spectrum . translating the spin configuration by one lattice vector and performing a global spin rotation by @xmath145 about the appropriate axis does not change the hamiltonian , but it does transform the wave function into an orthogonal one . hence , the electronic spectrum is manifestly doubly degenerate for all three states . by associating a distinct orthogonal spin component @xmath146 to each of the three @xmath122 momenta [ fig . [ fig : irrep](d ) ] , the resulting triple-@xmath56 states have noncoplanar spin configurations . up to a global @xmath144 spin rotation , these magnetic states are invariant under the reflection operations of the @xmath121 group . specifically , one can show that such a rotation is improper , i.e. , it gives a minus sign when translated into an @xmath2 rotation on the electron spin . this changes the symmetry of the electronic state to @xmath147 , @xmath116 and @xmath117 , respectively , after multiplication by @xmath147 . therefore , the electronic states corresponding to the three triple-@xmath56 spin ordered states transform as the representations @xmath147 , @xmath116 and @xmath117 of the hexagonal symmetry group @xmath121 . the noncoplanarity of the spins admits a discrete scalar order parameter characterizing the chirality of the structure . explicitly , a scalar chiral order parameter can be defined for each sublattice @xmath148 where @xmath149 refers to ordering vector @xmath58 . in general , the @xmath150 chiral order parameters of different sublattices are related to each other . more importantly , although the continuous spin @xmath2 symmetry can not be spontaneously broken in 2d , true long - range ordering of the discrete chirality is possible . consequently , one expects a finite temperature phase transition that breaks the @xmath150 symmetry . indeed , a similar transition is observed in the tetrahedral ordering transition in the triangular klm @xcite . in the rest of this section we will discuss the ground states observed in numerical simulations , which we find to be consistent with the predictions based upon symmetry analysis . we stress that our large - scale kpm - langevin numerical simulations make no _ a priori _ assumptions concerning the magnetic ordering . our kpm - langevin results are also confirmed by the variational monte carlo method in fourier space ; see appendix [ appendix1 ] for more details . ( a ) the triple-@xmath56 noncoplanar ( orthogonal ) spin order on kagome lattice that transforms according to the @xmath147 irreducible representation of the symmetry group . spins at the three inequivalent sublattices are indicated by different colors . this @xmath147 symmetry state is the ground state at filling fraction @xmath24 for small coupling and @xmath151 for intermediate coupling ( @xmath152 ) . the corresponding band structure is shown in panel ( b ) . the exchange coupling @xmath153 is used in the calculation . the dashed lines indicate the ferme levels at filling fractions @xmath24 and @xmath154 . ] the chiral structure of spin orders at van hove fillings in the kagome lattice . there is a vector associated with each triangle . the four different orientations of the vectors ( indicated by four different colors ) point to the corners of a regular tetrahedron . the vectors associated with each triangle correspond to the total spins @xmath155 in the @xmath147 symmetry order at @xmath59 filling [ fig . [ fig : vh1](a ) ] , and to the vector chirality @xmath156 in the @xmath116 symmetry order at the second van hove filling @xmath157 [ fig . [ fig : vh2](a ) ] . ] the first filling fraction we discuss is the van hove filling @xmath102 . the spin structure obtained from kpm - langevin simulations with @xmath158 is the @xmath115-symmetry state shown in fig . [ fig : vh1](a ) . our variational fourier - space minimization also confirms this spin order is the minimum - energy state at @xmath59 . the nonzero order parameter components at the three nesting wavevectors are : @xmath159 here \{@xmath160 , @xmath161 , @xmath162 } are three arbitrary orthogonal unit vectors . in this magnetic order , spins of the same sublattice are collinear with respect to each other , @xmath163 , @xmath164 , and @xmath165 , while spins at different sublattices are orthogonal to each other . the electronic bands for this magnetic ordering are shown in fig . [ fig : vh1](b ) . the band structure shows twelve bands , consistent with the requirement of ( at least ) double degeneracy due to a combined translational and rotational symmetry . we observe the opening of a spectral gap at @xmath24 filling for this noncoplanar order . we may view the three sublattices as utilizing three different nesting wavevectors @xmath106 to gap out the full fermi surface . a noncoplanar magnetic structure implies nonzero chiral order @xmath166 . the chiral structure of the noncoplanar order shown in fig . [ fig : vh1](a ) becomes apparent by considering the effective magnetic moment of each individual triangle . by defining the total spin for individual triangles as @xmath155 , we observe that this vector sum at the four inequivalent triangles points to the corners of a regular tetrahedron ( fig . [ fig : chiral ] ) . more specifically , the scalar chirality @xmath167 is nonzero in each triangle , deriving from a nontrivial solid angle subtended by the three spins . this implies that the electrons hopping on a triangle acquire a nonzero berry phase in the presence of such noncoplanar order . we therefore examine the electronic state in more detail . the @xmath147 symmetry spin order itself is found to preserve all rotations and reflections of the hexagonal group . the electronic state however , owing to the finite chirality ( @xmath168 ) , transforms as @xmath147 , breaking the reflection symmetries . in addition , time - reversal symmetry inverts all spins , an effect which can not be compensated by a rotation in case of a chiral spin configuration . we coclude that the electronic state admits a topological characterization in terms of the chern number @xcite . explicit evaluation shows that the chern number is nonzero for the fully gapped electronic state at @xmath59 . consequently , the electronic state corresponds to a qah state ( a chern insulator ) . the noncoplanar nature of this spin configuration implies that it breaks the discrete chiral part of the full @xmath144 symmetry ( @xmath169 ) . we note again that while the continuous @xmath170 spin symmetry can not be broken at finite temperatures in 2d , a discrete @xmath171 order parameter _ can _ develop long - range order at finite temperatures . one therefore expects a finite temperature phase transition associated with the chiral symmetry breaking . the magnetic order at @xmath59 filling can also be understood by a simple analysis of electron eigenstates at the three saddle points . we express the electron operators in terms of quasi - particle ones @xmath172 , where again the lattice site is labelled by @xmath108 , @xmath30 denotes the different sublattices and @xmath173 indicates the bravais lattice point . @xmath174 is the quasi - particle creation operator with momentum @xmath175 and @xmath176 is the eigenvector of the hopping matrix , i.e. @xmath177 ; it contains the sublattice weights of the quasi - particle . denoting the amplitude of the order parameters as @xmath178 , the exchange coupling term can then be written as @xmath179 the dominant contributions come from states at the three saddle points , i.e. for electrons with @xmath180 . at the three saddle points for filling fraction @xmath24 , we have @xmath181 , @xmath182 , and @xmath183 . consequently , the product @xmath184 ( no summation over @xmath109 ) , where @xmath185 is the antisymmetric tensor of rank 3 . this result immediately implies that the coupling will be maximized when @xmath186 , i.e. it has the @xmath115 irreducible representation of @xmath121 discussed above . the corresponding magnetic order thus transforms according to the @xmath147 irreducible representation , consistent with our numerical result . interestingly , we find numerically that the same chiral magnetic order is also the ground state at filling fraction @xmath151 . this is consistent with the observation that a gap is opened at this filling fraction for strong enough coupling [ fig . [ fig : vh1](b ) ] . snapshots of the spin configuration obtained from our large - scale kpm - langevin simulations are shown in fig . [ fig : snapshot1 ] for the two filling fractions . the coloring of triangles indicates the amplitude of the scalar chirality @xmath167 . both snapshots clearly show large domains of uniform chirality but , curiously , with different defect structures , indicating subtle differences in the effective spin - spin interactions . ( a ) the @xmath116 symmetry triple-@xmath56 noncoplanar spin order on kagome lattice . spins at the three inequivalent sublattices are indicated by different colors . this state is the ground state at filling fraction @xmath157 . panel ( b ) shows the corresponding band structure calculated with exchange coupling @xmath187 . the dashed line indicates the fermi level at @xmath157 filling . ] the numerically observed magnetic ordering at the second van hove singularity is shown in fig . [ fig : vh2 ] along with the corresponding dispersions . this @xmath116 symmetry spin state is also confirmed by the variational fourier - space calculation . again we find a vanishing uniform component @xmath188 . the non - zero order parameters at the nesting wavevectors are : @xmath189 here \{@xmath160 , @xmath161 , @xmath162 } are again three orthogonal unit vectors . each sublattice participates in two of the nesting order parameters , while spins in each sublattice are coplanar . the normal of the coplanar spins from different sublattices are orthogonal to each other . similar to the previous case , the electronic bands are doubly degenerate . the @xmath116 symmetry state at @xmath157 filling is rather robust for @xmath5 up to order @xmath6 . interestingly , there remains a quadratic fermi point at @xmath190 for @xmath191 filling ; see fig . [ fig : vh2](b ) . the lack of the spectral gap at this van hove filling as compared to the state at @xmath24 can also be understood from the couplings between the saddle points . as discussed in sec . [ sec : f1_4 ] , the effective coupling between electrons at the three inequivalent saddle points is proportional to @xmath192 [ eq . ( [ eq : coupling ] ) ] . for filling fraction @xmath157 , the sublattice weights obtained by diagonalizing the hopping matrix in eq . ( [ eq : hopping ] ) are @xmath193 . since the nesting wavevectors always connect different saddle points , the product @xmath194 for @xmath195 ( no summation over @xmath109 ) , hence the coupling coefficient vanishes at the @xmath53-points . consequently , the three saddle points remain degenerate , giving rise to a fermi point at @xmath190 in the reduced bz . the vanishing couplings between the saddle points are related to the nontrivial sublattice interference discussed in refs . although the susceptibility at the three @xmath106 still has a logarithmic divergence ( instead of @xmath196 ) due to the divergent dos at the saddle points , the triple-@xmath56 spin order now has to compete with simple ferromagnetism instability which also results from a divergent dos . indeed , the @xmath197 ferromagnetic order is the dominant sdw instability in hubbard - like model on kagome lattice @xcite . for kondo - lattice model , our results show that the triple-@xmath56 spin order is still favorable energetically , and the selection of the @xmath116 symmetry state shown in fig . [ fig : vh2](a ) is due to higher - order couplings on the nested fermi surface . it is worth noting that spins in each individual triangle ( both up and down ) form a 120@xmath198 coplanar structure in the @xmath116 symmetry spin order . this triple-@xmath56 state is thus one of the many classical ground states of the well - studied nearest - neighbor exchange interaction hamiltonian @xmath199 with antiferromagnetic coupling constant @xmath200 @xcite . since the nn interactions can be recast into the form @xmath201 up to an irrelevant constant . a triangle with three spins pointing 120@xmath198 to each other has a zero total spin @xmath202 . consequently , any spin configuration on kagome consisting of 120@xmath198 triangles is a ground state of @xmath203 and the @xmath116-symmetry state is one of them . in fact it is known that all such classical ground states with local @xmath202 form an extensively degenerate manifold @xcite . our large - scale kpm - langevin simulations of the klm at @xmath60 filling show that spins tend to form @xmath204 local order as temperature is lowered , indicating that the dominant effective spin interactions can be well - described by a nn exchange hamiltonian . to characterize the general ground states of @xmath203 with local @xmath202 , it is useful to consider the configuration of the vector chirality @xmath156 for individual triangles . the noncoplanar @xmath116 symmetry spin state shown in fig . [ fig : vh2](a ) corresponds to a specific tetrahedral ordering of the the vector chirality @xmath205 . as shown in fig . [ fig : chiral ] , the vector chirality at the four inequivalent triangles points to the four corners of a regular tetrahedron . the tetrahedral ordering of the vector chiralites thus results from the long - range ( beyond nn ) electron - mediated interactions in the klm , selecting this particular spin configuration out of the degenerate set of ground states of @xmath203 . although this triple-@xmath56 state is the ground state of the klm , hence is energetically favorable as temperature tends to zero , our large - scale kpm - langevin dynamics simulations find other coplanar 120@xmath198 states that are very close in energy . as shown in fig . [ fig : snapshot2 ] , the vector chiralities tend to colinearly align ( parallel or antiparallel ) at finite temperatures . these results are consistent with the fact that thermal fluctuations favor coplanar spins ( order - by - disorder ) in the nn exchange model @xcite . the electronic subsystem must have zero chern number because the scalar chirality is zero for coplanar spins in each triangle . alternatively , the presence of reflection symmetries in the representation @xmath116 , discussed in section [ symmetry ] , forces the chern number to be zero @xcite . however , the magnetic structure itself is chiral ( @xmath168 ) , as already discussed in section [ symmetry ] . this is also manifested by the tetrahedral order of the vector chirality . we now come to the fillings corresponding to the isolated band touching points of kagome tight - binding hamiltonian . in this section we investigate the magnetic ordering when the fermi surface shrinks to these special points at filling fractions @xmath40 and 2/3 . in the first case , the elementary excitations are dominated by electrons in the vicinity of two dirac points . the dirac points are connected by momentum vectors @xmath22 and coupling between the dirac point would therefore require magnetic ordering at these wavevectors . in contrast , at filling fraction @xmath50 , the fermi surface shrinks to a single point at the zone center ( @xmath44 ) , where the dispersive @xmath52 band touches the flat band . it is thus difficult to anticipate what the order parameter components will be other than the @xmath206 component . surprisingly , our large - scale simulations uncover another triple-@xmath56 magnetic ordering , which has different symmetry from those induced at the van hove fillings . we first discuss the dirac points and then come to the quadratic band crossing . ( a ) the @xmath207 coplanar spin order on kagome lattice at filling fraction @xmath40 . spins at the three inequivalent sublattices are indicated by different colors . ( b ) the corresponding band structure with exchange coupling @xmath187 . the dashed line indicates filling fraction @xmath40 . ] the fermi ` surface ' at filling fraction @xmath40 consists of two dirac points located at the coners of the brillouin zone . elementary excitations are then dominated by electrons in the vicinity of these two points . although the two fermi points are connected by wavevectors @xmath22 ( see fig . [ fig : dispersion ] ) , the corresponding susceptibility is not divergent and it only shows a finite maximum , instead of a divergence , due to the vanishing dos at the dirac points . indeed , the magnetic ordering obtained with our kpm - langevin dynamics has the nesting wavevectors @xmath208 shown in fig . [ fig : dirac](a ) . interestingly , this spin configuration coincides with the famous coplanar @xmath209 structure , which is also the ground state of the nn antiferromagnetic heisenberg hamiltonian @xmath203 ( spins in each triangle form a 120@xmath198 structure with zero total moment ) . this suggest that nn exchange is the dominant electron - mediated spin - spin interaction , similar to the case of the second van hove singularity . the @xmath207 structure with staggered vector chiralities @xmath205 on the two inequivalent triangles is selected by longer - range terms of the effective interaction . our kpm - langevin simulations show that this coplanar state is stable up to @xmath210 . a symmetry - based approach analogous to the one described in sec . [ symmetry ] for the @xmath53-point ordering can be applied to the @xmath39-point case , i.e. , the valleys at which the dirac points are located for filling fraction @xmath20 . the little group of the @xmath211 vectors is @xmath212 . in addition , @xmath213 modulo a reciprocal lattice vector . following the same reasoning , we first find all scalar orders which transform as @xmath124 irreducible representations of the kagome lattice symmetry group . we identify two distinct ordered states modulated by the valley momenta @xmath214 . by embedding them in a spinful setting , with the requirement of translational invariance modulo global rotations , we obtain a single spin ordered state : the coplanar @xmath209 state of fig . [ fig : dirac ] . this result can be intuitively understood by considering the two independent functions @xmath215 and @xmath216 defined on a triangular bravais lattice . these two functions transform as @xmath124 representations of @xmath121 . one can then embed them in a spinful setting as @xmath217 , where @xmath30 denotes the three sublattices of kagome lattice . the three unit vectors @xmath218 form a 120@xmath198 structure with zero vector sum , and @xmath219 , where @xmath220 is the normal of the coplanar @xmath221 and @xmath222 vectors . interestingly , the electron system remains gapless ; the @xmath42-point of the reduced brillouin zone remains doubly degenerate , in addition to the degeneracy required by the combined translation - rotation . this additional degeneracy can be understood by examining the interaction matrix between the two valley dirac points . the vector order parameter with momenta @xmath208 has the form @xmath223 . the exchange coupling in this case becomes @xmath224 where the eigenvectors at the two dirac points are @xmath225 and @xmath226 . substituting these eigenvectors into the above expression , we find that the interaction matrix in the valley - spin space has the form @xmath227 $ ] where @xmath92 is pauli matrices acting on the valley space and @xmath228 is the valley index . straightforward diagonalization of this @xmath229 matrix gives a double degenerate eigenvalue at @xmath230 and two nonzero @xmath231 , consistent with the numerical calculation of the electron band structure for the @xmath207 order [ fig . [ fig : dirac](b ) ] . the magnetic ordering of the klm at filling fraction @xmath50 represents a difficult degenerate perturbation problem . the dispersive @xmath52 band touches the flat band @xmath48 at the @xmath42-point . consequently , the two @xmath232 states ( with @xmath233 ) from the dispersive band can couple to the extensively degenerate states with nonzero momentum in the flat band . our kpm - langevin dynamics simulations show that the ground state at filling fraction @xmath50 is another triple-@xmath56 magnetic order , shown in fig . [ fig : qfp](a ) , that transforms according to the @xmath117 irreducible representation discussed above . this state is characterized by the following non - zero order parameters @xmath234 where @xmath235 are three orthogonal unit vectors . the magnetic moments are coplanar within each sublattice , similar to the state shown in fig . [ fig : vh2](a ) . the presence of reflection symmetries in this magnetic order again precludes topologically nontrivial electronic states @xcite . this is consistent with the fact that the system remains gapless [ fig . [ fig : qfp](b ) ] at the @xmath42-point @xcite . ( a ) the @xmath117 symmetry triple-@xmath56 noncoplanar spin order on the kagome lattice . spins at the three inequivalent sublattices are indicated by different colors . it is the ground state at filling fraction @xmath50 for weak couplings . the same magnetic order is also the ground state at filling fraction @xmath236 for intermediate couplings . ( b ) the corresponding band structure with exchange coupling @xmath237 . the two dashed lines indicate the fermi levels at @xmath236 and @xmath50 filling fractions . ] [ cols="<,<,<,<",options="header " , ] to shed light on the selection of the triple-@xmath56 order , we note that the couplings between electrons are proportional to the product @xmath238 [ eq . ( [ eq : coupling ] ) ] . at filling fraction @xmath50 the basis of the doubly degenerate @xmath44 states are @xmath239 and @xmath240 . any @xmath44 state with eigenenergy @xmath241 can be expressed as a linear combination of these two eigenvectors . the eigenvectors at the three @xmath53-points with energy @xmath242 are @xmath243 , @xmath244 and @xmath245 , which lie completely within this subspace and thus give rise to the largest overlap with the @xmath44 states . another interesting spectral feature of the @xmath117 symmetry triple-@xmath56 state is the appearance of dirac nodes at only _ one _ of the two inequivalent @xmath39-points [ fig . [ fig : qfp](b ) ] . since the two @xmath39-points are related by time - reversal symmetry , which valley point remains gapless depends on the sign of the magnetic order parameter . the fact that the dirac nodes should appear only at one of the two valleys is again in agreement with the symmetry - breaking of the @xmath117 irreducible representation . for large enough couplings , the remaining dirac point is an isolated crossing between the lowest two bands , i.e. there is no overlapping between the two bands . with the aid of unbiased kpm - langevin dynamics simulations , we find that the magnetic order shown in fig . [ fig : qfp](a ) is also the ground state at filling fraction @xmath236 for intermediate coupling constants @xmath246 . the fermi `` surface '' thus shrinks to the remaining dirac nodes at one of the @xmath39-points . further symmetry - breaking perturbation can gap out the residual dirac point and give rise to a topological insulating state . the dispersion of the @xmath52 band in the vicinity of this fermi point is quadratic . the finite dos at such a fermi point renders it susceptible to weak perturbations @xcite . moreover , the @xmath247 berry flux around this band - crossing point indicates that gapping out the fermi point would lead to a topologically nontrivial insulating state . however , the existence of a flat band is a rather special limit in real systems . indeed , inclusion of , e.g. next - nearest - neighbor hopping @xmath248 lifts the flat - band degeneracy . it is thus also interesting to investigate the magnetic ordering of the klm at the same @xmath50 filling in the presence of a small @xmath248 . to avoid unnecessary complications , we consider a small _ negative _ @xmath249 , such that the new @xmath48 band is bending upward , i.e. the minimum of @xmath48 is at the @xmath42-point . in this situation , we are left with an isolated quadratic band crossing point at @xmath44 . the magnetic ground state is expected to have a @xmath250 long - range order . by performing unbiased large - scale kpm - langevin dynamics simulations at , e.g. @xmath251 and @xmath252 , we indeed find a @xmath250 magnetic order , in which spins form a nearly 120@xmath198 umbrella structure in each triangle @xcite . although the out - of - plane canting angle is rather small ( @xmath253 ) , the quadratic fermi point is gapped out by the spin noncoplanarity and the electronic ground state is a qah insulator @xcite ; see appendix [ appendix2 ] for more details . the band structure of the kagome lattice gives rise to a rich variety of singular fermi surfaces at certain commensurate filling fractions . therefore , in this work we have addressed the question : what weak - coupling instabilities of the klm are triggered by these singularities ? we have presented a systematic symmetry - based approach has been shown to considerably limit the number of candidate spin ground states . the full variational space of potential magnetic orderings is determined by identifying the wave - vectors that maximize the magnetic susceptibility of the conduction electrons . this space is not necessarily small for lattices , such as kagome , that have more than one atom per unit cell . the potential magnetic orderings are classified according to the irreducible representations of the small group of the ordering vectors . we have demonstrated that imposing symmetry constraints reduces the large variational space to a small number of candidate ground states . our main results are summarized in table [ tab - summary ] . to determine the ground state magnetic ordering of the kagome klm for specific fillings , we have complemented our analytical study with two complementary numerical techniques . first , we have performed a @xmath254 variational fourier - space calculation , assuming a certain magnetic unit cell . second , we have performed large - scale numerical simulations based on the recently developed kpm - langevin algorithm . this method makes it practical to simulate the very large lattice sizes required to capture logarithmic divergences in the magnetic susceptibility . our numerical results coincide with the results of our analytical treatment when the latter is applicable , and prove that the high - symmetry spin configurations are good variational states . the kagome lattice was shown to have three distinct spectral features , giving rise to very interesting spin textures for the corresponding electron fillings . at the two van hove fillings , i.e. @xmath59 and @xmath60 , we have found spin states modulated by the three inequivalent nesting vectors , both noncoplanar and chiral . while the corresponding electron state at @xmath24 is quantum hall insulator , the other one preserves a fermi point at the @xmath8 filling . the filling fractions @xmath20 and @xmath43 exhibit dirac points and a quadratic band crossing , respectively . at the dirac point filling we find the well - known coplanar @xmath209 magnetic ordering . the triple-@xmath255 magnetic ordering obtained numerically for @xmath43 ( which is different from the triple-@xmath255 ordering observed at the van hove fillings ) is quite unexpected , if we naively assume that the magnetic ordering is triggered by the instability of a quadratic fermi point , which is located at the zone center . indeed , by lifting the full degeneracy of the flat - band with a finite second - nearest - neighbor hopping @xmath248 , we find a @xmath256 magnetic order , as expected from an isolated quadratic band crossing point . in addition to these special filling fractions , we have considered two other fillings , which are commensurate with a quadrupling of the magnetic unit cell implied by triple-@xmath255 ordering . they correspond to @xmath61 and @xmath62 . intriguingly , at @xmath62 we find the same triple-@xmath255 state as for @xmath59 . the electronic spectrum is gapped for large enough @xmath5 . at @xmath61 we find a magnetic ground state which leads to isolated degeneracies , i.e. dirac points . spontaneous chern insulators appear for different filling factors of the kagome lattice , i.e. @xmath59 , @xmath62 , and @xmath43 . by considering that similar results have been obtained for other lattices @xcite , we can safely conclude that kondo - lattice systems are rather strong candidates for realizing spontaneous qah states at ambient temperature . while realistic band structures are much more complex than the simple tight - binding models considered here , and in previous works , the weak - coupling instabilities that we have discussed here are only sensitive to singular features of the fermi surface . because these singular features also appear in more realistic band structures , our main conclusions remain relevant . we also note that noncoplanar magnetic ordering is a prerequisite for inducing the qah effect . the converse , however , is not true . not all noncoplanar spin orderings lead to a qah effect , as we have shown based on symmetry arguments . importantly , our results confirm that , in contrast to local moment magnets , noncoplanar orderings are quite ubiquitous in itinerant magnets . finally , the rather large lattice sizes amenable to the kpm - langevin dynamics method also allow us to investigate magnetic ordering at arbitrary filling fractions and coupling strength . since spin configurations at these conditions most likely are characterized by incommensurate wavevectors , large - scale unbiased numerical minimization ( which is insensitive to boundary conditions ) is required . indeed , we have found unusual incommensurate structures and skyrmion arrays in the kagome klm with intermediate coupling constant . a detailed characterization of these magnetic orders will be left for future studies . in this appendix , we outline the variational minimization method for magnetic orderings at the two van hove filling fractions . as discussed in the main text , the combination of divergent dos at the saddle points and perfect fermi surface nesting , fig . [ fig : dispersion](c ) , gives rise to a logarithmically squared divergent susceptibility at the three nesting vectors @xmath106 . one thus expects magnetic textures dominated exclusively by these three wavevectors . in real space , the most general spin state modulated by the three nesting vectors consist of a quadrupled unit cell with 12 spins . the four inequivalent up triangles of the kagome lattice are located at @xmath257 , where @xmath258 , ( @xmath259 , with @xmath260 integers ) and we have introduced @xmath261 for simplicity . in terms of the order parameters , the different spins are expressed as @xmath262 @xmath263 for completeness , we have included the @xmath264 component @xmath265 corresponding to uniform ordering at each individual sublattice . instead of working with the order parameters , we consider the most general spin configurations @xmath266 ( @xmath267 ) within the extended unit cell . this leaves us with 24 independent variational parameters @xmath268 characterizing the orientations of classical spins with fixed length . ( a ) the total energy ( arbitrary units ) as a function of the canting angle @xmath269 for the @xmath256 umbrella state . the inset shows the three spins of the umbrella state in a triangular unit cell . this curve is obtained using the following parameters : @xmath270 , @xmath271 . ( b ) the corresponding band structure with exchange coupling @xmath272 , @xmath271 and canting angle @xmath273 . the dashed line indicates the fermi level at @xmath50 filling . ] with the periodic spin structure specified in eq . ( [ eq : quadrupled ] ) , the electron hamiltonian can be diagonalized using fourier transformation . it contains two terms : @xmath274.\end{aligned}\ ] ] the @xmath175 summation is over the reduced bz . the matrix @xmath275 in the first hopping term is given in eq . ( [ eq : hopping ] ) , while the exchange coupling term is @xmath276 where @xmath277 denotes the basis vectors for the three sublattices . the factor @xmath278 encodes the momentum conservation ; it is symmetric with respect to the first two indices . @xmath279 when the three indices are all different and is zero otherwise . for a given spin state , the total energy at @xmath254 is given by @xmath280 , where the eigenenergies @xmath281 are obtained by diagonalizing the @xmath282 matrix : @xmath283 . we then employ the simulated annealing algorithm to minimize @xmath284 with respect to the angle parameters . starting from random initial configurations and using different rates of decreasing the effective temperatures , we robustly obtain the two triple-@xmath56 states ( @xmath147 and @xmath116 irreducible representations ) at the respective van hove filling fractions . here we consider the kagome kondo - lattice model with an additional second - nearest - neighbor hopping @xmath285 at filling fraction @xmath50 . as discussed in the main text , the flat band degeneracy is lifted by the additional @xmath248 , and an isolated quadratic band - crossing is left at the @xmath42 point . our large - scale langevin dynamics simulations find a @xmath197 state with spins in each triangle pointing almost at @xmath204 to one another . it is possible that the spins develop a uniform out - of - plane canting since such noncoplanar structure might completely gap out the quadratic fermi point at @xmath50 filling . in order to examine the ground - state spin order in more detail , we assume a @xmath197 umbrella structure with a uniform canting @xmath269 and compute the @xmath254 total energy as a function of the canting angle ; the result is shown in fig . [ fig : umbrella](a ) for @xmath270 and @xmath271 . interestingly , the energy minimum is reached at a nonzero small canting angle @xmath286 . in general the canting angle is rather small ; it depends on both the exchange coupling @xmath5 and the second - neighbor hopping @xmath248 . [ fig : umbrella](b ) shows the electronic band structure of the umbrella state . both the dirac and quadratic band crossing points are gapped out in this spin state . at @xmath50 filling , the system is an insulator with spontaneous quantum hall effect as pointed out in ref . @xcite . we thank useful discussions with s. ghosh , m. lawler , y. motome , and m. udagawa . work at lanl was carried out under the auspices of the u.s . doe contract no . de - ac52 - 06na25396 through the ldrd program . j.w.f.v acknowledges support from the dutch science foundation nwo , and from fom . the large - scale numerical simulations were performed using the ccs-7 darwin cluster at lanl . chang , j. zhang , x. feng , j. shen , z. zhang , m. guo , k. li , y. ou , p. wei , l .- l . wang , z .- q . ji , y. feng , s. ji , x. chen , j. jia , x. dai , z. fang , s .- c . zhang , k. he , y. wang , l. lu , x .- c . ma , q .- k . xue , science * 340 * , 167 ( 2013 ) .

we observed an entire 1.7 day orbit of the x - ray binary hercules x-1 with the far ultraviolet spectroscopic explorer ( _ fuse _ ) . changes in the ovi line profiles through eclipse ingress and egress indicate a keplerian accretion disk spinning prograde with the orbit . these observations may show the first double - peaked accretion disk line profile to be seen in the hercules x-1 system . doppler tomograms of the emission lines show a bright spot offset from the roche lobe of the companion star hz her , but no obvious signs of the accretion disk . simulations show that the bright spot is too far offset from the roche lobe to result from uneven x - ray heating of its surface . the absence of disk signatures in the tomogram can be reproduced in simulations which include absorption from a stellar wind . we attempt to diagnose the state of the emitting gas from the ciii , ciii , and niii emission lines . the latter may be enhanced through bowen fluorescence . = 1 based on observations made with the nasa - cnes - csa far ultraviolet spectroscopic explorer . fuse is operated for nasa by the johns hopkins university under nasa contract nas5 - 32985

0705.4319

hercules x-1 ( discovered by tananbaum et al . 1972 ) is one of the most frequently observed x - ray binary systems . the intermediate mass of the donor star , @xmath5 m@xmath6 , leads to a wealth of behavior seen in both low mass and high mass systems . although we suspect the mass flow is primarily through roche lobe overflow giving rise to an accretion disk , as in the low mass x - ray binaries ( lmxb ) , the variable p cygni profiles observed in uv lines suggest either a transient stellar wind or a stellar wind that is photoionized in some regions ( boroson , kallman , & vrtilek 2001 ) . there are theoretical grounds as well to expect that winds may arise in this system when x - rays from the neutron star heat the surface of the normal star or accretion disk ( arons 1973 ; davidson & ostriker 1973 ; basko et al . 1977 ; london , mccray , & auer 1981 ; begelman , mckee , & shields 1983 ; begelman & mckee 1983 ) . most lmxbs , including z - sources and atoll - sources , do not show persistent pulsations , perhaps because they have neutron stars with low magnetic fields . these sources thus lack an essential window into their kinematics . light travel - time delays in the 1.24 second her x-1 x - ray pulsation period determine that the neutron star is in a nearly circular 1.7 day orbit with semimajor axis a@xmath7 seconds . the orbital period is slowly lengthening from mass loss in the system ( deeter et al . 1991 ) . fortuitously , her x-1 has high orbital inclination ( @xmath8 degrees ) , allowing total x - ray eclipses to constrain the size of the donor star , and the high galactic latitude offers only small reddening , allowing the system to be observed at crucial uv and even euv wavelengths . the system s most mysterious variability is the 35-day x - ray high and low cycle ( giacconi et al . 1973 ) . for 811 days ( the `` main - on state '' ) of this cycle her x-1 emits x - rays with a @xmath9 erg s@xmath10 total x - ray luminosity . for a @xmath11 day cycle ( the `` short - on state '' ) , halfway through the 35 day cycle , her x-1 s output is several times lower . for the remaining portion of the 35 days , the off state , the observed x - ray flux is only several percent of that during the main - on state . the behavior of the optical flux and x - ray spectrum over the 35-day period shows that the 35-day variability is not isotropic . instead , the x - ray emission is merely obscured by a warped accretion disk whose shape precesses " globally throughout a 35-day period . the optical emission from the normal star varies with the 1.7 day binary orbit . its spectral type is variously classified as a through f , because x - rays from the neutron star heat the surface of the facing side of the roche lobe . the spectral type continues to change with the orbital period throughout the longer 35-day cycle , implying that x - rays continue to heat the surface of hz her . the transition between the off and main - on state is only a few hours , whereas the short - on state is entered into more gradually . this indicates that regions of the disk at different radii may obscure the central source , in which case the disk must precess globally with the same period . the variation in pulse shape with the long - term period is sometimes taken to imply that the innermost edge of the accretion disk precesses with the 35 day period , obscuring local regions on the neutron star ( scott , leahy , & wilson , 2000 ) . the physical cause of a 35-day global disk precessing disk warp , or a definitive relation between system parameters and disk precession period , remains unknown and is a major goal of investigations of the system . it has been proposed , but not confirmed , that reprocessed x - ray radiation pressure ( maloney & begelman 1997 ; wijers & pringle 1999 ) or an x - ray driven disk wind and corona ( meyer , meyer - hofmeister , 1984 ) could propagate the warped shape in the disk . similar " long - term periods have been observed in other x - ray binaries , including lmc x-4 . her x-1 serves as a prototype of this behavior . x - ray dips present another variability that has not been explained definitively . crosa & boynton ( 1980 ) showed that the dips recur every 1.65 days , near , but not at , the 1.62 day beat period between orbital and precessional periods . this was confirmed by long - term x - ray observations ( scott & leahy , 1999 ) . the dips are thought to be associated with the gas stream between the stars , but there is no consensus explanation for them or their period . as befits such a frequently observed , enigmatic , yet prototypical source , the methods of spectroscopy have been applied to the system at a wide range of wavelengths . these studies have borne fruit with determinations of the system s elemental abundances and the orbital motion of the normal star ( the neutron star s pulse delays indicate only the motion of one component of the binary ) . high resolution spectroscopy in the x - ray range using _ xmm _ ( jimenez - garate et al . 2002 ) shows a multitude of x - ray emission lines presumably from an accretion disk corona . the line ratios indicate that the gas is enhanced from cno processing from a massive progenitor . jimenez - garate et al . ( 2005 ) confirmed the cno enhancement by observations of more than two dozen emission lines originating in an accretion disk corona . model predictions of the disk corona s response to illumination by the central x - ray source are in reasonable agreement with the observed fluxes for low and moderate z elements ( o through s ) , but the fexxv , fexxvi lines are several times brighter than predicted ( jimenez - garate et al . 2005 ) . from observations of optical absorption lines ( reynolds et al . 1997 ) , and from optical pulsations that result from the x - rays from the neutron star periodically striking the surface of the normal star ( middleditch & nelson 1976 ) , we know that the normal star has a mass @xmath12 m@xmath6 and the neutron star has a mass 1.5@xmath13 m@xmath6 . the optical spectrum shows absorption lines from hz her , emission at the bowen blend near 4640 ( see schachter , filippenko , & kahn 1989 for a discussion of the formation of these lines ) , and emission from heii@xmath14 . still et al . ( 1997 ) performed doppler tomography on these optical emission lines but could not conclude that they arose in a symmetric accretion disk . the uv spectrum is crucial for an understanding of the accretion disk and the x - ray illuminated face of hz her . the continuum emission from both the illuminated star and the disk peaks in the uv and the disk contributes a greater fraction than at optical wavelengths ( cheng , vrtilek , & raymond 1995 ) . from models of the variable uv continuum as observed with the international ultraviolet explorer ( _ iue _ , vrtilek & cheng 1996 ) showed that a change in the accretion disk precession could explain an anomalously low period of x - ray emission . to study global accretion one should observe gas at ionization stages and temperatures resulting from x - ray illumination of the disk , and one would need to resolve doppler - shifted velocities that correspond to the orbital motion of the neutron star ( 160 km s@xmath10 ) , motion in the accretion disk ( expected to be @xmath15 km s@xmath10 at the edge of the disk ) , and the velocities expected in a disk or stellar wind ( @xmath16 km s@xmath10 ) . the strong uv resonance lines from nv , siiv , and civ , first seen with _ iue _ ( howarth & wilson 1983b ) , presumably result from photoionization of the accretion disk and hz her ( raymond 1993 , ko & kallman 1994 ) . these lines are much stronger than the strongest optical high ionization line , heii@xmath14 , both in absolute flux and equivalent width . observations with the faint object spectrograph ( fos ) aboard the hubble space telescope ( _ hst _ ) showed that these lines are still present at a few percent of maximum brightness during mid - eclipse when the disk and heated star should be entirely obscured ( anderson et al . the source of this emission may be an expanding wind . the goddard high resolution spectrograph ( ghrs ) on _ hst _ first resolved these emission lines at @xmath17 km s@xmath10 resolution ( boroson et al . , 1996 ) to discern variable broad and narrow emission components . the _ hst _ space telescope imaging spectrograph ( stis ) confirmed that the resonance lines have at least two components ( vrtilek et al . a broad component arises on the accretion disk while a narrow line component may be associated with hz her . prior to the stis observations , the accretion disk had only been observed as it contributed to the continuum light curve or obscured the central x - rays . during eclipse ingress and egress , the broad lines seen with stis behaved as expected for lines from an accretion disk rotating prograde with the orbital direction . the blue edge of the line was obscured first in eclipse ingress and appeared first in egress . observations with the far ultraviolet spectroscopic explorer ( _ fuse _ ) are complementary to the existing _ hst _ observations . long - term variability in the system makes it impossible to combine rigorously data from different epochs . analyzed separately , however , the _ fuse _ wavelength range of 900 - 1200 offers similar advantages and powerful consistency checks to analysis of the _ hst _ bandpass of 1200 - 1700 . both wavelength ranges have strong resonance lines that respond to x - ray photoionization . observations with the hopkins ultraviolet telescope ( _ hut _ ) showed that the ovi doublet has flux comparable to the nv doublet , the brightest near uv line ( boroson et al . 1997 ) . the resonance line doublets offer optical depth information through their doublet ratios , but if the lines are as broad as the doublet separation , it may be impossible to determine the individual contribution of each line where they overlap . the separations of civ@xmath18 , siiv@xmath19 , nv@xmath20 , ovi@xmath21 , svi@xmath22 correspond to doppler shifts of 500 , 1900 , 960 , 1650 , 3600 km s@xmath10 , respectively . the far uv lines ovi and svi compare favorably with the near uv lines ( i.e. have greater separation and are less likely to overlap ) , except for siiv , which suffers from confusion with an oiv blend near 1400 . for the present observations , we observed an entire 1.7 day binary orbit with _ fuse_. a major goal of this program was to apply the method of doppler tomography ( 7 ) , which has the advantage of diagnosing the accretion flow of different systems without bringing to bear more than a few assumptions . table 1 shows the measured physical parameters of the system and parameters we adopt for our models . with _ fuse _ ( the far ultraviolet spectroscopic explorer ) we extend observations in the uv spectral range to 900 - 1190 , a range observed only once before using _ hut _ , the hopkins ultraviolet telescope aboard the astro-1 , carried aboard the space shuttle but not placed in orbit ( boroson et al . 1997 ) . while _ hut _ had a resolution of @xmath23 , _ fuse _ has a resolution of @xmath24 . _ fuse _ is a nasa _ origins _ mission operated by the johns hopkins university . four aligned telescopes feed two identical far - uv spectrographs . with resolution r@xmath25 fuse approaches _ hst _ in its utility for our program ; the time coverage of the _ hst _ space telescope imaging spectrograph ( stis ) was limited because the detectors were turned off when the spacecraft passed through the south atlantic anomaly . the fuse mission is described in more detail in moos et al . ( 2000 ) and its on - orbit performance is described in sahnow et . al . ( 2000 ) . our _ fuse _ observations began on june 9 , 2001 at 7:47 ut . table 2 shows the log of exposures , each of which is integrated over each orbit of the earth , with gaps when her x-1 goes below the horizon . we use the orbital ephemeris of deeter et al . ( 1991 ) to determine the orbital phases of our observation . the exposure times listed in table 2 are in most cases equal to the raw observation time . however , for cases such as observation 13 , interrupted by a passage through the south atlantic anomaly and not occultation by the earth , the exptime listed is in general the minimum good exposure time for any _ fuse _ detector . the data were obtained through the lwrs aperature and in timetag mode . we used the calfuse pipeline software version 3.0.8 to extract and calibrate the data from all four _ fuse _ telescopes . below 1100 , where emission features are sharp , we added an offset to each wavelength scale , in intervals of 0.025 , so that the absorption and emission features from each detector best agreed . we tested our wavelength calibration against the interstellar siii@xmath26 absorption line , which we found to have a mean heliocentric velocity of @xmath27 km s@xmath10 . the standard deviation of the centroid of this line from orbit to orbit was @xmath28or @xmath29 km s@xmath10 . the s / n of the data was @xmath30 per 0.1 pixel in the continuum in the region @xmath31 and @xmath32 near 1100 . the s / n within the ovi line had greater variation with orbital phase , as the doublet changed both in shape and in strength . at @xmath33 , the peak s / n within the doublet was @xmath34 per 0.1 pixel , while at @xmath35 , the peak s / n was @xmath36 per pixel . for further analysis , we skip the region between 11201160 in the 1b lif detector . this region suffers from a systematic decrease in counts known as the worm .. in figures 1 and 2 , we show average observed fuse spectra at orbital phases when the emission is dominated by the disk and star , respectively . for the disk - dominated spectrum , we use observation number 20 , at @xmath37 . for the star - dominated spectra , we average observations number 7 through 11 ( phases @xmath38 ) . for the _ fuse _ spectra at orbital phases dominated by disk emission , we compare with the stis spectrum ( observation root name o4v401010 ) during a short - on state at @xmath39 . for the spectrum dominated by star emission , we use the stis observation with root name o4v452050 , observed during a main - on state approximately 5 months earlier , on january 24 , 2001 , starting at mjd 51933.779933 and lasting 2620 seconds . the mean orbital phase of this exposure was @xmath40 . this spectrum has not previously been published . we compare the spectra with the time - weighted average of our continuum models ( described in 5 ) . we indicate prominent uv emission lines from the star system . geocoronal ( airglow ) lines produced in the earth s atmosphere are also present ( see feldman et al . , 2001 ) : hilyman-@xmath41 973 , oi5 989 , hilyman-@xmath42+oi4 1026 - 1027 , and ni 1167 . interstellar features in the _ fuse _ spectra are interesting not only for the direct information they provide on the interstellar medium ( ism ) , but a proper accounting of these features can remove systematic errors from the analysis of the line and continuum emission from the system itself . the neutral hydrogen column density , n@xmath43 , has previously been determined to be @xmath44 @xmath45 ( boroson et al . 2000 ) from the wings of the saturated lyman @xmath46 line as observed with the _ hst _ stis . the _ fuse _ bandpass includes further saturated lines in the lyman series , and these are consistent with n@xmath47 @xmath45 . this n@xmath43 is also consistent with the e(b - v ) value of 0.018 according to the bohlin ( 1975 ) relation . from the galactic latitude of 37.52@xmath48 we should expect a sightline with less h and h@xmath49 than typically seen through the galactic disk . indeed , while absorption lines from rotational levels @xmath50 through @xmath51 are readily identified , they are not saturated . thus the level populations should be easy to measure and it should be easy to compensate for the effects of the absorption lines on the spectra . in figure 3 , we show a patch of the time - averaged _ fuse _ spectrum , with the absorption profiles expected from source with a flat spectrum absorbed by columns of ( 2.8 , 7.5 , 4.1 , 3.4)@xmath52 @xmath45 for absorption from rotational levels @xmath53 , respectively . we generate the h@xmath49 profiles from the templates of mccandliss ( 2003 ) , which are based on abgrall et al . ( 1993a , b ) . we have assumed a line velocity parameter @xmath54 km s@xmath10 and have convolved the profiles with a gaussian to simulate the _ fuse _ line spread function . the two lowest rotational energy levels should have populations given by a boltzmann factor , taking into account the statistical weights of the two levels , @xmath55 . the energy difference between the levels is @xmath56 k. we then estimate that the temperature of the intervening h@xmath49 gas is @xmath57 k. this is hotter than found for h@xmath49 gas from disk stars , consistent with the high galactic latitude of the line of sight ( gillmon , shull , tumlinson , & danforth 2006 ) . the average molecular fraction , defined by @xmath58 is therefore @xmath59 from the observed h@xmath49 lines , which have n(h@xmath49)@xmath60 @xmath45 . ratios of higher rotational levels than @xmath61 are determined not by temperature and collisional excitation but by background fuv radiation . the h2 column density observed for her x-1 places it just above the boundary at which h2 clouds start to become optically thick to the fuv radiation . gillmon et al . ( 2006 ) state that @xmath62 is a typical molecular fraction below this boundary . thus the first four levels of rotational excitation of h@xmath49 appear to match observed absorption features and to be caused by h@xmath49 clouds that are not out of the ordinary . the lines vary with the binary orbit in a manner similar to that previously observed in the near uv with _ the flux peaks generally near @xmath35 , when the x - ray heated face of hz her points toward the observer , although there may be a dip very close to @xmath35 as the accretion disk occults the star . in figure 4 a through i respectively , we show the photometric variation with orbital phase of the emission lines svi@xmath63 , svi@xmath64 , ciii@xmath65 , niii@xmath3 , ovi@xmath66 , ovi@xmath67 , siv@xmath68 , pv@xmath69 , and ciii@xmath70 . we have subtracted the background flux using a model that we present in 5 . for the niii line , we remove the flux of nearby airglow lines . in table 3 , we show the cross - correlation coefficients between the measured lightcurves of the spectral lines and the lightcurves of the disk and star contributions to the continuum lightcurve , as determined by our continuum model . the uv continuum varies , as the optical continuum does , with the 1.7 day orbital period . the continuum generally peaks near @xmath35 , when the x - ray heated face of the roche lobe points towards the viewer , but a portion of the normal star may be blocked by the accretion disk , and depending on long - term phase , the actual peak may occur within @xmath71 . the optical / uv orbital light curves do not repeat exactly over the 35-day disk precession cycle . the disk may block x - rays from heating portions of the star , and as it precesses it projects varying areas into the line of sight . the precessing disk should have different eclipse light curves throughout the 35 day cycle . to fit the uv continuum observed with _ fuse _ , we assume that x - ray heating of hz her and the accretion disk cause the entirety of continuum emission . the details of our simulation are similar to those of vrtilek et al . ( 1990 , especially the appendix ) , and howarth & wilson ( 1983a ) ( which was an elaboration on an earlier analysis of gerend & boynton , 1976 ) , and binary simulation codes by wilson & devinney ( 1971 ) . the method is summarized in appendix a. we fit the spectrum using a reddening @xmath72 and for the extinction curve , we use equation 5 in cardelli , clayton , & mathis ( 1989 ) . we also use a more modern determination of the distance to her x-1 ( reynolds et al . this greater distance requires a larger @xmath73 to reach the same continuum flux . thus the scale of our @xmath73 values are large compared with those reported earlier from _ iue _ and _ hst _ observations . the _ iue _ observations found a range of @xmath74 to @xmath75 ( given changes in the long - term phase during an anomalous low state ) , whereas the _ hst _ found @xmath76 . when fit our continuum model to the observed _ fuse _ spectrum from each _ fuse _ observation , we find a range of @xmath77 to 8.38 . for our fits we ignore wavelength regions that contain prominent emission or absorption lines . we correct for absorption from the first 4 rotational energy levels of interstellar h@xmath49 by multiplying the spectrum by the model presented in 3 . table 3 gives the best - fit values of the mass accretion rate , @xmath73 , which we allow to vary as a free parameter for each _ fuse _ orbit . we also list the reduced @xmath78 . we have used our continuum fits to examine the continuum photometry . the model allows us to separate the continuum emission into contributions from the heated face of hz her and from the accretion disk . in figure 5 we show the model flux near the ovi lines split into disk and stellar components . we made a simple model for emission from a symmetric precessing accretion disk undergoing eclipse by the roche lobe of its companion . we then allowed free parameters of that model to vary in order to fit the observed ovi profiles during eclipse ingress and egress . the disk and roche lobe geometry are based on the model of howarth & wilson ( 1983a ) , while the formation of the emission lines follows the model of horne ( 1995 ) , and is described in appendix b. we allow as free parameters the exponent of a radial power law of optical depth and elements of the mach turbulence matrix . we allow the normalization of the line flux at each orbital phase to vary as a free parameter . we calculate the eclipsing edge of the roche lobe of hz her assuming corotation with the orbit . before fitting the model to the data , we subtracted the mid - eclipse spectrum from all spectra . the origin of the uv mid - eclipse spectra was explored by anderson et al . ( 1994 ) who concluded that it probably does not arise in the accretion disk . the mid - eclipse ovi lines are broad , extending from -200 to @xmath79 km s@xmath10 heliocentric velocity . we also apply narrow gaussian absorption lines to our model spectra in order to simulate a possible cii interstellar line near 1036 , a possible cii@xmath80 line near 1037 , and a oi line near 1039 . we also include narrow interstellar absorption near the rest wavelengths of the ovi doublet . a narrow absorption line can be seen near the blue component of the doublet at @xmath81 , but the weaker line at 1037.6 can not be seen explicitly . we fix the wavelengths , widths , and optical depths of these lines and do not let them vary in our fit . we fix all the parameters describing the disk and orbit to those given in howarth & wilson ( 1983a ) , except the outer radius we fix to @xmath82 cm , following cheng et al . ( 1995 ) . the results are shown in figure 6 . the reduced @xmath78 of the fit was 2.8 with 1418 degrees of freedom . the power law index for the radial dependence on optical depth was @xmath83 , between the @xmath84 result found for the nv doublet by similar methods in boroson et al . ( 2000 ) and the @xmath85 value expected from the simulations of raymond ( 1993 ) . although we based our fits on the model of horne ( 1995 ) which allows for anisotropic turbulence , the fits do not unambiguously settle on particular values of the mach matrix . here , for simplicity , we present a model with isotropic mach=1 turbulence . as with the resonance doublets in the near uv , the observed blue component of ovi near eclipse is stronger than the red component , which suggests that much of the line flux is formed where @xmath86 . however , the doublet ratio is uncertain because strong interstellar cii and oi absorption lines affect only the red ovi doublet component . weaker interstellar lines , which we have not modeled , may be present as well , and may preferentially absorb one or the other ovi component . the model lines are double - peaked outside of eclipse , but one peak of the red component is absorbed by interstellar oi absorption . the observed line profiles are double - peaked at @xmath87 and double - peaked but broader at @xmath88 . observations at @xmath89 and @xmath90 , when narrow line emission predominates , show a dip in the flux at the position of the gap between peaks from the accretion disk line . the line normalizations we found from our disk eclipse model can be compared with the mass accretion rates inferred from our continuum fits . they do not appear correlated . at some of the phases investigated , we observe only a small fraction of the accretion disk , and none of the illuminated star , so that systematic errors in the line or continuum models may make accurate estimates of @xmath73 impossible . in particular , we have considered the disk flat when it is probably warped , we have not calculated the ionization structure of the disk , we have ignored possible emission by the accretion stream , and we have ignored the role of winds in this system . chiang ( 2001 ) presented models of accretion disk eclipse in her x-1 that considered a disk wind . these models were motivated by the lack of double - peaked line profiles in any observations of uv resonance lines . although the current model predicts double - peaked emission lines when the full disk is visible , the s / n is not strong enough to rule out a double - peaked structure . in addition , interstellar absorption lines coincide with three of the four peaks of the ovi doublet . it is clear that a model of a partially eclipsed accretion disk with parameters considered standard from previous work provides an excellent match to this data set . there appear to be too many ways to improve the current fit to justify singling one out . the fit at @xmath91 is particularly poor and we suggest that a process in addition to the partial eclipse of a keplerian accretion disk affects the line profile . figure 6 shows that even though more of the disk is eclipsed than at @xmath92 , the emission lines at @xmath91 are actually stronger . there also appears to be an additional absorption component near the rest wavelength of ovi@xmath93 . the continuum at @xmath91 has not increased over the continuum at @xmath92 as much as the line emission has increased . this is consistent with the earlier result that the x - ray illumination from our line and continuum models at other phases were not correlated . the doppler tomography method , developed by marsh and horne ( marsh & horne , 1988 , marsh 2005 ) takes as input an emission line which is broadened by line of sight motion . the line must be observed over a good sampling of the binary orbit . in analogy with medical tomography , a successful doppler tomogram builds a higher - dimensional view of the object from different observational slices . analysis of doppler shifted emission lines is prone to the confusion of different gas sources which have been projected onto the same observed velocity . the tomographic method presents a transform of the same data into features separated not only by projected velocity , but also by orbital variability . the output tomogram is not a physical image of an accretion disk , but rather an image in `` projected velocity space . '' a component of the emission that varies sinusoidally in velocity ( but remains constant in flux ) is placed at a point in velocity space along a circle with a radius equal to the orbital velocity of that emission component . the phase of the sinusoid determines its position along that circle . the method assumes that all of the variation in a spectral line must be caused by the orbit changing the projection of the motion into the line of sight . ( for example , the change in a spectrum must not be caused by a bright spot being obscured by an opaque object . ) even outside of the conditions in which it strictly applies , however , tomographic images can provide a common view of the data and may suggest directions for further analysis . technically , a doppler tomogram is an inversion of @xmath94 which gives the flux @xmath95 at the projected velocity @xmath96 and the orbital phase @xmath97 as a summation of the velocity space tomogram @xmath98 with a line broadening function @xmath99 which is usually narrow . this integral is an instance of a radon transform . the line broadening function for our fuse spectra is approximately 0.05and for tomographic analysis we assume it is a delta function . the relation between the radial velocity and the two axes of the doppler tomogram , @xmath100 and @xmath101 is given by @xmath102 where @xmath41 is the systemic velocity of the system . the formation of the spectrum from the tomogram can be thought of as follows . the phase @xmath97 determines a direction in the velocity space plane , and moving along that line , summing the intensity perpendicular to the line , one ideally forms the spectral line . at @xmath103 , one views from the right in our figures , at @xmath35 one views from the bottom , and at @xmath33 one views from the left . in this paper , we determine @xmath98 from @xmath95 by means of fourier - filtered back projection , which inverts equation [ eq : tomogram ] through @xmath104 where to calculate @xmath105 from @xmath106 , one takes its fourier transform , multiplies by a ramp filter and a wiener ( optimal ) filter based on the noise level , and then takes the inverse fourier transform . although it may seem as if this filtering complicates the method of back - projection , a ramp filter ( multiplying the fourier transform by the frequency ) is required in order for back projection to produce a rigorous inversion of equation [ eq : tomogram ] . otherwise , back projection would only produce a smeared version of the true tomogram . unfortunately , using a ramp filter also magnifies pixel to pixel noise . we apply a standard wiener filter that assumes a signal amid a noise component that is constant with frequency . we find the power spectrum of the line spectra is well fit by several components of the form @xmath107 ( with @xmath108 ) . the contribution of the signal is negligible at high frequencies @xmath109 , where the power spectrum is fit well by a constant noise level @xmath110 . the action of the wiener filter is then to multiply the power spectrum by @xmath111 , which is nearly 1 when the signal dominates the noise and negligible at high frequencies where the spectrum is almost entirely noise . we used the same form of @xmath112 for each line at each orbital phase , and made a visual comparison between the actual power spectrum and the function @xmath112 . because the filter adapts to different noise levels , the amount of smoothing varies from line to line . for the lines with highest s / n , such as ovi@xmath93 , the turnover frequency for the smoothing filter is on the order of 3 pixels , whereas for weaker lines such as svi@xmath113 , the smoothing is on the order of 10 pixels . we did not apply the popular method of maximum entropy regularization , which awards higher probabilities to those tomograms that provide the least information in the sense that they are smoothest . in figures 714 we show doppler tomograms of the emission lines of svi at 933 and 944 , niii@xmath114 , ovi at 1031 and 1037 , pv@xmath69 , siv@xmath115 , and ciii@xmath2 . the tomogram of ciii@xmath1 is not shown as it is contaminated by the presence of saturated absorption . we have interpolated between orbital phases for the integration , which we perform between @xmath116 and @xmath117 . the grayscale of the tomograms extends linearly from the minimum to the maximum . to understand the ciii multiplet , we consult _ chianti _ , a software package and atomic database described in dere et al . 1997 and landi et al . 2006 . the database , used extensively by stellar and solar astrophysicists , contains energy levels , wavelengths , radiative transition probabilities , and excitation data for many ions . the associated software package , written in idl ( interactive data language ) , can be used to examine how line ratios vary with temperature and density , subject to certain limiting assumptions . we find that ciii contains emission components at @xmath118 , and 367 km s@xmath119 relative to the 1175.59 component . we were able to reduce the artifacts by deconvolving with a triple - gaussian @xmath99 in equation [ eq : tomogram ] . we tried gaussians of equal weight at velocities @xmath120 km s@xmath10 ( figure 14 ) and find there is a single bright spot near the roche lobe . a further problem with the ciii multiplet is that the model of the continuum in this region predicts an absorption dip . to reduce artifacts , we use the simpler and smoother kurucz model atmospheres , with 1resolution for all of the tomograms , instead of the models based on actual stellar spectra observed with _ fuse _ , which we have computed with 0.1 resolution and which include some counting statistics noise . the signal in the tomograms , except as noted above , is concentrated in a peak offset from the roche lobe of hz her . the traditional signature of an accretion disk is not obviously present . accretion disks are expected to cause broad line emission that appears as a ring in a tomogram , with diminished flux within the velocity at the edge of the accretion disk , or about 300 km s@xmath10 for hercules x-1 . if the disk emission is symmetric , we expect it to be centered on the position of the neutron star in velocity space , @xmath121 . there is a simple interpretation of the tomograms in terms of p cygni lines commonly seen in stellar winds from massive stars ( and in _ hst _ spectra of her x-1 , boroson , kallman , & vrtilek 2001 ) . p cygni lines , caused by resonance scattering , are characterized by red - shifted emission and absorption blue - shifted by velocities common in the wind . there is a hint of blue - shifted absorption in the trailed spectrograms at @xmath122 , although there may be an interstellar absorption line as well at these wavelengths and the gap between the peaks of the accretion disk spectrum could also appear to be absorption . we note that for many tomograms in addition to the bright spot at @xmath123 , there is a dark spot at @xmath124 . ( for ovi@xmath125 , this absorption may have a darker lobe near the rest wavelength from interstellar ovi absorption . ) near the start of the phase range used by our tomograms , at @xmath126 , the observer s line of sight comes from the lower right in figure 7 . as the line of sight passes through both bright and dark spots , the observer at @xmath126 sees a diminished central peak in the spectral lines . at @xmath127 , the viewer looks at the system from the bottom of the plot , stacking pixels vertically . the result is red - shifted emission ( the bright spot ) and blue - shifted absorption ( the dark spot ) . although the blue - shifted absorption at @xmath122 may be explained through the analogy of p cygni lines , the red - shifted emission at these phases does not fit this explanation . in a p cygni line , except when a very hot wind is present ( as in a wolf - rayet star ) , all the emission is caused by spherically symmetric scattering , so that conservation of photons requires that the forward - scattered flux be nearly equal to the back - scattered ( `` absorbed '' ) flux ( some portion of the forward - scattered emission is blocked by the star itself ) . in contrast , any blue - shifted absorption in the her x-1 spectrum is only a small fraction of the emission . further , p cygni emission is only red - shifted because it overlaps with the absorption on the blue end ; the actual emission spectrum should be symmetric about zero velocity . it seems most natural to associate the bright spot with the x - ray heated face of the normal star , as it follows roughly the same light curve and appears near the star and far from the disk on the tomograms . yet the placement of the spot , or equivalently , the orbital variation with velocity , can not be accounted for easily by such a model . in figure 15 , we show the observed spectrum near ovi ( bold ) , the spectrum expected if the illuminated surface of hz her emits ovi doppler shifted by the local rotational velocity ( dashed ) , and an empirical gaussian model of the emission lines . the doppler shifts predicted from the rotation of the surface of hz her are less than observed in the narrow emission lines . to model the narrow ovi emission lines we adapt our model of the continuum emission . this model takes into account the x - ray shadow cast by the disk and the eclipse of portions of the normal star by the disk . the reprocessed ovi emission is proportional to the incident x - ray flux . we also experimented with a model in which the x - ray illumination was non - isotropic , illuminating hz her with greater luminosity at @xmath103 than at @xmath33 . this model still does not account for the observed velocities . we also made an empirical model of the lines as gaussians . the flux in the gaussian line emission at each phase was proportional to the flux in the stellar component of our continuum model using kurucz model spectra . the central velocity in this model varies sinusoidally with phase . we added gaussian absorption to correspond to features seen in _ hst _ stis spectra of the nv@xmath128 , siiv@xmath129 , and civ@xmath130 doublets near @xmath122 . we set the maximum projected redshift of the emission lines and blueshift of the absorption to occur near @xmath35 , but delayed by @xmath131 , corresponding to a deflection of a wind by the coriolis effect . the absorption line has maximum covering fraction near @xmath35 that smoothly varies to 0 at @xmath132 . ( any wind is probably confined to the cylinder from the star towards the disk , as the emission lines seen in mid - eclipse are only @xmath133% of the peak line strength , as observed with the _ hst _ fos by anderson et al . both emission and absorption components have constant width . both emission and absorption components vary about a central velocity given by the l1 point . when these features are added to the model of the disk spectrum in 6 , assumed to move with the neutron star s 169 km s@xmath10 orbit , and random counting statistics noise is added , the resulting tomogram ( figure 16 ) resembles the observed tomogram , while showing only a hint of the presence of an accretion disk . without the assumed phase delay of @xmath134 , the bright spot would appear rotated to @xmath135 . the _ fuse _ spectra show emission lines that may serve as diagnostics of density , temperature , or optical depth . the simultaneous measurement of ciii@xmath1 and ciii@xmath136 can provide us with an important diagnostic . the ciii 1176/977 ratio is sensitive to density in the range up to about @xmath137 @xmath138 . at higher densities ( up to @xmath139 ) , the populations of the ground and metastable levels are entirely determined by collisions . in that regime , the 1176/977 ratio depends only on temperature and the optical depths in the lines . we ignored flux from the _ fuse _ lif2b detector near the ciii@xmath1 line , even though this detector recorded flux down to exactly 977 . the edge of the detector causes the flux measured in this region was significantly lower than the other two detectors , so that including this data would have created an artificial dip at wavelengths @xmath140 . the fluxes we measure for ciii@xmath1 are lower limits because of a saturated interstellar absorption line at the rest wavelength . we correct the fluxes for interstellar reddening using e(b - v)=0.018 . we compare the fluxes of the two lines near @xmath141 , when the fluxes peak . if the ciii@xmath1 line behaves similarly to the ovi lines , then the bright narrow emission will , near this phase , have the greatest redshift . if so , the line may be redshifted away from the interstellar absorption and the flux measurement may be more accurate near this phase . if the ovi@xmath93 line were subject to similar interstellar absorption lines as the ciii@xmath1 line , the flux at @xmath141 would be diminished by @xmath14220% . if we measure the raw ciii@xmath1 flux from the two phases surrounding @xmath141 to be @xmath143 erg s@xmath10 @xmath45 , then we find a 1175/977 ratio of @xmath144 . if we measure the ciii@xmath1 flux at exactly the peak at @xmath141 and attempt to correct for interstellar absorption lines , we find a ratio of 0.60@xmath145 . these ratios are at the high end of those predicted by the _ chianti _ software , which uses atomic data from the _ chianti _ database and computes line ratios , given certain simplifying assumptions ( for example , the lines are optically thin and the gas is collisionally ionized and not photoionized ) . if we consider a density of @xmath146 @xmath138 , the lower ratio found above would restrict the temperature to @xmath147 k. the ratio is rendered uncertain observationally by the saturated interstellar absorption line , and theoretically because of optical depth effects . raymond ( 1993 ) presented models of line emission from an x - ray illuminated accretion disk which took into account optical depth through an escape probability formalism . one version of the model , cos " , assumed cosmic abundances while the other , cno " , assumed that abundances had been altered by cno processing . the cno models should be more appropriate for her x-1 . in the two cases the @xmath148 ratios were 1.3 and 1.1 , respectively . the observed emission at @xmath141 may , however , have an origin not on the disk but on the normal star . the bowen fluorescence process ( schachter et al . , 1989 ) arises because of the nearly perfect coincidence of the heii ly@xmath46 and the oiii 2p@xmath149 2p3d resonance line ( @xmath150304 ) , resulting in oiii near uv primary cascades at @xmath1513133 , 3444 and secondary cascades at @xmath150374 back to the ground state . if conditions are right , an additional fluorescence occurs , since the @xmath150374 line is almost coincident with the two 2p 3d resonance lines , resulting in n iii optical primary cascades at @xmath1504634 , 4641 , 4642 . the emission in all of these cascades is completely dominated by the bowen process ; detection of any of them is a clear confirmation . we note that in her x-1 the process requires substantial optical depth ( @xmath152 ) in the ly@xmath46 and @xmath150374 _ pumping _ lines , because otherwise they will simply escape without conversion to a bowen line . in the sun , raymond ( 1978 ) discovered @xmath150304 bowen emission . it was found that measurements of @xmath150703 , a _ ground - state _ bowen cascade which , therefore , can also be produced by collisional excitation , can serve as a density diagnostic , and compared favorably with other solar estimates of @xmath153 . we may use niii@xmath150990 , an analogous ground state bowen cascade produced as a result of the primary cascades ( `` @xmath1504640 '' ; all other and ground - state cascades lie below the lyman limit and hence are unobservable ) . the intensity of niii@xmath154 resulting from bowen fluorescence is related to the intensity of the bowen lines at 4640by @xmath155 where @xmath156 is the branching ratio of 2p-2s@xmath149p@xmath149 2d versus 3p-3s . from chianti , we take @xmath157 and @xmath158 . the remainder of the line intensity is therefore from collisional excitation and is given by @xmath159 . taking into account the reddening toward her x-1 implied by e(b - v)=0.018 , we find a peak flux in the 991 line of @xmath160 erg s@xmath10 @xmath45 . non - simultaneous measurements of the optical @xmath1504640 line have a peak flux at similar orbital phases of @xmath161 erg s@xmath10 @xmath45 ( still et al . the optical flux varies from orbit to orbit , and , moreover , probably includes contributions from ciii as well as niii . these values imply that 20% of the niii@xmath3 line flux is the result of the bowen fluorescence mechanism . we attempt to relate the density to the bowen flux by @xmath162 with @xmath163 the density of niii and @xmath164 s@xmath10 the excitation rate of the 991 transition at t=30,000 k. we estimate an upper bound on the oiii@xmath165 intensity ( the line may be optically thick ) from the oiii@xmath166 line observed with the faint object spectrograph on _ hst _ by anderson et al . we account for the other oiii branches leading to the @xmath165 line by using the observed oiii bowen spectrum of rr tel ( selvelli , danziger , & bonifacio 2007 ) . we then estimate @xmath167 , with @xmath168 kpc the distance to the system and @xmath169 the radius of the emitting region . boroson et al . ( 2000 ) found @xmath170 for the narrow lines from the ratios of lines observed with _ hst _ , and howarth & wilson ( 1983b ) found @xmath171 from _ iue _ observations . combining with equation [ eq : bowen ] , we find @xmath172 cm . this is smaller than the size of the accretion disk ( @xmath173 cm ) . we note that the tomograms show emission from a small region in velocity space . another test is provided by a comparison with theoretical models of niii@xmath3 emission in the absence of the bowen process . raymond ( 1993 ) found ratios between the niii line and c iii@xmath1 for the cos model of @xmath174 and for the cno model of @xmath175 . the observed ratio of @xmath176 requires enhancement by bowen fluorescence and provides an excellent match to the cno model if only 70% of the niii@xmath3 line results from collisional excitation . the far uv spectrum of hercules x-1 shows clear evidence for a keplerian accretion disk . a simple model such a disk can fit the broad ovi emission lines near eclipse . the origin of brighter , narrow emission lines is still unknown . doppler tomograms place this emission apart from the roche lobe of hz her . the bright narrow emission component generally follows the flux expected from the illuminated portion of the normal star . however , it appears brighter than expected at @xmath116 and the velocity excursion is also greater than is accounted for by the rotation of the roche lobe . a dense portion of a stellar wind that moves along the surface of the star , or that originates at the l1 point and flows back towards the star , would be red - shifted at @xmath35 and blue - shifted at @xmath103 and @xmath33 . although such a model can be made to reproduce the gross features of the line variation , there remain too many possible alternate explanations for this to be convincing . there could be a source of variable emission , absorption , or scattering in addition to the normal star , for example , the gas stream or the surface of the disk . in the presence of stellar winds , possibly transient or restricted in solid angle , tomography did not show a clear signature of an accretion disk in hercules x-1 . although we found that a simple model of an accretion disk generally fit the spectra during eclipse ingress and egress , there was an anomalous observation in which the ovi lines brightened as the disk eclipse progressed . the phase @xmath91 is in the @xmath177 range in which historically `` pre - eclipse dips '' of x - rays have been observed , and it is tempting to relate the temporary brightening of the lines to a physical cause or result of the dips . alternately , an anomalous change to the emission lines could result from a change in shadowing of the x - ray emission . boroson et al . ( 2001 ) concluded that anomalous _ absorption _ within the nv@xmath178 doublet changed too rapidly to result from material passing over the line of sight . the absorption changed so rapidly probably because of a rapid change in the shadowing of the x - rays which ionize the gas . future x - ray spectroscopy missions such as _ constellation x _ may be able to resolve lines from more highly ionized species than the uv and far uv resonance lines ( vrtilek et al . if more highly ionized species track disk material better and are not as prevalent in the narrow line region , we may be able to make reliable tomograms of the disk in this or similar systems . hydrodynamic models of this system may be required to cut down on the vast parameter space of possible gas flows responsible for the spectral signatures observed here . we would like to thank jeff bryant for helping us develop some _ mathematica _ routines for doppler tomography . chianti is a collaborative project involving the nrl ( usa ) , ral ( uk ) , mssl ( uk ) , the universities of florence ( italy ) and cambridge ( uk ) , and george mason university ( usa ) . we would like to thank the anonymous referee for suggestions that improved the text and presentation , for pointing us to the s vi@xmath22 lines , and prompting us to analyze the interstellar molecular absorption lines . with one free parameter , the mass accretion rate @xmath73 , we model the continuum emission from the accretion disk and x - ray illuminated face of the normal star . we do this not by means of radiative transfer calculations , but by calculating the heated temperatures of both surfaces , and co - adding spectra appropriate for those temperatures , either blackbody spectra , models of stellar spectra , or actual spectra of hot stars observed with _ fuse_. our model of the far uv continuum emission of her x-1 follows closely the methods of vrtilek et al . ( 1990 ) and cheng , vrtilek , & raymond ( 1995 ) . we describe the method here in detail , including the minor departures we have made to extend the spectral simulation to the far uv . the model of the accretion disk temperature as a function of radius , @xmath179 , includes the effects of both internal heating due to viscous forces and heating from x - rays from the neutron star . when we calculate the disk temperature as a function of radius , @xmath179 , we include both the heat from viscous forces and x - rays from the neutron star . the local energy generated by accretion is given by @xmath180 for stefan - boltzmann constant @xmath181 , neutron star mass @xmath182 and radius @xmath169 , and disk mass accretion rate @xmath73 ( shakura & sunyaev 1973 ) . for local energy balance , the energy emitted must equal the energy generated plus the x - ray energy absorbed : @xmath183 where we choose the albedo @xmath184 , and the disk height at radius @xmath185 is @xmath186 . the x - ray luminosity is related to the gravitational potential released by accreting matter : @xmath187 these equations , together with vertical hydrostatic equilibrium , can be solved numerically to determine @xmath179 . the surface of the noncollapsed star is also assumed to have an albedo @xmath184 . at each small region in the accretion disk , the temperature @xmath179 is compared with the critical disk temperature , @xmath188 . if @xmath189 , then we assume that the disk region radiates like a blackbody . if @xmath190 , then we interpolate through our library of actual stellar spectra to find a spectrum appropriate for temperature @xmath179 . from experience fitting _ hst _ fos spectra and fitting the balmer jump , we fix @xmath191 k so that in practice the disk almost always has blackbody spectra , except at its edge where it is not illuminated by x - rays . the stellar library from 1150 to 7500 is described in cannizzo & kenyon ( 1987 ) . the near uv spectra were obtained with _ iue _ and are described in wu et al . all the spectra are normalized to have constant flux in the v bandpass and are then scaled to their absolute magnitudes using their v - r colors and the barnes - evans relation ( barnes , evans , & moffett 1978 ) . for this paper , we have extended the library of stellar spectra further into the uv by using _ spectra of main sequence stars between o9.5 and a7 . these observations are described briefly in table 5 . stars of temperature greater than @xmath192 k are not used in the fits to her x-1 and are not included in the table . we have interpolated over those wavelengths we expect are affected by interstellar atomic or molecular absorption , and we have applied a filter designed to smooth interstellar lines while retaining stellar features . we de - reddened each spectrum and scaled the flux according to the star s visual magnitude v. we have tested the method and the use of these stellar spectra by using model stellar spectra as predicted by kurucz ( 1979 ) . plots of far uv spectra in our library , normalized to have a constant v magnitude , are shown in figure 17 . for the purpose of computing the disk shadow , the disk is assumed to have a fixed opening angle and a fixed tilt from the orbital inclination , although the direction of the disk normal precesses with the x - ray cycle in a manner described by gerend & boynton ( 1976 ) and howarth & wilson ( 1983a ) . because of the finite disk opening angle , the outer regions of the disk can occult the inner regions , and the disk edge itself can radiate , although it is not illuminated by the central x - rays . the noncollapsed star is assumed to fill its roche lobe , and its shape , visibility , and illumination by the x - ray source are treated according to methods evolved from wilson & devinney ( 1971 ) to howarth & wilson ( 1983a ) . following those references , we model the eclipse of the disk using the roche potential . we take into account both limb darkening and the gravity - darkening appropriate for a late - type star . if the temperature of some region on the surface of the star , after x - ray illumination , is greater than some critical temperature @xmath193 , we assume the region emits as a blackbody . if the local temperature is less than @xmath193 , we again interpolate through our stellar library to find the appropriate stellar spectrum with which the spot radiates . from earlier experience fitting her x-1 uv spectra , we fix @xmath194 ( cheng , vrtilek , & raymond 1995 ) . the model , as currently realized , does not account for emission or shadowing by a gas stream between the noncollapsed star and disk . accretion disks in theory have emission lines with a double - peaked shape , with peaks separated by a velocity given approximately by the projection into the line of sight of the orbital velocity at the edge of the accretion disk . for hercules x-1 , with typical neutron star mass , an nearly edge - on inclination , and an outer disk radius of @xmath195 cm , we expect the peaks to occur at @xmath196 km s@xmath10 . while greater velocities are found within the outer disk radius , each ring at a fixed radius emits down to a doppler velocity of 0 where the gas moves tangential to the line of sight . regions of high doppler velocity arise from progressively smaller regions in the disk , and contribute less to the overall line shape . we have calculated detailed models of emission from an accretion disk and have fit these to the profiles observed in her x-1 during eclipse ingress and egress . in reality , the disk is likely to be warped , and in future work we will test individual models of this warped shape . the importance of the demonstration in this paper is that it shows how easy it is to get general agreement with the observed line profiles given the standard picture of the accretion disk . the method in detail follows that of horne ( 1995 ) , which allows for nonisotropic turbulence and for keplerian shear , that is , the local dispersion in disk velocity as a result of the keplerian flow itself . this introduces free parameters for the mach matrix " describing the turbulent flow , in addition to the power law exponent that we use to as a phenomenological description the strength of the emission line as a function of radius in the disk . for simplicity , we keep the elements of the mach matrix constant with radius in the disk . the nonisotropic turbulence and shear enter through @xmath202 , given by @xmath203 or @xmath204 where @xmath205 is the orbital inclination , @xmath206 is the mean molecular weight , @xmath207 , and @xmath208 is the atomic weight . the shear parameter @xmath209 is given by @xmath210 and , for simplicity , is set to @xmath211 , the maximum value for an optically thin accretion disk . crrlrr star & spectral type & t@xmath214 ( k ) & fuse rootname & e(b - v ) & v + + hd 146813 & b1.5 v & 24000 & p1014901 & 0.02 & 9.06 + hd 74662 & b3 v & 18000 & a1290201 & 0.09 & 8.87 + ( interpolated ) & b4 v & 16500 & & + hd 92288 & b6 v & 14000 & z9012801 & 0.05 & 7.9 + ( hd 21672,hd 92536 ) & b8 v & 11500 & ( z9011401,z9012901 ) & ( 0.09,0.04 ) & ( 6.63,6.32 ) + hd 149630 & b9 v & 10800 & b0910101 & 0.04 & 4.2 + ( hd 109573,hd 181296 ) & a0 v & 10000 & ( b0910401,p2500101 ) & ( 0.0,0.01 ) & ( 5.78,5.03 ) + hd 31647 & a1 v & 9300 & c0380901 & 0.0 & 4.989 + hd 115892 & a2 v & 9050 & a0410505 & 0.0 & 2.75 + hd 43940 & a3 v & 8850 & c0380101 & 0.0 & 5.87 + hd 11636 & a5 v & 8500 & a0410101 & 0.0 & 2.64 + ( interpolated ) & a6 v & 8350 & + hd 187642 & a7 v & 8050 & d0990101 & 0.0 & 0.77 + ) and the _ hst _ stis ( @xmath215 ) near orbital phases when the emission is expected to be dominated by the accretion disk . prominent emission lines are indicated by dotted vertical lines . we show he average of our continuum models using actual fuse spectra of hot stars together with blackbodies . the component of the model resulting from the x - ray heated face of the normal star is indicated by a dotted curve.,title="fig : " ] ) and the _ hst _ stis ( @xmath215 ) near orbital phases when the emission is expected to be dominated by the accretion disk . prominent emission lines are indicated by dotted vertical lines . we show he average of our continuum models using actual fuse spectra of hot stars together with blackbodies . the component of the model resulting from the x - ray heated face of the normal star is indicated by a dotted curve.,title="fig : " ] ) and the _ hst _ stis ( @xmath215 ) near orbital phases when the emission is expected to be dominated by the x - ray heated face of the normal star . the spectrum is plotted as in figure 1 . the lines and continuum are brighter at these phases , and the lines are narrower.,title="fig : " ] ) and the _ hst _ stis ( @xmath215 ) near orbital phases when the emission is expected to be dominated by the x - ray heated face of the normal star . the spectrum is plotted as in figure 1 . the lines and continuum are brighter at these phases , and the lines are narrower.,title="fig : " ] level are shown . emission lines shown : ( a ) svi@xmath216 , ( b ) svi@xmath64 , ( c ) ciii@xmath65 , ( d ) niii@xmath3 , ( e ) o vi@xmath66 , ( f ) ovi@xmath67 , ( g ) siv@xmath115 , ( h ) p v@xmath69 , ( i ) ciii@xmath70,title="fig : " ] level are shown . emission lines shown : ( a ) svi@xmath216 , ( b ) svi@xmath64 , ( c ) ciii@xmath65 , ( d ) niii@xmath3 , ( e ) o vi@xmath66 , ( f ) ovi@xmath67 , ( g ) siv@xmath115 , ( h ) p v@xmath69 , ( i ) ciii@xmath70,title="fig : " ] level are shown . emission lines shown : ( a ) svi@xmath216 , ( b ) svi@xmath64 , ( c ) ciii@xmath65 , ( d ) niii@xmath3 , ( e ) o vi@xmath66 , ( f ) ovi@xmath67 , ( g ) siv@xmath115 , ( h ) p v@xmath69 , ( i ) ciii@xmath70,title="fig : " ] level are shown . emission lines shown : ( a ) svi@xmath216 , ( b ) svi@xmath64 , ( c ) ciii@xmath65 , ( d ) niii@xmath3 , ( e ) o vi@xmath66 , ( f ) ovi@xmath67 , ( g ) siv@xmath115 , ( h ) p v@xmath69 , ( i ) ciii@xmath70,title="fig : " ] level are shown . emission lines shown : ( a ) svi@xmath216 , ( b ) svi@xmath64 , ( c ) ciii@xmath65 , ( d ) niii@xmath3 , ( e ) o vi@xmath66 , ( f ) ovi@xmath67 , ( g ) siv@xmath115 , ( h ) p v@xmath69 , ( i ) ciii@xmath70,title="fig : " ] from @xmath116 to @xmath117 . the grayscale runs from the minimum to the maximum . top : the trailed spectrogram ( left ) and filtered trailed spectrogram ( right ) . middle : the back - projected tomogram ( left ) and filtered tomogram ( right ) . bottom : the trailed spectrogram inverted from the back - projected tomogram ( left ) and from the filtered tomogram ( right).,title="fig:",scaledwidth=80.0% ] from @xmath116 to @xmath117 . the grayscale runs from the minimum to the maximum . top : the trailed spectrogram ( left ) and filtered trailed spectrogram ( right ) . middle : the back - projected tomogram ( left ) and filtered tomogram ( right ) . bottom : the trailed spectrogram inverted from the back - projected tomogram ( left ) and from the filtered tomogram ( right).,title="fig:",scaledwidth=80.0% ] emission line averaged over every two _ fuse _ exposures ( thick curve ) , a model of ovi emission on the surface of the noncollapsed star ( dashed ) , and an empirical model of gaussian emission and absorption models . both models include flux from a model of the accretion disk.,scaledwidth=100.0% ] line as a sum of moving gaussians and an accretion disk model , with counting statistics included.,title="fig:",scaledwidth=85.0% ] line as a sum of moving gaussians and an accretion disk model , with counting statistics included.,title="fig:",scaledwidth=85.0% ]

supplementing existing survey data with vla observations , we have extendedray counterpart identifications down to decl . = -40 using our figure of merit approach . we find blazar counterparts for 70% of egret sources above decl . = -40 away from the galaxy . spectroscopic confirmation is in progress , and spectra for two dozen sources are presented here . we find evidence that increased exposure in the bulge region allowed egret to detect relatively faint blazars ; a clear excess of non - blazar objects in this region however argues for an additional ( new ) source class .

astro-ph0403692

the _ egret _ telescope on the compton gamma ray observatory ( cgro ) satellite has detected 271 sources in a survey of the @xmath3-ray ( 100 mev to 10 gev ) sky . roughly one quarter of these have previously been identified as blazars @xcite . recently , we have developed a new counterpart identification technique , which has pushed the identified fraction to @xmath270% in the northern hemisphere ( sowards - emmerd _ et al . _ 2003 , hereafter srm03 ) . here , we extend the application of this technique south ( -40@xmath1 @xmath4 decl . @xmath4 0@xmath1 ) , including the galactic bulge region . in order to rank blazar candidates in this zone using our ` figure of merit ' ( fom ) method , we have selected flat spectrum radio counterpart candidates in 3eg source error boxes . we have obtained spectroscopic confirmations and estimated redshifts for many of these candidate blazars with the hobby - eberly telescope ( het ) marcario low resolution spectrograph ( lrs ) and with the 2.7-m harlan j. smith telescope igi spectrograph at mcdonald observatory . for our final counterpart list , we used the vla to obtain compact 8.4 ghz fluxes similar to those previously mined from the class ( meyers _ et al . _ 2002 ) survey . this allows a robust selection of egret blazar - like counterparts and a statistical measure of the identification probability of each source . the blazar label is somewhat heterogeneous , but in the context of the unified agn model , these sources are believed to be viewed close to the axis of a powerful relativistic jet . as such they are compact flat spectrum radio sources , with apparent superluminal motion at vlbi scales . the optical counterparts exhibit significant polarization and ovv ( optically violently variable ) behavior @xcite . optical spectroscopy yields a dichotomy of sources : flat spectrum radio quasars with broad emission lines and continuum - dominated bl lac - type objects showing weak absorption features and occasionally narrow , weak emission lines . the broad - band spectral energy distribution ( sed ) is sometimes used to divide these into two classes , with ` red ' blazars showing a synchrotron peak in the ir - optical with a synchrotron self - compton ( ssc ) peak in the @xmath3-ray while ` blue ' blazars have a synchrotron component extending into the x - ray with a ssc peak inferred to extend to the tev range ( urry 1999 , and references therein ) . in our previous analysis ( srm03 ) , we quantified the correlation between flat spectrum radio sources and 3eg positions . this approach produces a quantitative evaluation of the likelihood of each counterpart candidate , based on radio flux , radio spectral index , x - ray flux , and location of the source within a 3eg error box . the independent functions used to generate this ` figure of merit ' ( fom ) statistic took the form of a fractional excess of sources found within the 3eg positional 95% confidence contours , in a given flux or spectral index bin , relative to the random background sources : @xmath5 these functions , combined with the positional likelihood @xmath6 , form our fom : @xmath7 the details of the fom analysis are described in srm03 . here it is worth noting that the adopted fitting functions take the radio fom to zero at @xmath8 85 mjy and the spectral index fom to zero at @xmath90.53 , where @xmath10 . to generate a fom for southern / bulge sources equivalent to those of the northern sample , accurate , compact 8.4 ghz fluxes were essential . source selection for 8.4 ghz vla a - array observations proceeded in a similar fashion to that for the class survey . we first selected nvss ( 1.4 ghz ; condon _ et al . _ 1998 ) sources within 90 arcseconds of pmn ( 4.85 ghz ; gregory _ et al . _ 1996 ) single dish source positions , for pmn sources in the ts maps of 3eg catalog sources in the range -40@xmath1 @xmath4 decl . @xmath4 0@xmath1 . the ` test statistic ' ts is the maximum likelihood estimator used in the egret source catalogs to quantify the probability that an egret point source is at a given location ; contours of @xmath11ts from the maximum define the source uncertainty region . for each 1.4 ghz source a spectral index was computed ; if multiple nvss sources matched one pmn source , the nvss fluxes were summed and the average 1.4/4.8 ghz spectral index was applied to each . sources with spectral indices steeper than 0.5 , the class spectral index cut , were dropped from consideration . additionally , a conservative spectral index lower limit of -2.0 was applied . such strongly inverted spectral indices are consistent with optically thick thermal emission and we expect these sources to be galactic ( planetary nebulae and hii regions ) . blazars generally do not exhibit spectra nearly this inverted . the class blazar survey @xcite contains 6 sources ( of 325 ) with @xmath12 and only one of these has been optically identified as a blazar . the deep x - ray blazar survey @xcite find only 2 sources ( of 298 ) with @xmath13 and none with @xmath12 . using these spectral indices , we then computed preliminary 8.4 ghz flux estimates . x - ray fluxes were mined from the rosat all - sky survey bright and faint catalogs . from the estimated 8.4 ghz flux , the spectral index , the x - ray flux , and the @xmath11ts at the position of the radio source , the fom was calculated . sources located at values of @xmath11ts @xmath14 13.5 were not considered . we have made a conservative cut of fom @xmath14 0.125 , half the threshold for inclusion as plausible identifications in srm03 , to allow for errors in the spectral index and the flux extrapolation to 8.4 ghz . blazars are significantly variable and our radio survey observations are not simultaneous , so it is important to note this is a significant source of error in the spectral index estimates . source confusion in the large pmn beam and measurement uncertainties for the fainter sources may also contribute . pmn observations were made in 1990 , while the nvss observations spanned 1993 - 1996 . sources meeting this preliminary fom @xmath14 0.125 criterion were targeted for vla 8.4 ghz follow - up . class 8.4 ghz fluxes are in general not available below decl.@xmath15 . thus , to establish a spectral index and flux for the sources equivalent to that used in srm03 , sources above the preliminary fom threshold were observed at 8.4 ghz with the vla a - array during 23 - 25 july 2003 ( program ar517 ) . observations were taken with a bandwidth of 50 mhz in each if and typically two 2 - 3 minute scans were taken per source . in practice , a number of targets in the original candidate list were not observed in this campaign . a handful of egret error boxes at low latitude contained very large numbers of sources , likely due to extended galactic emission . these were not observed , although a number of the candidates in these regions formally met our 1.4 ghz/4.85 ghz survey cuts . at higher latitude a few additional sources were missed due to clerical error . in the analysis that follows we use the 1.4/4.85 spectral indices and extrapolated fluxes for these objects . several of these sources are very likely to be thermal , but a few deserve interferometric follow up and our classification of these should be viewed as preliminary . archival 8.4 ghz fluxes were mined from the vla calibrator database for several of the brightest sources . several of the target sources were in fact used as vla phase calibrators during the observation run . a total of 102 sources were observed in this campaign . the data were edited and calibrated with the nrao astronomical image processing system ( aips ) package in the usual manner . maps were created , self - calibrated and cleaned using aips . a single - component model was employed in the self - calibration . self - calibration was not performed on a handful of the faintest ( few mjy ) sources . finally , peak intensities of the sources were fit . all of the bright sources were strongly core dominated , although about 20% show jets on 0.1 - 1 arcsec scales , consistent with the blazar classification . the typical noise level was less than 1 mjy / beam . except for the faintest sources , the flux errors are dominated by calibration uncertainties at these low declinations ( estimated at @xmath23% ) . fluxes of sources that meet our fom threshold appear in table 1 . figure 1 shows the images for our highest probability new identifications . precise positions and core fluxes for all observed sources are in table 4 . the remainder of the images will be provided upon request . several sources showed no compact flux , and a handful gave fluxes much lower than expected from our extrapolated values . indeed all but one of the sources with @xmath16 - 0.8 had little compact 8.4 ghz flux . several of these sources can be identified as extended hii regions and planetary nebulae . based on this result , we believe that the sources classified as ( g ) in table 1 are in fact galactic . we note that many of these objects have spectral indices consistent with an optically thick thermal spectrum , @xmath17 however , as halpern _ et al . _ ( 2001 ) have pointed out , blazars at low latitude can be masked in the radio by galactic objects , as in the case of 3eg j2016 + 3657 . to be conservative , we denote these as unclassified sources in our summary aitoff plot . it should be noted that 1.4 - 8.4 ghz spectral indices are generally steeper than 1.4 - 4.8 ghz spectral indices based on the pmn fluxes . this is also seen in the class blazar survey @xcite . the majority of our radio selected blazar candidates have archival optical classifications and redshifts . information for these objects were extracted from the simbad and ned databases . we have targeted the remainder of the sources for spectroscopic followup at mcdonald observatory . spectroscopy was obtained using the marcario lrs ( hill et al . 1998 ) on the 9.2 m het ( ramsey et al . 1998 ) . these targets were observed in regular queue operations between april 2003 and july 2003 . we obtained 2 @xmath18 300s exposures for most targets , and 3 @xmath18 300s for fainter targets . observations were made employing a 300 line mm@xmath19 grating and a 2@xmath20 slit , giving a dispersion of 4 per ( binned ) pixel and an effective resolution of 16 covering @xmath21 . due to the pointing limitations of the het ( decl . -11@xmath1 ) , the 2.7 m harlan j. smith telescope at mcdonald observatory was used to observe southern targets below this limit . observations were made during one observing run from 25 - 29 july 2003 . the imaging grism instrument ( igi ) spectrograph was used with a 6000 grism , and a 50 mm lens . a 2@xmath20 slit was employed for the first 1.5 nights , and a 2.5@xmath20 slit thereafter due to generally poor seeing . the image fwhm fluctuated between 1.5@xmath20 and 2@xmath20 on the best night , but typically held around 2 - 2.5@xmath20 . the igi setup covered a smaller wavelength range than the het / lrs , namely @xmath23 . due to a wide range of conditions and source brightnesses , total exposures ranged from 300 to 3600 seconds . standard ccd reductions were performed using iraf . spectra were optimally extracted from the 2-d images and calibrated . telluric corrections were applied to remove atmospheric absorption . redshifts were estimated by cross - correlation analysis with agn and galaxy spectral templates , using the iraf rvsao package . as of this publication , we have obtained 26 spectral classifications , 8 with the het / lrs and 18 with the mcdonald 2.7m / igi . the new redshifts and basic optical properties of these objects are listed in table 2 . marginal ( e.g. single line or low s / n ) redshift estimates are denoted by a colon . spectra taken at mcdonald observatory are plotted in figure 2 . since optical observations were initiated before the 8.4 ghz data were reduced , a number of sources in figure 2 and table 2 do not appear in table 1 , as their final fom are below the ` plausible ' threshold . sources observed at mcdonald observatory were classified based on their spectra , s / n permitting . the observed sources fall into the following 3 categories : bl lac objects ( bll ) , flat spectrum radio quasars ( fsrq ) , and narrow line radio galaxies ( nlrg ) . bl lac objects are defined here by the following properties ( marcha et al . 1996 ) : + 1 . h / k break contrast @xmath24 where @xmath25 and @xmath26 are the fluxes redward and blueward of the break . rest - frame emission line equivalent width @xmath4 5 . + a handful of objects were observed to have narrow emission lines , with kinematic widths @xmath41000 km / s , and equivalent widths larger than the bl lac threshold . these objects generally exhibit a weak continuum flux , and were consequently classified as nlrgs . the majority of observed sources were fsrqs with broad ( v@xmath27 1000 - 2000 km / s ) emission lines . for the sources with archival redshifts , we examined spectra from the literature whenever possible . when these were unavailable the ned / simbad classifications were adopted as a last resort . all of the 3eg error boxes located in the region -40@xmath1 @xmath4 decl . @xmath4 0@xmath1 are listed in table 1 , along with classifications and new counterpart candidates . we recover many of the counterparts identified in the third egret catalog @xcite and by mattox @xcite . those that did not meet our selection criteria are included for completeness ( italics ) . we also indicate in the classification column sources that are previously discussed pulsar / plerion candidates ( p ) . the spectral index listed in the table is calculated between 1.4 ghz and 8.4 ghz , where @xmath10 , with the exception of indices listed in parentheses which were calculated between 1.4 ghz and 4.8 ghz . a radio i d flag , rid , was included to indicate the origin of the radio flux : the vla calibrator survey , this campaign , or both . sources are divided into high and lower confidence identifications by their foms . lower confidence ( ` plausible ' ) identifications fall in the range 0.25 @xmath4fom@xmath4 1.0 , and higher confidence ( ` likely ' ) source identifications have fom @xmath28 1 . in the -40@xmath1 @xmath4 decl . @xmath4 0@xmath1 band we have identified 30 likely and 23 plausible counterparts with 8.4 ghz flux measurements ( 5 nominally ` likely ' counterparts lacking 8.4 ghz confirmation deserve follow - up , but as noted above the 28 additional sources classified as ` plausible ' based on extrapolated fluxes are mostly at low @xmath29 , are very likely galactic , and are designated ` g ' in table 1 ) . despite the rather cautious labels , the sources with 8.4 ghz observations are statistically highly significant identifications ; simulations based on the northern survey indicate a success rate of 92% for the likely sources and 82% for the plausible sources . thus among ` likely ' sources with 8.4 ghz measurements ( of which 7 are newly identified here ) we statistically expect 2 false positives . for the ` plausible ' sources ( 16 of which are new ) we expect no more than 4 false identifications . as in the north , we find evidence in this southern extension for multiple counterparts for individual 3eg sources , where the fainter radio counterpart(s ) would be ignored in the 3eg / mattox studies . figure 3 shows two cases , 3eg j1246 - 0651 and 3eg j1911 - 2000 . 3eg j1246 - 0651 contains two high - confidence ids in an elongated error region , with two likelihood maxima corresponding well to the candidate positions . 3eg j1911 - 2000 also contains two counterparts , one plausible and one high - confidence . again , the sources line up along the major axis of the error contour . it appears that the centroid of the uncertainty region is shifted slightly from the more likely source toward the fainter radio source , suggesting that a fainter @xmath3-ray contribution biases the ts localization . a comparison of the populations and luminosity functions of the candidates requires full treatment of the egret sensitivities and the radio / x - ray selection biases . we defer this population study to a future communication . however , it is already interesting to make a comparison of the areal densities of various source classes in the combined survey . the galactic aitoff projection in figure 4 summarizes the current status of the 3eg counterpart identification . in order to compare the different populations , we divide the sky above decl . @xmath14 -40@xmath1 into three regions : + galactic plane @xmath3010@xmath1 , + galactic bulge ( excluding the plane ) @xmath3110@xmath1 and @xmath4 30@xmath1 from @xmath32 , and + high latitude the remainder . + in the ` high latitude ' set , we find 88 blazar ids and 41 non - blazar ids , suggesting that 68% or more of the high latitude 3eg detections have radio - bright blazar counterparts . in the bulge region , we find 13 blazar ids and 15 non - blazar ids , compared to 4 blazar ids and 2 non - blazar ids predicted from the areal density at high latitude . the larger number of sources here , including a factor of 3 more blazars , may be attributed to deeper egret exposure ( 2.5@xmath18 larger 3eg exposure , averaged over area , in our bulge region relative to the high latitude region . ) even relative to this increase , there is clearly an excess of non - blazar identifications in the bulge ( 54% _ vs. _ 32% at high latitude ) , which likely represents a new bulge population of @xmath0ray sources . within the plane , we find 12 blazar ids and 49 non - blazar ids , compared to 18 blazar ids and 8 non - blazars predicted from high latitude . the apparent deficit of blazar identifications in the plane may be partly due to decreased egret sensitivity in this high background region , although source confusion limitations to counterpart identification may also play a role . the large number of unidentified sources here clearly represents a galactic population . grouping objects by type , we find 17% bl lacs and 83% fsrqs in the southern extension , very similar to the 19% bl lacs and 81% fsrqs in the northern survey . we can examine the redshift distributions of the northern and southern extension samples , since these should be independent of the initial candidate selection . before we do so , we include four additional redshifts determined for the northern sample since the publication of srm03 ( figure 5 , table 3 ) . since the bl lacs and fsrqs form two distinct populations in redshift space , we treat the two samples independently ( figure 6 ) . similar to the northern sample , we uncover a handful of @xmath33 sources in the southern extension . for the fsrq sample , we find a good agreement between the north and south , with a kolmogorov - smirnov probability of 0.27 that the populations are drawn from the same parent distribution . in the case of bl lacs , the north and south distributions are still consistent , but with a lower probability prob(k - s ) = 0.08 . however , it should be noted that spectroscopic identification is much more complete in the northern sample . 3eg j0340 - 0201_in addition to the likely association with j0339 - 0146 we find a plausible second counterpart in j0339 - 0133 . _ 3eg j0412 - 1853_in this large error region , @xmath24 degrees in diameter , we find a plausible counterpart , j0409 - 1948 , in addition to the likely i d j0416 - 1851 . _ 3eg j0530 - 3626_from the predicted radio flux and 1.4/4.8 ghz spectral index , we believe j0529 - 3555 to be the most likely counterpart , although this requires confirmation . the previous claim , j0522 - 3627 , with @xmath11ts = 17.9 lies well to the south of this error map , which is strongly elongated east to west . it does not meet our fom cut . _ 3eg j0542 - 0655_in addition to the previous association , j0541 - 0541 ( plausible ) , we find an equally likely counterpart based on 1.4/4.8 ghz spectral index and extrapolated flux ; again this should be interferometrically confirmed . _ 3eg j1246 - 0651_we find two likely associations , j1248 - 0632 and j1246 - 0730 , in this map . these sources present a strong case for two sources nearly resolved by egret . the ts map is shown in figure 3 . _ 3eg j1500 - 3509_j1457 - 3539 and j1505 - 3432 , neither of which were previously identified as @xmath3-ray counterparts , are both likely ids located within this error map . _ 3eg j1504 - 1537_three plausible sources occupy this error map in addition to the previous i d j1507 - 1652 , which is located well beyond the 99% contour . _ 3eg j1607 - 1101_two plausible and one likely source occupy this map based on extrapolated fluxes and 1.4/4.8 ghz spectral indices . j1605 - 1139 is claimed as a counterpart by tornikoski ( 2002 ) based on 90 ghz observations , but it does not match our survey criteria . _ 3eg j1612 - 2618_two likely sources are located in this error map , based on 1.4/4.8 ghz spectral indices and extrapolated fluxes . two plausible sources near the threshold for inclusion are also selected ; these are unlikely to pass the compact 8.4 ghz fom cut . _ 3eg j1718 - 3313_the likely counterpart , j1717 - 3342 , which we identify as a bl lac , is very reddened at its low latitude . _ 3eg j1746 - 2851_mclaughlin and cordes ( 2003 ) present evidence that psr j1747 - 2958 is the counterpart . _ 3eg j1800 - 2338_this @xmath3-ray source has been identified with the pulsar psr b1758 - 23 . _ 3eg j1809 - 2328_this @xmath3-ray source has been identified with a plerion @xcite . _ 3eg j1832 - 2110_counterparts j1832 - 2039 , plausible , and j1833 - 2103 , likely , share this compact , circular error box . _ 3eg j1911 - 2000_j1911 - 2006 , high - confidence , and j1911 - 1921 , plausible , are both associations in this error box . the @xmath3-ray flux of j1911 - 2006 is likely supplemented by that of j1911 - 1921 . the ts map is shown in figure 3 . _ 3eg j1937 - 1529_j1935 - 1602 , plausible , and j1939 - 1525 , likely , both lie in this elongated error box . our analysis does not support the previously claimed counterpart j1941 - 1524 . _ 3eg j2006 - 2321_we find the centrally located j2005 - 2310 , previously claimed as a new low radio flux i d by wallace ( 2002 ) , to be a likely counterpart . _ 3eg j2034 - 3110_one likely i d , j2030 - 3039 , and one plausible i d , j2039 - 3157 , reside in this large error box . neither have been previously associated with the egret source . the 95% contour is not closed in this map . from our radio follow - up , our fom method has selected 23 new ids , with 8.4 ghz fluxes , between -40@xmath1 @xmath4 decl . @xmath4 0@xmath1 . of these new identifications , seven sources have high confidence ` likely ' counterparts and the remainder are plausible . we expect no more than 3 of these new sources are false positives . a few additional identifications are based on 4.85 ghz extrapolations and require 8.4 ghz compact fluxes for confirmation . we also have spectroscopically identified 26 new optical counterparts selected for this sample . several of these targets fell below our fom after vla follow - up - the spectroscopic targets were selected before 8.4 ghz fluxes were available - but are included in table 2 for completeness . combined with the northern sample , for high latitude 3eg error boxes ( @xmath34 and @xmath1430@xmath1 from the galactic center ) , we find blazar ids for 88 of 129 sources , or 68% . our analysis has now been applied to 157 of the 186 3eg detections located above @xmath34 . extending the survey into the galactic bulge , we find a surplus of unidentified 3eg sources . these likely comprise a new set of galactic @xmath3-ray emitters . since the areal density of identified blazars in the bulge region is also larger than in the north , it seems that the longer egret exposure in this region has let us look lower on the blazar luminosity function , which should be helpful in refining the blazar contribution to the unresolved high latitude @xmath3-ray counts . we defer detailed calculation of the blazar luminosity function and number count predictions for glast to a later paper . the hobby - eberly telescope ( het ) is a joint project of the university of texas at austin , the pennsylvania state university , stanford university , ludwig - maximilians - universitt mnchen , and georg - august - universitt gttingen . the het is named in honor of its principal benefactors , william p. hobby and robert e. eberly . the marcario low resolution spectrograph is named for mike marcario of high lonesome optics who fabricated several optics for the instrument but died before its completion . the lrs is a joint project of the hobby - eberly telescope partnership and the instituto de astronomia de la universidad nacional autonoma de mexico . dse was supported by slac under doe contract de - ac03 - 76sf00515 and pfm acknowledges support from nasa contract nas5 - 00147 . we thank f. heatherington for assistance with reduction of the vla radio data . braje , t.m . , romani , r.w . , roberts , m.s.e . , & kawai , n. 2002 , , 565 , l91 condon , j.j . et al . _ 1998 , , 115 , 1693 gregory , p.c . , scott , w.k . , douglas , k. & condon , j.j . 1996 , apjs , 103 , 427 halpern , j.p . , eracleous , m. , mukherjee , r. , & gotthelf , e.v . 2001 , , 551 , 1016 hartman , r.c . , _ et al . _ 1999 , , 123,79 hill , g.j . , nicklas , h.e . , macqueen , p.j . , tejada , c. , cobos duenas , f.j . & mitsch , w. 1998 , 3355 , 375 laurent - muehleisen , s.a . , kollgaard , r.i . , ciardullo , r. , feigelson , e.d . , brinkmann , w. , & siebert , j. 1998 , , 118,127 marcha , m.j.m . , browne , i.w.a . , impey , c.d . , & smith , p.s . 1996 , , 281,425 marcha , m.j . , caccianiga , a. , browne , i.w.a , & jackson , n. 2001 , , 326 , 1455 mattox , j.r . , _ et al . _ 1996 , , 461 , 396 mattox , j.r . , hartman , r.c . , reimer , o. 2001 , , 135 , 155 mclaughlin , m.a . , & cordes , j.m . 2003 , astro - ph0310748 monet , d.g . _ et al . _ 2003 , , 125 , 984 myers , s.t . , jackson , n.j.,browne , i.w.a.,bruyn , a.g . , pearson , t.j . , readhead , a.c.s . , wilkinson , p.n . , biggs , a.d . , blandford , r.d . , fassnacht , c.d . , koopmans , l.v.e . , marlow , d.r . , mckean , j.p . , norbury , m.a . , phillips , p.m. , rusin , d. , shepherd , m.c . , & sykes , c.m . 2002 , astro - ph/0211073 perlman , e.s . , padovani , p. , giommi , p. , sambruna , r. , jones , l.r . , tzioumis , a. & reynolds , j. 1998 , , 115 , 1253 landt , h. , padovani , p. , perlman , e.s . , giommi , p. , bignall , h. & tzioumis , a. 2001 , , 323 , 757 ramsey , l.w . _ et al . _ 1998 , , 3352 , 34 roberts , m.s.e . , romani , r.w . , & kawai , n. 2001 , , 133 , 451 sowards - emmerd , d. , romani , r. w. & michelson , p. f. 2003 , , 590 , 109 tornikoski , m. _ et al . _ 2002 , , 579 , 136 urry , c.m . & padovani , p. 1995 , , 107 , 803 urry , c.m . 1999 , _ astpart . _ , 11 , 159 wallace , p.m. halpern , j.p . , magalhaes , a.m. , & thompson , d.j . 2002 , , 569 , 36 lllllllcccccl j0038 - 0949&b&*j0039 - 0942*&187&-0.12&0.46&3.58&2.101&*&1&&&f + j0130 - 1758&b&*j0132 - 1654*&952&-0.08&7.73&1.04&1.020&&3&&a&f + j0159 - 3603&n & & & & & & & & & & & + j0253 - 0345&n & & & & & & & & & & & + j0340 - 0201&b&j0339 - 0133&339&0.28&6.69&0.42&3.182&@xmath35*&1&&&f + & & * j0339 - 0146*&3010&-0.13&3.49&3.59&0.852&&3&+&a&f + j0412 - 1853 & b&j0409 - 1948&108&0.11&3.06&0.32&1.986&@xmath35&1&&&f + & & * j0416 - 1851*&780&0.23&1.91&2.30&1.536&&3&&a&f + j0422 - 0102&b&*j0423 - 0120*&4250&-0.2&4.81&4.03&0.915&&2&+&a&f + j0442 - 0033&b&j0442 - 0017&860&0.39&4.56&0.54&0.844&&2&+&a&f + j0456 - 2338&b&*j0457 - 2324*&1850&-0.0&0.95&7.42&1.003&&2&+&a&f + j0500 - 0159&b&*j0501 - 0159*&3700&-0.25&0.86&10.60&2.286&&2&+&a&f + j0530 - 3626&b&*j0529 - 3555*&(289)&(-0.42)&2.29&(4.07)&&(*)&0 & & & + & & _ j0522 - 3627_&5000&0.63&17.9&0.00&0.061&&2&&a&b + j0531 - 2940 & b&j0539 - 2839&760&0.06&9.83&0.32&3.104&&2&&a&f + j0542 - 0655&b&j0541 - 0541&980&-0.06&10.33&0.34&0.839&&2&&a&f + & & j0545 - 0539&(151)&(0.11)&6.48&(0.34)&&&0 & & & + j0616 - 0720&n & & & & & & & & & & & + j0616 - 3310&n & & & & & & & & & & & + j0622 - 1139&n&_j0619 - 1140_&210&0.47&3.76&0.13&&&1&&a & + j0706 - 3837&n & & & & & & & & & & & + j0747 - 3412&n & & & & & & & & & & & + j0812 - 0646&n&_j0808 - 0751_&2600&-0.29&14.19&0.05&1.837&&2&&a&f + j0852 - 1216 & b&*j0850 - 1213*&625&-0.37&2.67&5.04&0.57&&1&&a&b + j0903 - 3531&n & & & & & & & & & & & + j1134 - 1530&n&_j1130 - 1449_&3060&0.30&18.71&0.01&1.187&&2&&a&f + j1219 - 1520&b&j1222 - 1645&145&0.03&7.73&0.25&&*&1 & & & + j1230 - 0247 & n&_j1232 - 0224_&629&0.50&3.25&0.15&1.045&&1&&a&f + j1234 - 1318&n & & & & & & & & & & & + j1246 - 0651&b&*j1248 - 0632*&353&0.02&3.21&1.83&0.762&@xmath35&1&&&f + & & * j1246 - 0730*&630&-0.08&2.09&4.05&1.286&&2&&a&f + j1255 - 0549&b&*j1256 - 0547*&15600&-0.2&6.99&3.65&0.538&&2&+&a&b + j1310 - 0517 & b&j1312 - 0424&286&-0.15&7.85&0.67&0.824&*&1&&&f + j1314 - 3431 & b&j1316 - 3338&1100&0.07&12.5&0.14&1.210&&1&&a&f + j1339 - 1419 & b&*j1337 - 1257*&3850&-0.21&5.75&3.38&0.539&&3&+&a&b + j1409 - 0745&b&*j1408 - 0752*&630&0.00&3.50&2.19&1.494&&2&+&a&f + j1447 - 3936&n & & & & & & & & & & & + j1457 - 1903&b&j1459 - 1810&158&-0.24&9.82&0.25&&*&1 & & & + j1500 - 3509&b&*j1457 - 3539*&606&0.04&4.27&1.76&1.422&@xmath35*&1&&&f + & & * j1505 - 3432*&393&-0.59&4.32&3.37&&*&1 & & & + j1504 - 1537&b&j1502 - 1508&133&-0.18&5.25&0.62&&*&1 & & & + & & j1505 - 1610&130&-0.24&5.43&0.61&&*&1 & & & + & & j1508 - 1548&307&-0.03&8.50&0.41&2.499&*&1&&&f + & & _ j1507 - 1652_&2400.&0.05&13.97&0.03&0.876&&2&&a&f + j1512 - 0849&b&*j1512 - 0905*&2150&0.12&1.79&4.31&0.360&&2&+&a&f + j1517 - 2538 & b&j1517 - 2422&1930&0.02&9.18&0.44&0.049&&3&&a&b + j1527 - 2358 & b&j1532 - 2310&158&-0.09&5.82&0.64&2.289&*&1&&&f + j1600 - 0351&n & & & & & & & & & & & + j1607 - 1101&b&j1603 - 1007&(146)&(0.08)&6.75&(0.32)&&&0 & & & + & & j1605 - 1012&(127)&(-0.27)&4.15&(0.76)&&&0 & & & + & & * j1612 - 1133*&(169)&(-0.58)&4.91&(1.52)&&(*)&0 & & & + & & _ j1605 - 1139@xmath36 _ & & & & & & & & & & + j1612 - 2618&b&j1607 - 2656&(119)&(0.29)&2.55&(0.30)&&&0 & & & + & & * j1611 - 2612*&(113)&(-0.17)&0.32&(1.54)&&(*)&0 & & & + & & j1613 - 2750&(210)&(0.05)&8.54&(0.26)&&&0 & & & + & & * j1617 - 2537*&(153)&(-0.15)&3.06&(1.22)&&(*)&0 & & & + j1616 - 2221&n & & & & & & & & & & & + j1625 - 2955&b&*j1626 - 2951*&2250&-0.00&4.06&2.58&0.815&&2&+&a&f + j1626 - 2519&b&j1625 - 2527&1100&0.45&0.65&0.99&0.786&&2&+&a&f + j1627 - 2419&n & & & & & & & & & & & + j1631 - 1018&n & & & & & & & & & & & + j1633 - 3216&n & & & & & & & & & & & + j1634 - 1434&b&j1628 - 1415&275&0.14&6.85&0.54&1.025&@xmath35*&1&&&f + j1635 - 1751&b&j1629 - 1720&(245)&(-0.56)&9.70&(0.39)&&&0 & & & + j1638 - 2749&n & & & & & & & & & & & + j1646 - 0704 & b&j1644 - 0743&129&-0.08&4.73&0.55&0.139&*&1&&&r + j1649 - 1611&n & & & & & & & & & & & + j1652 - 0223&n & & & & & & & & & & & + j1653 - 2133&n & & & & & & & & & & & + j1709 - 0828&n & & & & & & & & & & & + j1714 - 3857&g&j1711 - 3908&(25100.)&(-1.98)&9.49&(0.71)&&&0&&&(g ) + & & j1712 - 3840&(468)&(-0.98)&4.87&(3.34)&&&0&&&(g ) + & & j1713 - 3900&(4340)&(-1.06)&0.00&(19.36)&&&0&&&(g ) + & & j1714 - 3810&(10100)&(-0.84)&9.06&(0.94)&&&0&&&(g ) + & & j1714 - 3937&(165)&(-1.31)&9.43&(0.28)&&&0&&&(g ) + & & j1714 - 3925&(203)&(-1.76)&6.56&(1.31)&&&0&&&(g ) + & & j1715 - 3913&(4300)&(-1.34)&1.81&(11.08)&&&0&&&(g ) + j1717 - 2737&n & & & & & & & & & & & + j1718 - 3313&b&*j1717 - 3342*&685&-0.06&2.87&3.18&&*&1&&&b + j1719 - 0430&n & & & & & & & & & & & + j1726 - 0807&n & & & & & & & & & & & + j1733 - 1313&b&*j1733 - 1304*&5110&0.07&3.90&2.86&0.902&&3&+&a&b + j1734 - 3232&n & & & & & & & & & & & + j1735 - 1500&b&*j1738 - 1503*&(247)&(0.15)&3.28&(1.06)&&(*)&0 & & & + j1736 - 2908&n & & & & & & & & & & & + j1741 - 2050&n & & & & & & & & & & & + j1741 - 2312&n & & & & & & & & & & & + j1744 - 0310&b&*j1743 - 0350*&6240&-0.84&3.99&5.98&1.054&&3&+&a&f + j1744 - 3011&n & & & & & & & & & & & + j1744 - 3934&n & & & & & & & & & & & + j1746 - 1001&n & & & & & & & & & & & + j1746 - 2851&n&psr j1747 - 2958 & & & & & & & & & & p + j1757 - 0711&n & & & & & & & & & & & + j1800 - 0146&n & & & & & & & & & & & + j1800 - 2338&n&psr b1758 - 23&&&&&&&&&&p + j1800 - 3955 & b&*j1802 - 3940*&1700&0.14&2.59&2.74&&&1&&a & + j1809 - 2328&n&rrk&&&&&&&&&&p + j1810 - 1032&n & & & & & & & & & & & + j1812 - 1316&n & & & & & & & & & & & + j1823 - 1314&g&j1822 - 1251&(753)&(-1.39)&12.48&(0.32)&&&0&&&(g ) + & & j1824 - 1251&(928)&(-1.01)&11.61&(0.46)&&&0&&&(g ) + & & j1825 - 1315&(15300)&(-1.33)&8.48&(1.36)&&&0&&&(g ) + j1824 - 1514&g&j1819 - 1501&(2260)&(-0.80)&13.25&(0.27)&&&0&&&(g ) + & & j1819 - 1530&(982)&(-1.28)&10.31&(0.66)&&&0&&&(g ) + & & j1821 - 1449&(6150)&(-1.98)&10.87&(0.66)&&&0&&&(g ) + & & j1822 - 1550&(316)&(-1.88)&10.00&(0.45)&&&0&&&(g ) + & & j1822 - 1602&(342)&(-1.98)&10.43&(0.43)&&&0&&&(g ) + & & j1825 - 1449&(2550)&(-1.05)&4.00&(5.62)&&&0&&&(g ) + & & j1826 - 1600&(269)&(-0.99)&10.00&(0.41)&&&0&&&(g ) + j1826 - 1302&g&j1824 - 1251&(918)&(-1.01)&6.32&(3.00)&&&0&&&(g ) + & & j1825 - 1315&(15100)&(-1.33)&3.82&(5.24)&&&0&&&(g ) + & & j1826 - 1338&(245)&(-1.26)&6.60&(1.53)&&&0&&&(g ) + j1832 - 2110&b&j1832 - 2039&1070&0.03&8.18&0.70&0.103&*&1&&&f + & & * j1833 - 2103*&6750.&0.25&4.31&1.29&2.510&&2&+&a&f + j1834 - 2803&n & & & & & & & & & & & + j1837 - 0423&g&j1837 - 0508&(276)&(-1.92)&10.74&(0.35)&&&0&&&(g ) + & & j1839 - 0419&(1890)&(-1.60)&5.43&(4.19)&&&0&&&pn + j1837 - 0606&n & & & & & & & & & & & + j1847 - 3219&n & & & & & & & & & & & + j1850 - 2652&b&*j1848 - 2718*&417&-1.26&2.38&5.85&&*&1 & & & + j1858 - 2137&n & & & & & & & & & & & + j1904 - 1124&b&j1905 - 1153&154&0.26&3.09&0.48&&*&1 & & & + j1911 - 2000&b&*j1911 - 2006*&2190&0.11&0.12&6.75&1.119&&3&+&a&f + & & j1911 - 1921&182&0.06&6.84&0.45&0.804&*&1&&&f + j1921 - 2015&b & j1923 - 2104&2190&0.19&8.94&0.32&0.871&&1&&a&f + j1937 - 1529&b&j1935 - 1602&286&0.04&6.68&0.73&1.460&*&1&&&f + & & * j1939 - 1525*&670&-0.0&1.69&4.75&1.657&&2&&a&f + & & _ j1941 - 1524_&&&&&0.452&&&&&r + j1940 - 0121&n & & & & & & & & & & & + j1949 - 3456&n & & & & & & & & & & & + j1955 - 1414&n & & & & & & & & & & & + j2006 - 2321&b&*j2005 - 2310@xmath37*&230&0.14&0.36&2.70&0.830&&1&&&f + j2020 - 1545&n & & & & & & & & & & & + j2025 - 0744&b&*j2025 - 0735*&727&0.33&0.89&2.20&1.388&&1&&a&f + j2034 - 3110&b&*j2030 - 3039*&229&-0.05&5.45&1.00&&*&1 & & & + & & j2039 - 3157&231&0.06&6.24&0.67&&*&1 & & & + j2158 - 3023&b&*j2158 - 3013*&204&0.44&0.36&1.21&0.117&&1&&a&b + j2251 - 1341&n & & & & & & & & & & & + j2321 - 0328 & b&*j2323 - 0317*&918&-0.00&1.97&5.12&1.411&&3&&a&f + lllllllll j0039 - 0942&3.58&00 39 06.28&-09 42 47.5& ... &20.6&2.1015&f + j0256 - 0453&0.02&02 56 50.35&-04 53 43.7&18.3&19.9&1.4410&f + j0339 - 0133&0.42&03 39 00.94&-01 33 18.1&&&3.1826&f + j1228 - 0304&0.20&12 28 36.91&-03 04 39.3&19.7&20.2&0.1930&r + j1309 - 0415&0.00&13 09 17.81&-04 15 15.2&18.8&19.0&0.5089&f + j1312 - 0424&0.67&13 12 50.92&-04 24 50.7&19.3&18.7&0.8249&f + j1441 - 3913&0.00&14 41 15.13&-39 13 48.4&14.9&15.2&0.1356&r + j1457 - 3539&1.76&14 57 26.76&-35 39 09.2&18.9&17.8&1.4222:&f + j1508 - 1548&0.41&15 08 35.69&-15 48 31.6&19.2&20.2&2.4990&f + j1532 - 2310&0.64&15 32 31.53&-23 10 32.2&19.7&21.4&2.2896&f + j1555 - 0326&0.16&15 55 30.80&-03 26 49.4&19.7&20.7&2.2996&f + j1612 - 2239&0.02&16 12 28.48&-22 39 46.8&19.5&19.6&1.5404&f + j1628 - 1415&0.54&16 28 46.71&-14 15 41.9& ... &20.4&1.0255&f + j1642 - 0621&0.00&16 42 02.22&-06 21 23.6&19.2&20.9&1.5143:&b + j1644 - 0743&0.55&16 44 52.03&-07 43 43.4& ... & ... &0.1389&r + j1717 - 3342&3.18&17 17 36.01&-33 42 08.1& ... & ... & ... &b + j1716 - 0452&0.12&17 16 26.57&-04 52 12.5&18.7&20.6&1.0258&f + j1722 - 0503&0.08&17 22 03.61&-05 03 25.8&18.5& ... &0.2606&r + j1832 - 2039&0.70&18 32 11.10&-20 39 47.9& ... &20.2&0.1033&f + j1911 - 1921&0.45&19 11 56.50&-19 21 52.0&18.3&18.5&0.8046&f + j1917 - 2110&0.13&19 17 08.66&-21 10 31.2& ... &21.5&0.7775&f + j1923 - 2104&0.32&19 23 32.20&-21 04 33.8& ... & ... &0.8719&f + j1935 - 1602&0.73&19 35 35.79&-16 02 32.3&19.3&20.1&1.4602&f + j1940 - 0039&0.01&19 40 09.00&-00 39 01.3&17.0&18.1&1.7110&f + j1947 - 3542&0.00&19 47 22.65&-35 42 04.2&19.2&19.9&2.600:&f + j2005 - 2310&2.70&20 05 56.61&-23 10 28.1&18.8&19.6&0.8301:&f + llllllll j0205 + 1444&2.58&02 05 13.12&+14 44 32.4& ... & ... &2.8504&f + j1226 + 4340&0.95&12 26 57.91&+43 40 58.4&19.2&19.3&2.0023&f + j1322 + 2148&0.29&13 22 11.40&+21 48 12.3&19.4&19.3&1.6803&f + j2102 + 6015&0.41&21 02 40.22&+60 15 09.8& ... & ... &4.5749:&f + lllll j0039 - 0942&187 @xmath38 6&00 39 06.291&-09 42 46.88 & + j0128 - 1722&117 @xmath38 4&01 28 01.393&-17 22 24.25&jet ? + j0132 - 1654&952 @xmath38 29&01 32 43.487&-16 54 48.52 & + j0239 - 0234&595 @xmath38 18&02 39 45.472&-02 34 40.91 & + j0256 - 0453&87.4 @xmath38 2.6&02 56 50.420&-04 53 43.50 & + j0339 - 0133&339 @xmath38 10&03 39 00.985&-01 33 17.60 & + j0339 - 0146&3010 @xmath38 90&03 39 30.937&-01 46 35.80 & + j0405 - 1906&31.0 @xmath38 0.9&04 05 49.678&-19 06 57.27 & + j0409 - 1948&108 @xmath38 3&04 09 40.549&-19 48 01.78 & + j0416 - 1851&779 @xmath38 23&04 16 36.544&-18 51 08.34&jet ? + j0522 - 2955&47.3 @xmath38 1.6&05 23 00.131&-29 55 17.36 & + j0527 - 3708&0 . & & & + j0531 - 3533&122 @xmath38 4&05 31 30.419&-35 33 32.65 & + j0615 - 3355&121 @xmath38 4&06 15 12.689&-33 55 53.07 & + j0619 - 1140&210 @xmath38 6&06 19 04.103&-11 40 54.89 & + j0621 - 1259&0.&&&pn + j0625 - 1133&183 @xmath38 6&06 25 49.339&-11 33 24.37&jet ? + j0703 - 3746&0 . & & & + j0710 - 3813&18.9 @xmath38 1.4&07 10 47.043&-38 13 45.72 & + j0713 - 3812&23.8 @xmath38 0.9&07 13 01.858&-38 12 28.22 & + j0747 - 3310&131 @xmath38 9&07 47 19.693&-33 10 47.03 & + j0754 - 3448&0 . & & & + j0848 - 1159&42.5 @xmath38 1.3&08 48 47.491&-11 59 53.68 & + j0850 - 1213&625 @xmath38 19&08 50 09.634&-12 13 35.39&jet ? + j0856 - 1105&357 @xmath38 11&08 56 41.805&-11 05 14.42&jet ? + j1222 - 1645&145 @xmath38 4&12 22 16.097&-16 45 54.89 & + j1223 - 1544&61.3 @xmath38 1.8&12 23 12.272&-15 44 06.63&jet ? + j1228 - 0304&128 @xmath38 4&12 28 36.916&-03 04 39.32 & + j1231 - 1236&80.6 @xmath38 2.4&12 31 50.265&-12 36 37.04 & + j1232 - 0224&629 @xmath38 19&12 32 00.014&-02 24 04.80&jet + j1246 - 0730&682 @xmath38 20&12 46 04.232&-07 30 46.57 & + j1248 - 0632&353 @xmath38 11&12 48 22.975&-06 32 09.80&jet ? + j1308 - 0500&0 . & & & + j1309 - 0415&84.7 @xmath38 2.5&13 09 17.831&-04 15 16.20 & + j1312 - 0424&286 @xmath38 9&13 12 50.900&-04 24 49.88 & + j1316 - 3338&1100 @xmath38 30&13 16 07.985&-33 38 59.17&jet ? + j1316 - 3429&0 . & & & + j1332 - 1402&154 @xmath38 5&13 32 30.929&-14 02 13.16&jet ? + j1332 - 1256&99.4 @xmath38 3.0&13 32 39.251&-12 56 15.34 & + j1337 - 1257&3850 @xmath38 120&13 37 39.782&-12 57 24.69 & + j1441 - 3913&68.9 @xmath38 2.1&14 41 15.088&-39 13 48.03 & + j1443 - 3908&36.7 @xmath38 1.1&14 43 57.197&-39 08 39.73 & + j1457 - 3539&606 @xmath38 18&14 57 26.711&-35 39 09.98&jet + j1459 - 1810&158 @xmath38 5&14 59 28.763&-18 10 45.19 & + j1502 - 1508&133 @xmath38 4&15 02 25.017&-15 08 52.50 & + j1505 - 1610&130 @xmath38 4&15 05 22.950&-16 10 40.60 & + j1505 - 3432&393 @xmath38 12&15 05 02.371&-34 32 56.83 & + j1507 - 1652&2090 @xmath38 60&15 07 04.786&-16 52 30.26&jet ? + j1508 - 1548&307 @xmath38 9&15 08 35.700&-15 48 31.51 & + j1513 - 2558&72.7 @xmath38 2.2&15 13 53.174&-25 58 30.16&jet + j1517 - 2618&92.2 @xmath38 2.8&15 17 26.621&-26 18 18.99 & + j1517 - 2422&1930 @xmath38 60&15 17 41.813&-24 22 19.47&jet + j1530 - 2410&55.6 @xmath38 1.7&15 30 17.018&-24 10 46.41 & + j1532 - 2310&158 @xmath38 5&15 32 31.529&-23 10 32.43&jet ? + j1555 - 0326&232 @xmath38 7&15 55 30.748&-03 26 49.51&jet ? + j1612 - 2239&107 @xmath38 3&16 12 28.444&-22 39 46.68 & + j1626 - 2951&1140 @xmath38 30&16 26 06.020&-29 51 26.97 & + j1627 - 2426&71.2 @xmath38 2.1&16 27 00.008&-24 26 40.44 & + j1627 - 0939&2.95 @xmath38 0.15&16 27 45.453&-09 39 45.58 & + j1628 - 1415&275 @xmath38 8&16 28 46.618&-14 15 41.82 & + j1632 - 1052&140 @xmath38 4&16 32 50.108&-10 52 31.94 & + j1634 - 1440&0 . & & & + j1639 - 0715&14.4 @xmath38 0.4&16 39 21.996&-07 15 33.12 & + j1644 - 0750&1.12 @xmath38 0.12&16 44 00.768&-07 50 00.54 & + j1644 - 0743&129 @xmath38 4&16 44 52.058&-07 43 43.10 & + j1653 - 0150&68.3 @xmath38 2.1&16 53 57.800&-01 49 59.32 & + j1716 - 0452&436 @xmath38 13&17 16 26.487&-04 52 11.94 & + j1717 - 3342&685 @xmath38 21&17 17 36.028&-33 42 08.91&jet + j1719 - 2720&27.0 @xmath38 0.8&17 19 59.708&-27 20 42.60 & + j1721 - 2711&0.&&&pn + j1722 - 0503&310 @xmath38 9&17 22 03.539&-05 03 25.00&jet + j1729 - 0735&103 @xmath38 3&17 29 34.946&-07 35 32.38 & + j1733 - 1304&5110 @xmath38 150&17 33 02.705&-13 04 49.54&jet ? + j1734 - 3314&0 . & & & + j1740 - 1515&55.5 @xmath38 1.7&17 40 03.344&-15 15 52.89 & + j1743 - 0350&6240 @xmath38 190&17 43 58.856&-03 50 04.61 & + j1800 - 3849&46.1 @xmath38 1.4&18 00 11.821&-38 49 52.50&extended + j1802 - 3940&1700 @xmath38 50&18 02 42.677&-39 40 07.90&jet ? + j1802 - 0207&1.36 @xmath38 0.14&18 02 49.678&-02 07 49.15 & + j1832 - 2039&1070 @xmath38 30&18 32 11.047&-20 39 48.20&jet + j1837 - 0629&0.&&&snr025.5 + 00.2 + j1837 - 0616&0 . & & & + j1848 - 2718&417 @xmath38 13&18 48 47.504&-27 18 18.09 & + j1850 - 2740&90.0 @xmath38 2.7&18 50 13.979&-27 40 20.92 & + j1905 - 1153&154 @xmath38 5&19 05 28.591&-11 53 32.38&jet + j1911 - 2006&2190 @xmath38 70&19 11 09.652&-20 06 55.10&jet ? + j1911 - 1921&182 @xmath38 5&19 11 56.517&-19 21 50.97&jet + j1917 - 2110&195 @xmath38 6&19 17 08.642&-21 10 30.78 & + j1923 - 2104&2200 @xmath38 70&19 23 32.189&-21 04 33.33 & + j1935 - 1602&286 @xmath38 9&19 35 35.795&-16 02 32.38 & + j1940 - 0039&89.1 @xmath38 2.7&19 40 09.000&-00 39 02.01&jet + j1947 - 3542&33.5 @xmath38 1.0&19 47 22.624&-35 42 03.69 & + j1952 - 3412&116 @xmath38 3&19 52 00.109&-34 12 26.45&jet + j2005 - 2310&230 @xmath38 7&20 05 56.594&-23 10 27.01 & + j2025 - 0735&727 @xmath38 22&20 25 40.659&-07 35 52.70&jet + j2030 - 3039&229 @xmath38 7&20 30 57.931&-30 39 24.36 & + j2039 - 3157&231 @xmath38 7&20 39 08.653&-31 57 03.90 & + j2158 - 3013&204 @xmath38 6&21 58 52.064&-30 13 32.10&jet + j2318 - 0352&81.2 @xmath38 2.4&23 18 15.623&-03 52 14.85&jet ? + j2323 - 0150&233 @xmath38 7&23 23 04.629&-01 50 48.10 & + j2323 - 0317&918 @xmath38 28&23 23 31.953&-03 17 05.02&jet ? + j2326 - 0202&226 @xmath38 7&23 26 53.776&-02 02 13.76&jet +

we present high resolution h41 hydrogen recombination line observations of the ( 3 pc ) region surrounding sagittarius a ( sgr a ) at 92 ghz using the millimeter array at the owens valley radio observatory ( ovro ) with an angular resolution of and velocity resolution of 13 . new observations of h31 , h35 , h41 and h44 lines were obtained using the nrao 12-m telescope , and their relative line strengths are interpreted in terms of various possible emission mechanisms . these nrao 12-m measurements are the most extensive recombination line survey of this region to date . these ovro data also represent the highest angular resolution and the highest sensitivity observations of sgr a west in hydrogen recombination line and continuum emission at the millimeter ( mm ) wavelengths . observations of hco transition at 89 ghz are obtained simultaneously with a 40% improved angular resolution and 4 - 15 times improved sensitivity over the previously published results , and the distribution and kinematics of the dense molecular gas in the circumnuclear disk ( cnd ) are mapped and compared with those of the ionized gas . the line brightness ratios of the hydrogen recombination lines are consistent with purely spontaneous emission from k gas with near lte condition . a virial analysis suggests that the most prominent molecular gas clumps in the cnd have mean densities of order , sufficient to withstand the tidal shear in the galactic center region . therefore , these clumps may survive over several dynamical times , and the cnd may be a dynamically stable structure . we estimate a total gas mass of about for the cnd . our combined analysis of the new high resolution h41 and hco images and our kinematic model demonstrates a widely spread physical and dynamical link between the molecular gas in the cnd and the ionized gas , including along the eastern rim of the cnd where a gap was previously suggested .

astro-ph0408136

studying radio recombination lines ( rrls ) is one of the most direct ways of probing ionized gas such as hii regions . ionized gas properties such as kinematics , electron temperatures , and the geometrical filling factor can be directly constrained by the recombination line flux density , line width , free - free continuum flux density , and size of the line - emitting regions . at centimeter ( cm ) wavelengths , rrls are primarily stimulated emission amplifying extended non - thermal synchrotron emission . therefore , distinguishing emission arising from diffuse , low density gas from that from clumpy , high density gas is difficult . also , pressure broadening of the line width can complicate the interpretation of the cm rrl observations . for instance , cm rrl emission from luminous starburst systems such as arp 220 may be dominated by internally stimulated emission from diffuse ionized gas @xcite while spontaneous emission process is important at higher frequencies ( e.g. * ? ? ? * ; * ? ? ? * ) . interpretation of recombination lines at millimeter ( mm ) wavelengths is more straightforward since stimulated emission and pressure broadening become negligible in most cases . recent studies of mm rrls in galactic h ii regions @xcite and starburst galaxies @xcite found higher line fluxes than expected from the lower frequency measurements . a steep rise in line flux density with frequency , characteristic of optically thin spontaneous emission , is seen in nearby starburst galaxies m82 and ngc 253 @xcite . the central 1 parsec region surrounding the putative central source in our galactic center , sgr a@xmath2 , is a frequent target of multi - wavelength investigations primarily because it is the nearest and the best laboratory for studying the active galactic nucleus ( agn ) phenomenon and its fueling process . currently this region appears to be relatively gas free . however , immediately surrounding sgr a@xmath2 is an array of ionized gas streamers ( sgr a west ; see review by * ? ? ? these streamers have been previously studied in the 12.8 @xmath11 m [ ne ii ] @xcite , p@xmath0 @xcite , and in cm rrl @xcite , and they consist of several distinct kinematic features . the total mass of the ionized streamers is estimated to be less than 100 m@xmath12 in total . recent ir and radio observations found evidence for _ neutral _ gas within this central `` cavity '' @xcite , suggesting a more substantial presence of gas in the central parsec region . surrounding and closely interacting with the ionized streamers is the dense and clumpy molecular ring commonly referred to as the circumnuclear disk ( cnd ) . the cnd is asymmetric about sgr a@xmath2 , indicating its possible origin related to a gravitational capture of a passing cloud or energetic disruption of a stable ring @xcite . this ring is believed to represent a circumnuclear accretion disk feeding the inner parsec although some have argued that the cnd is not a complete ring ( see * ? ? ? * ; * ? ? ? the ionized streamers in sgr a west are explained as cloudlets from the clumpy ring that after losing momentum have fallen toward the center and thereby have stretched and have been ionized by photo - dissociation within the inner parsec ( see * ? ? ? sensitive imaging studies in cm rrl @xcite suggest high temperature ( @xmath13 k ) , with a possible radial temperature gradient increasing toward sgr a@xmath2 . the large proper motions of the stars very close to sgr a@xmath2 suggest central dark mass of 2.5 - 4.1 @xmath14 m@xmath12 @xcite . motivated by the detection of extragalactic mm rrls and recent new observations of the galactic center region , we have imaged the h@xmath15 line at high angular resolution toward sgr a@xmath2 in order to study the nature of the ionized streamers and their relation to the dense neutral gas in and around the cnd . at high frequencies , most of the rrl and continuum emission comes from the densest ionized gas ( see * ? ? ? * ; * ? ? ? one of the primary goals of this study is to examine the existence of ionized gas denser than that traced by cm - wavelength rrls in this region . the nature of the radiation mechanism is investigated by analyzing the new h31@xmath0 , h35@xmath0 , h41@xmath0 , and h44@xmath0 line observations obtained at the nrao 12-m telescope at kitt peak , arizona . we further examine whether any dense _ neutral _ gas is directly associated with the ionized streamers using our sensitive , high resolution hco@xmath10 ( 10 ) images . our hco@xmath10 ( 10 ) data have superior sensitivity and angular resolution over the previously published observations of similar molecular species @xcite . in addition , a detailed analysis of the spatial distribution and kinematics of the dense molecular gas and ionized gas in the cnd is offered using a simple kinematical model . the h41@xmath0 and hco@xmath10 ( 10 ) observations were made using the low , equatorial and high resolution configurations of the millimeter telescope array at owens valley radio observatory . the observational parameters from the multi - configuration observations conducted during the spring and fall 1998 are given in table [ tab : ovrosummary ] . the ovro array consists of six 10.4 meter telescopes equipped with cooled sis receivers . interferometric measurements are made using a total of 45 independent baselines in order to maximize the sampling of the uv - plane and to achieve the highest imaging dynamic range possible . in the high resolution configuration the longest baseline used was 220 meters . the spectrometer modules were arranged so that each line was covered with a total of 60 channels at 4 mhz ( 13 ) resolution centered at the velocity of @xmath16 . by setting the lo frequency at 90.5 ghz , h41@xmath0 line at 92 ghz ( upper sideband ) and hco@xmath10 ( 10 ) line at 89 ghz ( lower sideband ) could be observed simultaneously . the primary beam ( field of view ) is about @xmath17 at these frequencies . a nearby calibrator nrao 530 was observed every 15 minutes to track the atmosphere and instrumental gain . the amplitude scaling is derived from the observations of 3c 273 , whose flux density is monitored daily relative to uranus and neptune with an accuracy of about 20% . the data were reduced using the ovro data reduction program mma @xcite . sgr a@xmath2 is a low declination source , @xmath18 , and the synthesized beam is highly elongated in the n - s direction . the resulting beam size is @xmath19 ( fwhm ) . the data were imaged and analyzed using difmap @xcite and astronomical image processing system ( aips ) provided by the national radio astronomy observatory ( nrao ) . of the total 60 channels in the spectra , 15 line - free channels ( 60 mhz total ) were used to construct the continuum image . unresolved strong continuum source sgr a@xmath2 ( @xmath20 jy beam@xmath21 ) and the ionized streamers were used for self - calibration in difmap , and the continuum was subtracted from the spectral line channels . all spectral and continuum images were corrected for primary beam attenuation . subsequent analysis of the spectral line data cubes were conducted using aips and groningen image processing system ( gipsy ) . we have obtained the h31@xmath0 ( 210.502 ghz ) , h35@xmath0 ( 147.047 ghz ) , h41@xmath0 ( 92.034 ghz ) , and h44@xmath0 ( 74.645 ghz ) hydrogen recombination line spectra of sgr a west using the nrao 12-m telescope at kitt peak during may 1998 . the nrao 12-m telescope is equipped with single sideband sis receivers , and typical system temperature ranged between 250 and 400 k ( ssb ) , despite the low elevation of the source . the beam size is inversely proportional to the frequency and is 69@xmath22 at 92 ghz . both the filter banks ( 500 mhz at 2 mhz resolution ) and hybrid digital spectrometer ( 600 mhz at 0.78 mhz resolution ) were used to record the data . a wide range of off positions were examined , mostly along the minima in co and hcn surveys of the galactic center region . after an extensive investigation , it was determined that the hydrogen recombination line emission is highly localized around sgr a@xmath2 , and the final spectra presented here were obtained with a simple azimuth offset of @xmath23 . from the observatory report on the telescope efficiency measurements , we adopt the corrected main beam efficiency @xmath24 of 0.48 , 0.85 , 0.90 , and 0.95 for the four observed frequencies , respectively @xcite . these efficiencies are used to convert the observed brightness temperature @xmath25 to the main beam temperature @xmath26 and then to flux density . the telescope efficiency measurements and the subsequent corrections are inherently uncertain , and we adopt an absolute uncertainty of 20% for the measured flux density . the 92 ghz continuum image of the sgr a complex at the resolution of @xmath19 ( pa = @xmath27 ) is shown in figure [ fig : continuum ] . the major components of the sgr a complex detected are the spiral structure known as sgr a west and the compact central source sgr a@xmath2 , marked by a cross . the bright central source , sgr a@xmath2 , has peak flux density of @xmath28 jy beam@xmath21 . the total 92 ghz continuum flux density detected inside the 85 diameter region is @xmath2912 jy . some of the extended continuum emission may be missed by our observations as the 19-pointing mosaic image by @xcite reported a total flux of @xmath2915 jy covering a slightly larger area while @xcite reported a total flux of 27 jy within a 2 radius of sgr a@xmath2from a combination of bima and nobeyama 45-m telescope data . the major continuum features of sgr a west are identified in figure [ fig : continuum ] , and their observed and derived properties are tabulated in table [ tab : continuum ] . the east - west oriented filamentary structure located immediately south of sgr a@xmath2 is the bar of sgr a west . the total continuum flux at 92 ghz from the bar is about 0.9 jy with a peak brightness of 0.7 jy beam@xmath21 . the bar appears to connect to the western arc and the eastern arc on each ends . running almost perpendicular to the bar is the northern arm with higher total integrated flux density of about 1.1 jy . the continuum flux is higher toward the southern part of the northern arm . to the west of the bar is the western arc that curves south . it also extends north almost reaching the northern arm . the continuum flux from the western arc is relatively weak . the eastern end of the bar curves along the eastern arm but is not detected further north . the morphology of the ionized gas around the sgr a complex shown in figure [ fig : continuum ] is remarkably similar to the 8.3 ghz continuum image by @xcite . the peak flux density measured outside sgr a@xmath2 range between 0.1 @xmath30 0.75 jy beam@xmath21 ( see table [ tab : continuum ] ) , which corresponds to peak brightness temperature of 1.1 @xmath30 6.0 k. assuming this continuum is largely thermal free - free emission from 7000 k gas ( see [ sec : l / c ] ) , one can infer an optical depth of @xmath31 if the emitting region fills the beam with a unit filling factor . even if the filling factor is not near unity , this inference of optically thin emission should be secure unless the continuum emission is entirely concentrated into discrete clumps with a diameter of 0.005 pc ( 1000 au ) or smaller . continuum optical depth is given by @xmath32 assuming @xmath33 k , the emission measure ( @xmath34 ) required to produce continuum opacity of @xmath35 at 92 ghz is @xmath36 @xmath37 pc . if the thickness of the filaments along the line of sight is comparable to the apparent width ( @xmath38 pc ) , then the inferred mean electron density is @xmath6 @xmath7 ( @xmath39 @xmath40 ) . if the ionized gas is clumped at much smaller scales , then the electron density would be correspondingly larger . in comparison , @xcite derived @xmath41 @xmath7 from their analysis of fine - structure line ratios . the consideration of collisional de - excitation of the observed [ ne ii ] emission places an upper limit on the electron density of about @xmath42 @xmath7 @xcite . therefore the geometry of the ionized gas must be fairly smooth over 0.01 - 0.1 pc scales . the velocity integrated line flux density ( moment zero " ) image , shown in figure [ fig : h41a ] , was generated by summing the continuum subtracted h41@xmath0 image cube over all velocity channels with line emission ( from @xmath43 to + 221 ; see figure 3 ) . the cross marks the position of sgr a@xmath2 . the main features seen in the line emission map are very similar to the 92 ghz continuum map shown in figure [ fig : continuum ] , except for the absence of the bright , compact source corresponding to sgr a@xmath2 . the emission from the western arc is weaker than seen in continuum . the northern arm is the brightest continuum feature besides sgr a@xmath2 , and the brightest h41@xmath0 emission region also occurs in the northern arm just northeast of sgr a@xmath2 . the central bar shows a wide range of emission intensity and structures . a summary of the measured parameters for all major features are given in table [ tab : h41a ] . the sizes tabulated for the listed regions have been calculated assuming 1@xmath44 = 2.5 pc @xcite . unlike the 92 ghz continuum that peaks at sgr a@xmath2 , the brightest feature in h41@xmath0 occurs in the northern arm with a peak brightness temperature of @xmath45 k. the western arc , however , appears relatively weak in h41@xmath0 line . the peak line brightness for all h41@xmath0 are @xmath46 k , and thus the h41@xmath0 emission is also optically thin , @xmath47 ( see below ) . the spatially averaged h41@xmath0 line profiles for the six prominent emission regions ( identified with boxes ) are shown in figure [ fig : l / c ] . the gipsy tasks profil and profit were used to calculate the line - to - continuum ( l / c ) ratios and to fit a gaussian profile to the data using a least squares procedure . the crosses indicate the data points , the solid lines are the gaussian profiles , and the dashed lines represent the residuals . calculations of gaussian profiles yield four parameters : peak line - to - continuum flux density ratio , velocity integrated line intensity , central velocity , and the line width . these parameters are summarized in table [ tab : te ] along with the derived lte electron temperature derived from the l / c ratios ( see further discussions in [ sec : l / c ] ) . a typical l / c ratio along the northern arm is around 0.4 . the l / c ratio is the largest ( @xmath48 ) in the western arc while the l / c ratio in the bar drops from 0.6 to 0.03 within the 2 to 4 beam radii of sgr a@xmath2 . the kinematics of the ionized gas traced by h41@xmath0 shown in figures 3 and [ fig : l / c ] are generally consistent with that of rotation about sgr a@xmath2 , similar to that of the cnd , in broad agreement with previous studies of sgra west in rrls @xcite and in 12.8 @xmath11 m [ ne ii ] @xcite . the central bar spans a very wide range of velocity , between + 33 and @xmath30293 ( see figure 3 ) , much larger than those associated with the northern arm ( between + 163 and + 7 ) and the western arc ( between @xmath307 and @xmath30111 ) . this large velocity gradient suggests that this gas is located deeper in the gravitational potential , much closer to the central mass concentration , than the rest of the ionized gas . citing the relatively cool dust temperature along the bar , @xcite have argued that the nearness of the bar to sgr a@xmath2 is largely in projection only , but this large spread in velocity and the correspondingly large velocity gradient requires that this gas feature has to be located much closer to the central mass concentration than the northern arm and the western arc . since the cnd is already viewed with a large inclination ( @xmath49 ) , placing a gas streamer in a more edge - on orbit can increase the line - of - sight velocity only by about 25% . therefore a viewing geometry alone can not account for the large observed velocity gradient . neither the angular resolution nor the overall s / n of the new h41@xmath0 data are better than the previous vla h92@xmath0 measurements , and the main new insight obtained is that the ionized gas traced in h41@xmath0 emission share the same kinematics as the gas responsible for the h92@xmath0 emission . the h41@xmath0 line widths of the ionized streamers range between 43 and 347 ( see table [ tab : te ] ) . the broadest line is seen toward sgr a@xmath2 with @xmath50v = @xmath51 ( see figure [ fig : l / c ] , box 4 ) , but the s / n of this measurement is quite small in each individual channels . the next broadest line occurs in the vicinity of bright h41@xmath0 region in the northern arm with @xmath50v @xmath52 ( box 6 in figure [ fig : l / c ] ) . the narrowest line is seen toward the southern end of the western arc ( box 3 in figure [ fig : l / c ] ) with @xmath50v @xmath53 . using 2 resolution vla h92@xmath0 images , @xcite have shown that crowding of distinct kinematical components along the line of sight can make the emission lines appear much broader than the intrinsic line width of about 50 . combined with the large velocity gradient expected near the central mass concentration , the large line width we derive from h41@xmath0 emission must be primarily the result of spatial averaging resulting from our large observing beam of @xmath54 . a typical line width measured in each pixel is between 40 - 75 . given the low s / n of the data , it is not physically meaningful to deconvolve the individual kinematic components by fitting multiple gaussians to the observed spectra . there is a clear trend of increasing l / c ratio with a decreasing line width . such a trend is expected if the line and continuum emission arises from the same ionized gas clouds and the emitting gas has a narrow range of electron temperature ( see [ sec : l / c ] ) . figure [ fig : hco+ ] shows the high resolution image of the molecular ring in hco@xmath10 emission ( grey - scale ) superimposed over the h41@xmath0 ( contour ) emission . the circumnuclear ring traced in hco@xmath10 appears more complete than seen in the previous hcn @xcite and hco@xmath10 @xcite images obtained using the bima array . more recently @xcite imaged hcn , hco@xmath10 , and h42@xmath0 in a wider field of view using a 19-pointing mosaic , offering a much better overall picture of the molecular gas distribution in the 10 pc region surrounding the sgr a@xmath2 . however , the latest bima data by wright et al . are not much better than the older bima data in resolution and sensitivity . these different images agree with each other broadly , but significant differences also exist in the details . one important difference is that the angular resolution of our new ovro data is 40% better than those of gsten et al . and marr et al . , and our sensitivity is also 4 and 15 times improved , respectively . the bima array is more sensitive to faint extended emission , and its data reduction procedure is tuned to emphasize such features . in contrast , our moment analysis is more sensitive to identifying bright emission clumps , and this accounts for the more complete appearance of the cnd in figure [ fig : hco+ ] , consisting of many discrete features . determining whether any dense _ neutral _ gas is associated with the ionized streamers surrounding sgr a@xmath2 is one of the major objectives for our study . while the new observations have revealed clear evidence for a close physical link between ionized and neutral molecular gas ( as discussed later in [ sec : cnd ] ) , no hco@xmath10 emission is found spatially and kinematically coincident with the ionized streamers traced in h41@xmath0 ( see figure 3 ) . actually hco@xmath10 absorption is seen in many channels , particularly at blueshifted velocities , but most of these features show little kinematic relation to the h41@xmath0 emission . the spatial distribution of hco@xmath10 absorption resembles the entire continuum structure , and therefore the absorbing gas must be located in the foreground , associated with the 100 pc scale molecular disk imaged by @xcite and others . among the redshifted velocity channels where little foreground absorption is present ( e.g. , + 117 to + 195 km / s ) , there is little evidence for hco@xmath10 absorption at the locations of the h41@xmath0 and continuum emission peaks . in these channels , the upper limit to the optical depth for hco@xmath10 absorption is about 0.3 ( @xmath56 ) at a spectral resolution of 13 . this limit corresponds to an upper limit on the hco@xmath10 column density of @xmath57 @xmath40 . for an assumed hco@xmath10 abundance of @xmath58 , this translates into an h@xmath59 column density of @xmath60 @xmath40 . if the hco@xmath10 abundance appropriate for a translucent cloud ( where ion chemistry is important , see * ) is adopted , the inferred upper limit on the @xmath61 column density becomes as little as @xmath62 @xmath40 . these upper limits are close to or smaller than the column density of @xmath63 @xmath40 inferred from the [ o i]/[c ii ] far - ir line ratio @xcite and the mid - ir color analysis @xcite . using the calculations by @xcite , @xcite have proposed that most of the molecules in the neutral gas inside the central cavity with @xmath64 @xmath40 are photo - dissociated and that the neutral gas is primarily atomic . our hco@xmath10 absorption measurements are consistent with these conclusions , and we conclude independently that sgr a west filaments are either highly ionized or are largely atomic . the hydrogen recombination line spectra obtained at the nrao 12-m telescope are shown in figure [ fig:12 m ] , and the observing frequency and the measured line flux are summarized in table [ tab:12 m ] . the mm frequency window is rich in spectral transitions , and studying faint emission lines in the galactic center region can be challenging . the h35@xmath0 line is not shown because it is completely swamped by the much stronger cs @xmath65 line at 146.969 ghz , and no useful information could be derived . the h44@xmath0 spectrum , shown at the bottom of figure [ fig:12 m ] , is the least confused and the highest s / n measurement among all transitions observed at the 12-m telescope . the observed line profile closely resembles the h41@xmath0 spectrum derived from the ovro interferometer data ( shown with a solid line in the middle of figure [ fig:12 m ] ) . the total velocity integrated line flux density is @xmath66 jy km s@xmath21 within the 84 beam of the 12-m telescope at 74.645 ghz . the h41@xmath0 spectrum obtained using the nrao 12 m telescope , shown with a dotted line in the middle of figure [ fig:12 m ] , shows a bright , narrow feature near + 200 km s@xmath21 . this is not seen in the h41@xmath0 spectrum obtained from the ovro data ( shown in solid line ) . we have mapped h41@xmath0 emission in a 2 by 3 region centered on sgr a@xmath2 in order to examine the spatial extent of this spectral feature . the broad emission feature seen in the ovro data and in the h44@xmath0 spectrum is spatially confined only to the central position while the bright , narrow feature appears everywhere with little velocity gradient . the narrow line width , little velocity gradient , and wide - spread distribution strongly indicate that this narrow component is a galactic foreground emission . a likely identification is ch@xmath67cn , which has transitions at 91.959 , 91.970 , 91.980 , 91.985 , and 91.987 ghz . the total velocity integrated h41@xmath0 line flux density from the ovro data is @xmath68 jy . this is a lower limit since some extended flux may be missed by the interferometer ( see below ) . the h31@xmath0 spectrum shown at the top of figure [ fig:12 m ] is obtained by summing 8 independent spectra covering the whole sgr a west complex . the telescope beam at 210.502 ghz is about 30 and spatially resolves the sgr a west structure . therefore a 3 @xmath69 3 grid map was obtained at full beam sampling . the ne grid point was not observed , and the 8 spectra are summed together to derive the integrated h31@xmath0 spectrum . the h41@xmath0 channel maps ( figure 3 ) suggest that our h31@xmath0 spectrum may be missing some redshifted emission at @xmath70 = 50 - 150 km s@xmath21 . nevertheless , the summed spectrum closely resembles the h41@xmath0 and h44@xmath0 spectra , and we made no attempt to correct for any missing flux . the nine brightest hco@xmath10 clumps within the cnd are identified as localized peaks of velocity integrated hco@xmath10 flux density map ( figure [ fig : hco+ ] ) , and their physical sizes and emission characteristics are summarized in table [ tab : hco+ ] . the peak line brightness temperature , averaged over the @xmath71 ( 0.29 pc @xmath69 0.15 pc ) beam , ranges between 10 and 30 k. in comparison , the gas temperature derived from nh@xmath67 is 60 - 70 k @xcite while the dust temperature of about 75 k is derived by latvakoski et al . ( 1999 ) from the far - infrared observations . since hco@xmath10 emission is strongly absorbed by the foreground gas along the line of sight and the beam filling factor is likely less than unity , the gas temperature inferred from the observed line brightness is strictly a lower limit . the measured peak hco@xmath10 line temperature is comparable to the values found by earlier bima studies ( see [ sec : hco+ ] ) with a lower angular resolution , and one can safely deduce that the optically thick molecular gas nearly fills the beam at @xmath72 pc scales in these cnd clumps . we have computed virial masses of the individual hco@xmath10 emitting clouds as , @xmath73 where @xmath74 is the radius of the cloud in pc and @xmath75 is the line width in km s@xmath21 . if these individual molecular gas clumps are self - gravitating , then their masses range between @xmath76 ( see table [ tab : hco+ ] ) . assuming a spherical geometry , the mean mass density ( @xmath77 ) is between @xmath78 g @xmath7 and mean h@xmath59 molecule density of @xmath79 @xmath7 . these densities are significantly larger than the values inferred from the excitation analysis ( @xmath80 @xmath7 , * ? ? ? * ; * ? ? ? * ; * ? ? ? for a central source with a mass of @xmath81 @xcite , the critical mass density required to survive the tidal shear at a radius of 1.5 pc is @xmath82 g @xmath7 or @xmath83 @xmath7 . contrary to the previous studies using excitation analysis , we conclude that the mean gas density inferred from the virial analysis is close to or exceeding the critical density for surviving the tidal shear if they are self - gravitating . also , the cnd may well be a structure that can survive several dynamical time scales ( @xmath84 years ) . the total gas mass for the cnd inferred from our virial analysis is around @xmath9 , which is only a few per cent of the total dynamical mass for the central one parsec region . the virial mass of each of these clumps are about two orders of magnitude larger than that of the ionized gas associated with the whole ionized spiral ( sgr a west ) . therefore , an infall of even a much smaller gas clump is sufficient to account for the observed ionized streamers . the new higher resolution 10-field mosaic observations of hco@xmath10 and hcn emission by @xcite suggest even higher mean density ( @xmath85 @xmath7 ) for these dense neutral clumps . the 92 ghz continuum around sgr a is primarily free - free emission , and a good spatial correspondence between the continuum ( fig . [ fig : continuum ] ) and the h41@xmath0 line ( fig . [ fig : h41a ] ) emission is naturally expected . despite the similar morphology , however , a comparison suggests that the line emission is significantly clumpier than the continuum , especially in the western arc . the central bar is also clumpier in the line emission . in addition , in the northern arm the line emission appears brighter than anywhere else with an increasing trend toward sgr a@xmath2 . these differences may not be intrinsic to the source . a lower signal - to - noise ratio in the line image probably accounts for some of the differences . in addition , the continuum image suffers more from the missing flux problem than the line images because of a larger effective source size . for optically thin gas in lte emitting thermal continuum and negligible pressure broadening , lte electron temperature @xmath86 can be derived from the line - to - continuum ratio as , @xmath87 where @xmath88 is the observing frequency in ghz , @xmath75 is the velocity width ( fwhm ) in km s@xmath21 , @xmath89 and @xmath90 are continuum and line flux densities , respectively , and @xmath91 is the singly ionized helium abundance @xcite . in the case of sgr a west the aforementioned assumptions may be reasonably made along with the interpretation that the continuum at 92 ghz is largely free - free emission . the singly ionized helium abundance @xmath91 of 10% is assumed for sgr a west @xcite . the h41@xmath0 line widths listed in table [ tab : te ] are larger than the typical h92@xmath0 line widths of 40 - 50 reported by @xcite , and this is the direct consequence of a poorer angular resolution of the h41@xmath0 observations . the resulting blending of emission features could conspire to produce unusual l / c ratios if gas temperature and density vary significantly along the ionized streamers . the high angular resolution observations by @xcite have shown that this is not the case , and the poorer angular resolution of our h41@xmath0 observations should not affect our @xmath92 analysis in most cases . an electron temperature map can be derived from the 92 ghz continuum map ( fig . [ fig : continuum ] ) , the velocity integrated h41@xmath0 map ( fig . [ fig : h41a ] ) , and the relation given in eq . 3 . the derived electron temperature map , shown in figure [ fig : te ] , reveals a fairly uniform distribution of @xmath92 around 7000 k throughout the sgr a west complex except near sgr a@xmath2 ( see table [ tab : te ] ) . in general @xmath93 derived at higher frequencies tends to be larger because of diminishing contribution from stimulated emission . however , electron temperature derived with h41@xmath0 line at 92 ghz agrees well with the derivation using h92@xmath0 line at 8.3 ghz ( @xmath94 k , * ? ? ? * ) , suggesting that the contribution from stimulated emission is negligible even frequencies as low as 8.3 ghz . detecting little h41@xmath0 emission in the direction of the bright continuum source sgr a@xmath2 is another clear evidence that stimulated emission is not important at 92 ghz ( see [ sec : rrl ] ) . the higher electron temperature derived along the central bar just south of sgr a@xmath2 has previously been noted by @xcite and @xcite , and whether this is a real effect or an artifact of data analysis has been a subject of recent debates . citing the discovery of highly blueshifted line emission near @xmath95 missed by these earlier studies , @xcite have suggested that the unusually low l / c ratio and thus the inferences of higher electron temperature may be erroneous . in general , a lower than expected l / c ratio can result from either ( 1 ) missing line flux or ( 2 ) extra continuum flux . the explanation by @xcite is the former , but a careful analysis using the actual measurements of the extremely blueshifted line by @xcite revealed that the missed flux alone can not fully account for the low l / c ratio . in addition , @xcite noted the presence of two optically thick continuum sources irs 2 and irs 13 whose additional continuum flux should contribute to the unusually low l / c ratio ( and higher @xmath92 ) observed . a broad coincidence of this position with that of the mini - cavity " @xcite and its possible impact on the calculation of @xmath93 are discussed in detail by @xcite . presence of significant structures at scales smaller than our synthesized beam violates the simple geometrical assumption going into eq . 3 , and this may account for part of the observed local `` enhancement '' in @xmath93 . the local enhancement in electron temperature near the location of irs 2 and irs 13 is clearly seen in figure [ fig : te ] in the same manner as the seen in figure 14 of @xcite . the persistence of this feature at 92 ghz implies that the presence of the two continuum sources as suggested by @xcite does not offer a full explanation for the observed low l / c ratio . @xcite estimated an optical depth of @xmath96 and @xmath97 at 8.3 ghz for the continuum sources irs 2 and irs 13 by comparing their 8.3 ghz image with the 5 and 15 ghz vla continuum images . free - free optical depth scales as @xmath98 ( see eq . 1 ) , and the continuum optical depth for irs 2 and irs 13 should be @xmath99 at 92 ghz . in comparison , optically thin free - free emission associated with the ionized filament should scale only as @xmath100 . as a result , the continuum contribution from irs 2 and irs 13 should be substantially reduced at 92 ghz , and the l / c ratio should reflect mostly the optically thin emission associated with the ionized gas filaments in sgra west . however , the observed l / c ratio at 92 ghz is still noticeably lower at this location . the area of a lower l / c ratio is not centered on these two continuum sources either . if continuum opacities for these two sources are much larger than estimated by roberts et al . ( i.e. , optically thick even at 92 ghz ) , then the explanation of irs 2 and irs 13 as additional sources of continuum is still plausible , although the spatial distribution is still a problem . in summary , the apparent enhancement in @xmath92 ( a low l / c ratio ) along the central bar just southwest of sgr a@xmath2is still not fully explained . the flux density ratio between h41@xmath0 and h92@xmath0 lines is shown in figure [ fig : h92a / h41a ] . the overall structure of the northern arm , the bar , western arc , and eastern arm is almost identical at both frequencies , and the line ratio is nearly constant , @xmath101 . this similarity is remarkable given the order of magnitude difference in the frequencies of these lines . a slightly higher ratio is seen along the western arc and toward the top where the western arc arches into the northern arm , suggesting brighter emission at centimeter wavelengths . this trend of comparatively brighter emission at longer wavelength was also noted in the comparison of h92@xmath0 to br@xmath102 ratios by @xcite . roberts & goss offered extinction of br@xmath102 emission by dust in the cnd as a possible explanation , but a higher @xmath103 ratio would require a different explanation . since the derived @xmath93 distribution is fairly constant ( see fig . [ fig : te ] and table [ tab : te ] ) , the enhanced @xmath103 ratio is not likely caused by any changes in the physical characteristics of the gas . the comparison of the total integrated h41@xmath0 line flux with those of other transitions measured with the nrao 12-m telescope suggest that some extended flux is missed by the interferometer ( see [ sec : multi - lines ] ) . the shortest baseline in the ovro observations of 3100 @xmath104 is about 4 times larger than that for the vla observations ( 830 @xmath104 ) , and the greater missing flux in the ovro data offers a plausible explanation for the apparent enhanced @xmath103 ratio along the western arc . the analysis of the 92 ghz continuum ( [ sec : continuum ] ) and the h41@xmath0 emission ( [ sec : h41a ] ) suggests high electron density ( @xmath6 @xmath7 ) and low optical depth ( @xmath105 ) for the ionized streamers . under these conditions , both internal and external stimulated emission should be negligible , and optically thin spontaneous emission should dominate the recombination line emission at millimeter wavelengths . for the optically thin case near lte , eq . 3 implies that the velocity integrated line flux density @xmath106 should increase linearly with frequency since @xmath107 for thermal free - free emission . in their eq . however , the @xmath108 factor , which they treated as a constant " term going from their eq . a8 to eq . a9 , actually behaves like @xmath109 ( @xmath110 $ ] ) , and this too leads to a linear relation , @xmath111 . ] a plot of the velocity integrated line flux density for h31@xmath0 , h41@xmath0 , h44@xmath0 , and h92@xmath0 lines is shown as a function of frequency shown in figure [ fig : lte ] , and indeed a nearly linear trend is seen . excluding the h41@xmath0 measurement from the ovro ( which is a lower limit because of the missing flux ) , a best fit power - law slope is @xmath112 , which is nearly exactly the theoretically expected value . the assumption of `` near lte '' is not strictly correct , however . a general expression for recombination line brightness in comparison to the lte value is @xmath113 where @xmath114 is the lte line brightness , @xmath115 is the departure coefficient , @xmath116 is continuum optical depth , and @xmath117 is the effective line absorption coefficient accounting for non - lte stimulated emission in the line . for the range of physical conditions we derive for the sgr a west ( @xmath5 k , @xmath118 @xmath7 ) , the computed values of these parameters are @xmath119 0.7 - 0.9 and @xmath120 @xcite . although the non - lte factor in the line absorption coefficient @xmath117 is large , it has little impact on the observed line brightness because a small optical depth ( @xmath31 , as shown in [ sec : continuum ] ) makes the second term in eq . 4 negligible , i.e. @xmath121 . the net effect is that the observed recombination line brightness is reduced by a factor of @xmath115 from the lte value ( i.e. , @xmath122 ) . the linear slope of the the observed power - law relation between the velocity integrated line flux density and frequency is unaffected as long as @xmath121 . as a result , the average flux density of the recombination line should increase linearly with frequency even when some non - lte effects are present as long as spontaneous emission dominates the radiation process . the comparison of the h41@xmath0 and h92@xmath0 spectra obtained toward sgr a@xmath2 ( figure [ fig : twospecs ] ) offers a definitive proof that the narrow h92@xmath0 line seen near @xmath123 by @xcite is indeed stimulated emission . citing the nonthermal nature of the sgr a@xmath2 continuum emission and the low observed l / c ratio , roberts & goss suggested that this h92@xmath0 emission must be stimulated emission arising from diffuse foreground ionized gas . the h92@xmath0 spectrum ( shown in dotted line ) and the h41@xmath0 spectrum ( shown in solid line ) have essentially identical shape at a position along the northern arm , 7 offset to the east of sgr a@xmath2 , suggesting the same ionized gas is responsible for both spectral features . at the location of sgr a@xmath2 , however , h92@xmath0 spectrum shows a distinct narrow component near @xmath123 km s@xmath21 , not seen in the h41@xmath0 spectrum . as shown in eq . 1 , the optical depth for h41@xmath0 line is expected to be more than 100 times smaller than the h92@xmath0 line , and h41@xmath0 is expected to be smaller by this large factor for optically thin stimulated emission . a review of the recent literature yields a coherent but complex picture for the composition , geometry , and kinematics of the neutral and ionized gas found in the central few parsec of our galaxy . the cnd , traced in hco@xmath10 and hcn , has a ring - like morphology with an inner radius of about 30 ( 1.25 pc ) and an outer radius of at least 45 - 60 ( 1.8 - 2.5 pc ) and is physically and kinematically linked with a larger scale disk with an extent of 10 pc or more @xcite . it may be made of a multi - phase medium as the estimates of gas temperature ( 50 - 250 k ) and density ( @xmath124 @xmath7 and @xmath125 @xmath7 in dense clumps , see [ sec : hco+ ] ) range widely @xcite . within it lies the `` central cavity '' which actually contains the ionized streamers sgr a west , made of about 100 @xmath126 of @xmath127 @xmath7 gas at @xmath128 k and as much as 1000 @xmath126 of neutral gas which is thought to be mostly atomic with @xmath129 @xmath7 and t = 100 - 300 k . by obtaining the most sensitive and the highest angular resolution images of h41@xmath0 , 92 ghz continuum , and hco@xmath10 ( 10 ) to date simultaneously , we are able to shed some new light on the physical link between the neutral and ionized gas in this region . the clearest demonstration of the physical connection between the neutral and ionized gas is the presence of recombination line peaks along the western arc , just inside of the dense gas clumps within the cnd as shown by the overlay of the velocity integrated h41@xmath0 and hco@xmath10 images ( see figure [ fig : hco+ ] ) . this radial offset between the ionized streamers and the molecular ring has been noted previously @xcite , and it is consistent with the suggestion that the western arc is the ionized inner edge of the molecular ring directly exposed to a central ionizing source @xcite . evidence for a similar link between neutral and ionized gas is also seen on the eastern rim of the cnd , between the eastern arm and the clump i shown in figure [ fig : hco+ ] . from their comparison of the 8 ghz radio continuum map and the 30 @xmath11 m optical depth map , @xcite postulated that the eastern arm is the ionized front of a protrusion of high - density material facing the central ionizing source . the clump i identified in figure [ fig : hco+ ] may indeed be this previously unseen high - density material , revealed for the first time . the extinction map derived from the hst / nicmos p@xmath0 observations by @xcite shows a hole in the obscuring material towards the northern arm and western arm , but high extinction regions with @xmath130 surround this hole , including the hcn clump corresponding to our clump i. possible ionizing source candidates range from the putative central engine sgr a@xmath2 to hot uv stars near the galactic center ( see discussion by * ? ? ? a more exotic possibility suggested by @xcite is that the central source was much brighter in x - ray some 300 years ago as a result of the accretion of infalling material pushed in by the forward shocks from the supernova explosion in sgr a east , leading to the ionization of the streamers . a close physical link between the neutral gas in the cnd and the ionized gas in sgr a west is a widely occurring phenomenon , and this fact can be demonstrated more convincingly in the comparison of the channel maps as shown in figure 3 . within the velocity ranges of @xmath131 km s@xmath21 , the bulk of h41@xmath0 and hco@xmath10 features trace the same rotation - like kinematics about sgr a@xmath2 . more importantly , the h41@xmath0 features frequently occur just inside of the hco@xmath10 features ( see the @xmath13291 , @xmath3039 , and + 65 km s@xmath21 channels ) . as in the comparison of the velocity integrated images ( figure [ fig : hco+ ] ) , this physical link is more evident along the western arc . there is a tendency for the h41@xmath0 features to show an azimuthal offset from the hco@xmath10 features , particularly for the features closer to sgr a@xmath2 ( e.g. v=@xmath3039 and v=+65 km s@xmath21 channels ) . these azimuthal offsets and a broad total line width ( @xmath133 ) for the h41@xmath0 line are naturally explained if the gas follows nearly keplerian orbits around a highly concentrated mass distribution . ( see the discussion in [ sec : model ] ) . constrained by excellent high spatial resolution imaging data in 12.8 @xmath11 m [ ne ii ] and radio recombination lines , several plausible and detailed dynamical models for the ionized streamers have been constructed ( e.g. * ? ? ? * ; * ? ? ? models of gas kinematics for the cnd have also been constructed , mostly to explain the features identified in low resolution images ( @xmath134 ) that are susceptible to confusion with ubiquitous foreground and background emission ( e.g. * ? ? ? * ; * ? ? ? taking advantage of the new high ( and matched ) resolution spectroscopic imaging data in h41@xmath0 and hco@xmath10 , we can now examine in detail the gas kinematics and the gravitational potential in the central few parsec radius region in a self - consistent manner . we have already concluded that some of the ionized gas is orbiting sgr a@xmath2 in the plane defined by the cnd ( see [ sec : cnd ] ) . using the velocity information , we can also isolate and examine other kinematic features proposed to be associated with the cnd , such as the `` + 50 streamer '' and the `` + 70 feature '' ( see * ? ? ? if a massive central source ( 2.5 - 4.1 @xmath14 m@xmath12 , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) dominates the gravitational potential in the central parsec region , the kinematics of the surrounding gas disk should be regular and well behaved . the neutral and ionized gas inside the cnd contributes only about 1000 @xmath126 while we estimate the total mass of the cnd to be about @xmath135 @xmath126 ( see [ sec : cnd ] ) . the mass of the central stellar cluster is more substantial and is highly centrally concentrated , doubling the enclosed total mass near 2 pc radius @xcite . we adopt a mass distribution of the form similar to eq . 3 of @xcite , after updating it using the most recent mass estimates of the massive central source by @xcite and the dynamical mass estimates for the central 10 pc radius by @xcite , @xmath136 where @xmath137 is radius in pc . the resulting keplerian - like rotation of the gas disk nicely reproduces the observed features in the channel maps such as the azimuthal displacement for radially distinct features as discussed in [ sec : cnd ] . to examine whether our interpretations of the gravitational potential , the geometry of the cnd , and the relationship between the ionized streamers and the dense neutral gas in the cnd are correct , we have created simulated channel maps of the cnd using a monte carlo realization . they are shown side - by - side in comparison with the observed channel maps in figure 11 . using matlab software package , we created a disk of 3000 particles with a @xmath138 radial distribution . the model disk has a radial extent from 0.1 pc to 2.5 pc and has a radially dependent thickness with an opening angle of @xmath139 . this particular disk geometry is adopted to allow a finite thickness which may play a role in the _ apparent _ gas distribution and kinematics in the projected view , but it is not motivated any particular observational constraints . opening angle , based on their own morphological and dust heating model . ] the model disk is viewed with an inclination angle of 68@xmath140 and is rotated by 60@xmath140 in the sky plane to match the aspect ratio and the position angle of the cnd . each disk particle is in rotation about the central potential described by eq . 5 and has a gaussian random velocity component of 10 . to produce the model channel maps shown in figure 11 , model disk particles with the line - of - sight velocity matching the particular channel velocity window are selected , and their distribution is convolved with the observing beam ( @xmath54 , pa@xmath141 ) . the only adjustment made to the model was adding + 13 to the systemic velocity in order to match the highest velocity hco@xmath10 features at @xmath142 and + 143 . even though our model disk is constructed for an illustrative purpose only and is not tuned to match the observations in detail , the side - by - side comparisons of the model and data shown in figure 11 offer significant insights on the gas distribution and kinematics : \(i ) as noted already , some of the ionized gas traced in h41@xmath0 is sharing the same orbital plane as the dense neutral gas in the cnd ( see [ sec : cnd ] ) . the model disk has no knowledge of temperature or ionization information , but presumably the inner part ( @xmath143 ) of the disk is either atomic or ionized . the clumps of the h41@xmath0 emission shown in contours on both panels in figure 11 are frequently overlaying the model emission regions shown in grey - scale , especially in the channels with velocities between @xmath144 91 . the northern arm is sometimes modeled as being displaced from the plane defined by the cnd ( e.g. , * ? ? ? * ) , but a large fraction of the northern arm gas seen in the velocity range between 0 and + 143 closely follow our kinematical model . \(ii ) the same comparison also shows that the gas in the eastern arm and the central bar _ do not _ follow the model . this leads us to believe that the ionized gas in the bar and the eastern arm is largely concentrated in a different orbital plane as suggested by others previously ( e.g. , * ? ? ? ( iii ) the similarity between the observed hco@xmath10 emission and the model disk emission is more striking . except for the channels with @xmath145 @xmath146 v @xmath147 where foreground absorption has severely altered the appearance of the emission features , most of the hco@xmath10 emission seen in the channel maps have matching features in the model . the hco@xmath10 emission arising from the cnd is frequently asymmetric about its major axis , unlike the model , and the cnd is probably not a solid ring or a solid disk and is likely highly clumped . two channels show substantial hco@xmath10 emission not predicted by our model , and these channels correspond to velocities for the `` + 50 streamer '' and the `` + 70 feature '' that are already known to deviate from the cnd rotation ( see below ) . the + 50 streamer is a band of emission running nearly east - west at a slightly oblique angle in the v = + 39 and v = + 65 channels , previously noted by @xcite and others . this feature extends well beyond the field of view of the ovro interferometer and is several arcmin in length in the single dish hcn ( 32 ) channel maps of @xcite . this is clearly a structure much larger than the cnd in size , and thus the + 50 streamer is most likely a foreground or a background structure , not directly associated with the cnd . the + 70 streamer is the 7 " shaped feature present in the v = + 91 channel . this structure is also seen in the single dish hcn ( 32 ) channel maps of @xcite , but it is more compact than the + 50 streamer . although this may be a chance projection , this feature lies nearly exactly over the cnd . however , its velocity differs from the cnd rotation by about 130 . it is not as certain that the + 70 streamer is another foreground structure . its long , linear appearance is suggestive of a large scale shock in origin . it is not seen in h41@xmath0 or in continuum to offer any further clues , and the lack of any ionized gas at the corresponding velocity is an important argument against the + 70 streamer being part of the cnd . one of the key arguments against the cnd being a coherent disk or a ring is the apparent gap along its eastern side in the previous hcn and hco+ maps . this gap now appears filled with several compact gas clumps such as the clump i in our new hco+ image shown in figure [ fig : hco+ ] . in addition , neutral gas with higher temperature and density also fills this gap as inferred by the observations of [ c ii ] and nh@xmath67 ( see * ? ? ? * ; * ? ? ? * ) . therefore earlier arguments based on the apparent morphology in the lower sensitivity hcn and hco+ maps are no longer valid . the mean velocity of the warm molecular gas found by herrnstein & ho ( their clump d ) is slightly redshifted ( @xmath148 ) than expected from our simple rotation model , but this velocity shift is in the correct sense if their average spectrum includes the emission from the entire north - south extent of their clump d. in summary , we have successfully demonstrated the continuity in the distribution and kinematics of the ionized and the neutral gas within the central 3 pc diameter region surrounding sgr a@xmath2 using our new observational data and a simple kinematical model . our kinematical model reproduces many of the observed neutral and ionized gas features , as shown in figure 11 . the success of this simple model strengthens the notion that most of the gas found in the central parsec of sgr a@xmath2partake in a simple rotation around the central source within the disk plane defined by the cnd . our kinematical model also identifies very clearly which gas components _ do not _ follow this simple rotation . in addition , our model suggests that some of the subtle observed effects such as the slight asymmetry between the near and far side of the disk are the products of the finite size , asymmetric observing beam and projected viewing of the 3-d cnd structure ( e.g. , see @xmath149 channel ) . the overall picture of the gas distribution and kinematics in the central 3 pc surrounding sgr a@xmath2 is not yet complete because information on the distribution and kinematics of atomic gas inside the cnd is quite incomplete . future arcsecond resolution imaging of dense atomic gas tracers such as [ c ii ] will provide important clues for completing our understanding of the gaseous environment in this region . the ovro millimeter array observations of the h41@xmath0 and 92 ghz continuum emission from the ionized streamers around sgr a@xmath2 ( sgr a west ) and the hco@xmath10 ( 10 ) emission from the circumnuclear disk ( cnd ) are presented along with a new survey of the hydrogen recombination lines at millimeter wavelengths ( h31@xmath0 , h41@xmath0 , and h44@xmath0 ) obtained using the nrao 12-m telescope . these observations are compared with the vla h92@xmath0 and the 8.3 ghz continuum data by roberts & goss ( 1993 ) . we summarize our results as following : \1 . the 92 ghz continuum image is essentially identical to the vla continuum image at 8.3 ghz in morphology . the total continuum flux detected inside the 85 diameter is about 12 jy , but some extended flux associated with structure larger than about 30 is missed . the brightest feature , sgr a@xmath2 , is unresolved at 3 resolution of our observations and has a peak flux density of @xmath28 jy beam@xmath21 . assuming the 7000 k electron gas fills our 0.005 pc ( 1000 au ) beam , we infer average continuum optical depth of @xmath105 and electron density @xmath6 @xmath7 . the kinematics and the morphology of h41@xmath0 emission are very similar to those observed in h92@xmath0 and 12.8 @xmath11 m [ ne ii ] lines . the brightest emission occurs along the north arm with a peak line brightness temperature of less than 10 k. this suggests a h41@xmath0 optical depth of @xmath150 averaged over the 13 channel width . the emission along the central bar displays a line - of - sight velocity ranging over 300 , and this gas may be located in the deepest part of the gravitational potential within the cnd region . the molecular cnd is imaged in hco@xmath10 with a 40% improvement in resolution and 4 - 15 times better sensitivity over the previous bima observations made in hcn and hco@xmath10 . the cnd consists of several discrete concentrations , and their peak brightness ranging between 10 and 30 k averaged over the 0.1 pc synthesized beam suggests that they originate from the warm ( 60 - 100 k ) gas responsible for nh@xmath67 and dust continuum emission in the cnd . a virial analysis suggests that these clouds have mean density of @xmath151 @xmath7 , sufficient to withstand the tidal shear in the region , and the cnd may be a long lived structure . we estimate a total gas mass of about @xmath9 for the cnd . our non - detection of hco@xmath10 absorption along the sgr a west places an upper limit to the gas column density of @xmath152 @xmath40 , or the neutral gas within the cnd is depleted of molecules . \4 . a line - to - continuum ratio analysis under the lte assumption yields a fairly uniform electron temperature of about 7000 k along the ionized streamers in the sgr a west complex , in an excellent agreement with the electron temperature calculated at 8.3 ghz using the h92@xmath0 observations @xcite . evidence for increased @xmath92 ( or an unusually low l / c ratio ) along the central bar just southwest of sgr a@xmath2 is found in our data as well even though our data include all line emission within @xmath153 . the low l / c ratio we find at 92 ghz challenges the optically thick continuum sources irs 2 and irs 13 as additional sources of continuum emission as suggested by @xcite . therefore , the elevated @xmath92 ( a lower l / c ratio ) seen near the location of irs 2 and irs 13 is still not fully understood . \5 . a comparison of the velocity integrated line flux densities for h31@xmath0 , h41@xmath0 , h44@xmath0 , and h92@xmath0 emission from the sgr a west complex shows line ratios characteristic of spontaneous emission from optically thin gas near lte . a detailed analysis reveals that observed recombination line brightness is reduced by 10 - 30% ( i.e. , @xmath154 ) from the lte value . \6 . a combined analysis of the new high resolution h41@xmath0 and hco@xmath10 images and a kinematic model clearly demonstrates a widely spread , close physical and dynamical link between the molecular gas in the cnd and the ionized gas , including along the eastern rim where a gap in the cnd was previously reported . the same analysis also clearly identifies several features that do not follow the disk rotation . we conclude that the `` + 50 streamer '' is not directly associated with the cnd . the morphology of `` + 70 streamer '' is suggestive of its association with the cnd , but its kinematics and absence of any ionized gas challenge the idea of their close association . we are grateful to d. roberts for providing us with the h92@xmath0 and 8.4 ghz continuum data from the vla for comparison . we are indebted to c. lang for her expert advice and help with the gipsy data reduction package and for many helpful suggestions as the referee of this article . in addition , we would like to acknowledge the direction gained from insightful discussions with w. m. goss and late k. r. anantharamaiah . we would like to thank the ovro and the nrao 12-m telescope staff for the support and help without which this project would not have been possible . the national radio astronomy observatory is a facility of the national science foundation operated under cooperative agreement by associated universities , inc . lc rest frequency of h41@xmath0 line & 92.034 ghz + rest frequency of hco@xmath4 line & 89.188 ghz + primary beam ( hpbw ) & 85@xmath155 + phase center & @xmath156 + & @xmath157 + center lsr velocity & 0.0 + synthesized beam ( hpbw ) & @xmath19 ( pa@xmath141 ) + spectrometer settings & @xmath158 mhz ( 13 ) + rms noise & 0.13 jy beam@xmath21 ( 0.21 k ) + llllcccc northern arm & 0.9 & @xmath159 & @xmath160 & @xmath161 + sgr a@xmath2 & 0.1 & @xmath162 & @xmath163 & @xmath164 + bar & 1.1 & @xmath165 & @xmath166 & @xmath167 + western arc & 1.5 & @xmath168 & @xmath169 & @xmath170 + eastern arm & 0.4 & @xmath171 & @xmath172 & @xmath173 + llllcccc northern arm & 0.9 & @xmath174 & @xmath175 & @xmath176 + bar & 1.1 & @xmath177 & @xmath178 & @xmath179 + western arc & 1.5 & @xmath180 & @xmath181 & @xmath182 + eastern arm & 0.4 & @xmath183 & @xmath184 & @xmath185 + ccccccc 1 & @xmath186 & @xmath187 & 78 @xmath144 5 & 0.50 @xmath144 0.04 & 6394@xmath144608 + 2 & @xmath188 & @xmath189 & 213 @xmath144 18 & 0.18 @xmath144 0.02 & 6473@xmath144891 + 3 & @xmath190 & @xmath191 & 43 @xmath144 5 & 0.80 @xmath144 0.13 & 7470@xmath1441372 + 4 & @xmath192 & @xmath193 & 347 @xmath144 48 & 0.03 @xmath144 0.01 & 18806@xmath1444770 + 5 & @xmath194 & @xmath195 & 70 @xmath144 5 & 0.66 @xmath144 0.07 & 5528@xmath144623 + 6 & @xmath196 & @xmath197 & 101 @xmath144 6 & 0.26 @xmath144 0.02 & 9057@xmath144825 + cccccccc a & 0.4 & @xmath198 & @xmath199 & 20.7 & 2.1 @xmath200 & @xmath201 + b & 0.3 & @xmath202 & @xmath203 & 17.3 & 1.1 @xmath200 & @xmath204 + c & 0.3 & @xmath205 & @xmath206 & 12.0 & 0.5 @xmath200 & @xmath207 + d & 0.4 & @xmath208 & @xmath209 & 29.9 & 4.5 @xmath200 & @xmath210 + e & 0.2 & @xmath211 & @xmath212 & 11.6 & 0.3 @xmath200 & @xmath204 + f & 0.3 & @xmath213 & @xmath214 & 19.4 & 1.4 @xmath200 & @xmath215 + g & 0.6 & @xmath216 & @xmath217 & 16.1 & 1.9 @xmath200 & @xmath218 + h & 0.4 & @xmath219 & @xmath220 & 19.0 & 1.8 @xmath200 & @xmath221 + i & 0.2 & @xmath222 & @xmath223 & 13.0 & 0.4 @xmath200 & @xmath224 +

we study radiative processes of uniformly accelerated entangled atoms , interacting with an electromagnetic field prepared in the minkowski vacuum state . we discuss the structure of the rate of variation of the atomic energy for two atoms travelling in different hyperbolic world lines . we identify the contributions of vacuum fluctuations and radiation reaction to the generation of entanglement as well as to the decay of entangled states . our results resemble the situation in which two inertial atoms are coupled individually to two spatially separated cavities at different temperatures . in addition , for equal accelerations we obtain that one of the maximally entangled antisymmetric bell state is a decoherence - free state .

1512.02886

superposition and entanglement are the distinguished properties of the quantum theory . entangled pure quantum states are defined as states that can not be factorized into product of pure states of subsystems . entanglement has attracted much interest since it is a essential feature underlying quantum information , cryptography and quantum computation @xcite . there exist several sources of entangled quantum systems , such as in solid state physics , quantum optics and also atoms in cavity electrodynamics . many ways were proposed in order to generate entangled states in systems of two - level atoms interacting with a bosonic field . see for example @xcite . in a seminal paper , yu and eberly studied the radiative processes of entangled two - level atoms coupled individually to two spatially separated cavities @xcite . a different scenario was discussed in ref . @xcite . in this work , radiative processes of a quantum system composed by two identical two - level atoms interacting with a massless scalar field prepared in the vacuum state in the presence of perfect reflecting flat boundaries were analyzed . the authors investigated the spontaneous transitions rates from the entangled states to its collective ground state induced by vacuum fluctuations . in the presence of a single boundary the transition rate for the symmetric state undergoes a strong reduction , whereas for the antisymmetric state the results indicate a slightly enhancement . following the discussion by ackerhalt , knight and eberly @xcite , which presents the possibility to interpret spontaneous decay as a radiation - reaction effect , in ref . @xcite we have discussed quantum entanglement between inertial atoms within the formalism developed by dalibard , dupont - roc , and cohen - tannoudji ( ddc ) @xcite . as discussed by milonni the effects of vacuum fluctuations and radiation reaction depend on a particular ordering chosen for commuting atomic and field operators @xcite . the ddc formalism asserts that there is a unique way in which such distinct contributions to the rate of change of an atomic observable are separately hermitian and hence they enjoy an independent physical meaning . this is achieved by choosing a preferred operator ordering , namely a symmetric ordering of commuting atom and field variables . within such a formalism the interplay between vacuum fluctuations and radiation reaction can be considered to maintain the stability of the atom in its ground state . we remark that this formalism was employed in many situations @xcite . for uniformly accelerated atoms , we remark that this method motivates in a quantitative sense the intuitive scenario outlined in ref . @xcite . recently there has been an extensive literature devoted to investigations of quantum information theory within the framework of quantum field theory and quantum field theory in curved space - time . this emerging field of relativistic quantum information has been attracting increasing interest within the scientific community . in its essence , this subject aims at understanding the role that relativistic settings play in quantum information processes . in some way , familiar notions in quantum information theory claim for a critique reappraisal when one brings up relativistic considerations . even though we do not intend to give an extensive bibliography of the subject , here we call attention on some of such important works . for instance , in reference @xcite it was shown that entangled states reduce to separable states at infinite acceleration . this phenomenon is sometimes called unruh decoherence and is closely related to the unruh effect @xcite . the idea is that the acceleration of the observer causes a sort of enviromental decoherence which narrows the fidelity of certain processes of primary importance in quantum information . relationships between entanglement and space - time curvature were investigated in ref . the degradation phenomenon of entanglement in a schwarzschild black hole was investigated in ref . @xcite . on the other hand , in contrast to this effect , in references @xcite the authors considered a situation of tripartite entanglement in which one of the three entities moves with a uniform acceleration with respect to the others ; their results indicate that entanglement does not completely vanish in the infinite acceleration limit . in turn , generation of quantum entanglement between two localized causally disconnected atoms was studied in references @xcite . similar results were found in references @xcite , but with different setups and within distinct scenarios . we also mention investigations of quantum teleportation between noninertial observers @xcite , relativistic approaches to the einstein - podolsky - rosen ( epr ) construction and also to bell s inequality @xcite , epr - type correlations in quantum fields @xcite and entanglement in fermionic fields @xcite . see also @xcite for more interesting discussions on the subject of relativistic quantum entanglement . as a general source for relativistic quantum information we also refer the reader the ref . @xcite which also indicates some open problems in such a research field . various aspects of this important research topic are reviewed in the special issue of _ classical and quantum gravity _ [ * 29 * ( 22 ) ( 2012 ) ] . one of the most important implications of such studies is the realization of the observer - dependency property of quantum entanglement @xcite . hence a quantitative understanding of such phenomena is highly required in discussions on certain quantum information processes between accelerated members of a system . in the description of the dynamics of a system coupled to a reservoir different formalisms have been used . generally the reservoir is described by an infinitely many degrees - of - freedom formalism . the standard framework is the traditional master equation approach . the evolution of the reduced density matrix of the atoms is usually written in the kossakowski - lindblad form . within this approach it is possible to quantify the degree of entanglement of a particular state . here since we are interested in apprehending the machinery responsible for generation ( or degradation ) of entanglement in radiative processes involving atoms , we choose a different route and use the ddc formalism aforementioned . moreover , this formalism easily enables one to discuss classical and quantum mechanical concepts . hence in the present paper we propose to generalize the results of ref . @xcite in which the authors investigate quantum entanglement between two uniformly accelerated atoms with the same acceleration . being more specific , we assume that both atoms travel in different hyperbolic trajectories . we consider that both atoms are coupled to a quantum electromagnetic field . in addition , we intend to identify the role of vacuum fluctuations and radiation reaction in the radiative processes of entangled atoms in uniformly accelerated motion . the organization of the paper is as follows . in section [ model ] we discuss the implementation of the ddc formalism in the situation of interest . in section [ atom ] we calculate in detail the distinct contributions of vacuum fluctuations and radiation reaction to the rate of variation of atomic energy in vacuum for atoms moving along different hyperbolic trajectories . specific discussion regarding the obtained results is given . conclusions and final remarks are given in section [ conclude ] . the paper includes appendices containing details of lengthy derivations . in this paper we use units @xmath0 . we are using the minkowski metric @xmath1 , @xmath2 and @xmath3 . all graphics were drawn using the mathematica package . let us suppose the case of two identical two - level atoms interacting with a common electromagnetic field . we are working in a four - dimensional minkowski space - time . in this paper we work in the multipolar coupling scheme which means that all interactions are realized through the quantum electromagnetic fields . this formalism is suitable for describing retarded dipole - dipole interactions between the atoms . let us consider that such atoms are moving along different hyperbolic trajectories in the @xmath4 plane . in this case , the atoms are said to be rindler observers . as well known , the rindler metric @xmath5 is the metric seen by an uniformly accelerated observer . rindler coordinates @xmath6 and minkowski coordinates @xmath4 are related by the known relations @xcite & & x_1 = a^-1e^aa + & & t = a^-1e^aa , [ acc ] where @xmath7 is a positive constant and @xmath8 . the coordinates @xmath6 cover only a quadrant of minkowski space , namely the wedge @xmath9 . lines of constant @xmath10 are straight , whereas lines of constant @xmath11 are hyperbolae : @xmath12 these represent world lines of uniformly accelerated observers . one has that the ( constant ) proper acceleration is given by a e^-a = . thus , different ( but constant ) @xmath11 s correspond to different hyperbolae and hence to different proper accelerations . the accelerated observers proper time @xmath13 is related to @xmath14 by = e^a . [ propt ] this implies that in general , the atoms will be moving along different world lines , so there will be two different proper times parameterizing each of these curves . we remark that the quantity @xmath7 remains the same for both atoms in this situation . here we would like to identify the distinct contributions of quantum field vacuum fluctuations and radiation reaction to the entanglement dynamics of the atoms . as discussed above , we consider both atoms on different stationary trajectories @xmath15 , where again @xmath16 denotes the proper time of the atom @xmath17 . usually one employs the proper time in order to describe time evolution of the atom - field system since it is directly measurable by the clocks of the observers . here , because the atoms are moving along distinct hyperbolic trajectories , one has two proper times . however , to derive the heisenberg equations of motion of the coupled system , we have to choose a common time variable . this implies that a certain amount of care is required when choosing the parameter which will describe the time evolution of the system . at first sight , it is most reasonable to refer generally to one of the atom s proper time @xmath18 and then relate the other proper time @xmath19 with @xmath18 . at any rate , an alternative simple choice is also available , namely a coordinate time . in the present context , one could choose the minkowski coordinate time . nevertheless , as will become clear throughout the text , a more sensible path is to adopt the rindler coordinate time @xmath10 . hence , in what follows we describe the time evolution with respect to such a parameter which , because of ( [ propt ] ) , has a functional relation with each of the proper times of the atoms . the stationary trajectories guarantees that the undisturbed atomic system has stationary states which in our particular case are given by equation ( [ sta ] ) . within the multipolar coupling approach the purely atomic part of the total hamiltonian describes the free hamiltonian of two two - level atoms . the hamiltonian that governs the time evolution of this atomic system with respect to @xmath10 can then be written as h_a ( ) & = & , [ ha ] where @xmath20 , @xmath21 , @xmath22 and @xmath23 being the ground states of the atoms isolated from each other and @xmath24 and @xmath25 their respective excited states . the space of the free two - atom system is spanned by four product states with respective energies & & e_gg = -_0|gg= |g_1|g_2 , + & & e_ge= 0|ge= |g_1|e_2 , + & & e_eg = 0|eg= |e_1|g_2 , + & & e_ee = _ 0|ee= |e_1|e_2 , [ sta ] where a tensor product is implicit . instead of working with this product - state basis , we can conveniently choose the bell state basis . in terms of the product states , one has : maximally entangled two - qubit bell states , and they form a convenient basis of the two - qubit hilbert space . even though we consider identical atoms we are employing the above notation in order to allow the possibility of a straighforward generalization of our results to the case of non - identical atoms . furthermore , we remark that the bell states @xmath26 are not eigenstates of the atomic hamiltonian @xmath27 . here we consider that the two - atom system is coupled with an electromagnetic field . by exploring the gauge invariance of the electromagnetic lagrangian , we impose the transversality condition ( the coulomb gauge ) . upon the usual decomposition of the electric and magnetic field operators in terms of creation and annihilation operators ( t,*x * ) & = & _ d^3 k _ ( * k * ) , ( t,*x * ) & = & _ d^3 k(*k*_(*k * ) ) the electromagnetic field hamiltonian @xmath28 $ ] acquires the traditional form h_f(t ) = _ _ * k*a^_*k*,(t)a_*k*,(t ) , [ hf ] where @xmath29 are the creation and annihilation operators of the electromagnetic field and we have neglected the zero - point energy . the index @xmath30 labels different polarizations of the field : the orthonormal vectors @xmath31 , @xmath32 correspond to the two polarization states of the photon . these obey known relations that can be found in many textbooks , see for instance @xcite . in the multipolar coupling scheme and within the so - called electric - dipole approximation one has that the hamiltonian which describes the interaction between the atoms and the field is given by h_i ( ) & = & - _ 1(_1 ( ) ) ( x_1(_1 ( ) ) ) + & - & _ 2(_2 ( ) ) ( x_2(_2 ( ) ) ) [ hi ] where @xmath33 ( @xmath34 ) is the electric dipole moment operator for the @xmath17-th atom . it should be observed that the electric field above is the measured electric field defined through the measured force it exerts on an atom . the appearence of terms as @xmath35 is a consequence of the fact that both atoms move along different stationary trajectories as discussed above . as asserted in this paper we describe time evolutions with a coordinate time parameter which is to be identified with the rindler coordiante time @xmath10 . moreover , we are assuming that for both atoms each component of the dipole moment is required to evolve in time in the same way as de witt s monopole in the case of a scalar field , as discussed by takagi @xcite . this would be the case only if the directions of the both dipole moments are kept fixed with respect to the proper frame of reference of the atoms ; otherwise the rotation of the dipole moment will bring in extra time dependence in addition to its intrinsic time evolution . in addition , one should bear in mind that , for each atom @xmath36 , where @xmath37 stands for the @xmath37-th spatial component with respect to the proper reference frame of the atom , which is the rindler orthonormal frame of reference . the expression for the atom - field interaction can always be put in a manifestly invariant form , that is , for each atom : @xmath38 , where @xmath39 is the four - velocity of the atom and @xmath40 is the four - vector associated with the dipole moment such that , in its proper reference frame , its temporal component vanishes and its spatial components are given by @xmath41 . we point out that the dipole moment operator can be written as _ i(_i ) & = & g_i |_i| e_i | g_i e_i| + + & & , [ dip ] ( no summation over repeated indices ) where we have assumed a proper choice of phases which allows the dipole matrix elements to be real and we have defined the dipole raising and lowering operators as @xmath42 and @xmath43 , respectively . in addition , since the atoms are presumed to be identical , the matrix elements @xmath44 are independent of the index @xmath17 and denoted simply by @xmath45 . the heisenberg equations of motion for the dynamical variables of the atom and the field can be derived from the total hamiltonian @xmath46 . we draw attention to the fact that the present work employs a coherent quantum evolution treatment instead of the more tradicional open quantum system approach in the analysis of quantum entanglement . as remarked in the previous section , the former allows one to discuss the contributions of vacuum fluctuations and radiation reaction in the radiative processes of entangled atoms . the rates of change of the atomic observables are governed by the heisenberg equations & = & - i [ _ 1(_1())(x_1(_1())),s_a^z(_a ( ) ) ] + & - & i[_2(_2())(x_2(_2())),s_a^z(_a ( ) ) ] , + [ bad1 ] @xmath21 , whereas for dynamical variables of the field one has a_*k*,(t ( ) ) & & = -i_*k*a_*k*,(t ( ) ) + & & - i[_1(_1)(x_1(_1)),a_*k*,(t ( ) ) ] + & & - i[_2(_2)(x_2(_2)),a_*k*,(t ( ) ) ] . + [ bad2 ] in eq . ( [ bad2 ] ) , @xmath47 and the proper times @xmath48 must be considered as functions of @xmath10 . in addition , in eq . ( [ bad1 ] ) one have taken into account that @xmath49 commutes with @xmath27 . in order to proceed one splits the solutions of the equations of motion in two parts , namely : the free part , which is present even in the absence of the coupling ; and the source part , which is caused by the interaction between atom and field and contains the coupling constant @xmath50 . therefore one can write @xcite s_a^z(_a ( ) ) & = & s_a^z , f(_a ( ) ) + s_a^z , s(_a ( ) ) , + a_*k*,(t ( ) ) & = & a^f_*k*,(t ( ) ) + a^s_*k*,(t ( ) ) . similar considerations can be applied for the the electric dipole moment operators of the atoms , namely : @xmath51 . in this way , up to first order in @xmath52 , the solutions of the equations of motion are given by : s_a^z , f(_a ( ) ) & = & s_a^z , f(_a(_0 ) ) + s_a^z , s(_a ( ) ) & = & - i _ b = 1 ^ 2__0^d [ ^f_b(_b( ) ) ^f(x_b(_b())),s_a^z , f(_a ( ) ) ] + a^f_*k*,(t ( ) ) & = & a^f_*k*,(t(_0))e^-i_*k * [ t ( ) - t(_0 ) ] + a^s_*k*,(t ( ) ) & = & -i_b = 1 ^ 2__0^d[^f_b(_b( ) ) ^f(x_b(_b())),a^f_*k*,(t ( ) ) ] . since one can construct from the annihilation and creation field operators the free and source part of the quantum electric field , one also has ( t(),*x * ( ) ) = * e*^f(t(),*x * ( ) ) + * e*^s(t(),*x * ( ) ) . however , as discussed in refs . @xcite , there is an ambiguity of operator ordering which arises in such a procedure . being more specific , when one considers the heisenberg equations of motion for an arbitrary atomic observable @xmath53 = [ a ( ) , h_a ( ) ] + [ a ( ) , h_i ( ) ] and proceeds with the split of the solutions for the quantum field into free and source parts . the feature that all atomic observables commute with @xmath54 is preserved in time because of the unitary evolution . however , this is no longer true for @xmath55 and @xmath56 separately : the source part of the field picks up contributions of atomic observables during its time evolution . in face of this fact , one must choose an operator ordering when discussing the effects of @xmath55 and @xmath56 separately . this operation does not modify the final results for physical quantities , yet it does produce distinct interpretations concerning the roles played by vacuum fluctuations and radiation reaction . nonetheless , there exists a preferred operator ordering by which the effects of such phenomena can posses independent physical meanings @xcite . in other words , such works reveal that only if a symmetric ordering of atomic and field variables is adopted , both contributions coming from vacuum fluctuations and radiation reaction are hermitian . therefore , adopting the symmetric ordering prescription , one can identify the contribution of the vacuum fluctuations to the above rate : ( ) _ vf & = & - _ a=1 ^ 2 \{*e*^f(x_a(_a()))+ [ _ a(_a()),a()]^f(x_a(_a ( ) ) ) } , [ vf ] where the subscript @xmath57 stands for the vacuum - fluctuation contribution , and the radiation reaction terms read ( ) _ rr & = & - _ a=1 ^ 2 \{*e*^s(x_a(_a()))+ [ _ a(_a()),a()]^s(x_a(_a ( ) ) ) } , [ rr ] with the subscript @xmath58 labeling the radiation - reaction effect terms . in the expressions above we have considered only the part due to the interactions with the field . this method is the most interesting instruction which allows the possibility of studying the interplay between vacuum fluctuations and radiation reaction in quantum entanglement . hence one can identify the contributions of vacuum fluctuations and radiation reaction in the evolution of the atoms energies , which are given by the expectation value of @xmath27 , given by equation ( [ ha ] ) . the free part of the atomic hamiltonian is constant in time so that the rate of change of @xmath27 consists only of the two contributions given by , respectively ( ) _ vf & = & - _ a=1 ^ 2 \{*e*^f(x_a(_a()))+ [ _ a(_a()),h_a()]^f(x_a(_a ( ) ) ) } , [ vfha ] and ( ) _ rr & = & - _ a=1 ^ 2 \{*e*^s(x_a(_a()))+ [ _ a(_a()),h_a()]^s(x_a(_a()))}. [ rrha ] in a perturbative treatment , we take into account only terms up to order @xmath59 . we also consider an averaging over the field degrees of freedom by taking vacuum expectation values . since they contain only free operators , only the electric field is affected . in turn , we are interested in the evolution of expectation values of atomic observables . accordingly , we also take the expectation value of the above expressions in a state @xmath60 with energy @xmath61 , which can be one of the states given by equation ( [ sta ] ) or one can also consider the bell states @xmath62 . more details concerning such calculations can be found elsewhere , see for instance @xcite . therefore , with the notation @xmath63 , one has the pivotal results concerning the vacuum - fluctuation contribution _ vf = _ _ 0^d _ a , b = 1 ^ 2d_ij(x_a(_a()),x_b(_b()))^ij_ab(_a(),_b( ) ) , [ vfha3 ] where ^ij_ab(_a(),_b( ) ) = | [ ^i , f_a(_a()),^j , f_b(_b())]| , a , b = 1,2 , [ susa ] is the linear susceptibility of the two - atom system in the state @xmath60 and @xmath64 is the hadamard s elementary function . since we are dealing with free fields , one can employ the results derived in the appendix [ a ] . on the other hand , for the radiation - reaction contribution , one has : _ rr = _ _ 0^d _ a , b=1 ^ 2_ij(x_a(_a()),x_b(_b( ) ) ) d^ij_ab(_a(),_b( ) ) , [ rrha3 ] where d^ij_ab(_a(),_b( ) ) = | \{^i , f_a(_a()),^j , f_b(_b())}| , a , b = 1,2 , [ cora ] is the symmetric correlation function of the two - atom system in the state @xmath60 and @xmath65| 0 \rangle,\,\,\,a , b = 1,2\ ] ] is the pauli - jordan function which , for the same reason as above , can be evaluated from appendix [ a ] . observe that we have employed an index notation in order to represent the dot product of two vectors @xmath66 , where the einstein summation convention is to be understood in the above equations and henceforth ( unless otherwise stated ) . therefore the vector character of the electric field is manifest in the appearance of indices in the correlation functions . note that @xmath67 and @xmath68 characterize only the two - atom system itself , unlike the statistical functions of the field which have to be evaluated along the trajectory of the atoms . we see from equations ( [ vfha3 ] ) and ( [ rrha3 ] ) that the rate of variation of the energy of the two - atom system presents contributions from the isolated atoms and also contributions due to cross correlations between the atoms mediated by the field . this interference is a consequence of the interaction of each atom with the field . such observations make clear that such a formalism enables one to discuss the generation of entanglement between atoms initially prepared , say , in the ground state as well as the stability of entangled states . using that @xmath69 @xmath21 , and the completeness relation for the atomic product states ( or the bell states ) : @xmath70 one can show that the statistical functions for the two - atom system can be put in the following explicit forms ^ij_ab(, ) & = & _ , [ susa1 ] and d^ij_ab(, ) & = & _ , [ cora1 ] where @xmath71 , @xmath72 being the energy associated with the state @xmath73 . if the initial state @xmath60 is one of the bell entangled states , then the sum over @xmath72 is carried out with respect to the product states given by eq . ( [ sta ] ) . on the other hand , if the initial state @xmath60 is one of the states of eq . ( [ sta ] ) , then such a sum is over the bell states of eq . ( [ bell ] ) . in the above the generalized atomic transition dipole moment @xmath74 is defined by ^ij_cd(, ) = |^i , f_c(0)| |^j , f_d(0)| . [ aa ] finally , observe that from ( [ propt ] ) one can easily perform a change of variables in equations ( [ vfha3 ] ) and ( [ rrha3 ] ) in order to describe the time evolution in terms of one of the proper times of the atoms . in this paper we wish to investigate the creation of entangled states [ currently represented by the bell states ( [ bell ] ) ] from atoms initially prepared in a separable state [ given by equation ( [ sta ] ) ] , and also the decay of the bell states to one of the product states . in order to study the degradation of entanglement between the atoms as a spontaneous emission phenomenon , assume that the atoms were initially prepared in one of the bell states @xmath62 . with this regard , equations ( [ sta ] ) and ( [ bell ] ) state that the only allowed transitions are @xmath75 , with @xmath76 and @xmath77 , with @xmath78 . on the other hand , suppose one wishes to address the generation of entanglement . assume that the atoms were initially prepared in the atomic ground state @xmath79 . consider the excitation rate to one of the bell states @xmath62 . hence @xmath80 . similarly , for atoms initially prepared in the state @xmath81 , only the transition @xmath82 is permitted , with @xmath76 . incidentally , we remark that , if we consider the excitation rate to one of the bell states @xmath26 , one would get @xmath83 , which implies that @xmath84 . hence it is not possible to generate such bell states out of the ground - state atoms . in the next section we will discuss in detail the rate of variation of atomic energy for atoms in uniformly accelerated motion . we consider our two - atom system in a situation where both atoms move along different hyperbolic trajectories in the @xmath4 plane . as explained , these correspond to different @xmath11 coordinates , which we call @xmath85 and @xmath86 and each @xmath11 is a constant . their world lines are parametrized by equations ( [ acc ] ) . for simplicity , we consider the case in which both atoms have the same spatial coordinates which are orthogonal to the direction of the acceleration . we meticulously evaluate the expressions concerning the contributions of vacuum fluctuations and radiation reaction to the rate of variation of atomic energy in appendix [ b ] . here we only present the main results . concerning the contributions from vacuum fluctuations , one has : & & _ vf = -__c , d = 1 ^ 2e^- a(_c + _ d)^2(_cd ) + & & \{j^_cd ( , ) + ( ) _ ij + & - & ( -)||_ij } , [ vacuum2 ] where @xmath87 . we have defined the following functions : ^_cd ( , ) & = & - _ a^du ( e^iu/_d + e^-iu/_c ) [ rev28 ] and ^ij_n , cd ( , ) & = & f^ij_n , cd ( , ; , ) + & = & nu^ij_cd(, ; , ) + v^ij_cd ( , ; , ) + [ rev29 ] where we also have introduced the following definitions : u^ij_cd ( , ) & = & u^ij_cd(, ; , ) + & = & a^ij_cd(, ) u_i , cd ( , ) [ rev30 ] with no implicit summation on repeated indices , and v^ij_cd ( , ) & = & v^ij_cd ( , ; , ) + & = & _ 0 ^ 2du e^-u/ + & & . [ rev31 ] the function @xmath88 is given in the appendix [ a ] whilst the functions @xmath89 and @xmath90 are properly defined in the appendix [ b ] , see equations ( [ b2 ] ) , ( [ d1 ] ) and ( [ d23 ] ) , respectively . now let us present the radiation - reaction contributions . one gets , with @xmath87 : _ rr & = & -__c , d = 1 ^ 2e^- a(_c + _ d)^2(_cd ) + & & \ { ( ) + & + & ( -)||}. [ radiation2 ] the prior consideration to be taken regarding the above results is that the effects of electromagnetic vacuum fluctuations over the rate of variation of energy of an accelerated atom are not purely thermal : there exist nonthermal corrections proportional to the acceleration of the atoms . a similar result is found in reference @xcite , where the authors study a multilevel hydrogen atom in interaction with the quantum electromagnetic field . with respect to the radiation - reaction effect , we also find a striking dependence on the acceleration . again this situation agrees with the result found in reference @xcite . for completeness we present the total rate of change of the energy of the two - atom system in a general situation . this is obtained by adding the contributions of vacuum fluctuations and radiation reaction given by equations ( [ vacuum2 ] ) and ( [ radiation2 ] ) , respectively . we assume sufficiently large observation time intervals @xmath91 in such a way that the term @xmath92 can be neglected . one has : _ tot & = & -__c , d = 1 ^ 2e^- a(_c + _ d)^2(_cd ) + & & _ ij , with @xmath93 , and _ tot & = & _ _c , d = 1 ^ 2e^- a(_c + _ d)^2(_cd ) || + & & _ ij , with @xmath94 . notice that in the ground state of the two - atom system there is not a balance between vacuum fluctuations and radiation reaction which prevents spontaneous excitations to higher levels . in our context this means that entanglement between the atoms can be generated via absorption processes and it persists even for asymptotic observation times . observe also that the cross correlations introduce strong temporal oscillations . in order to carefully discuss the thermal contributions , let us consider the following situation . imagine a family of particle detectors at rest in different points in rindler space - time . even though each detector measures different temperatures @xmath95 , one says that all of them are in thermal equilibrium due to the tolman relation @xmath96 @xcite . we find a similar structure here , namely rates possessing thermal contributions containing each of the temperatures felt by the atoms . by invoking the thermalization theorem @xcite , one concludes that in the situation with sufficiently high relative accelerations we have two atoms coupled individually to two spatially separated cavities at different temperatures . hence our results clearly indicates that such a situation bears a remarkable resemblance with the initial set up of ref . we infer that if a detailed investigation of a circumstance with high relative accelerations is carried out within a master equation approach , one should expect to obtain similar results as uncovered by the authors of such a reference . namely , that complete disentanglement can be achieved after finite time processes , yet local decoherence operations requires an infinite time to be accomplished . despite its importance , this interesting analysis is out of the scope of the present work . on the other hand , for a careful recent discussion on entanglement in a quantum system at finite temperature we refer the reader the reference @xcite . now let us discuss in detail the distinct contributions to the rate of variation of atomic energy for a specific transition . here we analyze the distinct contributions to the rate of variation of atomic energy associated with the transition @xmath97 . this means that we are investigating the creation of entanglement via absorption processes . hence in this context @xmath98 in equations ( [ vacuum2 ] ) and ( [ radiation2 ] ) . for the sake of the argument , we consider the generation of the symmetric bell state @xmath99 but it is possible to carry out a similar analysis for the antisymmetric bell state @xmath100 . we assume sufficiently large observation time intervals @xmath91 in such a way that the term @xmath92 can be neglected . it is convenient to write out the expressions in a more explicit way . concerning the vacuum - fluctuation contribution , one has & & _ vf = _ c=1 ^ 2e^a_c ( + 1)(1 + ) + & & + ^2(_12)_ij \{^i,12(_0,_1 , ) ^j(1 + ) + ^i,21(_0,_2 , ) ^j(1 + ) + & & + ^j } , whereas the radiation - reaction terms are given by _ rr = -_c=1 ^ 2(+1)e^a_c -^2(_12)_ij . in the above expressions we have defined : _ 1,cd(,_c , ) = _ 1\{_cd - } [ dd1 ] _ k , cd(,_c , ) & = & + & & \ { _ cd + _ cd ( ( ) ^2 ^2(_cd ) - 1 ) } , + [ dd23 ] @xmath101 , and _ i,12 ( , ) = _ 0 ^ 2du e^-u/ , [ ddef - v ] with no summation over the repeated index @xmath17 in this last expression . the functions @xmath102 are defined in the appendix [ a ] , with the property that @xmath103 . also , we have used the fact that the matrix elements are given by ^ij_11(g , ^ ) & = & + a^ij_22(g , ^ ) & = & + a^ij_12(g , ^ ) & = & [ a_21(g , ^)]^ * = . [ mel ] and @xmath86 . we choose @xmath104 , @xmath105 , @xmath106 and @xmath107 . we remark that all physical quantities are given in terms of the natural units associated with each transition . therefore , in this case @xmath85 , @xmath86 , @xmath10 and @xmath52 are measured in units of @xmath30 , @xmath7 are in units of @xmath108 and @xmath109 is given in units of @xmath110 , where @xmath111 . moreover , @xmath112 is measured in units of @xmath113 . for clarity we have multiplied each value of the rate by @xmath114.,width=332,height=302 ] . we consider the atoms with the same proper acceleration . we choose @xmath105 , @xmath107 and @xmath115 in natural units associated with the transition in study ( see caption below figure [ 1]).,width=302,height=302 ] figure [ 1 ] illustrates the behavior of each of the contributions to @xmath112 as functions of the rindler coordinates @xmath85 and @xmath86 and with isotropic polarizations @xmath116 . as discussed , each fixed @xmath11 represents the world line of an uniformly accelerated observer ; lines of large positive @xmath11 are associated with weakly accelerated observers , while the hyperbolae that have large negative @xmath11 have a high proper acceleration . in this context , we see from figure [ 1 ] that for large negative @xmath11 s the contribution of vacuum fluctuations and radiation reaction present significantly different behaviors . in this limit the former acquires large positive values whereas the latter does not change appreciably , within the considered values in the plot , as compared with vacuum fluctuations . this implies that the generation of entanglement between highly accelerated atoms is much more sensitive to vacuum fluctuations as compared with the radiation - reaction effect . for atoms with lower proper accelerations , both effects approximately yield contributions of the same order . for clarity one may consider the situation in which both atoms travel along the same hyperbolic trajectory @xmath117 . in this case , they have the same proper acceleration @xmath118 . in this limit @xmath119 , see eq . ( [ psi ] ) . expressions ( [ dd1 ] ) and ( [ dd23 ] ) reduce to : ^2(_12)_k,12 ( , , ) _ k with @xmath120 , @xmath121 in this limit and also @xmath122 . this situation allows one to easily express the rate in terms of the proper time of the atoms by employing relation ( [ propt ] ) . hence the expressions of vacuum fluctuations and radiation reaction simplify to , respectively _ vf = ( + 1 ) ( 1 + ) , and _ rr = - ( + 1 ) . let us investigate the behavior of each of the contributions to @xmath123 as functions of the parameter @xmath7 . this is depicted in figure [ 2 ] . we see from such a plot that for high values of @xmath7 ( i.e. , high proper accelerations ) the differences between vacuum - fluctuation and radiation - reaction contributions become increasingly clear . the consequence of these considerations is that in principle one could plainly distinguish the contributions of vacuum fluctuations and radiation reaction for the formation of entangled atoms only for highly accelerated atoms . incidentally we call attention for the fact that for the transition @xmath124 , the case with equal proper accelerations offers an interesting result : for both contributions of vacuum fluctuations and radiation reaction the cross correlations cancel the terms associated with isolated atoms . hence @xmath125 for this transition . the bell state @xmath100 can not be created from ground - state atoms with the same proper acceleration . we remark that for the case of the transitions @xmath75 and @xmath77 , i.e. degradation of entanglement , the results are qualitatively the same as those discussed above , including the fact that @xmath125 for the antisymmetric state in the situation of equal accelerations : the bell state @xmath100 is stable with respect to radiative processes . thus we recover a very known result in a complete different scenario , namely that atoms confined into a region much smaller than the optical wavelength , the antisymmetric bell state @xmath100 can be regarded as a decoherence - free state @xcite . furthermore , the results presented here should be compared with the outcomes of ref . @xcite , even though they are working with a scalar field . in this case in which both atoms have the same proper acceleration we expect to find similar qualitative conclusions as those of such a reference . there have been many works investigating quantum information theory within the framework of quantum field theory and quantum field theory in curved space - time . several of such studies were implemented in a framework of open quantum systems . following the discussion by dalibard , dupont - roc , and cohen - tannoudji , we proposed to analyze the distinct contributions of vacuum fluctuations and radiation reaction to the quantum entanglement between two identical atoms in uniformly accelerated motion . we assume that both atoms are coupled to a quantum electromagnetic field . we have shown that the contributions of vacuum fluctuations and radiation reaction to the rate of variation of atomic anergy substantially differ for highly accelerated atoms . in addition , the creation of entanglement is possible for atoms initially prepared in the ground state . it is possible to envisage our results as the situation of two atoms coupled individually to two spatially separated cavities at different temperatures . for equal accelerations , we have found that the antisymmetric bell state @xmath100 acts as a decoherence - free state a natural extension of this paper is to consider the same setup studied here in the presence of boundaries . since the presence of boundaries affect the vacuum fluctuations of the quantum field one should expect that the transition rates of atoms are modified in this situation . hence , for entangled states , which have a non - local nature , it is interesting to ask how the transition rates are modified in a situation where translational invariance is broken . in turn , a related scenario is the investigation of uniformly accelerated qubits interacting with a massless scalar field via monopole interaction . similar to the case studied here , we expect the radiation - reaction terms to possess some acceleration dependence as well as the appearance of non thermal contributions ( which also depend on the acceleration ) to the vacuum fluctuations of the field . however , we suspect that the emergence of such terms are connected solely with the presence of cross correlations between the atoms mediated by the scalar field , in contrast with the situation with the electromagnetic field . another possible direction is to consider a black - hole space - time . one may argue that , close to event horizon the schwarzschild metric takes the form of the rindler line element . therefore , in a certain sense the results discussed here can be extended to schwarzschild black holes assuming that the atoms are close to the event horizon . in spite of this , the topic is more subtle than it would appear at first . although entanglement is degraded near a schwarzschild black hole as predicted in @xcite , there are important discrepancies arising due to the difference between an event horizon in schwarzschild space - time and an acceleration horizon in rindler space - time . such subjects are under investigation and results will be published elsewhere . this paper was supported by conselho nacional de desenvolvimento cientifico e tecnolgico do brasil ( cnpq ) . in this appendix we shall briefly discuss free electromagnetic correlation functions . we consider a four dimensional unbounded minkowski space - time . in addition , as previously mentioned we will work in the coulomb gauge . here we will not discuss the quantization of the electromagnetic field in detail . for a thorough analysis of quantum electrodynamics in the coulomb gauge , we refer the reader the reference @xcite . in turn , details concerning the correlations of the electromagnetic field for the case of uniformly accelerated observers can be found in the comprehensive account by takagi @xcite and related references therein . instead , we limit ourselves to briefly discuss the evaluation of all relevant electromagnetic correlation functions in the vacuum employed in the text . we start our discussion by presenting the hadamard s elementary function associated with the electric field . it is given by d^ij(x , x ) & = & 0 |\{e^i(x ) , e^j(x)}| 0= ( ^ij - ) d^(1)(t - t,*x * - * x* ) , + & = & -(^ij - ) d^(1 ) ( , ) , where d^(1)(t - t,*x * - * x* ) = -\ { + } . after calculating the derivatives one gets d^ij(x , x ) & = & -\ { + & + & } . [ hada ] the pauli - jordan function of the electric field is given by ^ij(x , x ) & = & 0 |[e^i(x ) , e^j(x)]| 0= -i(^ij - ) d(t - t,*x * - * x* ) + & = & i(^ij - ) d ( , ) , where @xmath126 , @xmath127 and d(t,*x * ) & = & ( t)(|*x*|^2 - t^2 ) = + & = & _ 0 . [ pj ] a detailed analysis of the function @xmath128 can be found in reference @xcite . by evaluating the derivatives one obtains an explicit form for the pauli - jordan function in terms of the delta functions and their derivatives : ^ij(x , x ) & = & \ { ( - ^ij ) + & - & ( - ^ij ) } , [ pauli ] where @xmath129 and @xmath130 . the above derivatives of the delta function are to be understood as a distributional derivative . one can also express the pauli - jordan function based on the derivation of the hadamard s elementary function : ^ij(x , x ) & = & -\ { + & - & } . [ pauli - e ] now let us present the commutators associated with the magnetic field . from standard calculations , it is easy to show that @xmath131 and therefore 0 |[b^i(x ) , b^j(x)]| 0= 0 |[e^i(x ) , e^j(x)]| 0,0 correlation function of both fields is given by 0 |e^i(x ) b^j(x)| 0= -i_m^ijm d^+ ( , ) , where @xmath132 is the levi - civita tensor density and d^+(x , x ) = e^i [ * k*(*x*-*x*)-_*k*(t - t ) ] = - . therefore 0 |e^i(x ) b^j(x)| 0= _ m^ijm . [ eb ] hence , the symmetric correlation function is given by 0 |\{e^i(x ) , b^j(x)}| 0= _ m^ijm \ { + } , [ eb - sym ] and the commutator reads 0 |[e^i(x ) , b^j(x)]| 0&= & _ m^ijm \ { - } + & = & _ m^ijm , [ eb - comm ] where the last line follows from the limit @xmath133 . the statistical functions of the fields have to be evaluated along the trajectory of the atoms . from the results above one may evaluate the hadamard s elementary functions found in equation ( [ vfha3 ] ) . for the case in which the atoms are uniformly accelerated , the discussion is involved . as a basis of our analysis , we consider the following system of reference frames : the inertial laboratory frame @xmath134 with minkowski coordinates @xmath135 , a set of instantaneous inertial frames @xmath136 defined at each value of the proper time of atom 1 and another set of instantaneous inertial frames @xmath137 defined at each value of the proper time of atom 2 . the latter employ rindler coordinates @xmath138 . in this way , the transformations from the inertial frame @xmath134 to one of the frames @xmath139 , @xmath34 , for an arbitrary tensor field @xmath140 are given by standard lorentz transformations : t^ _ ( ) = ^_()^_()_^()_^()t^_(x ( ) ) , [ tensor ] where = ( cccc ( ) & - ( ) ( ) & 0 & 0 + - ( ) ( ) & ( ) & 0 & 0 + 0 & 0 & 1 & 0 + 0 & 0 & 0 & 1 ) , [ lambda ] where @xmath141 and @xmath142 . remember that @xmath10 and @xmath13 are related by expression ( [ propt ] ) . for instance , for the electromagnetic field strength tensor @xmath143 one has f^ ( ) = f^(x())^_()^ _ ( ) , [ tensor1 ] therefore one can obtain the associated transformation for the correlation function of the electromagnetic field strength tensor as follows 0 |f^ ( ) f^()| 0= 0 relations @xmath144 , and @xmath145 , and the results derived above one can obtain the correlation functions for the electromagnetic fields @xmath54 and @xmath146 . then it is straightforward to obtain the associated transformations for the hadamard s elementary functions and also for the pauli - jordan functions . for simplicity , we consider the case in which both atoms have the same spatial coordinates which are orthogonal to the direction of the acceleration . since @xmath147 and @xmath148 from equation ( [ hada ] ) and employing equation ( [ acc ] ) as well as the above coordinate transformations one has the following explicit form for the hadamard s elementary function ( @xmath149,@xmath150 ) appearing in equation ( [ vfha3 ] ) : d_ij[x_c(),x_d( ) ] & = & \ { + } , + [ hada - acc2 ] where there is no implicit summation over repeated indices and we have defined _ 1,cd(a(- ) , ) & = & ( a(- ) - ia)- _ cd + _ 2,cd(a(- ) , ) & = & _ 3,cd^ ( ) = ( a(- ) - ia)_cd - 1 with @xmath151 defined through the following equation : ( * x*_c , , _ c,*x*_d , , _ d ) = 1 + . [ psi ] in order to derive such expressions use has been made of known hyperbolic identities @xcite . in addition , we have absorbed a positive function of @xmath152 as well as a positive function of @xmath153 into @xmath154 . incidentally , we mention that in order to reach the above wightman functions we have redefined the regulator @xmath154 . for a finite time interaction between the atoms and the field , this leads to a non - trivial change in the interpretation of @xmath154 . for more details on this subject , see ref . available discussion of the renormalization of the epsilon - regularization and alternatives to it can also be found in refs . @xcite . on the other hand , from the above transformations one can compute the pauli - jordan function for this case which are present in equation ( [ rrha3 ] ) : _ ij[x_c(),x_d( ) ] & = & \ { -}. + [ pauli - acc2 ] ) and ( [ energy - rad ] ) ( @xmath155 ) . exceptionally , this figure was developed with the jaxodraw java program . ] in this appendix we calculate in detail the contributions of vacuum fluctuations and radiation reaction to the rate of variation of energy of the two - atom system . from the results derived in the appendix [ a ] , one may compute all the relevant correlation functions of the electric field which appears in equations ( [ vfha3 ] ) and ( [ rrha3 ] ) . namely , such contributions to spontaneous emission are computed by inserting in such expressions the statistical functions of the two atom system , given by equations ( [ susa1 ] ) and ( [ cora1 ] ) , and the electromagnetic - field statistical functions given by ( [ hada - acc2 ] ) and ( [ pauli - acc2 ] ) . initially let us present the contributions coming from the vacuum fluctuations . performing a simple change of variable @xmath156 , these can be expressed as , with @xmath157 : _ vf & = & -__c , d = 1 ^ 2 e^- a(_c + _ d ) + & & _ 0^adu ( e^iu/_d + e^-iu/_c ) , [ energy - flu ] where @xmath7 was absorbed into @xmath154 and we have defined t^ij_cd(, ; u , ) = a^ij_cd(, ) _ i , cd(u , ) , [ b2 ] with no implicit summation on repeated indices in equation ( [ b2 ] ) . the quantities @xmath158 and @xmath151 are defined in appendix [ a ] whereas the generalized atomic transition dipole moment @xmath74 is defined in equation ( [ aa ] ) . all integrals can be evaluated within the techniques of contour integration by noting that ( @xmath159 ) @xmath160\ , = \int_{0}^{z}\,dr\,[\cdots ] + \int_{z}^{\infty}\,dr\,[\cdots].\ ] ] the strategy goes as follows . consider the integrals : @xmath161 and @xmath162 let us take a look at @xmath163 . we use the contours in the complex @xmath164-plane illustrated in figure [ 3 ] . for @xmath165 , the poles located inside the contour considered are given by @xmath166 , where @xmath167 is an integer and @xmath168 . the associated contour integral can be splitted up as @xmath169 = \int_{0}^{r}du\,[\cdots ] + \int_{c_r}du\,[\cdots ] + \int_{ir}^{0}du\,[\cdots ] = 2\pi i\ , \sum_{n = 0}^{\infty}\textrm{res}(u_n),\ ] ] where @xmath170 is the quarter of the circle depicted in plot @xmath171 of fig . [ 3 ] . on the other hand , for @xmath94 the poles inside the contour are given by @xmath172 , where @xmath173 . similarly , such a contour can be splitted up as @xmath174 = \int_{0}^{r}du\,[\cdots ] + \int_{c_r'}du\,[\cdots ] + \int_{-ir}^{0}du\,[\cdots ] = -2\pi i\,\sum_{n = 1}^{\infty}\textrm{res}(u_n),\ ] ] where @xmath175 is the quarter of the circle depicted in plot @xmath176 of fig . the integrals along @xmath170 and @xmath175 vanish for @xmath155 in virtue of the jordan s lemma . similar considerations can be settled to @xmath177 . it is clear that same arguments apply for the integral associated with @xmath178 . hence , after some algebra one gets , for @xmath87 and @xmath133 & & _ vf = -__c , d = 1 ^ 2e^- a(_c + _ d)^2(_cd ) + & & \{j^_cd ( , ) + ( ) + & - & ( -)|| } , [ vacuum ] where we have defined ^_cd ( , ) = - _ a^du ( e^iu/_d + e^-iu/_c ) and also u^ij_cd ( , ) = a^ij_cd(,)u_i , cd ( , ) [ def - u ] v^ij_cd ( , ) = _ 0 ^ 2du e^-u/ , [ def - v ] with no implicit summation on repeated indices in equation ( [ def - u ] ) , and _ 1,cd ( , ) = -()[d1 ] _ 2,cd ( , ) = u_3,cd ( , ) = ( ) . [ d23 ] in the above equations we used the fact that @xmath179 . now let us evaluate the radiation - reaction contributions . as above we perform a simple change of variable @xmath156 . rr & & = -__c , d = 1 ^ 2e^- a(_c + _ d ) + & & _ 0^adu ( e^iu/_d - e^-iu/_c ) . [ energy - rad ] using the results derived above , one has , with @xmath87 and @xmath133 : _ rr & = & -__c , d = 1 ^ 2e^- a(_c + _ d)^2(_cd ) + & & \ { ( ) + & + & ( -)||}. [ radiation ] 99 , d. bonwnerster , a. ekert and a. zeilingue ( eds . ) ( springer verlag , berlin , 2000 ) . j. m. raimond , m. brune and s. haroche , rev . mod . phys . * 73 * , 565 ( 2001 ) . m. b. plenio , s. f. huelga , a. beige and p. l. knight , phys . a * 59 * , 2468 ( 1999 ) . a. m. bachanov , jetp . lett . * 75 * , 123 ( 2002 ) . z. ficek and r. tans , j. mod . op . * 50 * , 2765 ( 2003 ) . r. tans and z. ficek , j. opt . b * 5 * , 90 ( 2004 ) . l. amico , r. fazio , a. osterloh and v. vedral , rev . phys . * 80 * , 517 ( 2008 ) . t. yu and j. h. eberly , phys . * 93 * , 140404 ( 2004 ) . e. arias , j. g. dueas , g. menezes and n. f. svaiter , _ boundary effects on radiative processes of two entangled atoms _ [ arxiv:1510.00047 ] . ackerhalt , p.l . knight and j.h . eberly , phys . lett . * 30 * , 456 ( 1973 ) . g. menezes and n. f. svaiter , phys . a * 92 * , 062131 ( 2015 ) . j. dalibard , j. dupont - roc , and c. cohen - tannoudji , j. phys . ( paris ) * 43 * , 1617 ( 1982 ) . j. dalibard , j. dupont - roc , and c. cohen - tannoudji , j. phys . ( paris ) * 45 * , 637 ( 1984 ) . p. w. milonni , phys . rep . * 25 * , 1 ( 1976 ) . j. audretsch and r. mller , phys . rev . a * 50 * , 1755 ( 1994 ) . j. audretsch and r. mller , class . * 12 * , 2927 ( 1995 ) . r. passante , phys . a * 57 * , 1590 ( 1998 ) . z. zhu , h. yu and s. lu , phys . d * 73 * , 107501 ( 2006 ) . z. zhu and h. yu , phys . rev . a * 82 * , 042108 ( 2010 ) . l. rizzuto and s. spagnolo , phys . * t143 * 014021 ( 2011 ) . d. w. sciama , p. candelas e d. deutsch , adv . phys * 30 * , 327 ( 1981 ) . i. fuentes - schuller and r. b. mann , phys . * 95 * , 120404 ( 2005 ) . n. d. birrell e p. c. w. davis , _ quantum fields in curved space _ ( cambridge university press , new york , 1982 ) . b. f. svaiter and n. f. svaiter , phys . d * 46 * , 5267 ( 1992 ) ; phys . d * 47 * , 4802 ( 1993 ) . l. c. b. crispino , a. higuchi and g. e. a. matsas , rev . mod . phys . * 80 * , 787 ( 2008 ) . g. ver steeg and n. c. menicucci , phys . d * 79 * , 044027 ( 2009 ) . e. martn - martnez , l. j. garay and j. len , phys . d * 82 * , 064006 ( 2010 ) . d. k. park and e. jung , phys . a * 83 * , 012111 ( 2011 ) . hwang , e. jung and d. park , class . quantum grav . * 29 * , 224004 ( 2012 ) . b. reznik , found . phys . * 33 * , 167 ( 2003 ) . b. reznik , a. retzker , and j. silman , phys . a * 71 * , 042104 ( 2005 ) . d. braun , phys . a * 72 * , 062324 ( 2005 ) . s. massar and p. spindel , phys . d * 74 * , 085031 ( 2006 ) . j. d. franson , j. mod . opt . * 55 * , 2117 ( 2008 ) . y . lin and b. l. hu , phys . d * 79 * , 085020 ( 2009 ) . s .- y . lin and b. l. hu , phys . d * 81 * , 045019 ( 2010 ) . f. benatti and r. floreanini , phys . a * 70 * , 012112 ( 2004 ) . f. benatti and r. floreanini , j. opt . b : quantum semiclass . opt . * 7 * , s429 ( 2005 ) . j. zhang and h. yu , phys . d * 75 * , 104014 ( 2007 ) . j. hu and h. yu , jhep * 08 * , 137 ( 2011 ) . m. alsing and g. j. milburn , phys . * 91 * , 180404 ( 2003 ) p. m. alsing , d. mcmahon and g. j. milburn , j. opt . b : quantum semiclass . opt . * 6 * , s834 ( 2004 ) . k. shiokawa , e - print arxiv:0910.1715 [ gr - qc ] . m. czachor , phys . rev . a * 55 * , 72 ( 1997 ) . d. r. terno , phys . a * 67 * , 014102 ( 2003 ) . h. terashima and m. ueda , int . j. quantum inf . * 1 * , 93 ( 2003 ) . w. t. kim and e. j. son , phys . a * 71 * , 014102 ( 2005 ) . caban , j. rembielinski and m. wlodarczyk , phys . a * 79 * , 014102 ( 2009 ) . s. j. summers and r. werner , phys . a * 110 * , 257 ( 1985 ) . m. alsing , i. fuentes - schuller , r. b. mann and t. e. tessier , phys . a * 74 * , 032326 ( 2006 ) . a. bermudez and m. a. martin - delgado , j. phys . a * 41 * , 485302 ( 2008 ) . a. peres , p. f. scudo and d. r. terno , phys . lett . * 88 * , 230402 ( 2002 ) . r. m. gingrich and c. adami , phys . lett . * 89 * , 270402 ( 2002 ) . r. m. gingrich , a. j. bergou and c. adami , phys . a * 68 * , 042102 ( 2003 ) . y. shi , phys . d * 70 * , 105001 ( 2004 ) . m. czachor , phys . lett . * 94 * , 078901 ( 2005 ) . t. f. jordan , a. shaji and e. c. g. sudarshan , phys . a * 75 * , 022101 ( 2007 ) . g. adesso , i. fuentes - schuller and m. ericsson , phys . a * 76 * , 062112 ( 2007 ) . a. datta , phys . a * 80 * , 052304 ( 2009 ) . a. peres and d. r. terno , rev . phys . * 76 * , 93 ( 2004 ) . m. alsing and i. fuentes , class . quantum grav . * 29 * , 224001 ( 2012 ) . j. hu and h. yu , phys . rev . a * 91 * , 012327 ( 2015 ) . c. cohen - tannoudji , j. dupont - roc and g. grynberg , _ photons and atoms : introduction to quantum electrodynamics _ ( wiley - vch , 1997 ) . s. takagi , prog . phys . suppl . * 88 * , 1 ( 1986 ) . r. c. tolman and p. ehrenfest , phys . * 36 * , 1791 ( 1930 ) . s. fulling , phys . d * 7 * , 2850 ( 1973 ) . p. c. w. davies , j. phys . * 8 * , 609 ( 1975 ) . hsiang and b. l. hu , phys . b * 750 * , 396 ( 2015 ) . z. ficek e r. tans , phys . rep . * 372 * , 369 ( 2002 ) . m. abramowitz and i. a. stegun , _ handbook of mathematical functions with formulas , graphs , and mathematical tables _ ( dover , new york , 1972 ) . p. langlois , _ imprints of spacetime topology in the hawking - unruh effect _ , phd thesis ( university of nottingham , 2005 ) [ arxiv : gr - qc/0510127 ] . e. martn - martnez , a. r. h. smith and d. r. terno , phys . d * 93 * , 044001 ( 2016 ) .

we present analytic and numerical results which illustrate the effects of jupiter s accretion of nebular gas and the planet s radial migration on its trojan companions . initially , we approximate the system by the planar circular restricted three - body problem and assume small trojan libration amplitudes . employing an adiabatic invariant calculation , we show that jupiter s thirty - fold growth from a core to its present mass causes the libration amplitudes of trojan asteroids to shrink by a factor of about 2.5 to of their original size . the calculation also shows that jupiter s radial migration has comparatively little effect on the trojans ; inward migration from 6.2 to 5.2 au causes an increase in trojan libration amplitudes of . in each case , the area enclosed by small tadpole orbits , if made dimensionless by using jupiter s semimajor axis , is approximately conserved . similar adiabatic invariant calculations for inclined and eccentric trojans show that jupiter s mass growth leaves the asteroid s eccentricities and inclinations essentially unchanged , while one au of inward migration causes an increase in both of these quantities by . numerical integrations confirm and extend these analytic results . we demonstrate that our predictions remain valid for trojans with small libration amplitudes even when the asteroids have low , but nonzero , eccentricities and inclinations and/or jupiter has an eccentricity similar to its present value . the integrations also show that trojans with large libration amplitudes , including horseshoe orbits , are even more strongly affected by jupiter s mass growth and radial migration than simple scaling from our analytic results would suggest . further , the numerical runs demonstrate that jupiter s predicted mass growth is sufficient to cause the capture of asteroids initially on horseshoe orbits into stable tadpole orbits . thus , if jupiter captured most of its trojan companions before or while it accreted gas , as seems probable , then jupiter s growth played a significant role in stabilizing trojan objects by systematically driving them to lower libration amplitudes . key words : + asteroids , dynamics + celestial mechanics + jupiter + origin , solar system + resonances + astronomy department , university of maryland + heather j. fleming + astronomy department + university of maryland + college park , md 20742 - 2421 + tel : ( 301 ) 474 - 7582 + fax : ( 301 ) 314 - 9067 + e - mail : [email protected] + douglas p. hamilton + astronomy department + university of maryland + college park , md 20742 - 2421 + tel : ( 301 ) 405 - 1548 + fax : ( 301 ) 314 - 9067 + e - mail : [email protected] + submitted to _ icarus _ : june 16 , 1999 + revised : + 41 pages + 13 figures and 2 tables + running head : origin of the trojan asteroids + corresponding author : + douglas p. hamilton + astronomy department + university of maryland + college park , md 20742 - 2421 + tel : ( 301 ) 405 - 1548 + fax : ( 301)314 - 9067 + e - mail : [email protected] +

astro-ph0007296

the trojans are a distant group of asteroids dynamically linked to jupiter by a 1:1 mean motion resonance which causes them to librate about stable lagrangian equilibrium points located @xmath4 in front of ( l4 ) and behind ( l5 ) jupiter along its orbit . because trojans orbit far from the sun and also have low albedos , they suffer from a low discovery rate and hence are underrepresented among numbered asteroids . extrapolating from the identified trojan objects , shoemaker _ et al . _ ( 1989 ) estimate that the total trojan population contains nearly half as many asteroids as the main belt . the large number of trojans and their strong dynamical connection to jupiter make determining their origins an important goal ; by understanding the early history of the trojans , we may also gain insight into jupiter s formation and early evolution . clues about the origins of the trojan asteroids may be found in their current physical and orbital properties , which include the overlapping signatures of mechanisms which captured them into librating orbits , as well as processes which have contributed to the population s evolution over time . some distinctive characteristics of the current trojan asteroids include a small mean eccentricity of @xmath5 , a small mean libration amplitude of @xmath6 , and a large mean inclination of @xmath7 ( @xcite , @xcite ) . also , nearly twice as many asteroids have been observed librating about the l4 point as about the l5 point . this , however , may simply be the result of observational selection effects ( @xcite ) . there are many competing theories for the origin and evolution of the trojan asteroids . it has been suggested that the trojans may have originally been comets ( @xcite ) or near - jupiter planetesimals ( _ e.g. _ @xcite , @xcite ) . a number of mechanisms have been considered for capturing these objects into trojan orbits , including collisions between objects , drag forces , and mass accretion by jupiter . et al . _ ( 1989 ) theorized that collisional emplacement of fragments of near - jupiter planetesimals during the dispersion of the planetesimal swarm may have provided most of the trojan objects . more recently , numerical modeling of the collisional evolution of the trojan population ( marzari _ et al . _ 1997 ) has shown that collisions are largely responsible for shaping the current size distribution of the smaller trojans , as well as having caused the escape of some of the trojan objects into chaotic orbits . long - term numerical integrations by levison _ et al . _ ( 1997 ) have also shown that the trojan population is dynamically unstable and that dynamical diffusion has contributed and continues to contribute to the loss of trojan objects . various types of drag forces acting on jupiter and/or the trojan precursors have also been examined . kary and lissauer ( 1995 ) considered solar nebular gas drag . they showed numerically that gas drag could cause planetesimals to be captured into 1:1 resonance with a protoplanet . interestingly , they found that such capture is rare for planets on circular orbits , but quite common for planets with appreciable eccentricities . gas drag may also have played a significant role in evolving the trojan population into its present form , provided that the trojan precursors were captured before the dispersion of the solar nebula ( peale 1993 ) . yoder ( 1979 ) looked at the effects of dynamical friction during jupiter s dispersal of the planetesimal swarm , which caused a slight inward migration of jupiter . he argued that this migration would cause a decrease in the libration amplitudes of jupiter s trojan companions ( see , however , table i ) . the possibility that a change in jupiter s mass could be responsible for the capture of the trojan asteroids was investigated by rabe ( 1954 ) who argued analytically that a decrease in jupiter s mass could cause its satellites to move onto trojan orbits . more recently , marzari and scholl ( 1998 ) showed numerically that an increase in jupiter s mass could cause the capture of planetesimals into librating orbits . they used a proto - jupiter on a `` best guess '' orbit growing simultaneously with saturn over a period of @xmath8 or @xmath9 years , and found that a large fraction of the planetesimals initially on horseshoe orbits and a small percentage of those initially orbiting near the 1:1 resonance were captured into tadpole orbits . past changes in jupiter s mass have also long been considered as a potentially significant evolutionary mechanism for creating the current distribution of trojan asteroids ; however , attempts to predict the exact form of the effects of jupiter s growth on the trojans have thus far been contradictory . for an increase in jupiter s mass , rabe ( 1954 ) theorized that the trojan libration amplitudes would increase , horedt ( 1974a , 1974b , 1984 ) argued that they would not be appreciably affected , and yoder ( 1979 ) argued that they would decrease ( table i ) . in this paper , we focus on the changes which jupiter underwent early in its history . we investigate the significance of jupiter s mass growth and radial migration as trojan capture mechanisms and especially as mechanisms for evolving the trojan population . we focus on these two processes in isolation in order to fully characterize their behavior . other potentially important processes including gas drag , the gravity of other planets , and collisions amongst trojans are not considered here , because models which include all of these effects would have a large number of poorly - determined free parameters . for example , with gas drag , what is the gas density as a function of distance from the sun ? when and exactly how does jupiter form a gap in the gas distribution ? how sensitive are trojan asteroids to different gas drag models ? these questions are important and need to be studied in depth . there are many open questions like these in the full trojan formation problem , probably more than can be addressed in a single paper . accordingly , we have chosen to study individual processes first for later incorporation into a more general model . a strong advantage to this approach is that investigating individual processes in detail will ultimately lead to a deeper physical understanding of results from more complicated models . adopting this approach , we first present consistent analytic and numerical results which show the effects of jupiter s growth and radial migration on its trojan companions in the limit of a slowly - changing jupiter , thus settling the previous controversy . we then explore the working of these mechanisms for a wide range of timescales , initial trojan libration amplitudes , jupiter eccentricities , and asteroid eccentricities and inclinations . consider the planar circular restricted three - body problem , in which two massive bodies move about each other in circular orbits due to their mutual gravitation and a third body of infinitesimal mass moves in the orbital plane of the two massive objects . this system admits five equilibrium points where a test particle can have zero velocity and zero acceleration in the frame which corotates with the primary masses about their common center of mass ( danby 1988 ) . three of these points lie along the line through the two primaries and are unstable . the other two , l4 and l5 , lie at the tips of the equilateral triangles whose bases are the line connecting the primary masses ( see fig . 1 ) . these are called the triangular lagrangian equilibrium points and are stable to small oscillations so long as the mass ratio of the primaries , @xmath10 ( where @xmath11 and @xmath12 are the larger and smaller of the primary masses , respectively ) , satisfies @xmath13 ( @xcite ) . this condition is met for all sun - planet and planet - moon pairs in the solar system , with the exception of pluto and charon . the planar circular restricted three - body problem is a reasonable approximation for the system consisting of the sun , jupiter , and a trojan asteroid , since the asteroid s mass is insignificant in comparison to either jupiter or the sun , and jupiter s eccentricity is relatively small ( @xmath14 currently ) . if we make the further approximations that the trojan s oscillations about its equilibrium position are small and that the asteroid is on a nearly circular orbit , then , noting that @xmath15 where @xmath16 is the mass of the sun and @xmath17 is the mass of jupiter , the motion of the asteroid is well approximated by the equation @xmath18 ( @xcite ) , where @xmath19 is the difference between the mean longitudes of the asteroid and jupiter , @xmath20^{1/2 } \label{nj}\ ] ] is the mean motion of jupiter ( @xmath21 is jupiter s semimajor axis and @xmath22 is the gravitational constant ) , and an overdot signifies differentiation with respect to time . the observed librational motions of the trojan asteroids are well approximated by the solution of eq . [ tmotion ] , @xmath23 where @xmath24 @xmath25 is time , and @xmath26 and @xmath27 are constants . note that @xmath28 is the amplitude of the @xmath19 oscillations . the libration amplitude , @xmath26 , is defined to be the total angular extent of these oscillations ( see fig . 1 ) . equivalently , the system can be described by the hamiltonian : @xmath29 the canonical variables for this hamiltonian are @xmath30 and @xmath31 , so that @xmath32 , and @xmath33 reproduces the equation of motion ( eq . [ tmotion ] ) . when changes are made to this system ( _ e.g. _ mass growth of one of the primaries or the addition of an external drag force ) , the hamiltonian is no longer conserved . however , if the changes to the hamiltonian system are slow enough , it can be shown that the action , @xmath34 , is approximately conserved ( @xcite ) . such slow changes are called adiabatic changes , and @xmath35 is an adiabatic invariant . using the expressions for @xmath36 , @xmath37 , and @xmath19 above , we determine the action for the three - body system : @xmath38 where we have evaluated the integral by using the adiabatic approximation that @xmath21 , @xmath26 , and @xmath39 are constant over one libration period . the conservation of the action can be written in the useful form : @xmath40 where the subscripts @xmath41 and @xmath42 indicate the initial and final values of the variables , respectively . for radial migration of jupiter , we can find the resulting change in the trojan s libration amplitude directly from eq . [ analyticresult ] ; since changing @xmath21 has no effect on @xmath17 , the factor @xmath43 is equal to 1 . then , for a physically reasonable inward radial migration of jupiter from approximately 6.2 au to 5.2 au ( see section 4 ) , eq . [ analyticresult ] predicts an increase in the trojan s libration amplitude of only @xmath2 . determining the effects of jupiter s mass growth on the trojan s libration amplitude is a bit more subtle , since increasing @xmath17 may affect @xmath21 as well as @xmath26 . in order to determine how changing @xmath17 affects @xmath21 , we digress briefly to discuss a second adiabatic invariant calculation . if we consider the two - body system of jupiter and the sun moving about one another on elliptical orbits due to their mutual gravitation , we can calculate an invariant action for slow changes in the mass of either body . noting that , in this case , adiabatic means that changes to the system are negligible over one orbital period , we find @xmath44 this result was contemplated nearly one hundred years ago when strmgen ( 1903 ) considered what effect mass accretion by the earth would have on the orbit of the moon ( see also @xcite ) . for our case , the invariance of @xmath45 means that if the mass of either jupiter or the sun were slowly decreased , jupiter s orbit would drift outward . this effect is well known ( _ e.g. _ horedt 1984 ) and was observed recently in numerical simulations by duncan and lissauer ( 1998 ) in which the orbits of the outer planets were seen to expand as the mass of the sun was decreased to a small fraction of its original value . similarly , if either jupiter or the sun slowly accretes mass , jupiter s semimajor axis will decrease . note that eq . [ 2bconstant ] implies that adding a jovian mass of material to either jupiter or the sun produces the same change in the semimajor axis of the system . returning to our discussion of the effects of jupiter s mass growth on its trojan companions , we see from eq . [ 2bconstant ] that if we change the mass of jupiter adiabatically , the semimajor axis of jupiter s orbit will be altered according to @xmath46 substituting this into eq . [ analyticresult ] , we see the full effect which altering @xmath17 has on the trojan s libration amplitude : @xmath47 since @xmath48 , the second term in parentheses is very nearly equal to 1 , and the change in @xmath26 is given sufficiently accurately by eq . [ analyticresult ] if we simply take @xmath49 . we find that the growth of jupiter by gas accretion from a @xmath50 core to its present mass of @xmath51 causes @xmath26 to decrease to @xmath52 of its original value . thus , if librating trojan asteroids were already present when jupiter was a @xmath0 core , then the effects of jupiter s mass increase would be substantial and would dominate over the effects of its radial migration . the effect of jupiter s radial migration is comparable to its mass accretion only if the planet moves inward by a factor of @xmath53 , _ e.g. _ from 150 au to 5 au , substantially more than current theories predict . there has been some significant confusion in the literature about the effects of jupiter s mass growth on the libration amplitudes of its trojan companions ( see table i ) . our analytic result for the effects of jovian mass growth disagrees with the findings of rabe ( 1954 ) and horedt ( 1974a , 1974b , 1984 ) , but agrees exactly , in both direction and magnitude , with the conclusions of yoder ( 1979 ) . our prediction for the effects of jupiter s radial migration , however , disagrees with yoder s 1979 result ( see table i ) . statements in yoder _ et al . _ ( 1983 ) about the tidal evolution of the saturnian satellites janus and epimetheus , however , are inconsistent with yoder s 1979 calculations , but agree , at least in sign , with our radial migration results . in order to dispel this confusion , we carefully verify our analytic predictions with numerical simulations in sections 3 and 4 below . we can make further use of the two - body adiabatic invariant for the sun - jupiter system ( eq . [ 2bconstant ] ) to gain additional insight into the meaning of the three - body adiabatic invariant ( eq . [ taction ] ) . when @xmath26 and @xmath54 are both very small , the trojan asteroid orbit , viewed in the frame which corotates with jupiter , looks like a little ellipse centered on one of the lagrangian equilibrium points with the ratio of its semimajor to semiminor axis equal to @xmath55 ( murray and dermott 1999 ) . the area within this ellipse is @xmath56 for a small tadpole orbit , the libration amplitude times the semimajor axis of jupiter s orbit is approximately equal to the major axis of the ellipse ( @xmath57 ) . thus we can rewrite eq . [ area1 ] as @xmath58 where we have used the two- and three - body adiabatic invariants defined in eqs . [ 2bconstant ] and [ taction ] , respectively . so , we find : @xmath59 thus , the invariance of the action for the three - body system implies that the dimensionless area enclosed by a small - amplitude trojan orbit ( _ i.e. , _ the area enclosed by the trojan orbit divided by the square of jupiter s semimajor axis ) , remains constant for an adiabatic change to the system . for example , as jupiter accretes mass , the resulting change in @xmath21 is relatively small , so as @xmath26 decreases , the radial width of the trojan orbit must increase in order to keep the area nearly constant ( see fig . 1 ) . note , however , that the area within the tadpole orbit in fig . 1 is not strictly conserved because the libration amplitude of the tadpole is quite large . finally , we can use the two - body adiabatic invariant for the sun - jupiter system ( eq . [ 2bconstant ] ) to determine what effect changing the mass of the sun has on the libration amplitudes of jupiter trojans . the dependence of @xmath26 on @xmath16 is present in eq . [ analyticresult ] through the factor @xmath60 . we know from @xmath45 ( eq . [ 2bconstant ] ) that if the mass of the sun is altered adiabatically , @xmath21 will change according to @xmath61 combining this with eq . [ analyticresult ] , we find @xmath62 for an adiabatic change in @xmath16 . thus , slowly increasing the mass of the sun increases the libration amplitude of the trojan , which is the opposite of the effect on @xmath26 caused by adding mass to jupiter . additional adiabatic invariant calculations determine the effects of changing jupiter s mass and semimajor axis on a trojan asteroid s eccentricity , @xmath63 , and inclination , @xmath64 . for the case when @xmath65 , we approximate the orbit of the trojan as an inclined circle . then , the motion of the asteroid in the @xmath66-direction can be well represented by simple harmonic motion with a restoring force equal to the sum of the @xmath66-components of the gravitational forces of the sun and jupiter acting on the asteroid . if we assume that the asteroid s inclination and libration amplitude are small , the distances between jupiter , the asteroid , and the sun are all approximately equal to @xmath21 . then , the restoring force per unit mass acting on the asteroid is @xmath67 where @xmath66 is the distance of the trojan asteroid above the plane of jupiter s orbit . notice that the vertical oscillation frequency is the same as jupiter s orbital frequency ( eq . [ nj ] ) . the hamiltonian for the vertical motion is @xmath68 which has canonical variables @xmath69 and @xmath70 . as in section 2.1 , for an adiabatic change to the system , the action is conserved . noting that here adiabatic means that changes to the system are negligible over one orbital period , we find @xmath71 where we have used the relation @xmath72 and integrated over a full orbital period . from this result , we see that if an external force adiabatically changes the semimajor axis of jupiter s orbit about the sun , the inclination of the trojan will change according to the relation @xmath73 if instead the mass of jupiter or the sun is varied , eq . [ inclconst ] tells us that some combination of @xmath64 and @xmath21 must change to keep the action constant . we determine how the variation is shared between @xmath64 and @xmath21 by using the invariant action for the two - body sun - jupiter system , @xmath45 ( eq . [ 2bconstant ] ) . dividing eq . [ inclconst ] by eq . [ 2bconstant ] , we find @xmath74 for an adiabatic change in either @xmath17 or @xmath16 . the adiabatic calculation for an eccentric asteroid orbit parallels that for an inclined orbit , however the radial oscillation frequency is given by @xmath75^{1/2 } \label{rfreq}\ ] ] ( murray and dermott 1999 , page 94 ) rather than by eq . [ nj ] . accordingly , the action , evaluated for small eccentricities has the form @xmath76 and @xmath63 will respond to an adiabatic change in @xmath21 according to the relation @xmath77 this result is in agreement with analytic work by gomes ( 1997 ) who showed that @xmath63 would increase when @xmath21 was decreased . for variations in @xmath16 and/or @xmath17 , we combine eqs . [ 2bconstant ] and [ jecc ] , and find @xmath78 thus an increase in jupiter s mass should lead to a slight increase in a trojan asteroid s eccentricity . in summary we find , using the invariance of the action for adiabatic changes to the sun - jupiter and sun - jupiter - trojan systems , that if the semimajor axis of jupiter is decreased , the asteroid s libration amplitude ( eq . [ analyticresult ] ) , eccentricity ( eq . [ eccaresult ] ) , and inclination ( eq . [ inclaresult ] ) will all increase . we find further that if we increase jupiter s mass , @xmath21 will decrease ( eq . [ amj ] ) and the trojan s libration amplitude will decrease ( eq . [ libampmj ] ) , its eccentricity will increase slightly ( eq . [ eccmresult ] ) , and its inclination will remain unchanged ( eq . [ inclmresult ] ) . finally , if the sun s mass is increased , @xmath21 will be decreased ( eq . [ ams ] ) and the asteroid will be dragged inward with jupiter , @xmath26 will increase by the same amount it would were @xmath21 altered by an external force ( eq . [ libampms ] ) , @xmath63 will decrease slightly ( eq . [ eccmresult ] ) and @xmath64 will not change ( eq . [ inclmresult ] ) . these results are also summarized in table ii . in this section and the next , we confirm the analytic results above and explore their range of validity by numerically integrating the three - body system consisting of the sun , jupiter , and a massless asteroid . the full equations of motion , consisting of the gravitational force of each body acting on the other two , are integrated in inertial coordinates . we use bulrisch - stoer and runge - kutta methods with adaptive stepsize ( @xcite ) , having initially decided against faster symplectic methods ( @xcite ) because of the difficulty in handling close approaches with jupiter . although fast symplectic methods for handling close approaches do exist ( @xcite ) , the speed of our routines was adequate . we ran extensive tests on our code , including checking that the bulrisch - stoer and runge - kutta integrators converged to the same solutions , producing plots which matched specific trojan asteroid orbits illustrated in murray and dermott ( 1999 , pages 97 and 98 ) , and verifying that the jacobi constant was conserved to sufficient accuracy for the circular restricted three - body problem with constant jupiter mass and semimajor axis . we begin our numerical exploration by integrating the three - body system as jupiter grows from @xmath79 to its current mass . we experimented with growing jupiter both exponentially ( @xmath80 ) and linearly ( @xmath81 ) , where @xmath82 is the initial mass of jupiter and @xmath83 and @xmath84 are constants . we find , as expected from section 2 , that our results were not significantly affected by the form of mass growth so long as the growth was slow enough to be adiabatic ; accordingly all results presented below are for an exponentially - growing jupiter . additionally , orbits librating around the lagrangian equilibrium point l4 behaved similarly to orbits around l5 , as suggested by our analytic analysis . thus , the results presented below apply to objects librating about either of these points . placing the sun , jupiter , and the asteroid all on initially circular coplanar orbits , we carry out a set of integrations in which jupiter grows on timescales ranging from @xmath85 to @xmath9 years . we monitor the changes to the asteroid s libration amplitude and plot our results in fig . 2 . for long mass growth timescales ( @xmath86 years ) , the numerical asteroid orbits agree well with the analytic prediction . initially , the numerically - determined points track the analytic prediction precisely , but as jupiter s mass grows the points begin to scatter more . the increase in amplitude of the oscillation of the numerical points about the analytically - predicted line is due to our method of calculating the libration amplitude . we use an analytic approximation ( yoder 1983 , shoemaker _ 1989 ) which makes the following assumptions : 1 ) a planar three - body system with all the objects on circular orbits , 2 ) @xmath87 , and 3 ) @xmath26 is very small . the jupiter - trojan system is reasonably approximated by these assumptions ; however , as @xmath88 grows , the error in the approximation increases , thereby causing an increased scatter of the calculated points . furthermore , the changing mass of jupiter itself leads to additional effects which are not accounted for in the simple theory ; these effects are larger for more rapid growth timescales . runs with slower growth timescales of @xmath89 years ( not shown ) exhibit the same behavior as the @xmath8 year runs shown here , since both of these slow growth rates represent changes to the system that are well within the adiabatic limit where eq . [ analyticresult ] is valid . for the faster jupiter growth rates ( @xmath85 and @xmath90 years ) , the numerical curves deviate substantially from the analytic prediction of eq . [ analyticresult ] ( fig . 2 ) . in these cases , significant changes in the mass of jupiter occur on timescales comparable to the libration period of the trojan asteroid , @xmath91 ( with @xmath39 given by eq . [ omega ] ) , which is @xmath92 years for @xmath93 and @xmath94 years for jupiter s current mass . thus , the change in the trojan orbit depends on its initial conditions , _ i.e. _ , exactly where along the tadpole orbit the asteroid starts . the initial conditions for the @xmath90-year timescale run shown in fig . 2 are chosen so that the curve exhibits maximum deviation from the analytic result . note that the oscillations in this curve are not due to inaccuracies in the analytic approximation used to calculate the libration amplitude ; they are real effects due the asteroid s librational motion and occur at the libration frequency . the dependence of short timescale runs on initial conditions is clearly illustrated by the curves for two extreme choices of asteroid starting point for the @xmath85-year timescale plotted in fig . 2 . the curve which is everywhere above the analytic line corresponds to an asteroid which was started at the point along the tadpole orbit farthest away from jupiter ( maximum @xmath19 , see fig . 1 ) . the basic characteristics of this curve can be understood using a simple harmonic oscillator analogy : we treat the librational motion of the asteroid as a one - dimensional oscillation in the @xmath19 direction , and the rapid growth of jupiter in @xmath85 years is approximated as instantaneous growth . using this analogy , growing jupiter when the asteroid starts at the farthest point from jupiter is equivalent to increasing the restoring force of the harmonic oscillator instantly when the oscillator is at its maximum extension . this affects the period of the oscillations , but leaves the amplitude unchanged . for this reason , the upper @xmath85-year curve in fig . 2 is initially horizontal , indicating no change in the libration amplitude . since increasing jupiter s mass over @xmath85 years is not truly an instantaneous change , the libration amplitude of the trojan asteroid eventually decreases as it begins to move toward jupiter . the @xmath85-year timescale curve in fig . 2 which is everywhere below the analytic line corresponds to an asteroid started at the point on the tadpole orbit farthest from the sun ( see fig . 1 ) where @xmath95 is maximum . again using the harmonic oscillator analogy , this is equivalent to increasing the restoring force when the oscillator has its maximum velocity . this causes the amplitude of the oscillations to decrease . thus , we see an initially rapid decrease in the trojan s libration amplitude which slows as the asteroid moves closer to jupiter . in the adiabatic limit , the effect of the mass change is averaged over all points along the asteroid s orbit . thus , the decrease in the asteroid s libration amplitude for adiabatic growth lies between the two extremes just discussed , as can be seen in fig . 2 . next , we explore the role of the asteroid s initial libration size in determining how its libration amplitude will change as jupiter s mass grows . recall that the analytic prediction ( eq . [ analyticresult ] ) was derived with the assumption of small libration amplitude . we study a set of tadpoles with different initial libration amplitudes ; in each integration the timescale for jupiter s mass growth is @xmath9 years , well within the adiabatic limit . we find that for initial libration amplitudes @xmath96 , our numerical results agree well with the analytic prediction , as expected ( fig . 3 ) . the curve for the @xmath97 tadpole shows a slight deviation from the analytic result ; tadpoles with larger initial amplitudes show even greater departures . as in fig . 2 , the spread of the numerical points is due to the analytic method used to determine the libration amplitude . the larger tadpole orbits shrink more rapidly than eq . [ analyticresult ] predicts . furthermore , the larger the libration amplitude is , the faster it shrinks , as can be seen by examining the initial slopes of the curves for the @xmath98 , @xmath99 , and @xmath100 tadpoles in fig . this can be understood by considering the effective potential in the corotating frame ( fig . 1 ) . the effective potential changes much more steeply at the head of the tadpole near jupiter than at the tail which lies further away ( erdi 1997 ) . the larger the tadpole orbit is , the further it extends away from jupiter , and the less steep is the potential in which the tail end of the orbit lies . when the potential is flatter , the location of the turning point of the orbit changes by a greater amount for the same change in energy , and so the orbits whose tails lie in the shallowest potential , _ i.e. _ those with the largest libration amplitude , shrink the fastest ( see fig . once the initially large tadpole orbits become small enough , they shrink at the rate predicted by eq . [ analyticresult ] . this is illustrated in fig . 3 where , by the end of the integrations , all the curves are tending to the same slope as the analytic result . we also examine the effects of jupiter s mass growth on very large `` horseshoe orbits '' whose librations encompass both the l4 and l5 equilibrium points . figure 4 shows the results of an integration in which the asteroid has an initial libration amplitude of @xmath101 . the horseshoe orbit shrinks slowly until @xmath102 years when it transitions to an l4 tadpole . the tadpole then continues to shrink . although horseshoe orbits are well outside the range of libration amplitudes described by our adiabatic calculation , they too shrink under the influence of a growing jupiter . the rate of decrease in libration amplitude is , however , different for horseshoe orbits and tadpole orbits . looking at the lower edge of the plot in fig . 4 , we see that the turning point of the horseshoe orbit pulls away from jupiter at a nearly constant rate for exponential growth of jupiter s mass . furthermore , the corresponding turning point in the tadpole orbit ( the tip which lies nearest jupiter ) pulls away approximately linearly as well , but with a shallower slope . for a doubling of jupiter s mass , one tip of the horseshoe orbit recedes from jupiter by @xmath103 , while for the same mass change , the near - jupiter tip of the tadpole moves by only @xmath104 . in this way , horseshoe orbits shrink even faster than tadpole orbits . further integrations demonstrate that the growth of jupiter to its current mass captures most asteroids which were initially on horseshoe orbits into tadpole orbits . figure 5 shows the results of a set of integrations which determine the fate of asteroids placed into different - sized horseshoe orbits at various points during the growth of jupiter . we find that if the asteroids are placed on horseshoes when jupiter is a @xmath0 core , orbits with initial libration amplitudes as large as @xmath105 are captured into tadpole orbits by the time jupiter has reached its present mass . when the asteroids are started on horseshoes after jupiter has already grown part - way to its current mass , fewer of the orbits shrink sufficiently to transition to tadpoles ; however , even if the asteroids are started when jupiter has reached three - quarters of its final mass , some of the smaller horseshoe orbits still become tadpoles , at least temporarily , during the course of the integration . in fig . 5 , we can clearly see the chaotic nature of the evolution of the horseshoe orbits : there is mixing between the orbits which escape the system and those which remain in horseshoes for the entire time . we also observe mixing at the boundary for capture into tadpoles . small differences in the initial asteroid orbits can significantly alter the effects of jovian perturbations , vastly changing the final asteroid orbit . additionally , note that the smallest possible horseshoe orbits occur at @xmath106 , nearly independent of the mass of jupiter . this result is in agreement with the analytic work of horedt ( 1984 ) . the transition to a tadpole orbit occurs whenever a horseshoe orbit shrinks to this minimum size . data from the integrations plotted in fig . 5 confirms that the observation made from fig . 4 , that each tip of a horseshoe orbit recedes from jupiter at a roughly uniform rate of @xmath103 per jupiter doubling , holds for general horseshoe orbits . figure 6 shows the decrease in the libration amplitudes of the horseshoe orbits plotted against the number of doublings of jupiter s mass . the solid line , fit by eye to the data , has a slope of @xmath107 per doubling ( indicating that each of the two turning points recedes at @xmath108 per doubling ) . the slight flattening of the data points near ( 0,0 ) on the plot indicates that the horseshoe orbits shrink a bit more slowly when they are near the transition to tadpole orbits . we verified this observation with additional simulations not shown here . in order to get the best value for the slope in fig . 6 , we ignored the points in the flattened tail of the plot and did not insist that our fitted line go through ( 0,0 ) . next , we look at the final sizes of the tadpoles produced by the transitions from horseshoes shown in fig . 5 to see if these objects could , in fact , contribute to the current population of jupiter trojans . in fig . 7 , we see that a small number of the asteroids placed in horseshoe orbits when jupiter is @xmath109 of its final mass become tadpole orbits which are stable for a significant fraction of the age of the solar system . a substantially larger number of the horseshoe orbits started when jupiter is @xmath110 of its final mass become tadpoles that remain stable for at least a hundred million years . thus , assuming that objects resided on horseshoe orbits when jupiter was accreting mass , a small fraction of these objects may still survive in the trojan swarm today . furthermore , since these asteroids undergo relatively frequent collisions ( @xcite ) , they almost certainly produced some fragments which were ejected into more stable tadpole orbits . to further explore the range of validity of our analytic results from section 2 above , we add eccentricity and inclination to the trojan orbit , at first leaving jupiter on a circular orbit . we integrate the three - body system for a number of different small values of eccentricity and inclination ( @xmath111 , @xmath112 ) with the trojan on a small ( @xmath113 ) tadpole orbit and jupiter growing over @xmath9 years . we find that for these low values of eccentricity and inclination , our analytic expression for the change in libration amplitude ( eq . [ analyticresult ] ) still approximates the behavior of the system well . figure 8 shows the results of one of our numerical integrations . the asteroid is given an initial eccentricity of @xmath114 and an initial inclination of @xmath115 . the analytic method of yoder _ et al . _ ( 1983 ) for determining the libration amplitude , @xmath26 , fails for non - zero @xmath63 and @xmath64 , so we resort to a simpler but more time consuming method . the values of libration amplitude are determined from the numerical data by taking the average of the local maximum and minimum values of @xmath19 at the turning points ( @xmath116 ) of the tadpole orbit . the range in values of @xmath19 at the turning points is due to both the eccentricity and inclination of the orbit . the averaging process takes advantage of the difference in orbital and librational timescales , and is essentially the guiding centre approximation for the orbit ( murray and dermott 1999 ) . the remaining uncertainty in the numerical points is due primarily to the difficulty in determining the maximum and minimum @xmath19 values from a sample of discrete points . further , as the libration period becomes smaller , we sample only parts of the elliptical orbit during the turning point . despite these difficulties , we find that the numerical points in fig . 8 follow the analytic curve well . we also explore the effects of jupiter s mass growth on the eccentricity and inclination of trojan orbits . figure 9 shows the eccentricity and inclination of the trojan orbit whose libration amplitude is plotted in fig . we see that , to first order , the eccentricity and inclination are constant during jupiter s growth , as predicted in section 2.2 . a closer inspection reveals a slight increase in the mean eccentricity which is in good agreement with the analytical predictions of eq . [ eccmresult ] . finally , we add eccentricity to jupiter s orbit while keeping the trojan on an inclined and eccentric orbit . figure 10a shows the eccentricity and inclination of a trojan when jupiter has a constant mass and an eccentricity , @xmath117 . the asteroid s eccentricity oscillates around a constant `` forced '' component which is equal to the eccentricity of jupiter . the `` free '' component of the eccentricity is approximately the amplitude of the oscillations ( @xmath118 in fig . the trojan s forced inclination is zero , since the inclination is measured relative to jupiter s orbit . the high - frequency oscillations visible in the plot of @xmath64 are due primarily to the trojan s librations , while the lower - frequency oscillations in @xmath64 are correlated to the oscillations of the trojan s eccentricity . starting with the same initial orbits for both jupiter and the trojan , but allowing jupiter to accrete material over @xmath119 years , we obtain the results shown in fig . the values of the forced and free eccentricities and inclinations are not significantly altered by jupiter s growth ; however , the frequencies of the oscillations in both @xmath63 and @xmath64 increase as the mass of jupiter grows , and the strength of its gravitational perturbation increases . the slight drift in the mean trojan inclination seen in fig . 10b is simply part of a periodic oscillation in @xmath64 correlated to the drift of the orbital node . so , as predicted in section 2.2 , we find that the eccentricity and inclination of a trojan orbit remain essentially unchanged as jupiter s mass grows , even when jupiter is on an eccentric orbit . thus , the primordial eccentricities and inclinations of some large trojan objects may have been preserved during jupiter s accretion of mass . we now turn to a series of integrations of the three - body sun - jupiter - asteroid system in which jupiter undergoes 1 au of inward radial migration . for our integrations , we set jupiter s initial semimajor axis to @xmath120 au and leave its mass constant at @xmath50 . because the effects of jupiter s radial migration are independent of jupiter s mass and depend only on the ratio of the initial and final semimajor axes ( see eqs . [ analyticresult ] , [ inclaresult ] , and [ eccaresult ] ) our numerical results are applicable to migration of jupiter at any point during its history . the amount of radial migration that jupiter underwent as a rocky core and growing gas giant due to tidal interactions with the gas and planetesimal disks is poorly constrained , but might be several au ( @xcite ) . furthermore , after attaining its present mass , jupiter continued to experience radial migration due to scattering of planetesimals by the giant planets ( dynamical friction ) ; recent models for the formation of the oort cloud ( @xcite , @xcite ) predict several tenths of an au of radial migration at this stage . our results may be scaled to either or both of these scenarios . we artificially cause jupiter s orbit to shrink by applying a drag force of the form @xmath121 ( where @xmath122 is jupiter s heliocentric velocity and @xmath123 is the drag constant ) which acts only on jupiter . this form of drag affects jupiter s semimajor axis , but not its eccentricity , providing a `` simplest case '' for studying the effects of jupiter s radial migration on trojan objects . recall that our analytic work shows that , in the adiabatic limit , changes in the trojans libration amplitudes are independent of the exact form of the drag force ( see eq . [ analyticresult ] ) ; thus our choice for the form of the drag force is reasonable . as in the integrations for jupiter s mass growth , we find no apparent dependence on the choice of lagrangian equilibrium point , so the results below are equally applicable to orbits about the l4 and l5 points . we initially place the sun , jupiter , and the asteroid all on circular coplanar orbits . we carry out a set of integrations with the trojan on a small tadpole orbit using different drag coefficients , @xmath123 , to cause jupiter to migrate inward by @xmath124 au on timescales ranging from @xmath125 to @xmath89 years . for slow evolution , we observe that the trojans are always dragged inward with jupiter . as in the case of jupiter s mass growth ( section 3.1.1 ) , we find that for timescales significantly larger than the trojan libration period ( @xmath126 years for @xmath127 and @xmath128 au ) , the libration amplitude of the trojan increases in the manner predicted by eq . [ analyticresult ] . also , as in the mass accretion case , when the migration timescale approaches the libration period , the change in libration amplitude deviates from our analytic prediction , with the direction and amount of deviation depending on the asteroid s initial librational phase . for very fast migration rates , we find that the asteroid is not pulled inward with jupiter , but instead is ejected from its tadpole orbit . for migration of jupiter by @xmath124 au in @xmath90 years , one sample integration shows an initially small tadpole orbit transforming into a horseshoe orbit near the end of the integration . in another example , we change jupiter s semimajor axis in 500 years and find that the asteroid escapes entirely from the 1:1 resonance after only @xmath129 years . next , we do a set of integrations starting the trojan asteroid on different - sized tadpole orbits . in each run , we cause jupiter to move from @xmath120 to @xmath130 au over @xmath9 years so that its migration is adiabatic . the results of several of these integrations are shown in fig . 11 . it is difficult to produce a plot similar to fig . 3 because yoder s ( 1979 ) formula for calculating the libration period assumes that there are no drag forces . thus , we estimate the initial and final libration amplitudes by measuring the difference between the minimum and maximum values of @xmath19 for the first and last complete librations in the integration . we note that the trojans librate about shifted equilibrium points due to the effects of the drag force ( _ cf . _ murray 1994 ) . in fig . 11 we see that for small tadpoles ( @xmath131 ) the numerical results agree well with the prediction of eq . [ analyticresult ] . the steplike appearance of the first five points reflects difficulties inherent in our measurement technique . as the initial libration amplitude becomes larger , however , the numerical points deviate increasingly from the analytic result . this is expected since the larger tadpoles break the assumption of small libration amplitude which was made during the derivation of eq . [ analyticresult ] . as in the case of jupiter s mass growth , we see that for larger initial tadpoles , the change in libration amplitude is greater in magnitude but in the same direction as is predicted analytically . also , we see that the deviation of @xmath132 from the analytic prediction increases more steeply as the initial orbits become even larger . as was discussed in section 3.1.2 , this is caused by the shallow slope of the effective potential near the tail end ( the end farthest from jupiter ) of large tadpole orbits . next , we add eccentricity and inclination to the asteroid orbit , leaving jupiter on a circular orbit . as in the case of jupiter s mass growth , we find that small trojan eccentricities and inclinations ( @xmath133 , @xmath134 ) do not cause the behavior of the libration amplitude to deviate significantly from the analytic prediction of eq . [ analyticresult ] . however , unlike the mass growth case , jupiter s migration does affect the eccentricity and inclination of the trojan . figure 12 shows the semimajor axis , eccentricity , and inclination of a trojan asteroid on a small tadpole orbit as jupiter migrates from @xmath120 to @xmath130 au in @xmath9 years . during that time , the asteroid s inclination increases from @xmath135 to @xmath136 and its eccentricity grows from 0.00999 to 0.01043 , giving us @xmath137 and @xmath138 . these agree well with the analytic prediction that @xmath139 which we calculate using eqs . [ inclaresult ] and [ eccaresult ] . these changes in the asteroid s eccentricity and inclination resulting from @xmath124 au migration of jupiter are quite small , but are systematic . finally , we explore the general case of non - zero jovian eccentricity and a trojan on an eccentric and inclined orbit . figure 13 shows the semimajor axis , eccentricity , and inclination of a sample trojan asteroid as jupiter , with @xmath140 , migrates from @xmath120 to @xmath141 au in @xmath142 years . note that we allow jupiter to move by a much greater amount than in our earlier simulations so that the effects of migration can be more easily observed and quantified . as in section 3.3 , the asteroid has a forced eccentricity equal to jupiter s eccentricity and a forced inclination of zero . the asteroid in fig . 13 also has a free component of its eccentricity which is initially @xmath143 , and a free inclination which is initially @xmath144 . as jupiter migrates inward , the free components of both the asteroid s eccentricity and inclination increase systematically . from fig . 13 , we find @xmath145 and @xmath146 . both of these results agree well with the analytic prediction that @xmath147 ( eqs . [ inclaresult ] and [ eccaresult ] ) for an adiabatic change in jupiter s semimajor axis from 6.2 to 2.6 au . thus , our analytic results ( eqs . [ inclaresult ] and [ eccaresult ] ) hold even when jupiter has a low eccentricity if @xmath63 and @xmath64 are interpreted as the free components of the asteroid s eccentricity and inclination . we have shown with both a simple adiabatic calculation and numerical simulations , that slow changes to the mass and semimajor axis of jupiter cause the trojan libration amplitude to vary according to the relation : @xmath148 for trojans with small libration amplitudes , small eccentricities , and small inclinations , and jupiter with a small eccentricity . for inclined and eccentric trojan objects , we find that jupiter s mass growth does not significantly affect either the trojan s eccentricity or inclination ; however , jupiter s radial migration causes a change in the free component of both of these quantities by a factor of @xmath149 . applying our results to the core accretion model for the early evolution of jupiter , we find that the planet s growth by gas accretion from a @xmath150 core to its present mass would cause a decrease in the libration amplitude of any trojan companions on small tadpole orbits to @xmath1 of their original size . our representative choice for jupiter s radial migration from @xmath120 to @xmath130 au would result in an increase in the trojans libration amplitudes , eccentricities , and inclinations of only @xmath3 . even for radial migrations of several tens of au , the effects of jupiter s mass growth dominate over the effects of its migration . thus the combined result of mass accretion and radial migration is to stabilize trojan objects by systematically driving them to lower libration amplitudes . also , our numerical integrations show that the libration amplitudes of trojans on larger orbits shrink at an even faster rate . further , the shrinking of horseshoe orbits due to jupiter s growth could place additional objects or perhaps fragments of objects onto stable tadpole orbits . thus , jupiter s growth by mass accretion most likely played a significant role in the capture and evolution of the trojan asteroid population . our results for the evolution of trojan libration amplitudes , eccentricities , and inclinations are quite general and can be applied to other objects within the solar system . for example , eqs . [ analyticresult ] , [ inclaresult ] , and [ eccaresult ] predict that @xmath26 , @xmath151 , and @xmath41 will all decrease substantially if the secondary body undergoes significant outward radial migration . uranus and neptune probably underwent more substantial radial migration due to dynamical friction with planetesimals than jupiter , moving outward by as much as several au ( @xcite , @xcite ) . this would have caused a decrease in @xmath26 , @xmath63 , and @xmath64 of possible trojan - like companions by about @xmath152 . radial migration effects might be even more significant for some planetary satellites , notably our moon . the moon is believed to have formed via a giant collision which produced a temporary ring of debris around the earth ( @xcite and @xcite ) . such a process would have most likely captured some debris in librating orbits about the moon s lagrangian equilibrium points . over the subsequent @xmath153 years , the moon migrated outward to about 30 times its initial orbital radius . ignoring other effects , this migration should have decreased the libration amplitudes , eccentricities , and inclinations of the debris particles to @xmath1 of their original values , stabilizing these objects in 1:1 resonance with the moon . since we observe no such objects today , either the 1:1 resonance was never populated , or other effects ( such as solar gravity ) caused them to be unstable . another potential area for study is the satellite system of saturn . the many resonances in this system are believed to have formed during the significant outward migration of these satellites due to both tidal interactions with saturn and ring torques . one unexplained characteristic of the current saturnian system is that of the six largest satellites near saturn , the middle two , tethys and dione , have a total of three trojan companions but the others , mimas , enceladus , rhea , and titan , have none . this is curious because there is no obvious reason why the middle satellites should be the best at capturing trojan companions . also , the results of this paper suggest that the objects which migrate outward by the greatest amount should be the best at capturing and stabilizing their trojan companions . this suggests that the inner two satellites , which have migrated farthest , should be most likely to have trojan companions . it is likely that the probability of trojan capture and the stabilization of trojan orbits is complicated by the presence of resonances between the satellites . the inner four saturnian satellites are all currently locked in resonances with each other , and they may have passed through various other resonances in the past . an exploration of the interactions between the migration process explored in this paper , and the effects of other resonances , discussed by morais and murray ( 1999 ) , may provide insight into this unexplained characteristic of the saturnian system . it may also provide insight into how the unusual pair of coorbital satellites , janus and epimetheus , which librate on horseshoe orbits , was formed in the saturnian system . our work is also directly relevant to the capture of planetary satellites during the growth of the giant planets , as was suggested by heppenheimer and porco ( 1977 ) . equation [ 2bconstant ] clearly shows that distant satellites would be drawn inward as a planet grows . their migration would cease when the planet reached its final mass . thus , this mechanism provides a natural way for a giant planet to pull in distant satellites without causing them to collide with the planet . press , w. h. , b. p. flannery , s. a. teukolsky , and w. t. vettering 1987 . _ numerical recipes in c : the art of scientific computing _ , cambridge university press , cambridge . analytical prediction with increasing libration amplitude , since the rabe , e. 1954 . the trojans as escaped satellites of jupiter . _ astron . j. _ * 59 * , 433 - 439 . rabe , e. 1972 . orbital characteristics of comets passing through the 1:1 commensurability with jupiter . in _ motion , evolution of orbits , and origin of comets . proceedings , iau symposium no . 45 leningrad _ , ( e.i . chevotarev et al . , springer - verlag , new york / berlin , 55 - 60 . shoemaker , e. m. , c. s. shoemaker , and r. f. wolfe 1989 . trojan asteroids : population , dynamical structure , and origin of the l4 and l5 swarms . in _ asteroids ( r. p. binzel , t. gehrels , and m. s. matthews , eds . ) univ . of arizona press , tucson , 487 - 523 . three sample trojan asteroid orbits are plotted in the frame which corotates with jupiter about the center of mass of the sun - jupiter system . the l4 ( leading ) and l5 ( trailing ) lagrangian equilibrium points are each indicated by the symbol @xmath154 . the orbits shown , which enclose either the l4 or l5 point , but not both , are called tadpoles due to the shape of their librations . the tadpole around the l4 point was integrated with jupiter at twice its present mass , which makes it wider and thus easier to view . the libration amplitude , @xmath155 , is indicated for this orbit . the two tadpoles about the l5 point are the initial ( long , thin tadpole with @xmath156 ) and final ( short , fat tadpole with @xmath157 ) orbits for a trojan as jupiter s mass grows slowly from one - half to twice its current value . the tadpole both shortens and widens as jupiter s mass increases . the trojan libration amplitude , @xmath26 , ( normalized to its initial value , @xmath158 ) is plotted against jupiter s mass , @xmath17 , ( normalized to its initial value , @xmath159 ) on log - log scale for a series of integrations during which jupiter grows from @xmath150 to its current @xmath51 . the initial trojan orbits are small ( @xmath160 ) tadpoles . the analytic prediction of eq . [ analyticresult ] is plotted as a solid line . the numerical curve for the @xmath8 year growth rate ( dotted line ) agrees well with the analytic result . curves for longer growth timescales ( not shown ) agree equally well . the curves for shorter timescales , however , deviate significantly from the analytic curve in a manner which depends on the initial conditions of the orbit . note that the two curves for the @xmath85-year timescale differ only in the initial librational phase of the orbit . the trojan libration amplitude ( normalized to its initial value ) is plotted on a log - log scale for a variety of different - sized initial tadpole orbits as jupiter grows from @xmath150 to its present mass . our analytic result ( eq . [ analyticresult ] ) is plotted as a heavy solid line . the upper dotted curve , which overlays the theoretical curve , is for a tadpole with @xmath161 initially . all smaller tadpoles ( not shown ) agree even better with the analytic prediction . the lower curves in this figure are tadpoles with ( from top to bottom ) @xmath98 , @xmath99 , and @xmath100 initial libration amplitudes . the large tadpole orbits shrink faster than our analytic work predicts . the oscillations in the numerical curves are a result of the analytic approximation used to calculate the libration amplitude and occur at the libration frequency . here we show the behavior of @xmath19 , the longitude of the asteroid in the frame which corotates with jupiter , as jupiter grows from @xmath162 to its present mass in @xmath163 years . jupiter is at @xmath164 , and the initial asteroid orbit is an @xmath165 horseshoe . at @xmath166 years the orbit jumps to an l4 tadpole with @xmath167 . the trojan s libration period is initially about @xmath168 years , and decreases to @xmath169 years when the horseshoe transfers to a tadpole . afterwards , the tadpole s libration period continuously shortens in accordance with eq . [ omega ] . note that the tadpole orbit shrinks most rapidly when its tail ( _ i.e. _ its away - from - jupiter turning point ) is near @xmath170 , since the effective potential is flattest there . further , observe that the lower edge of the plot has a steeper slope for the horseshoe orbit than for the tadpole orbit ; the straight line overlaying the plot is fit by eye to the edge of the horseshoe orbit and has a slope of @xmath103 per jupiter mass doubling . this plot shows the final state of asteroids started on horseshoe orbits at different points during jupiter s growth to its current size of @xmath171 . the vertical axis shows the initial libration amplitude of the horseshoe orbit and the horizontal axis shows the mass of jupiter when the asteroid was placed in that orbit . asteroids which escaped from the 1:1 resonance during the @xmath9 years of the integration are indicated by open circles . objects which remained in horseshoe orbits for the entire integration are shown as filled triangles , and asteroids which were captured into tadpole orbits are shown as stars . most , but not all , of the orbits which became tadpoles remained tadpoles for the rest of the integration . the mixing of final states shown in this figure is an indication of the chaotic nature of the orbits . this plot shows the decrease in the libration amplitudes of those horseshoe orbits from fig . 5 which did not escape the 1:1 resonance plotted against the number of secondary mass doublings . a line with a slope of @xmath107 per doubling was fitted by eye to the data . the low scatter of the points about this line indicates that the decrease in @xmath26 by @xmath107 per doubling is a general property of horseshoe orbits , independent of the mass of the secondary . this plot shows the final libration amplitudes of the asteroids from fig . 5 which were on tadpole orbits at the end of the @xmath9-year integration . the horizontal axis is the mass of jupiter at the time when the asteroids were placed into their initial horseshoe orbits . the orbits represented by points below the long - dashed line are stable for more than @xmath172 years , while those below the short - dashed line are stable for greater than @xmath173 years . these timescales are estimated from the work of levison _ et al . _ ( 1997 ) for tadpoles with zero eccentricity . the square data points show our numerical measurements of the change in the libration amplitude of a trojan asteroid with @xmath174 , @xmath175 , and @xmath176 as jupiter ( on a circular orbit ) grows from @xmath150 to its current mass in @xmath9 years . representative error bars are shown for two data points . the solid line is the analytic prediction for @xmath177 ( eq . [ analyticresult ] ) . the numerical points agree with the analytic curve to within the error bars . this plot shows the eccentricity and inclination of the tadpole orbit whose libration amplitude is plotted in fig . the mean values of the asteroid s eccentricity and inclination are essentially unchanged ( note the vertical scale ) by the growth of jupiter s mass , although the tiny increase in the mean eccentricity is real . the mean eccentricity rises by approximately one small tick mark over @xmath9 years , which is consistent with the @xmath178% increase predicted by eq . [ eccmresult ] . the high frequency oscillations visible in the traces of both @xmath63 and @xmath64 are due to the trojan s librational motion and thus become more rapid as jupiter s mass increases , in accordance with eq . [ omega ] . the low frequency oscillations in the inclination are correlated to the precession of the asteroid s pericenter ( not shown ) driven by the disturbing effects of jupiter . since all the oscillations are caused by jovian perturbations , they increase in amplitude as jupiter s mass grows . these plots compare the orbital evolution of a trojan asteroid a ) without and b ) with jovian mass growth . for both cases , the trojan asteroids start with identical initial conditions and jupiter has an eccentricity of @xmath118 . in a ) , the mass of jupiter is kept constant at @xmath150 , while in b ) jupiter s mass grows from @xmath150 to its current mass in @xmath119 years . both @xmath63 plots show oscillations about a forced eccentricity equal to jupiter s eccentricity . the plots of @xmath64 show the free inclination ( since @xmath179 ) . the low frequency oscillations in @xmath64 are correlated to the oscillations in @xmath63 . the high frequency oscillations visible in the plots of @xmath64 are due to the trojan s librations . the free and forced eccentricities and inclinations of the trojan asteroid are essentially unchanged by jupiter s growth , as predicted in section 2.2 ; however , the frequencies of all the observed oscillations in @xmath63 and @xmath64 increase as jupiter s mass grows . this plot shows the ratio of final to initial libration amplitude for several different - sized tadpole orbits affected by the migration of jupiter from @xmath120 to @xmath130 au over @xmath9 years . jupiter is on a circular orbit . the results of numerical integrations are shown as solid dots with error bars which reflect the difficulties in measuring the libration amplitude . the horizontal line at @xmath180 represents the analytic prediction of eq . [ analyticresult ] which is valid only for small initial libration amplitudes . the numerical results agree well with the analytic prediction for @xmath181 , but deviate increasingly from the prediction for larger initial libration amplitudes . this plot shows the change in the semimajor axis , eccentricity , and inclination of a trojan asteroid on a slightly eccentric and inclined orbit as jupiter ( on a circular orbit ) migrates radially from @xmath182 to @xmath130 au in @xmath9 years . the trojan s eccentricity and inclination increase by a factor of @xmath183 as is predicted by eqs . [ inclaresult ] and [ eccaresult ] . note that the tiny oscillations in @xmath63 and @xmath64 are due primarily to the trojan s librational motion . this plot shows the semimajor axis , eccentricity , and inclination of a trojan asteroid on a slightly eccentric and inclined orbit when jupiter has an eccentricity of @xmath118 and migrates radially from @xmath120 to @xmath141 au in @xmath142 years . the forced components of both the asteroid s eccentricity and inclination remain constant ( @xmath184 and @xmath179 ) ; however , the free components of both @xmath63 and @xmath64 increase like @xmath185 , as predicted by eqs . [ inclaresult ] and [ eccaresult ] .

for a class of martingales , this paper provides a framework on the uniform consistency with broad applicability . the main condition imposed is only related to the conditional variance of the martingale , which holds true for stationary mixing time series , stationary iterated random function , harris recurrent markov chains and processes with innovations being a linear process . using the established results , this paper investigates the uniform convergence of the nadaraya watson estimator in a non - linear cointegrating regression model . our results not only provide sharp convergence rate , but also the optimal range for the uniform convergence to be held . this paper also considers the uniform upper and lower bound estimates for a functional of harris recurrent markov chain , which are of independent interests .

1402.0966

let @xmath1 with @xmath2 , be a sequence of random vectors . a common functional of interests @xmath3 of @xmath4 is defined by @xmath5,\qquad x\in r^d,\ ] ] where @xmath6 is a certain sequence of positive constants and @xmath7 is a real function on @xmath8 . such functionals arise in non - parametric estimation problems , where @xmath9 may be a kernel function @xmath10 or a squared kernel function @xmath11 and the sequence @xmath12 is the bandwidth used in the non - parametric regression . the uniform convergence of @xmath3 in the situation that the @xmath4 satisfy certain stationary conditions was studied in many articles . liero @xcite , peligrad @xcite and nze and doukhan @xcite considered the uniform convergence over a fixed compact set , while masry @xcite , bosq @xcite and fan and yao @xcite gave uniform results over an unbounded set . these work mainly focus on random sequence @xmath13 which satisfies different types of mixing conditions . investigating a more general framework , andrews @xcite gave result on kernel estimate when the data sequence is near - epoch dependent on another underlying mixing sequence . more recently , hansen @xcite provided a set of general uniform consistency results , allowing for stationary strong mixing multivariate data with infinite support , kernels with unbounded support and general bandwidth sequences . kristensen @xcite further extended hansen s results to the heterogenous dependent case under @xmath14-mixing condition . also see wu , huang and huang @xcite for kernel estimation in general time series settings . in comparison to the extensive results where the @xmath15 comes from a stationary time series data , there is little investigation on the the uniform convergence of @xmath3 for the @xmath15 being a non - stationary time series . in this regard , gao , li and tjstheim @xcite derived strong and weak consistency results for the case where the @xmath15 is a null - recurrent markov chain . wang and wang @xcite worked with partial sum processes of the type @xmath16 where @xmath17 is a general linear process . while the rate of convergence in gao , li and tjstheim @xcite is sharp , they impose the independence between @xmath18 and @xmath15 . using a quite different method , wang and wang @xcite allowed for the endogeneity between @xmath18 and @xmath15 , but their results hold only for the @xmath19 being in a fixed compact set . the aim of this paper is to present a general uniform consistency result for @xmath3 with broad applicability . as a framework , our assumption on the @xmath13 is only related to the conditional variance of the martingale , that is , @xmath20 $ ] . see assumption [ assumption2.3 ] in section [ sec2 ] . this of course is a `` high level '' condition , but it in fact is quite natural which holds true for many interesting and important examples , including stationary mixing time series , stationary iterated random function and harris recurrent markov chain . see sections [ sec22 ] and [ sec23 ] for the identification of assumption [ assumption2.3 ] . this condition also holds true for @xmath0 processes with innovations being a linear process , but the identification is complicated and requires quite different techniques . we will report related work in a separate paper . by using the established result , we investigate the uniform convergence of the nadaraya watson estimator in a non - linear cointegrating regression model . it confirms that the uniform asymptotics in wang and wang @xcite can be extended to a unbounded set and the independence between the @xmath21 and @xmath13 in gao , li and tjstheim @xcite can be removed . more importantly , our result not only provides sharp convergence rate , but also the optimal range for the uniform convergence to be held . it should be mentioned that our work on the uniform upper and lower bound estimation for a functional of harris recurrent markov chain is of independent interests . this paper is organized as follows . our main results are presented in next section , which includes the establishment of a framework on the uniform convergence for a class of martingale and uniform upper and lower bound estimation for a functional of harris recurrent markov chain . an application of the main results in non - linear cointegrating regression is given in section [ sec3 ] . all proofs are postponed to section [ sec4 ] . throughout the paper , we denote constants by @xmath22 which may be different at each appearance . we also use the notation @xmath23 . we make use of the following assumptions in the development of uniform convergence for the @xmath3 defined by ( [ in1 ] ) . recall @xmath24 where @xmath25 is an integer . [ assumption2.1 ] @xmath26 is a martingale difference , where @xmath27 , satisfying @xmath28 , a.s . , for some @xmath29 specified in assumption [ assumption2.4 ] below . [ assumption2.2 ] @xmath7 is a real function on @xmath8 satisfying @xmath30 and @xmath31 for all @xmath32 and some constant @xmath33 . [ assumption2.3 ] there exist positive constant sequences @xmath34 and @xmath35 with @xmath36 for some @xmath37 such that @xmath38 = { \mathrm{o}}_p(c_n).\ ] ] [ assumption2.4 ] @xmath39 and @xmath40 , where @xmath41 is defined as in assumption [ assumption2.3 ] and @xmath42 is defined as in assumption [ assumption2.1 ] . we remark that assumption [ assumption2.1 ] ensures that @xmath43 is a martingale for each fixed @xmath19 and is quite weak . clearly , assumption [ assumption2.1 ] is satisfied if @xmath21 is a sequence of i.i.d . random variables , which is independent of @xmath44 , with @xmath45 and @xmath46 . the lipschitz condition used in assumption [ assumption2.2 ] is standard in the investigation of uniform consistency , where we do not require the @xmath7 to have finite compact support . assumption [ assumption2.3 ] is a `` high level '' condition for the @xmath15 . we use it here to provide a framework . in sections [ sec22 ] and [ sec23 ] , we will show that this condition is in fact quite natural which holds true by many interesting and important examples . assumption [ assumption2.4 ] provides the connections among the moment condition required in assumption [ assumption2.1 ] , the condition ( [ m1 ] ) and the bandwidth @xmath12 . in many applications , we have @xmath47 , where @xmath48 and @xmath49 is a slowly varying function at infinite . see section [ sec23 ] and examples [ example1][example3 ] in section [ sec22 ] . in the typical situation that @xmath47 , if there exists a such that @xmath50 , the @xmath42 required in assumption [ assumption2.1 ] can be specified to @xmath51 + 1 $ ] . we have the following main result . [ th4 ] under assumptions [ assumption2.1][assumption2.4 ] , we have @xmath52 \biggr| = { \mathrm{o}}_p \bigl[(c_n \log n)^{1/2 } \bigr].\ ] ] if ( [ m1 ] ) is replaced by @xmath53 = { \mathrm{o}}(c_n),\qquad\mbox{a.s.},\ ] ] the result ( [ m10 ] ) can be strengthened to @xmath54 \biggr| = { \mathrm{o}}\bigl[(c_n \log n)^{1/2 } \bigr],\qquad\mbox{a.s.}\ ] ] theorem [ th4 ] can be extended to uniform convergence for the @xmath55 $ ] over unrestricted space @xmath8 . this requires additional condition on the @xmath15 and the tail decay for the function @xmath7 . [ th4a ] in addition to assumptions [ assumption2.1][assumption2.4 ] , @xmath56 $ ] and there exists a @xmath57 such that @xmath58.\ ] ] then , @xmath59 \biggr| = { \mathrm{o}}_p \bigl[(c_n \log n)^{1/2 } \bigr].\ ] ] similarly , if ( [ m1 ] ) is replaced by ( [ m1a9 ] ) and ( [ p09 ] ) is replaced by @xmath60,\qquad\mbox{a.s.},\ ] ] then @xmath61 \biggr| = { \mathrm{o}}\bigl[(c_n \log n)^{1/2 } \bigr],\qquad\mbox{a.s.}\ ] ] [ rem21 ] theorems [ th4][th4a ] allow for the @xmath13 to be a stationary or non - stationary time series . see examples [ example1][example3 ] and section [ sec23 ] below . more examples on non - stationary time series will be reported in a separate paper . the rates of convergence in both theorems are sharp . for instance , in the well - known stationary situation such as those appeared in examples [ example1][example3 ] , the @xmath41 can be chosen as @xmath62 . hence , when there are enough moment conditions on the @xmath21 ( i.e. , @xmath42 is large enough ) , we obtain the optimal rate @xmath63 , by taking @xmath64 . in non - stationary situation , the rate of convergence is different . in particular we have @xmath65 for the @xmath13 to be a random walk given in corollary [ th3 ] . the reason behind this fact is that the amount of time spent by the random walk around any particular point is of order @xmath66 rather than @xmath67 for a stationary time series . for more explanation in this regard , we refer to wang and phillips @xcite . this section provides several stationary time series examples which satisfy assumption [ assumption2.3 ] . examples [ example1 ] and [ example2 ] come from wu , huang and huang @xcite , where more general settings on the @xmath13 are established . example [ example3 ] discusses a strongly mixing time series . this example comes from hansen @xcite . by making use of other related works such as peligrad @xcite , nze and doukhan @xcite , masry @xcite , bosq @xcite and andrews @xcite , similar results can be established for other mixing time series like @xmath68-mixing and near - epoch - dependent time series . in these examples , we only consider the situation that @xmath69 . the extension to @xmath70 is straightforward and hence the details are omitted . throughout examples [ example1][example3 ] , we use the notation @xmath71 . example on the harris recurrent markov chains , which allows for stationary ( positive recurrent ) or non - stationary ( null recurrent ) series , is given in section [ sec23 ] . in the section , we also consider the uniform lower bound , which is of independent interests . more examples on @xmath0 processes with innovations being linear processes will be reported in a separate paper . [ example1 ] let @xmath72 be a linear process defined by @xmath73 where @xmath74 is a sequence of i.i.d . random variables with @xmath75 and a density @xmath76 satisfying @xmath77 and @xmath78 where @xmath79 denotes the @xmath80-order derivative of @xmath81 . suppose that @xmath82 and @xmath83 , and in addition assumption [ assumption2.2 ] , @xmath7 has a compact support . it follows from section 4.1 of wu , huang and huang @xcite that , for any @xmath84 and @xmath85 , @xmath86 \biggr| = { \mathrm{o}}\biggl[\sqrt{\frac{\log n}{nh } } + n^{-1/2 } l(n ) \biggr ] , \qquad\mbox{a.s.},\ ] ] where @xmath49 is a slowly varying function . note that @xmath13 is stationary process with a bounded density @xmath87 under the given conditions on @xmath88 . simple calculations show that @xmath89 = { \mathrm{o}}_p(nh),\ ] ] that is , @xmath13 satisfies assumption [ assumption2.3 ] . [ example2 ] consider the non - linear time series of the following form @xmath90 where @xmath91 is a bivariate measurable function and @xmath88 are i.i.d . this is the iterated random function framework that encompasses a lot of popular non - linear time series models . for example , if @xmath92 , it is the threshold autoregressive ( tar ) model ( see tong @xcite ) . if @xmath93 , then it is autoregressive model with conditional heteroscedasticity ( arch ) model . other non - linear time series models , including random coefficient model , bilinear autoregressive model and exponential autoregressive model can be fitted in this framework similarly . see wu and shao @xcite for details . in order to identify assumption [ assumption2.3 ] , we need some regularity conditions on the initial distribution of @xmath94 and the function @xmath95 . define @xmath96 denote by @xmath97 the conditional density of @xmath98 at @xmath19 given @xmath94 . further let @xmath99 and @xmath100^{1/2 } \quad\mbox{and}\quad j(x ) = \biggl [ \int _ { { r } } \biggl| \frac{\partial}{\partial x } g'(y\mid x ) \biggr|^2 \,{\mathrm{d}}y \biggr]^{1/2},\ ] ] @xmath101 and @xmath102 can be interpreted as a prediction sensitivity measure . these quantities measure the change in 1-step predictive distribution of @xmath98 with respect to change in initial value @xmath94 . suppose that : there exist @xmath103 and @xmath104 such that @xmath105 < 0 \quad\mbox{and}\quad e l_{\varepsilon _ 0}^2 < 1;\ ] ] @xmath106 < \infty$ ] ; in addition to assumption [ assumption2.2 ] , @xmath7 has a compact support . it follows from section 4.2 of wu , huang and huang @xcite that , for any @xmath84 and @xmath85 @xmath86 \biggr| = { \mathrm{o}}\biggl[\sqrt{\frac{\log n}{nh } } + n^{-1/2 } l(n ) \biggr ] , \qquad\mbox{a.s.},\ ] ] where @xmath49 is a slowly varying function . note that @xmath13 has a unique and stationary distribution under the given condition ( i ) and ( ii ) . see diaconis and freedman @xcite , for instance . simple calculations show that @xmath89 = { \mathrm{o}}_p(nh),\ ] ] that is , @xmath13 satisfies assumption [ assumption2.3 ] . [ example3 ] let @xmath107 be a strictly stationary time series with density @xmath87 . suppose that : @xmath13 is strongly mixing with mixing coefficients @xmath108 that satisfy @xmath109 where @xmath110 and @xmath111 ; @xmath112 for some @xmath113 satisfying @xmath114 and there is some @xmath115 such that for all @xmath116 , @xmath117 where @xmath118 is the joint density of @xmath119 ; in addition to assumption [ assumption2.2 ] , @xmath7 has a compact support . it follows from theorem 4 ( with @xmath120 ) of hansen @xcite that , for any @xmath84 and @xmath121 with @xmath122 , @xmath123 \biggr| = { \mathrm{o}}_p \biggl[\sqrt{\frac{\log n}{nh } } \biggr].\ ] ] if in addition @xmath124 , the result ( [ exa3 ] ) can be strengthened to almost surely convergence . simple calculations show that @xmath89 = { \mathrm{o}}_p(nh),\ ] ] that is , @xmath13 satisfies assumption [ assumption2.3 ] . let @xmath107 be a harris recurrent markov chain with state space @xmath125 , transition probability @xmath126 and invariant measure @xmath127 . we denote @xmath128 for the markovian probability with the initial distribution @xmath129 , @xmath130 for correspondent expectation and @xmath131 for the @xmath132-step transition of @xmath107 . a subset @xmath133 of @xmath134 with @xmath135 is called @xmath133-set of @xmath107 if for any @xmath136 , @xmath137 where @xmath138 and @xmath139 . as is well - known , @xmath133-sets not only exist , but generate the entire sigma @xmath140 , and for any @xmath133-sets @xmath141 and any probability measure @xmath142 on @xmath125 , @xmath143 where @xmath144 . see nummelin @xcite , for instance . let a @xmath133-set @xmath133 and a probability measure @xmath145 on @xmath146 be fixed . define @xmath147 } \nu p^k(d),\qquad t\ge0.\ ] ] by recurrence , @xmath148 . by virtue of ( [ 13 ] ) , the asymptotic order of @xmath149 depends only on @xmath107 . as in chen @xcite , a harris recurrent markov chain @xmath107 is called @xmath150-regular if @xmath151 where @xmath152 . it is interesting to notice that , under the condition ( [ d1 ] ) , the function @xmath149 is regularly varying at infinity , that is , there exists a slowly varying function @xmath153 such that @xmath154 . this implies that the definition of @xmath150-regular harris recurrent markov chain is similar to that of @xmath150-null recurrent given in karlsen and tjstheim @xcite and gao , li and tjstheim @xcite , but it is more natural and simple . the following theorem provides uniform upper and lower bounds for a functional of @xmath13 . the upper bound implies that @xmath13 satisfies assumption [ assumption2.3 ] , allowing for the @xmath13 being stationary ( @xmath155 , positive recurrent markov chain ) and non - stationary ( @xmath156 , null recurrent markov chain ) . the lower bound plays a key role in the investigation of the uniform consistency for the kernel estimator in a non - linear cointegrating regression , and hence is of independent interests . see section [ sec3 ] for more details . both upper and lower bounds are optimal , which is detailed in remarks [ rem22 ] and [ rem23 ] . [ th1 ] suppose that : @xmath107 is a @xmath150-regular harris recurrent markov chain , where the invariant measure @xmath127 has a bounded density function @xmath157 on @xmath91 ; in addition to assumption [ assumption2.2 ] , @xmath158 . then , for any @xmath159 satisfying @xmath160 for some @xmath161 , we have @xmath162 = { \mathrm{o}}_p\bigl[a(n ) h \bigr],\ ] ] where @xmath163 can be any finite integer . for a given sequence of constants @xmath164 , if there exists a constant @xmath165 such that , uniformly for @xmath67 large enough , @xmath166\ge a(n ) h / c_0,\ ] ] then , for any @xmath159 satisfying @xmath160 for some @xmath161 , we have @xmath167 \biggr\}^{-1 } = { \mathrm{o}}_p \bigl\{\bigl[a(n ) h\bigr]^{-1 } \bigr\}.\ ] ] [ rem22 ] the result ( [ 12a ] ) implies that , for any @xmath168 , there exists a constant @xmath169 such that @xmath170 \ge a(n ) h / c_\eta \biggr ) \ge1-\eta.\ ] ] this makes both bounds on ( [ 12 ] ) and ( [ 12a ] ) are optimal . on the other hand , since the result ( [ 15 ] ) implies that @xmath171\ge a(n ) h ( 1-\eta ) /c_{\eta}\ ] ] for any @xmath168 , the condition ( [ wond ] ) is close to minimal . note that random walk is a @xmath172-regular harris recurrent markov chain . the following corollary on a random walk shows the range @xmath173 can be taken to be optimal as well . [ th3 ] let @xmath174 be a sequence of i.i.d . random variables with @xmath175 , @xmath176 and the characteristic function @xmath177 of @xmath178 satisfying @xmath179 . write @xmath180 . if in addition to assumption [ assumption2.2 ] , @xmath181 , then , for @xmath182 and @xmath183 where @xmath184 , we have @xmath185 = { \mathrm{o}}_p(\sqrt n h)\ ] ] for any integer @xmath186 , and @xmath187 \biggr\}^{-1 } = { \mathrm{o}}_p \bigl\{(\sqrt n h)^{-1 } \bigr\}\ ] ] for any @xmath188 . [ rem23 ] for a random walk @xmath13 defined as in corollary [ th3 ] , it was shown in wang and phillips @xcite that @xmath189 \to_d\int f^2(s ) \,{\mathrm{d}}s l_w(1,y),\ ] ] where @xmath190 is a local time of a brownian motion @xmath191 , and @xmath192 if @xmath193 and @xmath194 if @xmath195 . since @xmath196 for any @xmath197 , it follows from ( [ ez1 ] ) that the range @xmath198 in ( [ m15 ] ) can not be extended to @xmath199 for any @xmath200 . [ rem24 ] as in examples [ example1][example3 ] , we may obtain a better result if @xmath201 is stationary ( positive null recurrent ) and satisfies certain other restrictive conditions . indeed , kristensen @xcite provided such a result . let @xmath202 be a time - homogeneous , geometrically ergodic markov chain . denote the 1-step transition probability by @xmath203 , such that @xmath204 . also denote the @xmath205-step transition probability by @xmath206 , such that @xmath207 . since @xmath13 is geometrically ergodic , it has a density @xmath87 . further suppose that : ( strong doeblin condition ) there exists @xmath208 and @xmath209 such that for all @xmath210 , @xmath211 @xmath212 exists and is uniformly continuous for all @xmath19 , for some @xmath213 , @xmath214<\infty$ ] for some @xmath215 , in addition to assumption [ assumption2.2 ] , @xmath7 has a compact support . it follows from kristensen @xcite that , for any @xmath216 and @xmath217 , @xmath218 - g(x ) \int f^2(s ) \,{\mathrm{d}}s \biggr| = { \mathrm{o}}_p \biggl[h^r + \sqrt { \frac{\log n}{nh } } \biggr],\ ] ] which yields ( [ 12 ] ) with @xmath219 and ( [ 12a ] ) with @xmath219 and @xmath220 , where @xmath221 is a constant such that @xmath222 . [ rem25]it is much more complicated if @xmath13 is a null recurrent markov chain , even in the simple situation that @xmath13 is a random walk defined as in corollary [ th3 ] . in this regard , we have ( [ ez1 ] ) , but it is not clear at the moment if it is possible to establish a result like @xmath223 - \int f^2(s ) \,{\mathrm{d}}s l_w(1,x ) \biggr| = { \mathrm{o}}_p ( c_n)\ ] ] for some @xmath224 and @xmath225 . note that ( [ ez2 ] ) implies that @xmath226\to_p\int f^2(s ) \,{\mathrm{d}}s l_w(1,0)\ ] ] for any fixed @xmath227 . this is a stronger convergence than that given in ( [ ez1 ] ) . our experiences show that it might not be possible to prove ( [ ez2 ] ) without enlarging the probability space in which the @xmath13 hosts . consider a non - linear cointegrating regression model : @xmath228 where @xmath229 is a stationary error process and @xmath230 is a non - stationary regressor . let @xmath231 be a non - negative real function and set @xmath232 where @xmath233 . the conventional kernel estimate of @xmath234 in model ( [ mo1 ] ) is given by @xmath235 the point - wise limit behavior of @xmath236 has currently been investigated by many authors . among them , karlsen , myklebust and tjstheim @xcite discussed the situation where @xmath230 is a recurrent markov chain . wang and phillips @xcite and cai , li and park @xcite considered an alternative treatment by making use of local time limit theory and , instead of recurrent markov chains , worked with partial sum representations of the type @xmath237 where @xmath17 is a general linear process . in another paper , wang and phillips @xcite considered the errors @xmath21 to be serially dependent and cross correlated with the regressor @xmath13 for small lags . for other related works , we refer to kasparis and phillips @xcite , park and phillips @xcite , gao _ et al . _ @xcite , marmer @xcite , chen , li and zhang @xcite , wang and phillips @xcite and wang @xcite . this section provides a uniform convergence for the @xmath236 by making direct use of theorems [ th4 ] and [ th1 ] in developing the asymptotics . for reading convenience , we list the assumptions as follows . [ assumption3.1 ] @xmath107 is a @xmath150-regular harris recurrent markov chain defined as in section [ sec3 ] , where the invariant measure @xmath127 has a bounded density function @xmath157 on @xmath91 ; @xmath26 is a martingale difference , where @xmath238 , satisfying @xmath239 , where @xmath240 for some @xmath241 . [ assumption3.2 ] the kernel @xmath10 satisfies that @xmath242 , @xmath243 and for any @xmath244 , @xmath245 [ assumption3.3 ] there exists a real positive function @xmath87 such that @xmath246 uniformly for some @xmath247 and any @xmath248 , where @xmath249 can be chosen sufficiently small and @xmath250 . assumption [ assumption3.1 ] is similar to , but weaker than those appeared in karlsen , myklebust and tjstheim @xcite , where the authors considered the point - wise convergence in distribution . assumption [ assumption3.2 ] is a standard condition on @xmath231 as in the stationary situation . the lipschitz condition on @xmath231 is not necessary if we only investigate the point - wise asymptotics . see remark [ rem32 ] for further details . assumption [ assumption3.3 ] requires a lipschitz - type condition in a small neighborhood of the targeted set for the functionals to be estimated . this condition is quite weak , which may host a wide set of functionals . typical examples include that @xmath251 ; @xmath252 ; @xmath253 ; @xmath254 . we have the following asymptotic results . [ th31 ] suppose assumptions [ assumption3.1][assumption3.3 ] hold , @xmath216 and @xmath255 where @xmath241 is given as in assumption [ assumption3.1 ] . it follows that @xmath256^{-1/2 } \log^{1/2}n + h^{\alpha } \delta_n \bigr\},\ ] ] where @xmath257 , @xmath258 and @xmath35 satisfies that @xmath259\ge a(n ) h / c_0\ ] ] for some @xmath165 and all @xmath67 sufficiently large . in particular , for the random walk @xmath13 defined as in corollary [ th3 ] , we have @xmath260 where @xmath261 for some @xmath188 and @xmath262 . [ rem31 ] when a high moment exists on the error @xmath21 , the @xmath263 can be chosen sufficiently small so that there are more bandwidth choices in practice . it is understandable that the results ( [ q1 ] ) and ( [ q1a ] ) are meaningful if only @xmath264 , which depends on the tail of the unknown regression function @xmath234 , the bandwidth @xmath12 and the range @xmath265 . when @xmath234 has a light tail such as @xmath266 , @xmath267 may be bounded by a constant . in this situation , the @xmath268 in ( [ q1a ] ) can be chosen to be @xmath269 for some @xmath188 . in contrast to theorem [ th1 ] and remark [ rem23 ] , this kind of range @xmath270 might be optimal , that is , the @xmath268 can not be improved to @xmath271 , for any @xmath200 , to establish the same rate of convergence as in ( [ q1a ] ) . [ rem32 ] both results ( [ q1 ] ) and ( [ q1a ] ) are sharp . however , a better result can be obtained if we are only interested in the point - wise asymptotics for @xmath272 . for instance , as in wang and phillips @xcite with minor modification , we may show that , for each @xmath19 , @xmath273 whenever @xmath13 is a random walk defined as in corollary [ th3 ] . furthermore @xmath272 has an asymptotic distribution that is mixing normal , under minor additional conditions . more details are referred to wang and phillips @xcite . [ rem33 ] wang and wang @xcite established a similar result to ( [ q1a ] ) with the @xmath13 being a partial sum of linear process , but only for the @xmath19 being a compact support and imposing a bounded condition on @xmath21 . the setting on the @xmath13 in this paper is similar to that given in gao , li and tjstheim @xcite , but our result provides the optimal range for the uniform convergence holding true and removes the independence between the error @xmath21 and @xmath13 required by gao , li and tjstheim @xcite . proof of theorem [ th4 ] we split the set @xmath274 into @xmath275 balls of the form @xmath276 where @xmath277 $ ] , @xmath278 and @xmath279 are chosen so that @xmath280 . it follows that @xmath281 \biggr| \nonumber \\ & & \quad\le \max_{0\le j\le m_n}\sup_{x\in a_{nj } } \sum _ { t=1}^{n } |u_t| \bigl|f\bigl[(x_t+x)/h \bigr]-f\bigl[(x_{t}+y_j)/h\bigr ] \bigr| \nonumber\\[-8pt]\\[-8pt ] & & \qquad { } + \max_{0\le j\le m_n } \biggl|\sum_{t=1}^{n } u_t f\bigl[(x_{t}+y_j)/h\bigr ] \biggr| \nonumber \\ & & \quad:= \lambda_{1n}+\lambda_{2n}.\nonumber\end{aligned}\ ] ] recalling the assumption [ assumption2.2 ] , it is readily seen that @xmath282-f \bigl[(x_{t}+y_j)/h\bigr ] \bigr| \nonumber \\ & \le & c \bigl(h m_n'\bigr)^{-1 } \sum _ { t=1}^n|u_t| \\ & \le & c ( c_n \log n)^{1/2 } \frac1n \sum _ { t=1}^n|u_t| = { \mathrm{o}}\bigl[(c_n \log n)^{1/2}\bigr ] , \qquad\mbox{a.s.}\nonumber\end{aligned}\ ] ] by the strong law of large number . in order to investigate @xmath283 , write @xmath284 $ ] and @xmath285 . recalling @xmath286 and @xmath287 , we have @xmath288 \biggr| \nonumber \\ & & { } + \max_{0\le j\le m_n } \biggl|\sum_{t=1}^{n } \bigl[\bigl|u_t - u_t'\bigr|+e\bigl(\bigl|u_t - u_t'\bigr| \mid{\mathcal f}_{t-1}\bigr ) \bigr ] f\bigl[(x_{t}+y_j)/h \bigr ] \biggr| \nonumber\\[-8pt]\\[-8pt ] & \le & \max_{0\le j\le m_n } \biggl|\sum_{t=1}^{n } u_t^ * f\bigl[(x_{t}+y_j)/h\bigr ] \biggr| + c \sum _ { t=1}^{n } \bigl[\bigl|u_t - u_t'\bigr|+e \bigl(\bigl|u_t - u_t'\bigr|\mid{\mathcal f}_{t-1}\bigr ) \bigr ] \nonumber \\ : \!\!&=&\lambda_{3n } + \lambda_{4n}.\nonumber\end{aligned}\ ] ] routine calculations show that , under @xmath289 and @xmath40 , @xmath290 \nonumber \\ & \le & c \biggl(\frac{c_n}{\log n } \biggr)^{(1-{2p})/2 } \sum _ { t=1}^n \bigl[|u_t|^{2p}+e \bigl(|u_t|^{2p}\mid{\mathcal f}_{t-1}\bigr ) \bigr ] \nonumber\\[-8pt]\\[-8pt ] & \le & c ( c_n \log n)^{1/2 } \frac1n \sum _ { t=1}^n \bigl[|u_t|^{2p}+e \bigl(|u_t|^{2p}\mid{\mathcal f}_{t-1}\bigr ) \bigr ] \nonumber \\ & = & { \mathrm{o}}\bigl[(c_n \log n)^{1/2 } \bigr],\qquad\mbox{a.s.}\nonumber\end{aligned}\ ] ] by the strong law of large number again . we next consider @xmath291 . as @xmath292\le2 ( e [ |u_t|^{2p}\mid{\mathcal f}_{t-1}])^{1/p}$ ] , a.s . , assumptions [ assumption2.1 ] and [ assumption2.3 ] imply that @xmath293e \bigl [ \bigl(u_t^*\bigr)^2\mid{\mathcal f}_{t-1 } \bigr]={\mathrm{o}}_p(c_n).\ ] ] hence , for any @xmath294 , there exists a @xmath295 such that @xmath296 where @xmath297e [ ( u_t^*)^2\mid{\mathcal f}_{t-1}]$ ] , whenever @xmath67 is sufficiently large . this , together with @xmath298 and the well - known martingale exponential inequality ( see , e.g. , de la pea @xcite ) , implies that , for any @xmath294 , there exists a @xmath299 ( @xmath132 is as in assumption [ assumption2.3 ] ) such that , whenever @xmath67 is sufficiently large , @xmath300 \nonumber \\ & & \quad\le p \biggl[\lambda_{3n } \ge m_0 ( c_n \log n ) ^{1/2 } , \max_{0\le j\le m_n } \sum_{t=1}^{n } \sigma_{tj}^2\le m_0 c_n \biggr]+ \eta \nonumber\\[-8pt]\\[-8pt ] & & \quad\le \sum_{j=0}^{m_n } p \biggl[\sum _ { t=1}^{n}u_t^ * f \bigl[(x_{k}+y_j)/h\bigr]\ge m_0 ( c_n \log n ) ^{1/2 } , \sum_{t=1}^{n } \sigma_{tj}^2 \le m_0 c_n \biggr ] + \eta \nonumber \\ & & \quad\le m_n \exp\biggl\{-\frac{m_0 ^ 2 c_n \log n } { 6 m_0c_n } \biggr\}+\eta\le m_nn^{-m_0/6}+\eta\le2 \eta,\nonumber\end{aligned}\ ] ] where we have used the following fact : @xmath301^d \le c_1 n^{(k+2)d}\ ] ] as @xmath302 and @xmath217 . this yields @xmath303 $ ] . combining ( [ p6])([p10 ] ) , we establish ( [ m10 ] ) . to prove ( [ m11 ] ) , by checking ( [ p6])([p9 ] ) , it suffices to show that @xmath304,\qquad\mbox{a.s.}\ ] ] under the alternative condition ( [ m1a9 ] ) . in fact , as in ( [ * ] ) , it follows from ( [ m1a9 ] ) that @xmath305e \bigl [ \bigl(u_t^*\bigr)^2\mid{\mathcal f}_{t-1 } \bigr]={\mathrm{o}}(c_n),\qquad\mbox{a.s.}\ ] ] similarly to proof of ( [ p10 ] ) , we have for sufficiently large @xmath306 ( @xmath307 , say ) , @xmath308 \nonumber \\ & & \quad= p \biggl[\lambda_{3n } \ge m_0 ( c_n \log n ) ^{1/2 } , \max_{0\le j\le m_n } \sum_{k=1}^{n } \sigma_k^2\le m_0 c_n\mbox { , i.o . } \biggr ] \nonumber\\ & & \quad\le \lim_{s\to\infty } \sum_{n = s}^{\infty } p \biggl[\lambda_{3n } \ge m_0 ( c_n \log n ) ^{1/2 } , \max_{0\le j\le m_n } \sum_{k=1}^{n } \sigma_k^2\le m_0 c_n \biggr ] \\ & & \quad\le\lim_{s\to\infty } \sum_{n = s}^{\infty } m_n \exp\biggl\ { -\frac{m_0 ^ 2 c_n \log n } { 6 m_0c_n } \biggr\ } \nonumber \\ & & \quad\le c \lim_{s\to\infty } \sum_{n = s}^{\infty } n^{(k+2)d } n^{-m_0/6}=0,\nonumber\end{aligned}\ ] ] which yields ( [ p12 ] ) . the proof of theorem [ th4 ] is now complete . proof of theorem [ th4a ] we only prove ( [ m10a ] ) . it is similar to prove ( [ m11a ] ) and hence the details are omitted . we may write @xmath309\nonumber\\ & & \quad=\sum_{t=1}^{n } u_t f\bigl[(x_t+x)/h\bigr]i\bigl(\|x_t\|\le b_n/2 \bigr ) \nonumber\\[-8pt]\\[-8pt ] & & \qquad { } + \sum_{t=1}^{n } u_t f \bigl[(x_t+x)/h\bigr]i\bigl(\|x_t\|>b_n/2 \bigr ) \nonumber\\ & & \quad=\lambda_{5n } ( x)+\lambda_{6n}(x)\qquad \mbox{say}.\nonumber\end{aligned}\ ] ] it is readily seen from ( [ m10 ] ) and @xmath310\\ & & { } + \sup_{\|x\| > b_n/2 } \bigl|f(x / h)\bigr|\sum_{t=1}^{n } \\ & \le & { \mathrm{o}}_p \bigl[(c_n \log n)^{1/2 } \bigr]\end{aligned}\ ] ] as @xmath311 , a.s . by the strong law . as for @xmath312 , we have @xmath313 \\ & \le & c \sum_{t=1}^{n } p\bigl(\|x_t \|>b_n/2 \bigr ) \le c b_n^{-k_0 } \sum _ { t=1}^n e\|x_t\|^{k_0 } \\ & = & { \mathrm{o}}\bigl [ ( c_n \log n)^{1/2 } \bigr],\end{aligned}\ ] ] which yield @xmath314 $ ] . taking these estimates into ( [ h1 ] ) , we obtain ( [ m10a ] ) . the proof of theorem [ th4a ] is complete . proof of theorem [ th1 ] first , assume there exists a @xmath315 such that @xmath316 for some @xmath317 and probability measure @xmath145 on @xmath125 with @xmath318 . under this addition assumption , theorem [ th1 ] can be established by using the so - called split chain technique . to this end , define new random variables @xmath319 and @xmath320 by @xmath321 where @xmath322 . as easily seen , @xmath323 is a harris recurrent markov chain with state space @xmath324 and @xmath325 has the same transition probability @xmath126 as those of @xmath326 . since our result is free of the initial distribution , @xmath327 can be assumed to be identical with @xmath328 , that is , @xmath94 has the distribution @xmath145 . further define @xmath329 , @xmath330 @xmath331 , and @xmath332,\qquad z_{jn}(x ) = \sum _ { k=\rho_{j-1}\wedge n+1}^{\rho_{j}\wedge n } f^2 \bigl[(x_k+x)/h\bigr]\ ] ] for @xmath333 . it is well known that the blocks @xmath334 are i.i.d . blocks , @xmath335 having the distribution @xmath145 . hence , for each @xmath12 and @xmath19 , @xmath336 , where @xmath337 or @xmath338 is a sequence of i.i.d . random vectors . furthermore , by recalling that @xmath127 has a bounded density function @xmath157 , @xmath339 and @xmath340 , we have @xmath341 \pi({\mathrm{d}}s ) \nonumber\\[-8pt]\\[-8pt ] & = & b h \int_{-\infty}^{\infty } f^2(s ) p(-x+sh ) \,{\mathrm{d}}s \le c^ * h\nonumber\end{aligned}\ ] ] for any @xmath342 and @xmath343 for any integer @xmath132 . see lemma 5.2 of karlsen and tjstheim @xcite or lemma b.1 of gao , li and tjstheim @xcite . we also have the following lemma . [ lem1]suppose that @xmath344 , where @xmath165 is a constant , and all @xmath345 , are different , where @xmath346 and @xmath347 for some @xmath348 . then , @xmath349 \biggr| = { \mathrm{o}}_p\bigl[n^{-\varepsilon_0/4 } a(n ) h\bigr ] , \\ \label{96 } \delta_{n } & : = & \max_{0\le j\le m_n}e \biggl|\sum _ { k=0}^{d_n } \bigl[z_k^*(y_j)-ez_k^*(y_j ) \bigr ] \biggr| = { \mathrm{o}}\bigl[n^{-\varepsilon_0/4 } a(n ) h\bigr],\end{aligned}\ ] ] where @xmath263 is a constant such that @xmath350 . only consider @xmath351 , as the situation that @xmath352 is similar . to this end , write @xmath353 and @xmath354 . we have @xmath355 \biggr| + \max_{0\le j\le m_n } \sum _ { i=1}^{d_n } \bigl [ \widehat z _ { i}(y_j)+ e \widehat z _ { i}(y_j ) \bigr ] \nonumber\\[-8pt]\\[-8pt ] : \!\!&= & r_{1n}+r_{2n}.\nonumber\end{aligned}\ ] ] by taking @xmath356 in ( [ fi2 ] ) and noting @xmath357 , simple calculations show that @xmath358 which yields @xmath359 $ ] . as for @xmath360 , by using ( [ fi2 ] ) with @xmath361 and noting @xmath362 for any @xmath363 and some @xmath165 , the standard markov inequality implies that @xmath364 \biggr|\ge m n^{-\varepsilon_0/4}a(n)h \biggr ) \nonumber\\[-8pt]\\[-8pt ] & & \quad\le c m_n \exp\bigl(-m t n^{-\varepsilon_0/4 } a(n)h + c_{\varepsilon } a(n)t^2h\bigr ) \nonumber \\ & & \quad\le c m_n \exp\bigl(-m n^{\varepsilon_0/4}/4 \bigr)\to0\nonumber\end{aligned}\ ] ] as @xmath365 . hence , @xmath366 $ ] . combining ( [ b1])([l1 ] ) , we prove ( [ 95 ] ) . the proof of ( [ 96 ] ) is similar except more simpler . indeed , by independence of @xmath367 , we obtain @xmath368 \biggr| \\ & & { } + 2\max_{0\le j\le m_n } \sum_{i=1}^{d_n}e \widehat z _ { i}(y_j ) \\ & \le & 2 \max_{0\le j\le m_n } d_n^{1/2 } \bigl[e\widetilde z _ { 1}^2(y_j)\bigr]^{1/2 } + c n^{-\varepsilon_0/2}a(n ) h \\ & \le & c n^{-\varepsilon_0/4}a(n ) h,\end{aligned}\ ] ] due to the fact : @xmath369 the proof of lemma [ lem1 ] is complete . we are now ready to prove ( [ 12 ] ) and ( [ 12a ] ) under the additional condition ( [ 1 ] ) . ( [ 12 ] ) first . as in proof of ( [ p6 ] ) and ( [ p7 ] ) , but letting @xmath370 - 1+j / m_n',j=0 , 1,2,\ldots , m_n , $ ] where @xmath371 $ ] and @xmath372 + 1)m_n'$ ] , we have @xmath373 \le\max_{0\le j\le m_n}\sum _ { k=0}^n f^2 \bigl[(x_k+y_j)/h\bigr]+ c a(n ) h.\ ] ] note that @xmath374 \le\sum_{k=0}^{\rho _ { n(n+1 ) } } f^2\bigl[(x_k+x)/h\bigr ] = \sum _ { i=1}^{n(n+1 ) } z _ { i}(x),\ ] ] and @xmath375 is bounded in probability . see , for example , chen @xcite . for each @xmath376 , there exist @xmath377 such that @xmath378 whenever @xmath67 is sufficiently large . consequently , for each @xmath379 and @xmath67 large enough , @xmath380\ge a \biggr)\le p \biggl(\max_{0\le j\le m_n } \sum_{i=1}^{c_\varepsilon a(n ) } z _ { i}(y_j)\ge a \biggr)+\varepsilon.\ ] ] this , together with ( [ 95 ] ) with @xmath351 , implies ( [ 12 ] ) under ( [ 1 ] ) , since @xmath381.\ ] ] we next consider ( [ 12a ] ) under ( [ 1 ] ) . to this regard , let @xmath382 - 1+j / m_n ' , j=0 , 1,2,\ldots , m_n , $ ] where @xmath383 $ ] and @xmath384 + 1)m_n'$ ] . since @xmath385 } \sum_{t=1}^{n } \bigl|f^2\bigl [ ( x_t+x)/h\bigr]-f^2\bigl [ ( x_t+y_j)/h\bigr ] \bigr| \\ & & \quad \le c n h^{-1 } \max_{0\le j\le m_n-1}|y_{j+1}-y_j| \le cn^{-\varepsilon_0/2 } a(n)h,\end{aligned}\ ] ] it is readily seen that @xmath386 \ge\delta_{1n}-{\mathrm{o}}_{p } \bigl[n^{-\varepsilon_0/2 } a(n)h\bigr],\ ] ] where @xmath387 $ ] . furthermore , by recalling ( [ a99 ] ) and noting that @xmath374 \ge\sum_{k=0}^{\rho _ { n(n)}\wedge n } f^2\bigl[(x_k+x)/h\bigr ] = \sum _ { i=1}^{n(n ) } z _ { in}(x),\ ] ] we have , for each @xmath388 and @xmath67 large enough , @xmath389 on the other hand , it follows from ( [ 95 ] ) with @xmath352 that @xmath390\\[-8pt ] & \ge & c_{1\varepsilon } a(n)\inf_{0\le j\le m_n}ez _ { 1n}(y_j ) -{\mathrm{o}}_{p}\bigl[n^{-\varepsilon_0/4 } a(n)h\bigr].\nonumber\end{aligned}\ ] ] combining ( [ 145])([147 ] ) , the result ( [ 12a ] ) under ( [ 1 ] ) will follow if we prove : there exists a @xmath391 such that @xmath392 for all @xmath67 sufficiently large . to prove ( [ 148 ] ) , first note that there exists a @xmath393 such that @xmath394 . see lemma 3.3 of karlsen and tjstheim @xcite , for instance . therefore , by taking @xmath395 + 1 $ ] , where @xmath396 is chosen later , we have for some @xmath391 , @xmath397 \\ & \ge&\frac{1}{d_n } \inf_{0\le j\le m_n}e \biggl(\sum _ { t=1}^ { n } f^2\bigl [ ( x_t+y_j)/h\bigr ] -i(\rho_{d_n } \le n ) \sum _ { t=1}^ { \rho_{d_n } } f^2\bigl [ ( x_t+y_j)/h\bigr ] \biggr ) \\ & \ge&\frac{1}{d_n } \biggl(\inf_{|x|\le b_n+1}e\sum _ { k=1}^nf^2\bigl [ ( x_t+x)/h \bigr]-m_n \biggr ) \\ & \ge&\frac{1}{d_n } \bigl[a(n)h / c_0-m_n \bigr ] \\ & \ge & b_0 h,\end{aligned}\ ] ] whenever @xmath67 is sufficiently large , where we have used the condition ( [ wond ] ) and the fact : it follows from ( [ 45 ] ) , ( [ 96 ] ) and @xmath398 if and only if @xmath399 that @xmath400 \biggr ] \\ & = & \max_{0\le j\le m_n } e \biggl[i(\rho_{d_n } \le n ) \sum _ { i=1}^{d_n } z_i(y_j ) \biggr ] \\ & \le & d_n \max_{0\le j\le m_n } ez_1(y_j ) p \bigl(n(n)\ge d_n\bigr)+ \max_{0\le j\le m_n } e \biggl|\sum _ { i=1}^{d_n } \bigl[z_i(y_j)- ez_i(y_j ) \bigr ] \biggr| \\ & \le & b_2^{-1 } c^*h a^{-1}(n ) en^2(n)+{\mathrm{o}}\bigl[n^{-\varepsilon_0/4}a(n)h\bigr ] \\ & \le & c_0^{-1 } a(n)h/2\end{aligned}\ ] ] by choosing @xmath401 and @xmath67 sufficiently large . this proves ( [ 148 ] ) and also completes the proof of ( [ 12a ] ) under ( [ 1 ] ) . we now consider general situation . let @xmath402 be fixed . define a transition probability @xmath403 on @xmath125 by @xmath404 let @xmath405 be an i.i.d . bernoulli random variables with the common law @xmath406 and assume @xmath405 and @xmath202 are independent . define a renewal sequence @xmath407 by @xmath408 with these notations , @xmath409 is a harris recurrent markov chain with the invariant measure @xmath127 . the transition probability @xmath403 of @xmath409 satisfies the additional condition ( [ 1 ] ) and @xmath410 see chen @xcite , for instance . by virtue of these facts , it follows from the first part proof of ( [ 12 ] ) that , for any fixed @xmath186 and @xmath159 , @xmath411 = \sup_{|x|\le n^m } \sum_{k=1}^n f^2\bigl[(x_{\sigma(k)}+x)/h\bigr]= { \mathrm{o}}_p \bigl[a_t(n ) h\bigr].\ ] ] now by noting @xmath412)/n\to_{\mathrm{a.s . } } \lambda/(1-t)$ ] by the strong law and taking @xmath413 such that @xmath414 , simple calculations show that @xmath415 \le \sup_{|x|\le n^m } \sum_{k=1}^{\sigma([\lambda n ] ) } f^2\bigl[(x_{\sigma ( k)}+x)/h\bigr]={\mathrm{o}}_p\bigl[a(n ) h \bigr].\ ] ] similarly , @xmath416 = { \mathrm{o}}_p\bigl[a(n ) h\bigr]\ ] ] and hence the result ( [ 12 ] ) under general situation follows . the proof of ( [ 12a ] ) under general situation is similar and hence the details are omitted . proof of corollary [ th3 ] we first notice that : * @xmath417 is a harris null recurrent markov chain , satisfying ( [ 1 ] ) , @xmath418 and the invariant measure @xmath127 is the lebesgue measure . due to the fact ( f ) , ( [ 12a1 ] ) follows immediately from theorem [ th1 ] . to prove ( [ m15 ] ) , by theorem [ th1 ] , it suffices to show that ( [ wond ] ) holds true with @xmath419 and @xmath420 . in fact , under the conditions of corollary [ th3 ] , @xmath421 has a density @xmath422 , satisfying @xmath423 , as @xmath424 , where @xmath425 , due to the central limit theorem . this implies that @xmath426 for some @xmath427 and all sufficiently large @xmath132 . hence , for @xmath428 and @xmath67 sufficiently large , we have @xmath429 & = & \inf_{|x|\le \tau_n \sqrt n+1 } \int_{-\infty}^{\infty } f^2 \bigl[(\sqrt k y+x)/h\bigr ] p_k(y)\,{\mathrm{d}}y \\ & \ge&\frac{h}{\sqrt k}\inf_{|x|\le\tau_n \sqrt n+1}\int_{-\infty } ^{\infty } f^2(y ) p_k\bigl[(yh - x)/\sqrt k\bigr]\,{\mathrm{d}}y \\ & \ge&\frac{h}{\sqrt k}\inf_{|x|\le3 \tau_n } p_k(x)\int _ { |y|\le m_1 } f^2(y)\,{\mathrm{d}}y \\ & \ge&\frac{a_0 h}{2\sqrt n } \int_{|y|\le m_1 } f^2(y)\,{\mathrm{d}}y,\end{aligned}\ ] ] where @xmath430 is chosen such that @xmath431 . consequently , there exists a constant @xmath165 such that @xmath432\ge\inf_{|x|\le\tau_n \sqrt n+1 } \sum _ { k = n/2}^n e f^2 \bigl[(x_k+x)/h\bigr ] \ge\sqrt n h / c_0\ ] ] as required . the proof of corollary [ th3 ] is now complete . proof of theorem [ th31 ] we may write @xmath433 as @xmath434k_{h}(x_{t}-x)}{\sum_{t=1}^{n}k_{h}(x_{t}-x ) } \nonumber\\[-8pt]\\[-8pt ] : \!\!&= & \theta_{1n}(x)+\theta_{2n}(x).\nonumber\end{aligned}\ ] ] note that , for any @xmath265 , there exists a @xmath165 such that @xmath435=0 $ ] if @xmath436 . it follows from assumption [ assumption3.3 ] that , whenever @xmath67 is sufficiently large , @xmath437}{\sum _ { t=1}^{n}k[(x_t - x)/h ] } \le ch^{\alpha } \delta_n.\ ] ] this , together with ( [ 12a ] ) [ taking @xmath438 in theorem [ th1 ] , implies that ( [ q1 ] ) will follow if we prove @xmath439 = { \mathrm{o}}_p \bigl[\bigl[a(n)h \bigr]^{1/2 } \log^{1/2}n \bigr].\ ] ] in fact , with @xmath240 and @xmath440 , we have @xmath441 since @xmath442 . now , by recalling ( [ 12 ] ) , it is readily seen that the conditions of theorem [ th4 ] hold for @xmath443 and @xmath444 . the result ( [ mp1 ] ) follows from ( [ m10 ] ) in theorem [ th4 ] . the authors thank associate editor , two referees and professor jiti gao for helpful comments on the original version . wang acknowledges the partial research support from the australian research council .

we give a bound for the geometric dimension for the family of virtually cyclic groups in mapping class groups of an orientable compact surface with punctures , possibly with nonempty boundary and negative euler characteristic .

1606.00306

let @xmath0 be an orientable compact surface with finitely many punctures ( possibly zero punctures ) and negative euler characteristic . the mapping class group , @xmath1 , of @xmath0 is the group of isotopy classes of orientation preserving diffeomorphisms of @xmath0 that fix point - wise the boundary . let @xmath2 be an integer and @xmath3 be the congruence subgroup of @xmath1 , this is the subgroup of those elements that act trivially on @xmath4 . classifying spaces , @xmath5 , for the family of finite subgroups of a group @xmath6 , have been extensively studied . for the mapping class group , it is well - known that the teichmller space @xmath7 is a model for @xmath8 by results of kerckhoff given in @xcite . later , j. aramayona and c. martnez proved in @xcite that the minimal dimension for which there exists a model for @xmath9 coincides with the virtual cohomological dimension @xmath10 , this has been computed by j. l. harer in @xcite . finite dimensional models for classifying spaces @xmath11 for the family of virtually cyclics have been constructed for word - hyperbolic groups ( juan - pineda , leary @xcite ) , for groups acting in cat(0 ) spaces ( farley @xcite , lck @xcite , degrijse and petrosyan @xcite ) , and many other groups . a method to construct a model for @xmath11 is to start with a model for @xmath5 and then try to extend this to obtain a model for @xmath11 . in @xcite lck and weiermann gave a general construction using this idea . the smallest possible dimension of a model of @xmath11 is denoted by @xmath12 and is called the geometric dimension of @xmath6 for the family of virtually cyclic subgroups . we prove that @xmath13 is finite . degrijse and petrosyan proved this fact in @xcite for closed surfaces of genus at least 2 , although our method produces a larger bound , it is more general since it includes surfaces with boundary and our techniques are different from theirs . this paper is organized as follows : we review the fundamental material about classifying spaces and mapping class groups in sections [ sec - classif ] and [ sec - mcg ] respectively . in section [ sec - commen ] , we develop the analysis of commensurators of infinite virtually cyclic subgroups in the mapping class groups and we prove : [ intro - prop - comm ] let @xmath0 be an orientable closed surface with finitely many punctures and @xmath14 . let @xmath15 be fixed . let @xmath16 be infinite cyclic and @xmath17 such that @xmath18 . then @xmath19= n_{\mcg}(d)\ ] ] where @xmath20 $ ] is the commensurator of @xmath21 and @xmath22 is the normalizer of @xmath23 . furthermore , the subgroup @xmath23 may be chosen to be maximal in @xmath24 . from proposition [ intro - prop - comm ] and a description of normalizers of infinite cyclic subgroups that we develop in section [ sec - commen ] , we prove the following : let @xmath0 be an orientable compact surface with finitely many punctures and @xmath14 . then @xmath25 , that is , the mapping class group @xmath1 admits a finite dimensional model for @xmath26 . our main result is the following : let @xmath0 be an orientable compact surface with finitely many punctures and @xmath14 . let @xmath15 , then 1 . @xmath27 ; 2 . let @xmath28 $ ] be the index of @xmath3 in @xmath29 , then @xmath30 \cdot \gdvc \gamma_m(s)\\ & \leq [ \gamma(s):\gamma_m(s ) ] \cdot ( vcd(\gamma(s ) ) + 1 ) .\end{aligned}\ ] ] where @xmath31 is the virtual cohomological dimension of @xmath29 . we point out that the bound in @xmath32 is sharp . a key result in finding the above is the following proposition which we could not find in the literature and we think is one of the most interesting contributions of this work , see proposition [ prop - modm - maximality ] . let @xmath0 be an orientable compact surface with finitely many punctures and @xmath14 . let @xmath15 , then the group @xmath33 satisfies the property that every infinite virtually cyclic subgroup is contained in a unique maximal virtually cyclic subgroup . we emphasize that degrijse and petrosyan proved in @xcite the finiteness of @xmath34 by other methods . in this work we outline a method that gives a precise description of a model for the classifying space for virtually cyclics of mapping class groups . it depends on certain teichmller spaces and mapping class groups of subsurfaces of @xmath0 . acknowledgment : a. trujillo would like to thank john guaschi for many fruitful conversations . let @xmath6 be a group . a _ family _ @xmath35 of subgroups of @xmath6 is a set of subgroups of @xmath6 which is closed under conjugation and taking subgroups . the following are natural examples of families of @xmath6 : @xmath36 let @xmath35 be a family of subgroups of @xmath6 . a model of the _ classifying space @xmath37 for the family @xmath35 _ is a @xmath6-cw - complex @xmath38 , such that all of its isotropy groups belong to @xmath35 and if @xmath39 is a @xmath6-cw - complex with isotropy groups belonging to @xmath35 , there is precisely one @xmath6-map @xmath40 up to @xmath6-homotopy . in other words , @xmath38 is a terminal object in the category of @xmath6-cw complexes with isotropy groups belonging to @xmath35 . in particular , two models for @xmath37 are @xmath6-homotopy equivalent . let @xmath6 be a group , @xmath41 and @xmath38 a @xmath6-set . @xmath42-fixed point set _ @xmath43 is defined as @xmath44 ( * ? ? ? 1.9 ) [ thm - esp - clas - defi ] a @xmath6-cw - complex @xmath38 is a model of @xmath37 if and only if the @xmath42-fixed point set @xmath43 is contractible for @xmath45 and is empty for @xmath46 . the smallest possible dimension of a model of @xmath47 is called the _ geometric dimension of @xmath6 for the family @xmath48 _ and is usually denoted as @xmath49 . when a finite dimensional model of @xmath47 does not exist , then @xmath50 . + we abbreviate @xmath51 and call it the _ universal _ @xmath6-cw-_complex for proper @xmath6-actions _ , and we abbreviate @xmath52 . denote by @xmath53 and @xmath54 . it is known that for any groups @xmath55 , @xmath56 for a subgroup @xmath41 , and a family @xmath48 of @xmath6 , denote by @xmath57 the family of subgroups of @xmath42 induced from @xmath35 . a model of @xmath58 is given by restricting the action of @xmath6 to @xmath42 in a model of @xmath37 . then @xmath59 we will use the construction given by w. lck and m. weiermann in @xcite . in particular for the families @xmath60 , it is as follows : let @xmath61 be the collection of infinite virtually cyclic subgroups of @xmath6 . define an equivalence relation , @xmath62 , on @xmath63 as @xmath64 for @xmath65 and @xmath66 in @xmath67 , where @xmath68 denotes the cardinality of the set @xmath69 . let @xmath70 $ ] denote the set of equivalence classes under the above relation and let @xmath71 \in [ \vci_g ] $ ] be the equivalence class of @xmath42 . define @xmath72 this is the isotropy group of @xmath71 $ ] under the @xmath6-action on @xmath70 $ ] induced by conjugation . observe that @xmath73 $ ] is the commensurator of the subgroup @xmath74 . define a family of subgroups of @xmath75 by @xmath76 the method to build a model of @xmath11 from one of @xmath5 is given in the following theorem . * thm.2.3 ) [ thm - luck - weierm ] let @xmath67 and @xmath62 be as above . let @xmath77 be a complete system of representatives , @xmath71 $ ] , of the @xmath6-orbits in @xmath70 $ ] under the @xmath6-action coming from conjugation . choose arbitrary @xmath73$]-cw - models for @xmath78)$ ] , @xmath79}(n_g[h])$ ] and an arbitrary @xmath6-cw - model for @xmath80 . define @xmath38 a @xmath6-cw - complex by the cellular @xmath6-pushout @xmath81\in i } g \times_{n_g[h ] } \underline{e}(n_g[h ] ) \ar[d]^{\coprod_{[h]\in i } id_g \times_{n_g[h]}f_{[h ] } } \ar[r]^(0.75){i } & \underline{e}(g ) \ar[d ] \\ \coprod_{[h]\in i } g \times_{n_g[h ] } e_{\g[h]}(n_g[h ] ) \ar[r ] & x } \end{aligned}\ ] ] such that @xmath82}$ ] is a cellular @xmath73$]-map for every @xmath71\in i$ ] and @xmath83 is an inclusion of @xmath6-cw - complexes , or such that every map @xmath82}$ ] is an inclusion of @xmath73$]-cw - complexes for every @xmath71\in i$ ] and @xmath83 is a cellular @xmath6-map . then @xmath38 is a model for @xmath84 the maps in theorem [ thm - luck - weierm ] are given by the universal property of classifying spaces for families and inclusions of families of subgroups . observe that if @xmath85 is finite and both @xmath86 $ ] and @xmath87 $ ] are uniformly bounded , then @xmath12 is finite . observe that if @xmath88 and both @xmath89 and @xmath90 are infinite cyclic subgroups of @xmath42 and @xmath91 respectively , let @xmath62 as in ( [ rel - eq ] ) @xmath92 let @xmath93 be the set of infinite cyclic subgroups @xmath21 of @xmath6 . then the equivalence relation @xmath62 given in ( [ rel - eq ] ) can be defined in @xmath93 , we will denote by @xmath94 $ ] the set of equivalence classes . let @xmath0 be an orientable compact surface with a finite set @xmath95 , called punctures , of points removed from the interior . we will assume that the surface has negative euler characteristic . references for this section are @xcite , @xcite and @xcite . let @xmath96 denote the group of orientation preserving difeomorphisms of @xmath0 that restrict to the identity on the boundary @xmath97 . we endow this group with the compact - open topology . the _ mapping class group of _ @xmath0 , denoted @xmath1 , is the group @xmath98 that is , the group of ( smooth ) isotopy classes of elements of @xmath99 where isotopies are required to fix the boundary pointwise . _ congruence subgroups . _ let @xmath100 , @xmath101 . we denote by @xmath24 the kernel of the natural homomorphism @xmath102 defined by the action of diffeomorphisms on the homology group , @xmath24 is called the congruence subgroup of @xmath1 . note that this subgroup has finite index in @xmath1 . + _ complex of curves . _ an _ essential _ curve is a simple closed curve of @xmath0 that is not homotopic to a point , a puncture , or a boundary component . the _ complex of curves _ , denoted by @xmath103 , is the abstract simplicial complex associated to @xmath0 such that , ( i ) vertices are isotopy classes of essential curves , we denote by @xmath104 the set of vertices ; ( ii ) @xmath103 has a @xmath105-simplex for each @xmath106-tuple of vertices , where each pair of corresponding isotopy classes have disjoint representants . the _ realization _ of a simplex is the union of mutually disjoint curves that represent its vertices . the mapping class group acts on @xmath104 : if @xmath107 , the action is given by @xmath108 . then @xmath1 acts on @xmath103 , since this action sends simplicies into simplicies . + _ dehn twists . let @xmath109 be isotopy classes of simple closed curves in @xmath0 . we will denote the dehn twist about @xmath110 as @xmath111 . let @xmath112 and @xmath113 . we have the following properties of dehn twists , see ( * ? ? ? 3.3 ) , 1 . @xmath114 iff @xmath115 and @xmath116 ; 2 . @xmath117 ; 3 . @xmath118 iff @xmath119 . we will first assume that @xmath0 has empty boundary . we assume the reader is familiar with the theory of transverse singular foliations . see @xcite or @xcite for references . + _ pseudo - anosov diffeomorphisms . _ a _ diffeomorphism _ @xmath120 is called _ pseudo - anosov _ if there exists a pair of transverse measured foliations @xmath121 , @xmath122 of @xmath0 and a real number @xmath123 such that * @xmath124 * the 1-prongs singularities of these foliations belong to the set of punctures . the measure foliation @xmath121 is called the _ stable foliation for @xmath125 _ and @xmath122 is called the _ unstable foliation for @xmath125 _ , and @xmath126 is the dilatation of @xmath125 . an element @xmath112 is called _ pseudo - anosov _ if it is represented by a pseudo - anosov diffeomorphism . an element @xmath127 is called _ reducible _ if @xmath127 fixes some simplex of @xmath103 and _ irreducible _ otherwise . among the irreducible elements , those of finite order are periodic and those of infinite order are pseudo - anosov . there is the following classification theorem for elements of the mapping class group , see ( * ? ? ? 13.2 ) . ( nielsen - thurston classification ) [ classif - mcg ] let @xmath128 . let s be an orientable surface of genus @xmath129 and @xmath130 punctures . each @xmath112 is either periodic , reducible , or pseudo - anosov . further , pseudo - anosov mapping classes are neither periodic nor reducible . a periodic element is represented by a finite order diffeomorphism . _ surfaces with boundary . _ if @xmath131 has non - empty boundary , we define @xmath112 to be pseudo - anosov if @xmath127 restricts to a pseudo - anosov diffeomorphism on the punctured surface obtained by removing the boundary @xmath97 . + let @xmath0 be an orientable closed surface with finitely many punctures . let @xmath132 be an orientable compact subsurface of @xmath0 , the inclusion @xmath133 induces a natural homomorphism @xmath134 let @xmath135 and @xmath136 be a representative difeomorphism of @xmath129 . then @xmath137 is defined as the isotopy class of the difeomorphism which coincides with @xmath138 in @xmath132 and is the identity in @xmath139 . is the identity on the shaded regions and coincides with @xmath127 on the unshaded regions . ] let @xmath140 with vertices @xmath141 and @xmath21 be its realization in @xmath0 . let @xmath142 be an open regular neighborhood of @xmath21 in @xmath0 and denote by @xmath143 , where each @xmath144 is a connected component . let @xmath145 and @xmath146 denote the two boundary components of @xmath142 that are isotopic to @xmath147 in s. [ rem - euler - k - r ] suppose that the surface @xmath0 has genus @xmath129 and @xmath130 punctures . note that @xmath148 and @xmath149 , therefore @xmath150 and @xmath151 , see @xcite page 249 . see ( * ? ? ? 3.18 ) as a reference for the following homomorphisms . + _ cutting the surface . _ from the inclusions @xmath152 and @xmath153 denote the induced homomorphisms by @xmath154 with @xmath155 . observe that @xmath156 and any element in the image @xmath157 leaves each subsurface @xmath144 of @xmath0 invariant . + _ corking . _ we glue a 1-punctured disk in each boundary component of @xmath158 , that is , we corked all boundary components , and denote this surface by @xmath159 , where the @xmath160 are the connected components of @xmath161 ( see figure [ fig - corking ] ) . note that the surface @xmath162 has the same euler characteristic as @xmath0 . from the inclusion @xmath163 , the induced homomorphism defined as ( [ homom - injection ] ) is called the _ corking homomorphism of @xmath162 _ , and it is denoted by @xmath164 with @xmath165 . let @xmath166 be the set of punctures in @xmath160 coming from boundary components of @xmath144 . note that @xmath167 is the subgroup of @xmath168 that fixes pointwise all @xmath169 . since @xmath156 and @xmath170 , then @xmath171 note that @xmath172 can be seen as the subgroup of @xmath173 that fixes each subsurface @xmath160 and fixes pointwise the punctures in @xmath174 . let @xmath0 be an orientable compact surface with finitely many punctures . a diffeomorphism @xmath175 is called _ pure _ if there exists a @xmath176-submanifold @xmath21 ( possibly empty ) of @xmath0 such that the following conditions are satisfied : 1 . @xmath21 is the realization of an element @xmath177 or @xmath21 is empty ; if @xmath178 , @xmath138 fixes @xmath21 , it does not rearrange the components of @xmath179 , and it induces on each component of @xmath179 a diffeomorphism isotopic to either a pseudo - anosov or the identity diffeomorphism . we call an element @xmath180 _ pure _ if the isotopy class of @xmath127 contains a pure diffeomorphism . we call @xmath181 pure if @xmath42 consists of pure elements . 1.5 and 1.8 ) [ thm - modm - pure ] if @xmath182 , then @xmath24 is a torsion free group and a pure subgroup of @xmath1 . from now on , we will consider the subgroup @xmath24 with @xmath182 . + _ canonical reduction system . _ let @xmath183 . an isotopy class @xmath184 is called an essential reduction class for @xmath6 if the following conditions are satisfied ( i ) @xmath185 for all @xmath186 ; ( ii ) if @xmath187 with @xmath188 , and @xmath189 is denoted by @xmath190 . ] then @xmath191 for some @xmath192 . the set of essential reduction classes of @xmath6 is a simplex of @xmath103 and it is called _ a canonical reduction system _ of @xmath6 and it is denoted by @xmath193 . in general , for a subgroup @xmath181 define @xmath194 and for @xmath112 define @xmath195 . 7.2 , 7.3 ) [ lem - red - syst ] let @xmath112 , and @xmath196 . + ( i ) if @xmath197 has finite index , then @xmath198 , + ( ii ) @xmath199 . [ gs = s - reduct ] let @xmath0 be an orientable compact surface with finitely many punctures . if @xmath200 are such that @xmath201 for some @xmath202 , then @xmath203 . the hypothesis entails that @xmath204 , by lemma [ lem - red - syst ] @xmath205 then @xmath206 and by lemma [ lem - red - syst ] @xmath207 , therefore @xmath203 . _ canonical form . _ by cutting @xmath0 along a reduction system @xmath208 of an element in @xmath1 and applying the nielsen - thurston classification theorem to each subsurface of @xmath209 we can obtain a decomposition of the element as follows , see figure [ fig - canonicalf ] . 13.3 ) [ canonical - descom - mod ] let @xmath210 and @xmath211 be its canonical reduction system with vertices @xmath141 . let @xmath212 be as before and let @xmath213 be the union of pairwise disjoint closed neighborhoods @xmath214 of curves representatives of the @xmath147 . then there is a representative @xmath215 of @xmath127 that permutes the @xmath144 , so that some power of @xmath125 leaves invariant each @xmath144 . moreover , there exists an integer @xmath216 so that @xmath217 for all @xmath83 and @xmath218 where each @xmath219 is either pseudo - anosov or the identity and @xmath220 for @xmath221 . when @xmath222 and @xmath15 the integer @xmath223 can always be taken to be one . , where each subsurface is fixed . a shaded region indicates a pseudo - anosov component or a dehn twist and a unshade region indicates an identity component . ] [ sec - stabilizer ] in this section , we shall consider @xmath0 with empty boundary . the stabilizers @xmath224 will be used to understand the normalizers of reducible elements of @xmath1 . let @xmath140 and @xmath225 be as in section [ sec - prelim ] . denote the stabilizer subgroup of @xmath226 in @xmath1 by @xmath227 ( * ? ? ? 3.20)[cutting - homom ] let @xmath140 with vertices @xmath141 and @xmath228 . then there is a well defined homomorphism @xmath229 where @xmath230 is the free abelian subgroup generated by dehn twists about the curves @xmath231 . let @xmath232 be the finite index subgroup of @xmath224 that fixes each @xmath147 with orientation , since elements in @xmath232 are orientation - preserving , it follows that they also preserve the sides of each curve @xmath147 in @xmath0 , thus they fix each subsurface @xmath144 . denote the restriction @xmath233 , then @xmath234 therefore we have that @xmath235 is surjective and @xmath236 . [ rem - image - pseudo ] by ( * ? ? 1.2 ) , elements in @xmath237 do not rearrange the components of @xmath158 and fix each curve of @xmath226 , so @xmath238 . let @xmath222 , @xmath211 with vertices @xmath141 , and let the canonical form of @xmath127 be as follows @xmath239 moreover , assume that @xmath240 . note that @xmath241 and @xmath242 since @xmath243 and @xmath244 is well defined , then @xmath245 hence for each @xmath83 , @xmath246 , therefore @xmath247 is pseudo - anosov or the identity . in order to build a model @xmath248 as mentioned in sec . [ section - build ] , we need to know the commensurators of the infinite virtually cyclic subgroups in @xmath1 . using the following condition , we may give a description of the commensurators of infinite virtually cyclic subgroups . 1 . for every @xmath249 , with @xmath250 , and @xmath251 , @xmath252 [ cond - c - mod ] let @xmath0 be an orientable compact surface with finitely many punctures and @xmath14 . the group @xmath1 satisfies condition ( c ) . * case i. * the proof when @xmath0 has empty boundary is given in propositions [ panosov - cond - c ] and [ reducible - cond - c ] , since elements with infinite order in @xmath1 are reducible or pseudo - anosov . + * case ii . * assume @xmath253 and it has @xmath254 connected components . let @xmath255 , with @xmath256 such that @xmath257 for some @xmath258 . let @xmath259 be the corking homomorphism of @xmath0 as in ( [ corking - h ] ) , then @xmath260 is the free abelian group generated by dehn twists about curves isotopic to the boundary components of @xmath0 . since @xmath261 is central in @xmath1 , if @xmath262 we conclude that @xmath263 . + on the other hand , if @xmath264 , applying @xmath265 , we have @xmath266 and @xmath267 , applying case i we conclude that @xmath268 . about powers of a pseudo - anosov diffeomorphism we have the following : suppose that @xmath0 has empty boundary . let @xmath210 be a pseudo - anosov ele - ment and @xmath215 be a pseudo - anosov diffeomorphism in its class , with @xmath269 , @xmath270 the stable and unstable foliations of @xmath271 respectively and @xmath272 its dilatation . for @xmath273 , @xmath274 the reader may consult @xcite and @xcite as references . [ panosov - cond - c ] let @xmath0 be an orientable closed surface with finitely many punctures and @xmath14 . let @xmath275 and @xmath276 be pseudo - anosov mapping classes , suppose that there exist @xmath277 and a permutation @xmath278 such that @xmath279 for some @xmath280 , then @xmath281 suppose that @xmath282 and let @xmath283 . for each @xmath284 , let @xmath285 be pseudo - anosov diffeomorphisms in the class @xmath286 , for each @xmath83 there exists @xmath287 in the class @xmath288 , such that @xmath289 , this follows by the uniqueness of pseudo - anosovs ( * ? ? ? + let @xmath290 and @xmath291 be the stable and unstable foliations , respectively , of @xmath285 , with dilatation @xmath292 , @xmath293 . by ( [ power - p - a - fol ] ) , if @xmath294 ( or @xmath295 ) , @xmath296 and @xmath297 are the stable and unstable ( unstable and stable ) foliations respectively of @xmath298 with dilatation @xmath299 ( or @xmath300 ) . suppose that @xmath301 . then @xmath302 sends the stable and unstable foliations of @xmath303 to the stable and unstable foliations of @xmath304 respectively ( ( * ? ? ? 16 , exp . 12 ) ) . since the foliations are uniquely ergodic , the measure is up to a constant , that is , @xmath305 with @xmath306 . thus @xmath307 then the diffeomorphisms @xmath308 and @xmath309 have the same dilatation . then for each @xmath83 , we have @xmath310 . since @xmath311 , for all @xmath83 , @xmath312 has only one real positive @xmath313-root , then we can conclude that @xmath314 if for some @xmath83 , @xmath315 , then we conclude that @xmath263 . if @xmath316 for all @xmath83 , let @xmath317 be the minimum positive integer such that @xmath318 . since @xmath319 for all @xmath83 , then @xmath320 thus @xmath321 . since @xmath322 , we conclude that @xmath263 . on the other hand , if @xmath323 , we have that @xmath324 , and in a similar way , we conclude @xmath325 . [ reducible - cond - c ] let @xmath0 be an orientable closed surface with finitely many punctures and @xmath14 . let @xmath112 be a reducible element such that @xmath201 for some @xmath326 and @xmath202 , then @xmath327 . let @xmath211 , suppose that @xmath226 has vertices @xmath141 . by hypothesis and by lemma [ gs = s - reduct ] , @xmath328 , so @xmath129 lies in the stabilizer @xmath224 . let @xmath329 as in theorem [ cutting - homom ] , @xmath330 * case 1 . * suppose that @xmath331 , that is , @xmath332 ; without loss of generality we can suppose that @xmath333 . note that @xmath129 may permute @xmath141 , let @xmath334 such that @xmath335 for all @xmath293 . since the @xmath336 have disjoint realizations in @xmath0 , the dehn twists @xmath337 commute , thus by hypothesis and because @xmath338 ( see ( * ? ? ? 3.3 ) ) , we have the following , @xmath339 then we conclude that @xmath340 @xmath341 , we can regard these as vectors @xmath342 and @xmath343 in @xmath344 and @xmath345 . then @xmath346 and @xmath347 are in the same line of @xmath348 , and since @xmath347 is obtained by a permutation of the coordinates of @xmath346 , we conclude that @xmath349 or @xmath350 , therefore @xmath268 . + * case 2 . * since @xmath351 is of finite index , we may assume that @xmath352 . suppose that @xmath353 . since @xmath352 , @xmath354 fixes each subsurface @xmath355 of @xmath161 . let @xmath356 , for every @xmath357 , let @xmath358 note that @xmath359 may permute the subsurfaces @xmath160 of @xmath161 , let @xmath360 for each @xmath357 and let @xmath361 such that @xmath362 , @xmath363 . by hypothesis we have that @xmath364 , then @xmath365 since @xmath352 and @xmath366 , for some @xmath367 , @xmath368 is pseudo - anosov ( see remark [ rem - image - pseudo ] ) . let @xmath369 be the minimum positive integer such that @xmath370 , from ( [ conjugate - red ] ) it follows that @xmath371 then @xmath372 is pseudo - anosov for each @xmath373 . observe that @xmath374 and @xmath375 are homeomorphic for all @xmath376 , then we can apply proposition [ panosov - cond - c ] , therefore @xmath268 . we denote by @xmath377 and @xmath378 the centralizer and normalizer respectively of the subgroup @xmath379 in @xmath6 . 6.1)[f^m = g^m ] let @xmath0 be an orientable compact surface with finitely many punctures . let @xmath380 be a pure subgroup . if @xmath381 are such that @xmath382 for some @xmath383 , then @xmath384 . [ lem - n(f)=n(fk ) ] let @xmath0 be an orientable compact surface with finitely many punctures . let @xmath352 , @xmath385 and let @xmath386 be either @xmath1 or @xmath24 , then @xmath387 suppose that @xmath383 . since @xmath388 , we need to prove that @xmath389 . if @xmath390 , we have @xmath391 for some @xmath392 . since @xmath393 and @xmath24 is pure ( thm . [ thm - modm - pure ] ) , we can apply theorem [ f^m = g^m ] and we conclude that @xmath394 therefore , @xmath395 . thus @xmath24 is a normal subgroup of @xmath1 , we can use theorem [ f^m = g^m ] in similar way to prove that @xmath396 . we have the proof for the centralizers by taking @xmath397 . [ lem - norm - dehnt ] let @xmath0 be an orientable compact surface with finitely many punctures . let @xmath177 with vertices @xmath398, ... ,@xmath399 , and @xmath400 , with @xmath401 , then for any @xmath402 , @xmath403 let @xmath404 , we will prove that @xmath405 . by lemma [ lem - red - syst ] , @xmath406 and @xmath129 may permute the classes @xmath141 . let @xmath334 such that @xmath335 for all @xmath83 . by the results about dehn twists given in section [ sec - prelim ] and since @xmath407 , for some @xmath408 , we have @xmath409 then we conclude that @xmath410 @xmath411 , so @xmath412 for all @xmath83 . on the other hand , we have that @xmath413 since @xmath412 then @xmath414 therefore @xmath405 . 4.2 ) [ lem - norm - cond - c ] suppose that @xmath6 satisfies condition ( c ) . then , for any @xmath415 there is a nested sequence of subgroups @xmath416 where @xmath417 is the subgroup of @xmath21 given by @xmath418 , observe that @xmath419=\bigcup_{k\geq 1 } n_g(k!c).\ ] ] the subgroup @xmath420 denotes the normalizer of @xmath417 in @xmath6 . we will follow the same notation as section [ section - build ] . let @xmath93 be the set of infinite cyclic subgroups of @xmath6 . [ prop - comm - mcgm ] let @xmath0 be an orientable closed surface with finitely many punctures and @xmath14 . let @xmath15 be fixed . let @xmath421 and @xmath17 such that @xmath422 . then @xmath19=n_{\mcg}(g^n).\ ] ] furthermore , the subgroup @xmath423 can be choosen maximal in @xmath424 . let @xmath425 , @xmath426 $ ] its class and suppose that @xmath427 . from lemma [ lem - norm - cond - c ] , we have that @xmath428 @xmath429=\bigcup_{k\geq 1 } n_{\mcg}(g^{k ! } ) . \end{aligned}\ ] ] let @xmath17 such that @xmath422 , by lemma [ lem - n(f)=n(fk ) ] , we have that @xmath430 for any @xmath431 , and @xmath432 , then @xmath19=n_{\mcg}(g^n).\ ] ] we will prove in proposition [ prop - modm - maximality ] that @xmath423 is contained in a unique maximal @xmath433 , therefore @xmath19=n_{\mcg}(g^n)=n_{\mcg}(\overline{c}).\ ] ] by lemma [ lem - norm - dehnt ] , we have : [ cor - comm - dehn ] let @xmath0 be an orientable compact surface with finitely many punctures . let @xmath177 with vertices @xmath398, ... ,@xmath399 , and @xmath400 , with @xmath401 . then @xmath434=n_{\mcg}(f).\ ] ] for surfaces with empty boundary , by the nielsen - thurston classification theorem , infinite order elements are pseudo - anosov or reducible . [ normalizer - pseudoa ] ( * ? ? ? 1 ) let @xmath0 be an orientable closed surface with finitely many punctures . let @xmath112 be a pseudo - anosov mapping class . the centralizer @xmath435 is a finite extension of an infinite cyclic group . the normalizer , @xmath436 is either equal to @xmath435 or contains @xmath435 as a normal subgroup of index 2 . since @xmath24 is torsion free for @xmath15 , for @xmath437 a pseudo - anosov mapping class , we know that @xmath438 is an infinite cyclic group . + let @xmath352 be reducible and @xmath211 with vertices @xmath141 . by lemma [ gs = s - reduct ] , @xmath435 and @xmath436 are subgroups of the stabilizer @xmath224 . recall that @xmath232 is the subgroup of @xmath224 that fixes each @xmath147 with orientation . denote by @xmath439 and @xmath440 , the finite index subgroups of @xmath435 and @xmath436 respectively . [ prop - centralizer ] let @xmath0 be an orientable closed surface with finitely many punctures . let @xmath352 with @xmath240 , then @xmath441 & \z^r \ar[r ] & c_{\mcg}(f)^0 \ar[r]^(0.4){\rhoso } & \prod_{i=1}^k c_{\gamma(\widehat{s}_i,\mathcal{q}_i)}(f_i ) \ar[r ] & 1 , } \end{aligned}\ ] ] and @xmath442 has index @xmath443 in @xmath444 . write @xmath127 in its canonical form as in ( [ canonical - descomp ] ) , @xmath445 by remark [ rem - image - pseudo ] we have @xmath446 let @xmath447 , following the method of theorem [ canonical - descom - mod ] , since @xmath129 fixes each subsurface @xmath144 of @xmath162 and each @xmath147 with orientation , @xmath129 can be written as @xmath448 and @xmath449 can be reducible , periodic or pseudo - anosov , for each @xmath83 . in a similar way as in remark [ rem - image - pseudo ] , @xmath450 thus @xmath451 then @xmath452 the result follows since by the definition of @xmath453 , its kernel is generated by elements of the form @xmath454 and any commutator has no dehn twists about boundary components . now , if @xmath455 , then @xmath456 which is generated by dehn twists about boundary components of @xmath144 , then we have @xmath457 , therefore @xmath458 which follows by the equality given in ( [ eq - theta - i ] ) . + since there are no restrictions for @xmath459 with @xmath460 , we conclude ( [ sucesion - centralizer ] ) . + from the equality ( [ norm = cent ] ) , if some @xmath461 , then @xmath462 . + on the other hand , if @xmath463 , since each @xmath247 is the identity or pseudo - anosov , in case @xmath247 is pseudo - anosov , @xmath464 is a subgroup of index @xmath176 or @xmath465 in @xmath466 ( theorem [ normalizer - pseudoa ] ) , therefore we conclude that @xmath442 has index @xmath443 in @xmath444 . by proposition [ prop - centralizer ] and theorem [ normalizer - pseudoa ] , if we rename the subsurfaces @xmath144 as necessary , we have : [ prop - centralizer - vcy ] let @xmath0 be an orientable closed surface with finitely many punctures . let @xmath352 with @xmath467 where @xmath468 are pseudo - anosov . then @xmath469 & \z^r \ar[r ] & c_{\mcg}(f)^0 \ar[r]^(0.4){\rhos } & \prod_{i=1}^a \gamma(\widehat{s}_i,\mathcal{q}_i ) \prod_{j = a+1}^k v_j \ar[r ] & 1 , } \end{aligned}\ ] ] where @xmath470 is virtually cyclic for each @xmath471 . surfaces with boundary . suppose that @xmath0 has @xmath472 boundary components @xmath473 , let the corking homomorphism @xmath474 with kernel @xmath475 , generated by dehn twists about curves isotopic to boundary components of @xmath0 . let @xmath476 be the set of punctures of @xmath477 which comes from the boundary components of @xmath0 . then @xmath478 which is the subgroup of @xmath479 that fixes pointwise the set @xmath476 . let @xmath255 , suppose that @xmath480 and @xmath481 commute , then @xmath482 is in @xmath483 , but @xmath482 has no dehn twists about boundary components , so @xmath484 . therefore @xmath485&\z^b \ar[r ] & c_{\mcg}(f ) \ar[r]^(0.45){\theta_{s } } & c_{\gamma(\widehat{s},\mathcal{r})}(\theta_{s}(f ) ) \ar[r ] & 1 . } \end{aligned}\ ] ] observe that elements in @xmath1 leave invariant a regular neighborhood of the boundary @xmath97 . then , if @xmath127 has a non - zero power of a dehn twist @xmath486 , for any @xmath326 , @xmath487 has the same power of @xmath486 , then @xmath488 moreover , if @xmath127 has no dehn twist about curves @xmath473 , then @xmath489 then @xmath490&\z^b \ar[r ] & n_{\gamma(s)}(f ) \ar[r]^(0.45){\theta_{s } } & n_{\gamma(\widehat{s},\mathcal{r})}(\theta_{s}(f ) ) \ar[r ] & 1 } .\end{aligned}\ ] ] in section [ sec - geom - dim - mcg - mcgm ] we prove that @xmath492 , and in section [ sec - bounds - gdvc ] , we will give bounds for @xmath493 and @xmath494 . let @xmath0 be an orientable compact surface with finitely many punctures and @xmath14 . it is well - known that the teichmller space @xmath7 is a finite dimensional space which is contractible , on which @xmath1 acts properly and it is a model for @xmath8 by results of kerckhoff given in @xcite . on the other hand , j. aramayona and c. martnez proved in @xcite the following : ( * ? ? ? 1.3 ) [ thm - aramayona ] let @xmath0 be an orientable compact surface with finitely many punctures . then there exist a cocompact model for @xmath495 of dimension equal to the virtual cohomological dimension @xmath31 . we will use the same notation of section [ section - build ] . we will prove that there exist finite dimensional models for @xmath498 $ ] and @xmath499 } n_{\mcg}[c]$ ] and a uniform bound on @xmath500}n_{\gamma}[c]$ ] for any @xmath426\in [ \c_{\mcg}]$ ] , with these results , theorem [ thm - luck - weierm ] and the fact that @xmath501 is finite , we have : [ thm - gdf - mcg ] let @xmath0 be an orientable compact surface with finitely many punctures and @xmath14 . then @xmath25 , that is , the mapping class group @xmath1 admits a finite dimensional model for @xmath26 . 5.16)[seq - efin ] let @xmath502 be an exact sequence of groups . suppose that @xmath42 has the property that for any group @xmath503 which contains @xmath42 as subgroup of finite index , @xmath504 . if @xmath505 , then @xmath506 . 4.3 ) [ prop - guido ] let @xmath510 be an exact sequence of groups with @xmath511 finite . then @xmath6 admits an @xmath130-dimensional cocompact @xmath5 homeomorphic to @xmath512 , with @xmath6 acting by affine maps . [ rem - w - ec ] let @xmath426\in [ \c_{\mcg}]$ ] . from proposition [ prop - comm - mcgm ] , we could assume that @xmath21 is a cyclic maximal subgroup in @xmath424 and @xmath20=n_{\mcg}(c)$ ] . let @xmath513 and @xmath514 , the projection . + from theorem [ thm - esp - clas - defi ] , a model for @xmath515 with the @xmath20$]-action induced from the projection @xmath223 is a model for @xmath499 } n_{\mcg}[c]$ ] . + then it is sufficient to consider models for @xmath516 and @xmath517 , of maximal infinite cyclic subgroups @xmath21 in @xmath424 . _ proof of theorem _ [ thm - gdf - mcg ] : suppose that the surface @xmath0 has genus @xmath129 , @xmath254 boundary components and @xmath130 punctures . + * part 1 . * _ for any @xmath426\in [ \c_{\mcg}]$ ] , @xmath518 $ ] is finite . _ we may assume that @xmath519 , it follows that @xmath20=n_{\mcg}({c})$ ] . since @xmath520 from the properties given in ( [ gd - subg ] ) and theorem [ thm - aramayona ] , we conclude that @xmath518\leq \gdf \mcg\leq vcd \mcg$ ] , which is finite . + * part 2 . * _ we will prove that there exist @xmath521 , such that for any @xmath426\in [ \c_{\mcg}]$ ] , @xmath500 } n_{\mcg}[c]\leq z.$ ] _ + ( i ) _ the surface @xmath0 has empty boundary . _ let @xmath426\in \c_{\mcg}$ ] , with @xmath522 . by the nielsen - thurston classification theorem , @xmath127 is either , a pseudo - anosov class or a reducible element . + ( a ) if @xmath127 is pseudo - anosov , then @xmath20=n_{\mcg}(f)$ ] is virtually cyclic and @xmath523 $ ] is the family of all subgroups of @xmath436 , hence a point is a model for @xmath524}n_{\mcg}[c]$ ] , therefore @xmath500}n_{\mcg}[c]=0 $ ] . + ( b ) if @xmath127 is reducible and @xmath525 , with @xmath401 , where @xmath141 are the vertices of @xmath211 . by corollary [ cor - comm - dehn ] , @xmath20=n_{\mcg}(f)$ ] and we can suppose that @xmath526 . + following the same idea as in remark [ rem - w - ec ] , a model for @xmath527 with the induced action of the projection @xmath528 is a model for @xmath499}n_{\mcg}[c]$ ] . we will prove that @xmath529 is finite . + note that any element @xmath447 commutes with @xmath127 , because @xmath129 fixes each class @xmath147 , then @xmath530 . on the other hand , we have @xmath531 , therefore @xmath232 is a finite index subgroup of @xmath436 and it is normal , then @xmath469 & \mcg_{\sigma}^0 \ar[r ] & n_{\mcg}(f)\ar[r ] & b \ar[r ] & 1 , } \end{aligned}\ ] ] where @xmath532 is finite , we will obtain a uniform bound for the order of @xmath532 . since @xmath533 , we have @xmath469 & \mcg_{\sigma}^0/\langle f\rangle \ar[r ] & n_{\mcg}(f)/\langle f\rangle \ar[r ] & b \ar[r ] & 1 . } \end{aligned}\ ] ] since the index @xmath534\leq ( 2r)!$ ] , then @xmath535 . by theorem [ finite - index - gd ] , @xmath536 from ( [ cutting - homom-0 ] ) , since @xmath537 we have @xmath469 & \langle t_{\alpha_1}, .... ,t_{\alpha_r}\rangle /\langle f\rangle \ar[r ] & \mcg_{\sigma}^0/\langle f\rangle \ar[r ] & \prod_{i=1}^k \gamma(\widehat{s}_i,\mathcal{q}_i ) \ar[r ] & 1 , } \end{aligned}\ ] ] since @xmath538 and @xmath127 is identified with the point @xmath539 via that isomorphism , then the quotient @xmath540 , because @xmath333 . then @xmath469 & \z^{r-1 } \ar[r ] & \mcg_{\sigma}^0/\langle f\rangle \ar[r ] & \prod_{i=1}^a \gamma(\widehat{s}_i,\mathcal{q}_i ) \ar[r ] & 1 , } \end{aligned}\ ] ] we apply theorem [ seq - efin ] , the properties given in ( [ efin - product ] ) and ( [ gd - subg ] ) , and theorem [ thm - aramayona ] to conclude that @xmath541 observe that each @xmath542 has at least one puncture , no boundary components and negative euler characteristic , thus from theorem [ thm - vcd - harer ] we can see that @xmath543 , then @xmath544 from inequalities ( [ ineq - w - t ] ) and ( [ ineq - wt2 ] ) , we conclude @xmath545 by remark [ rem - euler - k - r ] @xmath546 , therefore @xmath547 note that the bound only depends on the surface . + ( c ) now suppose that @xmath127 is reducible , @xmath211 has vertices @xmath141 , and @xmath548 is not trivial . since @xmath20=n_{\mcg}(f^n)$ ] , for some @xmath549 such that @xmath550 , then we may assume that @xmath352 and @xmath551 is maximal in @xmath424 . we will apply remark [ rem - w - ec ] again . + note that @xmath552 is of finite index , then @xmath469 & c_{\mcg}(f)^0 \ar[r ] & n_{\mcg } ( f ) \ar[r ] & f \ar[r ] & 1 , } \end{aligned}\ ] ] with @xmath553 , this follows by proposition [ prop - centralizer ] and because the index @xmath534\leq ( 2r)!$ ] . since @xmath352 , then @xmath554 , and so , @xmath469 & c_{\mcg}(f)^0/\langle f\rangle \ar[r ] & n_{\mcg } ( f)/\langle f\rangle \ar[r ] & f \ar[r ] & 1 . } \end{aligned}\ ] ] by theorem [ finite - index - gd ] @xmath555 let @xmath240 and we rename the @xmath542 such that @xmath556 for @xmath557 and @xmath558 is pseudo - anosov for @xmath559 , then by proposition [ prop - centralizer - vcy ] , @xmath469 & \langle t_{\alpha_1}, .... ,t_{\alpha_r}\rangle \ar[r ] & c_{\mcg}(f)^0 \ar[r]^(0.35){\rhoso } & \prod_{i=1}^a \gamma(\widehat{s}_i,\mathcal{q}_i ) \prod_{j = a+1}^k v_j \ar[r ] & 1 , } \end{aligned}\ ] ] where @xmath560 is virtually cyclic for each @xmath561 . + denote the group @xmath562 by @xmath563 , then we have the following homomorphism @xmath564 which is well - defined because @xmath565 , @xmath138 is a homomorphism and since @xmath244 is onto , then @xmath138 is onto too . thus @xmath566 is a free abelian subgroup isomorphic to @xmath344 . moreover , @xmath567 since @xmath568 is of finite index , @xmath569 , if @xmath570 , @xmath571 & \pi_{j = a+1}^k \langle f_j\rangle/\langle \check{f}\rangle \ar[r ] & \pi_{i = a+1}^k v_j/ \langle \check{f}\rangle \ar[r ] & \pi_{i = a+1}^k f_j \ar[r ] & 1 , } \end{aligned}\ ] ] with @xmath572 finite @xmath569 , thus from proposition [ prop - guido ] and ( [ vj ] ) we have that @xmath573 . thus applying theorem [ seq - efin ] to the exact sequence given by @xmath138 , the properties given in ( [ efin - product ] ) and ( [ gd - subg ] ) , and theorem [ thm - aramayona ] , we have : @xmath574 \nonumber \\ & \leq r+ [ \gdf ( \pi_{i=1}^a\gamma(\widehat{s}_i ) ) + ( k - a-1)]\nonumber \\ & \leq r+ [ \sum_{i=1}^a \gdf ( \gamma(\widehat{s}_i ) ) + ( k - a-1)]\nonumber \\ & \leq r+ [ \sum_{i=1}^a vcd ( \gamma(\widehat{s}_i ) ) + ( k - a-1 ) ] \nonumber \\ & \leq r+ [ \sum_{i=1}^a ( -2\chi(\widehat{s}_i ) ) + ( k - a-1 ) ] \nonumber \\ & \leq r+ [ ( -2\chi(s ) ) + ( k - a-1 ) ] \nonumber \\ & \leq \frac{-3\chi(s)-n}{2}+ ( -3\chi(s)-1 ) , \label{eq - c0f}\end{aligned}\ ] ] the last inequality ( [ eq - c0f ] ) holds because @xmath150 ( see remark [ rem - euler - k - r ] ) and @xmath575 . from ( [ ineq - red ] ) and ( [ eq - c0f ] ) we conclude that @xmath576 note that the bound only depends on the surface . from ( a ) and the inequalities ( [ ineq - partc ] ) and ( [ ineq - partb ] ) , for any @xmath426\in [ \c_{\mcg}]$ ] , @xmath577 } n_{\mcg}[c]\leq ( -6\chi(s)-n ) ( ( -3\chi(s)-n)!).\end{aligned}\ ] ] + ( ii ) _ the surface @xmath0 has non - empty boundary . _ following in a similar way as in part ( i ) , from ( [ centraliz - boundary]),([norm - boundary - eq ] ) and ( [ normaliz - boundary ] ) , we conclude that there exist @xmath521 , such that for any @xmath426\in [ \c_{\mcg}]$ ] , @xmath500 } n_{\mcg}[c]\leq z$ ] . + we follow the same notation as in section [ sec - prelim ] and [ sec - stabilizer ] . the homomorphism @xmath329 is given in section [ section - stabilizers ] , and @xmath581 . by ( * ? ? ? 1.2 ) , elements in @xmath237 do not rearrange the components of @xmath158 and fix each curve of @xmath226 , then @xmath238 . note that if @xmath582 and @xmath583 , by definition of @xmath584 and because @xmath129 is pure , each @xmath288 is pure , then the image @xmath585 is a pure subgroup of @xmath586 . let @xmath587 be the projection of the image @xmath585 over @xmath588 for each @xmath83 , then each @xmath587 is torsion free and we have @xmath589 where @xmath590 is a free abelian subgroup of @xmath591 , observe that @xmath592 as @xmath24 is of finite index and @xmath590 is of finite index in @xmath593 . as a reference , see ( * ? 8.7 ) [ lem - abelian ] suppose that @xmath0 has empty boundary . let @xmath6 be a subgroup of @xmath24 , @xmath15 . let @xmath594 , and suppose that @xmath595 , where @xmath302 denotes the projection of @xmath596 over @xmath588 for each @xmath83 . the group @xmath6 is abelian if and only if each @xmath302 is either trivial or an infinite cyclic group . * case i * : suppose that @xmath0 has empty boundary . since @xmath24 is torsion free for @xmath15 , then @xmath598 is the set of infinite cyclic subgroups of @xmath24 . + it is well - known that for surfaces @xmath0 with empty boundary , periodic elements of @xmath1 are of finite order . by the nielsen - thurston classification theorem , each element of @xmath599 is either reducible or pseudo - anosov . + let @xmath600 . observe that if @xmath601 and @xmath602 , then @xmath603 for some @xmath604 . by lemma [ lem - n(f)=n(fk ) ] , @xmath605 , thus @xmath606 . actually , we will prove that @xmath91 lies in a free abelian subgroup of @xmath24 ( which does not depends on @xmath91 but only on @xmath42 ) , then @xmath42 must be contained in an unique maximal subgroup of @xmath424 . if @xmath127 is pseudo - anosov , by theorem [ normalizer - pseudoa ] , @xmath607 and it is the unique maximal subgroup of @xmath424 containing @xmath42 . on the other hand , suppose that @xmath127 is reducible and let @xmath211 be its canonical reduction system , by lemma [ gs = s - reduct ] , @xmath608 . suppose that @xmath226 has vertices @xmath141 , let @xmath609 and @xmath584 as in ( [ cutting - hom - m ] ) , @xmath610 where @xmath611 is torsion free for each @xmath83 and @xmath612 is a free abelian subgroup of @xmath613 . note that each @xmath337 commutes with @xmath127 , if @xmath614 , then @xmath615 & \z^s \ar[r ] & c_{\mcgm}(f ) \ar[r]^(0.45){\rhosm } & \pi_{i=1}^k c_{\gamma_i}(f_i)\ar[r ] & 1 } , \end{aligned}\ ] ] furthermore each @xmath247 is either the identity or pseudo - anosov , see remark [ rem - image - pseudo ] . let @xmath616 , then we have that * @xmath617 , * @xmath618 , @xmath247 is an @xmath130-th root of @xmath619 because @xmath127 is an @xmath130-th root of @xmath105 , * if @xmath620 for some @xmath561 , then @xmath621 , because each @xmath622 is torsion free . let @xmath623 , such that @xmath624 if only if @xmath625 is pseudo - anosov . by theorem [ normalizer - pseudoa ] , @xmath626 is an infinite cyclic subgroup for each @xmath624 . we regard @xmath627 as subgroup of @xmath628 , let @xmath629 then we have @xmath615 & \z^s \ar[r ] & g \ar[r]^(0.35){\rhosm } & \pi_{j=1}^d c_{\gamma_{l_j}}(f_{l_j})\ar[r ] & 1 } , \end{aligned}\ ] ] by lemma [ lem - abelian ] , we conclude that @xmath6 is a free abelian subgroup and by construction , we have that @xmath630 . since @xmath91 was taken arbitrarily , we conclude that there exists a unique maximal subgroup @xmath631 containing @xmath42 . * case ii . * suppose that @xmath0 has non empty boundary and that @xmath0 has @xmath472 connected boundary components @xmath473 . let @xmath632 , be the corking homomorphism as in ( [ corking - h ] ) , with @xmath633 by definition of @xmath265 we have that @xmath634 . since each @xmath486 acts trivially on @xmath635 , @xmath636 , then @xmath637 . then we have the following , @xmath638 & \z^b \ar[r ] & \mcgm \ar[r]^{\theta_{s , m } } & \gamma_m(\widehat{s } ) . } \end{aligned}\ ] ] it is well known that @xmath1 is torsion free when @xmath0 has non - empty boundary , then @xmath639 is the set of infinite virtually cyclic subgroups of @xmath24 . + let @xmath640 , if @xmath641 is such that @xmath642 , then @xmath643 . as in case i , we will prove that @xmath532 lies in a free abelian subgroup of @xmath24 . + by case i , we have that there exists a free abelian subgroup @xmath644 such that if @xmath645 and @xmath646 , then @xmath647 . let @xmath648 , note that @xmath6 is a free abelian subgroup and @xmath649 , because @xmath650 . + let @xmath651 , @xmath652 and @xmath653 , since @xmath654 and @xmath655 commute , by definition of @xmath265 , it follows that @xmath656 and @xmath656 must commute , therefore @xmath657 is a free abelian subgroup . since @xmath532 was taken arbitrarily , we conclude that @xmath658 is contained in an unique maximal subgroup @xmath659 . + note that the free abelian subgroup does not depends on @xmath532 but only on @xmath658 .

the ranking problem is to order a collection of units by some unobserved parameter , based on observations from the associated distribution . this problem arises naturally in a number of contexts , such as business , where we may want to rank potential projects by profitability ; or science , where we may want to rank variables potentially associated with some trait by the strength of the association . most approaches to this problem are empirical bayesian , where we use the data to estimate the hyperparameters of the prior distribution , then use that distribution to estimate the unobserved parameter values . there are a number of different approaches to this problem , based on different loss functions for mis - ranking units . however , little has been done on the choice of prior distribution . typical approaches involve choosing a conjugate prior for convenience , and estimating the hyperparameters by mle from the whole dataset . in this paper , we look in more detail at the effect of choice of prior distribution on bayesian ranking . we focus on the use of posterior mean for ranking , but many of our conclusions should apply to other ranking criteria , and it is not too difficult to adapt our methods to other choices of prior distributions . empirical bayes ; posterior mean ranking ; choice of prior

1610.08779

suppose we have a collection of units we want to rank by a certain feature of each unit : for example , we may wish to rank genes by the risk they cause of a particular condition ; we may wish to rank sportsmen by their success - rate at particular standardised trials ; we may wish to rank business opportunities by the profit they will generate . this is a very common inference problem first studied as a formal statistical problem by bechhofer ( 1954 ) and by gupta ( 1956 ) . typically , for each unit we wish to rank , we will have some data on the associated feature , but will not know the true value of that feature . based on our data , we will have a point estimate for the feature , and an associated error distribution . the amount of data we might have for different units can vary wildly , meaning that the associated error distributions can be very different for different units . this means that when we select the top units using only our point estimates , the units for which we have largest errors have a higher chance of appearing among the top units , because a large error increases the chance of the point estimate being large . we are therefore likely to select a large number of false positives if we select based solely on the point estimates . we can illustrate this with a simple example . suppose we have 300 coins , we toss 100 of them six times each , 100 of them eight times each , and the remaining 100 of them ten times each , and rank the coins by the proportion of heads observed . if the coins are all fair , then among the 100 that we toss six times each , there is likely to be at least one that achieves 100% heads . among the 100 that we toss eight times , there might be one that achieves 100% heads , and there are likely to be several that achieve 87.5% heads . among the 100 that we toss ten times each , it is fairly unlikely than any will exceed 80% , so the highest ranked units will almost certainly come from among the coins that we toss only six times . that is , the highest ranked units are almost all false - positives arising only out of chance . this is still true , even if one or more of the coins that are tossed ten times have a slightly higher probability of heads . on the other hand , if our main aim is to avoid false positives , we could use a testing - based approach , where for each unit , we perform an hypothesis test of whether the unit has some null status for example whether the probability of heads is 0.5 . we can then rank by the @xmath0-values of these tests . this has the advantage of minimising false positives , but in many cases there are a large number of true positives , but only a few of them are truly important . if we apply the testing approach , we will often select the units on which we have collected most data , simply because the more data we have , the more evidence that they are not null cases . this may lead to neglecting some units which have much higher underlying value , but for which we have less data . other approaches to the problem mainly take a bayesian approach . they assume that the true values of the relevant feature fall under some distribution . we can estimate this underlying distribution from all the data points . then for each observed unit , we use this distribution as a prior to estimate a posterior distribution of the true value for this unit . we then perform our ranking based on these posterior distributions and a choice of loss function . there are a range of different methods based on different loss functions . for example , posterior expected rank ( laird and louis , 1989 ) use a loss function the squared difference between the true rank of a unit ( based on the actual value of the feature ) and the estimated rank . the @xmath1-values method ( henderson and newton , 2015 ) , corresponds to a loss function the sum of absolute differences between estimated rank and true rank . both of these loss functions are based entirely upon ranks , with no consideration of the actual true values . that is , they consider mis - ranking two units with almost identical true values to be as bad as mis - ranking units with very different true values . for the vast majority of practical ranking problems , this will not be the case . gelman and price ( 1999 ) present the interesting case of looking for spatial patterns among the top - ranked units , where artificial patterns can arise from patterns in available sample sizes . for their purposes , the ideal ranking method would be in such a way that the distribution of rank is the same for all values of standard error . for a known prior , it is possible to calculate this rank , though we are not aware of any work applying such a ranking method . however , methods with loss functions based only on rank , rather than value might be expected to perform better on this criterion , since all errors in ranking can cause this issue equally . the aim of a ranking analysis is often to maximise the average true value from the selected units . for instance , in the business profit example , the aim would be to maximise the expected total profit . for these purposes , the loss function is the difference between the largest true values and the true values of selected units . this loss function is introduced in gupta and hsiao ( 1983 ) , with some additional thought given to the situation where the loss is different for the case of omitting a variable that should be included , from the case of including a variable that should be omited . they show that for this loss function with known prior the bayes rule is to rank by posterior mean ( though they are not very explicit about this , and include some unnecessary hypotheses ) . this posterior mean ranking is used for example in aitkin and longford ( 1986 ) . a range of other loss functions have also been considered , for example , lin _ et al_. ( 2006 ) summarise a range of choices of loss function . for this paper , we will be focussing on the posterior mean ranking method , and its corresponding loss function , although many of our methods can be easily adapted to other bayesian ranking methods . the key difficulty in bayesian ranking methods is to choose the form of the prior . two common choices are the conjugate prior ( which for normal error is normal ) , and a non - parametric prior , which can be calculated using the results of laird ( 1978 ) . figure [ motivatingproblem ] shows the sort of problem that can arise with this approach . the lines on that figure show points that are ranked equally by posterior mean under a normal prior estimated from the whole dataset . as can be seen in that plot , a lot of emphasis gets placed on points with small variance . the reason is that the normal prior is light - tailed , so large true values are deemed implausible , and discounted . however , the true prior distribution seems to be more heavy - tailed than the normal , so larger values should not be discounted so much . for example , consider the point in the red circle . while it does have a larger standard error , it is very significantly non - zero , and it is likely that the true log - odds ratio is high . intuitively , we would probably want to rank this data point among the very top - ranked units . however , the posterior mean under the normal prior ranks it below a lot of other points which , while certainly significantly non - zero , have very small effect size . for practical purposes , this is not desirable . we are usually interested in units with a large effect size . the aim of this paper is to study the effect that choice of prior can have on the ranking problem , and determine suitable choices of prior for such analyses . despite a fair amount of literature on bayesian ranking methods , there has been a noticeable lack of work on the question of choice of prior . in view of the fact that selecting a suitable model for the prior distribution is a very difficult problem in model selection , it is important to consider the effects of a misspecified prior distribution . as will become apparent later , certain choices of prior are inherently more robust to misspecification than others . furthermore , some choices of prior are more sensitive to parameter estimation than others . we describe the objective more formally as follows . a ranking problem consists of a collection of units with unobserved parameters @xmath2 . for each unit , we have a point estimate @xmath3 for @xmath2 . we assume that @xmath3 is normally distributed with mean @xmath2 and variance @xmath4 , where @xmath5 is known . it is straightforward to adapt our approach to a number of other error distributions , but for this paper , we will focus on the normal error case . we assume that the unobserved values @xmath2 follow what we will refer to as the _ true prior _ distribution . we will rank by posterior mean using what we will refer to as the _ estimating prior _ , which may or may not be the same as the true prior . we are interested in how choice of the estimating prior affects the ranking . the structure of this paper is as follows : in section [ theorysection ] , we develop some theory behind posterior mean ranking , and the loss from using the wrong estimating prior . in section [ sectionisotaxes ] , we give a visual representation of the effect of choice of estimating prior on posterior mean ranking . in section [ sectionnonparametric ] , we show that using the non - parametric mle as an estimating prior for posterior mean ranking produces a robust ranking . in section [ simulations ] , we apply our theory to some examples of misspecified estimating priors , and perform a simulation study to confirm the results are as expected . we show that an exponential estimating prior is a good general - purpose choice for posterior mean ranking . in section [ sectionrealdata ] , we apply this to some real data examples where we show the difference in the ranking between using a normal distribution for the estimating prior and using an exponential distribution . in section [ sectionconclusions ] , we make some concluding remarks and suggestions for further investigations . we suppose that our true prior distribution is continuous and has density function @xmath6 . suppose that we have a point estimate @xmath7 , whose error distribution is normal with variance @xmath8 , where @xmath9 is small . since @xmath9 is small , values of @xmath10 that are far from @xmath7 are extremely implausible , and contribute little to the posterior mean for most choices of @xmath6 . we therefore focus on the form of @xmath6 for values of @xmath10 close to @xmath7 . taking a first order taylor expansion about @xmath7 gives @xmath11 using this approximation to @xmath6 gives that the posterior mean is @xmath12 this means that the key part of choice of estimating prior is to estimate the quantity @xmath13 . for the tail of the distribution , this quantity is positive , and asymptotically approaches the hazard rate . for an exponential distribution , it is constant . for heavier - tailed distributions it tends to zero as @xmath14 . for light - tailed distributions , it tends to infinity as @xmath14 . suppose we should estimate the posterior mean as @xmath15 , but in fact , we estimate it as @xmath16 , for some particular value of @xmath7 . the question is what is the average loss function resulting from this . for a ranking of all the observations , we can consider the total loss as the sum of losses due to individual mis - rankings . that is , suppose we rank the observations @xmath17},\theta_{[2]},\ldots,\theta_{[n]}$ ] , when the correct ranking is @xmath18 . if we choose our selection cutoff as the first @xmath19 units , then the loss function is @xmath20})\\ & = \left(\sum_{\substack{i\leqslant k\\\theta_{(i)}\not\in\{\theta{[1]},\ldots,\theta{[k]}\ } } } \theta_{(i)}\right)-\left(\sum_{\substack{j\leqslant k\\\theta_{[j]}\not\in\{\theta{(1)},\ldots,\theta{(k)}\ } } } \theta_{[j]}\right)\\\end{aligned}\ ] ] we can move from the correct ranking to the estimated ranking by a series of transpositions of adjacent units in the current ranking . for example , if the correct ranking is @xmath21 and the estimated ranking is @xmath22 , we can change from the correct ranking to the estimated ranking via the following sequence : @xmath23 for each such transposition , exchanging the position of @xmath24 in the @xmath25th postion , with @xmath26 in the @xmath27th position , the change in loss function is @xmath28 the total loss from this mis - ranking is then given by the sum of the loss functions for each transposition . we see that the loss for each transposition is non - negative for each value of @xmath19 , so we can analyse the overall loss of a misranking by looking at the loss of each pairwise misranking . if we consider the overall loss as the total of the loss functions for all values of @xmath19 , we see that this loss function is just the sum of the loss functions for each transposition . furthermore , whatever sequence of transpositions is performed , there will be one transposition for each misranked pair . therefore the total loss function is the sum of the losses from each misranked pair . we can therefore study the total loss function by studying the misranking loss for any pair of observations . in practice , we will often consider only the loss of the upper tail of the distribution . that is , we will choose some cutoff @xmath29 and evaluate the sum of the loss function for all @xmath19 such that @xmath30 . for this we have the following proposition ( proof in appendix [ appproofoflossfunction ] ) suppose the true prior distribution of the parameter @xmath10 has density function @xmath6 , and that we have two observations @xmath31 and @xmath32 which are normally distributed with means @xmath33 and @xmath34 and standard deviations @xmath35 and @xmath36 respectively , where @xmath33 and @xmath34 are random samples from the true prior distribution , and @xmath35 and @xmath36 are assumed to be small . 1 . the expected loss when the estimating prior and the true prior are the same ( which we will refer to as the _ optimal expected loss _ ) is approximately given by @xmath37 2 . when the estimating prior has density @xmath38 , the difference between the expected loss and the optimal expected loss is approximately given by @xmath39 where @xmath40 . the difference between the expected loss from using the point estimate @xmath3 and the optimal expected loss is approximately given by + @xmath41 we see that for @xmath42 , @xmath43 is bounded by @xmath44 , which is the expected information of @xmath10 , and is bounded for most distributions . this means that if the estimating prior is too heavy - tailed , we can do no worse than ranking by point estimators alone . on the other hand , if we have @xmath45 , then the integral can approach @xmath46 , which can be unbounded if the true prior has a heavy tail , but the estimating prior has a light tail . in most cases , the expression will not be unbounded . for example , for a normal estimating prior and a pareto true prior , we have that @xmath47 and @xmath48 , so @xmath49 which diverges whenever @xmath50 . thus for very heavy - tailed true priors , the loss from using a light - tailed estimating prior can diverge . we see that there is a risk of this unbounded loss whenever @xmath51 diverges . this can happen for any estimating prior with a lighter tail than an exponential distribution . we therefore suggest using an exponential distribution for the estimating prior to ensure the loss is not too great . this has the added mathematical convenience that the posterior mean is easily calculated as @xmath52 for some constant @xmath53 . if we use an improper exponential prior with density proportional to @xmath54 for all @xmath10 ( not just @xmath55 ) then this formula for the posterior mean is exact . indeed the posterior distribution is given by @xmath56 which is the density of a normal distribution with mean @xmath15 and variance @xmath8 . in this proposition , part ( ii ) gives the measure of the cost of using the wrong estimating prior . ( i ) and ( iii ) give measures of the overall difficulty of the ranking problem . ( i ) is the irreducible cost of misranking . ( iii ) is the additional cost from using the point estimates to rank , instead of using the posterior mean . it is an indication of the extent to which the ranking can be improved by using bayesian methods . henderson and newton ( 2015 ) describe different ranking methods in terms of the shapes of what they refer to as `` threshold functions '' , namely the functions @xmath57 which are the smallest value of @xmath7 , such that the observation @xmath58 is ranked in the top @xmath59 proportion under the ranking method in question . these threshold functions are curves joining points of equal rank : we will therefore refer these curves as _ isotaxes _ ( singular : _ isotaxis _ , from greek _ iso _ meaning equal , and _ taxis _ meaning rank ) . henderson and newton ( 2015 ) then describe their @xmath1-values procedure directly by calculating the shape of these isotaxes . we will examine the shape of the isotaxes as a method to better determine the effect of the estimating prior on ranking . for bayesian methods , the shape of these isotaxes depends heavily on the choice of estimating prior . for the normal estimating prior with mean 0 and variance @xmath60 , for an observation @xmath7 with standard error @xmath9 , the posterior mean is @xmath61 , so isotaxes are given by solutions to @xmath62 for constant @xmath63 , or to @xmath64 . when plotted on a graph of @xmath8 against @xmath7 , these are lines of varying slope , with shallower slope at higher ranks . ( indeed , these lines all pass through the point @xmath65 . ) for an exponential estimating prior with hazard rate @xmath53 , as mentioned above , the posterior mean is given by @xmath15 . the isotaxes are therefore given by the equation @xmath66 , or @xmath67 , so they are lines of constant slope . for a heavy - tailed distribution , recall that we have posterior mean approximately @xmath68 . therefore the isotaxes are functions of the form @xmath69 . a typical example is @xmath70 , so that @xmath71 . this means the isotaxes are curves of the form @xmath72 which gives a parabola . we plot the shapes of the isotaxes for these estimating prior distributions in figure [ fig_isotaxis ] . 0.33 0.33 0.33 we see that for the exponential and heavy - tailed estimating priors , the slopes of isotaxes are bounded away from zero , so the posterior mean can not be very far from the point estimate for @xmath7 . since by assumption , the true value also will not be so far from the point estimate , this means that the posterior mean can not be too far from the true value . from the shapes in figure [ fig_isotaxis ] , we see that for the normal estimating prior , the standard error becomes increasingly important as we move towards the tail of the distribution , and that the posterior mean can be arbitrarily far away from the true value . for the exponential estimating prior , the standard error remains equally important throughout . for the heavy - tailed estimating prior , the standard error becomes less important as we move to the tail of the distribution . furthermore , the standard error is most important for small standard error , and differences in standard error become less important as the standard error increases . it is also possible to calculate a non - parametric maximum likelihood estimate for the prior distribution . it was shown by laird ( 1978 ) that the prior in this case is a discrete distribution with finite support . an implementation of this non - parametric prior estimation is given in the ` rvalues ` package in ` r ` . however , this implementation is buggy , so we were unable to compare this method in section [ simulations ] . we show that for such a choice of estimating prior , provided the support of the prior distribution includes points sufficiently close to all the observed data , then the posterior mean estimators are robust . proofs of the following lemmas are in appendix [ nppriorproof ] . let @xmath73 be a discrete distribution with probability at least @xmath74 in the interval @xmath75 $ ] for some @xmath76 . let @xmath77 be the posterior mean for an observation @xmath7 with standard error @xmath9 . then @xmath78 this means that provided the prior distribution assigns some probability to a region near to each observed value of @xmath7 , then the posterior mean estimate will have some robustness to model misspecification . for a sample of @xmath79 datapoints and their corresponding standard errors , the non - parametric mle estimate for the prior distribution always assigns probability at least @xmath80 to the interval @xmath81 , for every observed data point @xmath82 . from the preceding lemmas , we conclude that ranking based on posterior mean under the non - parametric mle estimate for the prior is relatively robust , with @xmath83 we also know that for large @xmath79 , the non - parametric mle estimate is consistent , so the ranking will be optimal with the non - parametric mle . overall , we conclude that non - parametric estimation of prior provides a reasonable compromise between efficiency and robustness . however , as is typically the case with non - parametric methods , there is a trade - off between bias and variance . for the non - parametric method , the estimated ranking is asymptotically unbiassed , but can have fairly large variance for smaller sample sizes . figure [ npsimulationexample ] gives an illustration of this . 0.33 0.33 0.33 isotaxes for the upper tail of simulated data . 500 data points were simulated with the true means following a normal distribution with mean @xmath84 and variance 1 . variances for the observed data points are simulated following a gamma distribution with shape parameter 2 and scale parameter 0.1 . plot ( a ) shows the isotaxes for the non - parametric mle estimate for the prior distribution . plot ( b ) shows the isotaxes for an exponential estimating prior . plot ( c ) shows the isotaxes for the true prior . points are numbered according to their rank by posterior means under the true prior . note that some points are outside the region shown , hence the missing numbers . the isotaxes shown are the ones passing through observed data points . we can see that while the non - parametric approach has the isotaxes in approximately the right direction for larger variances , they are somewhat distorted for smaller variances . this is particularly observable at the tail , because the support of the mle ( which is discrete by laird ( 1978 ) ) is fairly sparse around the tail . this has a big effect on the posterior mean estimates for points with small standard error . however , it is worth noting that this distortion usually has limited influence on the estimated ranking . the reason for this is that the distortion is only for small standard error , compared with the standard error of the data points , so if some of the data points have small standard error , the isotaxes for posterior mean ranking based on the mle prior will be close to the correct isotaxes except for very small standard error . meanwhile , if the standard errors are large , the mle isotaxes will become further from the correct isotaxes , but not many of the observed data points will be included in this region where the isotaxes are far from optimal . the example given in figure [ npsimulationexample ] is a typical example where the non - parametric mle prior gives a poor ranking . there are other typical examples where the mle prior does not give such a poor ranking . let @xmath73 be a discrete distribution with probability at least @xmath74 in the interval @xmath75 $ ] for some @xmath76 . let @xmath77 be the posterior mean for an observation @xmath7 with standard error @xmath9 . then @xmath78 let the support of @xmath73 be the values @xmath156 , with probabilities @xmath157 . then for the posterior distribution of @xmath10 , the probability of @xmath156 is @xmath158 the posterior mean is therefore @xmath159 we see that the difference @xmath160 is maximised when the @xmath161 all have the same sign , which we will w.l.o.g . assume to be positive . it is clear that @xmath77 is maximised by setting all the @xmath161 in the interval @xmath162 $ ] to equal @xmath29 , since this both minimises the posterior probability of the interval @xmath163 $ ] and maximises the posterior mean conditional on lying in this interval . we will therefore assume that @xmath164 , and @xmath165 , then we have @xmath166 for fixed @xmath161 , and fixed @xmath167 , if @xmath168 then @xmath169 @xmath170 this gives that the contours are linear functions in @xmath73 , so the maximum value of @xmath63 occurs at a vertex with only one non - zero value of @xmath157 . the value of @xmath171 is maximised subject to @xmath164 , @xmath165 by setting @xmath172 , @xmath173 and choosing the value of @xmath174 to maximise the resulting quantity . in this case , we have @xmath175 substituting @xmath176 , @xmath177 , this expression becomes @xmath178 the derivative of @xmath179 with respect to @xmath180 is @xmath181 . we see that for @xmath182 , @xmath183 and @xmath184 , we have @xmath185 so @xmath179 is increasing . therefore @xmath186 is decreasing . this means that assuming @xmath187 , we have that @xmath188 for all @xmath182 . meanwhile , we always have @xmath189 , so we always have @xmath190 , and therefore @xmath191 for a sample of @xmath79 datapoints and their corresponding standard errors , the mle estimate for the prior distribution always assigns probability at least @xmath80 to the interval @xmath81 , for every observed data point @xmath82 . suppose the mle assigns probability @xmath157 to point @xmath161 . we will separate the points @xmath161 into points that are in the interval @xmath192 , and points that are not . suppose the first @xmath25 points are in the interval @xmath193 and the remaining points are not . we are aiming to show that @xmath194 . suppose this is not the case . we will then show that the distribution assigning probability @xmath157 to each point @xmath161 is not the mle by constructing a prior distribution with larger likelihood . let @xmath195 . let @xmath196 be the data set , and let @xmath197 be a data point . the log - likelihood of the data can be represented as @xmath198 , i.e. as the likelihood of the point @xmath82 plus the likelihood of the remainder of the data points . for a data point @xmath199 , we will use @xmath200 to represent the conditional likelihood of @xmath201 given that its corresponding value of @xmath10 is contained in @xmath193 , and @xmath202 for the conditional likelihood of @xmath201 given that its corresponding value of @xmath10 is not contained in @xmath193 . we have that the likelihood of @xmath201 is @xmath203 . if we change the prior to have probability @xmath59 at @xmath7 and @xmath204 times the previous prior , then the log - likelihood is larger than @xmath205 the increase in log - likelihood is therefore @xmath206 for this to be an increase , we need @xmath207 if we substitute @xmath208 , where @xmath209 , then @xmath210 for large @xmath79 . we therefore need @xmath211 now we know that @xmath212 , and @xmath213 , since the prior probability of the interval @xmath193 is at most @xmath135 , and for each point @xmath161 , the likelihood is @xmath214 so @xmath215 . also @xmath216 so @xmath217 . therefore , provided that @xmath218 we will have an improvement in likelihood . we see that as @xmath219 , @xmath220 . for small @xmath221 , the left - hand side is approximately @xmath222 . for the right - hand side , we are given that @xmath223 , so the right - hand side is less than @xmath224 so the required inequality holds . we use three simulation distributions for the priors ( both the true priors and the estimating priors ) : a normal distribution with known mean 0 and variance @xmath60 ; an exponential distribution with hazard rate @xmath53 ; and a pareto distribution with density function @xmath85 for @xmath86 where we take @xmath87 as known . ( we have taken one parameter as known for the normal and pareto distributions , so that each prior has one hyperparameter to be estimated . ) for the true priors in the simulation , we set @xmath88 for the normal distribution , @xmath89 for the exponential distribution and @xmath90 for the pareto distribution . for each simulation distribution , we simulate datasets of size 1000 , 10000 , and 100000 . we simulate the standard error @xmath9 for each data set as following an exponential distribution . we present results for the mean of this exponential distribution equal to 0.02 . results for mean 0.01 , 0.05 and 0.1 are presented in the supplementary materials . the values of @xmath9 are independent of the values of @xmath10 and values of @xmath9 for different data points are independent . to avoid some computational issues caused by values of @xmath9 too close to 0 , we added 0.0001 to all values of @xmath9 . we do not expect this to significantly impact the results , but we found that some numerical integration routines produced errors when the value of @xmath9 was very close to zero . for each simulated dataset , we analyse with each of the normal , exponential and pareto distributions as the estimating prior . we will assess the performance of the ranking by the average increase in the loss function from using the given estimating prior compared to using the true prior . that is , the loss function is : @xmath91})\ ] ] where @xmath24 is the true value of @xmath10 for the @xmath92th ranked unit under the true prior , and @xmath93}$ ] is the true value of @xmath10 for the @xmath92th ranked unit under the estimating prior . in order to better understand issues related to parameter estimation , we examine the loss function for both optimal parameter estimates ( based on minimising the expected loss function ) and estimated parameters ( estimated from the upper tail of the data ) . we do not compare the effect of estimating the hyperparameters from the whole data set because two of the distributions used for analysis had support only on the positive real numbers , so estimating these based on the whole data including negative values might lead to strange results . even if the supports were all the same , estimating the parameters for the estimating prior based on the whole data set when the focus is on the ranking of the top units leads to suboptimal results in ranking . .expected loss functions [ cols="<,<,<",options="header " , ] we look at several real data sets . these datasets were studied by henderson & newton ( 2015 ) for their work on @xmath1-values . the first data set consists of gwas data for log odds ratio between snps and type-2 diabetes from morris _ et al . _ ( 2006 ) . the data are available from http://diagram-consortium.org/downloads.html . the data consists of 137,899 snps from 12,171 type 2 diabetes cases and 56,862 controls . for each snp , an odds ratio is available along with a 95% confidence interval . following henderson & newton ( 2015 ) , we have taken the value as the log - odds ratio , assuming this estimate follows a normal distribution , and that the standard deviation of this distribution is one - quarter of the width of the log of the 95% confidence interval provided . the resulting positive data points and isotaxes for a normal and exponential estimating prior are shown in figure [ diabetesisotaxes ] . 0.35 0.35 0.35 from this figure , we see that using a normal estimating prior with naively estimated variance , the estimated variance is small , leading to isotaxes passing close to the origin . this makes the ranking focus on values with small variance , and rank values with larger observed value and larger variance behind values with smaller variance . the exponential estimating prior provides a ranking that selects many more of the points with large estimated effect size . we can improve the performance of a normal prior by artificially inflating the variance to match the tail better . figure [ diabetesisotaxes_c ] shows the effect of this . this does select a lot more of the points with large effect sizes , and indeed the top 1% isotaxis is very similar to the 1% isotaxis for the exponential estimating prior . on the other hand , the higher isotaxes do put too much weight on having smaller standard error , ranking a number of points with smaller estimated effect size ahead of the point with largest estimated effect size ( circled in red ) . it is extremely implausible that this ranking is correct . overall , the ranking based on an exponential estimating prior appears more plausible to us for this dataset . next we look at the gene expression data relating to breast cancer from west _ et al . _ this dataset is available in the rvalues package . the data set consists of gene expression measurements of 7129 genes accross 49 breast tumour samples 25 oestrogen receptor ( er)+ samples and 24 er@xmath94 samples . for each gene , the difference in means between the er@xmath95 and er@xmath94 groups is calculated , along with its appropriate standard error . in theory the error distribution should be modelled as a @xmath96-distribution with 47 degrees of freedom . however , for the purpose of this paper , we have used a normal distribution . the loss of accuracy should be fairly small . the resulting plot of variance against estimated effect size , and isotaxes for posterior mean with an exponential and a normal estimating prior are shown in figure [ bcisotaxes ] . 0.35 0.35 0.35 as for the diabetes data , we see that the normal estimating prior results in very flat isotaxes , and therefore gives a high ranking to observations with small variance . meanwhile , the exponential estimating prior puts a lot more weight on points with large estimated effect size . again , we see that using a normal estimating prior with inflated variance results in isotaxes that are more similar to the exponential estimating prior . in this case , the differences between the two rankings are not so clear - cut as the previous case , where some of the rankings using a normal estimating prior with increased variance were completely implausible . in this case , both the rankings for the exponential estimating prior and the normal estimating prior with inflated variance seem reasonable . we have seen that choice of estimating prior can have a very large effect on bayesian ranking methods . for the majority of ranking problems , we are particularly interested in ranking at the upper tail of the distribution . the ranking of the upper tail can be particularly affected by choice of estimating prior . using a light - tailed estimating prior for posterior mean ranking can lead to very bad results . if the true prior is heavy - tailed , the posterior mean can be far away from the truth . conversely , if the estimating prior is too heavy - tailed , the posterior mean estimated will be between the posterior mean under the true prior and the point estimate . this can not be too far from the true posterior mean . this means that using an exponential or heavier - tailed distribution as the estimating prior should be more robust to model misspecification . in addition to being less robust to model misspecification , light - tailed estimating priors can be more sensitive to estimated parameter values . in cases where the estimating prior is misspecified , using mle estimates for hyperparameters can also be far from optimal , so this can lead to bad results even in cases with large datasets . since we are usually particularly interested in the top units , it is usually advisable to choose parameter values that fit the tail of the distribution well . using a non - parametric prior is robust , in that there is an upper bound on how far the posterior mean can be from the posterior mean under the true prior . however , using a non - parametric prior can be inefficient for smaller sample sizes , and can lead to some strange rankings . we have confirmed our results by simulation studies and real data examples . in the simulation study , we found that an exponential estimating prior performed relatively well regardless of the true prior . our simulation study also studied the effect of estimating hyperparameters on the performance . as expected , estimating hyperparameters does cause some loss . the estimation in this simulation was done by maximum likelihood . however , since the loss function we are aiming to minimise is not the standard squared error loss , this is not the optimal estimation method . we based our parameter estimation on the upper 10% of the data points . it is common for analyses which use the whole data to estimate hyperparameters . doing this could lead to far worse results when the estimating prior is misspecified . even in the case where the true prior is not heavy - tailed , uncertainty in the hyperparameters can lead to a heavy - tailed prior when we apply model averaging over different possible values for the hyperparameters . we saw this effect in our simulation , when we looked at the results with estimated hyperparameters . particularly in the case where the prior distribution is not the true prior , estimation of hyperparameters can have a big effect on the estimated ranking . overall , unless there is good evidence otherwise , we suggest an exponential estimating prior will be a good compromise between robustness and efficiency in most cases . it also offers easy computation of posterior mean . the most obvious direction to need improvement in this research is hyperparameter estimation . we have seen in our simulation study that estimation by mle can lead to bad ranking results . this is because the loss function from prior misspecification is different from the loss function that mle estimation aims to minimise . this suggests that a different method of estimating the hyperparameters is needed a method specifically targeted at optimising ranking estimates . we know the loss function that we are aiming to minimise , so it should be possible to find an explicit way to solve this and derive a procedure for estimating the hyperparameter . given our recommendation to use an exponential estimating prior in most cases , finding the best hyperparameters should not prove too challenging a problem . our study has a number of limitations . we have considered only cases where ranking is by posterior mean and the error distribution is normal . in future work , we should study the problem for different error distributions , not just normal . further estimation is also needed into cases where the variance of the error distribution depends on the parameter @xmath10 . this can allow certain approximations to be applied . for example a poisson distribution can be approximated by a normal distribution where the variance depends on the mean . we should also study the problem for different methods and objective functions , e.g. @xmath1-values , posterior expected rank . we also have not considered the effect of model selection on ranking . if the estimating prior distribution is chosen based on certain model selection criteria , this may improve the ranking . however , model selection for mixture models can be difficult , so it might not provide the improvements we hope for . model selection also depends upon a good set of candidate models . our research suggests that the form of the function @xmath13 is most crucial in our choice of estimating prior , so including a sufficient range of models to allow flexibility in this function should allow us to obtain good ranking results , provided the model selection criteria are well chosen to be related to our objective function . suppose the true prior distribution of the parameter @xmath10 has density function @xmath6 , and that we have two observations @xmath31 and @xmath32 which are normally distributed with means @xmath33 and @xmath34 and standard deviations @xmath35 and @xmath36 respectively , where @xmath33 and @xmath34 are random samples from the true prior distribution , and @xmath35 and @xmath36 are assumed to be small . 1 . the expected loss when the estimating prior and the true prior are the same ( which we will refer to as the _ optimal expected loss is approximately given by @xmath37 2 . when the estimating prior has density @xmath38 , the difference between the expected loss and the optimal expected loss is approximately given by @xmath97 where @xmath40 . the difference between the expected loss from using the point estimate @xmath3 and the optimal expected loss is approximately given by + @xmath41 \(i ) suppose that the true parameter values are @xmath33 and @xmath34 respectively . let @xmath98 . now the loss from mis - ranking is @xmath99 if the points are mis - ranked and 0 if they are not misranked . the points are misranked if either @xmath100 and @xmath101 or if @xmath102 and @xmath103 . since the points will not plausibly be misranked if @xmath104 is large ( since @xmath31 and @xmath32 will then with high probability be far apart ) , we will assume that @xmath104 is small , so that we have @xmath105 . we will denote this common value @xmath53 . now for fixed @xmath33 and @xmath34 , suppose @xmath100 ; we want to calculate @xmath106 . we know that @xmath107 is normally distributed with mean @xmath104 and variance @xmath108 . therefore , @xmath109 on the other hand , if @xmath102 , we have @xmath110 now suppose that we fix @xmath33 , and we want to take the expected loss over the distribution of @xmath34 . this is given by @xmath111 since the probability of misranking is negligible for large @xmath104 , we will consider only small values of @xmath104 . for these small values , we can take the taylor expansion @xmath112 substituting this into the above loss function gives @xmath113 we recall that @xmath114_{-c}^\infty+\int_{-c}^\infty \frac{2\sigma}{\sqrt{2\pi}}\xi e^{-\frac{\xi^2}{2\sigma^2}}\,d\xi\\ & = \frac{1}{\sqrt{2\pi}}(\sigma c^2 + 2\sigma^3)e^{-\frac{c^2}{2\sigma^2}}\\ \int_{-c}^\infty \frac{\xi^2 e^{-\frac{\xi^2}{2\sigma^2}}}{\sqrt{2\pi}\sigma}\,d\xi&=\left[-\frac{\sigma}{\sqrt{2\pi}}\xi e^{-\frac{\xi^2}{2\sigma^2}}\right]_{-c}^\infty+\int_{-c}^\infty \frac{\sigma}{\sqrt{2\pi } } e^{-\frac{\xi^2}{2\sigma^2}}\,d\xi\\ & = \sigma^2\phi\left(\frac{c}{\sigma}\right)-\frac{\sigma c e^{-\frac{c^2}{2\sigma^2}}}{\sqrt{2\pi}}\\ \int_{-c}^\infty \frac{\xi e^{-\frac{\xi^2}{2\sigma^2}}}{\sqrt{2\pi}\sigma}\,d\xi&=\frac{\sigma e^{-\frac{c^2}{2\sigma^2}}}{\sqrt{2\pi}}\\ \int_{-c}^\infty \frac{e^{-\frac{\xi^2}{2\sigma^2}}}{\sqrt{2\pi}\sigma}\,d\xi & = \phi\left(\frac{c}{\sigma}\right)\\\end{aligned}\ ] ] hence we calculate @xmath115_0^\infty+\int_0^\infty \frac{\delta^3}{3}\frac{e^{-\frac{(c-\delta)^2}{2\sigma^2}}}{\sqrt{2\pi}\sigma}\,d\delta\\ & = \int_{-c}^\infty \frac{(\xi+c)^3}{3}\frac{e^{-\frac{\xi^2}{2\sigma^2}}}{\sqrt{2\pi}\sigma}\,d\xi\\ & = \frac{1}{3}\left(\frac{1}{\sqrt{2\pi}}(\sigma c^2 + 2\sigma^3)e^{-\frac{c^2}{2\sigma^2}}+3c\left(\sigma^2\phi\left(\frac{c}{\sigma}\right)-\frac{\sigma c e^{-\frac{c^2}{2\sigma^2}}}{\sqrt{2\pi}}\right)+3c^2\frac{\sigma e^{-\frac{c^2}{2\sigma^2}}}{\sqrt{2\pi}}+c^3\phi\left(\frac{c}{\sigma}\right)\right)\\ & = \frac{1}{3}\left((\sigma c^2 + 2\sigma^3 - 3c^2\sigma+3c^2\sigma)\frac{e^{-\frac{c^2}{2\sigma^2}}}{\sqrt{2\pi}}+(3c\sigma^2+c^3)\phi\left(\frac{c}{\sigma}\right)\right)\\ & = \frac{1}{3}\left((2\sigma^3+\sigma c^2)\frac{e^{-\frac{c^2}{2\sigma^2}}}{\sqrt{2\pi}}+(3c\sigma^2+c^3)\phi\left(\frac{c}{\sigma}\right)\right)\\ \int_{0}^\infty \delta\phi\left(\frac{c-\delta}{\sigma}\right)\,d\delta&=\left[\frac{\delta^2}{2}\phi\left(\frac{c-\delta}{\sigma}\right)\right]_0^\infty+\int_0^\infty \frac{\delta^2}{2}\frac{e^{-\frac{(c-\delta)^2}{2\sigma^2}}}{\sqrt{2\pi}\sigma}\,d\delta\\ & = \int_{-c}^\infty \frac{(\xi+c)^2}{2}\frac{e^{-\frac{\xi^2}{2\sigma^2}}}{\sqrt{2\pi}\sigma}\,d\xi\\ & = \frac{1}{2}\left(\left(\sigma^2\phi\left(\frac{c}{\sigma}\right)-\frac{\sigma c e^{-\frac{c^2}{2\sigma^2}}}{\sqrt{2\pi}}\right)+2c\frac{\sigma e^{-\frac{c^2}{2\sigma^2}}}{\sqrt{2\pi}}+c^2\phi\left(\frac{c}{\sigma}\right)\right)\\ & = \frac{1}{2}\left((\sigma^2+c^2)\phi\left(\frac{c}{\sigma}\right)+\frac{\sigma c e^{-\frac{c^2}{2\sigma^2}}}{\sqrt{2\pi}}\right)\\\end{aligned}\ ] ] in the loss function , we let @xmath116 and @xmath117 . substituting these into the loss function gives : @xmath118 if we let @xmath119 , so that @xmath120 , then we have @xmath121 ( for anyone thinking at this point that the dimensions do not work in this formula , it is worthwhile to remember that @xmath53 and @xmath73 are inversely proportional to changes in the scale of @xmath10 . that is , if we change the units so that the value of @xmath10 doubles , the values of @xmath53 and @xmath73 will be halved . ) with @xmath122 assumed to be small , we can neglect the @xmath123 term to get @xmath124 if we assume that @xmath125 is negligible , then our expression becomes @xmath126 we take the expectation of this over the distribution of @xmath33 to get that the expected loss is approximately @xmath127 we have assumed that @xmath128 is small with respect to this second term , so the expected loss is approximately @xmath129 . \(ii ) since @xmath35 and @xmath36 are small , we can assume that @xmath130 . we will let @xmath53 denote this common value . suppose that @xmath31 is the observed value of @xmath131 and @xmath32 is the observed value of @xmath132 , and that @xmath133 , but @xmath134 so that @xmath31 and @xmath32 are mis - ranked compared to the ranking under the true prior . let @xmath10 be the underlying parameter value for @xmath31 , and let @xmath135 be the underlying parameter value for @xmath32 . the expected increase in the loss function due to this misranking , compared to using the true prior , is then @xmath136 ( where @xmath137 and @xmath138 are the posterior distributions of @xmath10 and @xmath135 given observations @xmath31 and @xmath32 respectively , under the true prior ) . similarly , if @xmath139 , then the expected increase in loss is @xmath140 . now suppose we fix @xmath141 and take the expectation of the loss over @xmath132 . the expected loss due to mis - ranking them is @xmath142 where @xmath143 is the marginal density of @xmath32 . in the case where @xmath144 , we get this by calculating expected misranking loss over all values for which @xmath31 and @xmath32 are misranked ( compared to posterior mean ranking using the true prior ) . in the case where @xmath145 , calculating the expected misranking loss over all values where @xmath31 and @xmath32 are misranked gives @xmath146 and by reversing the limits and negating the integrand , we get the formula from equation [ eqexplossfixedx1 ] in this case also . since @xmath147 and @xmath148 are both small , we can assume that @xmath73 , @xmath149 and @xmath143 are approximately constant arround @xmath31 , so that @xmath150 for all @xmath32 in the relevant range . the integral is then approximately @xmath151 for the overall mis - ranking loss , we take the expectation of this over @xmath31 . we are usually particularly interested in the mis - ranking loss of the upper tail , that is the expected loss due to all misrankings in the upper tail , so we usually take the expectation over the distribution of @xmath31 for values @xmath152 for some chosen @xmath29 . this is given by @xmath153 \(iii ) we calculate this loss by substituting @xmath154 into our expression for the additional loss , we get @xmath155 recall that for a distribution with density function @xmath6 , we define @xmath225 , and that the best choice of estimating prior to use for ranking is chosen to minimise @xmath226 where @xmath227 and @xmath6 are for the true prior , while @xmath228 is for the estimating prior . we evaluate this loss function for each combination of priors . if the true prior is normal with mean 0 variance @xmath60 , and the estimated prior has mean 0 , variance @xmath229 , then the loss function is given by @xmath230_a^\infty}{4\pi}+\int_a^\infty \frac{e^{-\frac{\theta^2}{\tau^2}}}{4\pi}\,d\theta\right)\\ & = \left(\frac{1}{\tau^2}-\frac{1}{\hat{\tau}^2}\right)^2\left(\frac{a e^{-\frac{a^2}{\tau^2 } } } { 4\pi}+\frac{\tau}{4\sqrt{\pi}}\left(1-\phi\left(\frac{\sqrt{2}a}{\tau}\right)\right)\right)\\\end{aligned}\ ] ] for the normal estimating prior , we have @xmath238 . for the pareto true prior , we have @xmath239 . the objective function is therefore @xmath240_a^\infty\\ & = \frac{\alpha^2\eta^{2\alpha}}{a^{2\alpha+3}}\left(\frac{(\alpha+1)^2}{(2\alpha+3)}-\frac{2(\alpha+1)a^2}{(2\alpha+1)\hat{\tau}^2}+\frac{a^4}{(2\alpha-1)\hat{\tau}^4}\right)\\\end{aligned}\ ] ] we have @xmath242_a^\infty+\int_a^\infty e^{-\frac{\theta^2}{\tau^2}}\,d\theta\\ & = ae^{-\frac{a^2}{\tau^2}}+\sqrt{\pi}\tau\left(1-\phi\left(\frac{\sqrt{2}a}{\tau}\right)\right)\\ \int_a^\infty \frac{e^{-\frac{\theta^2}{\tau^2}}}{2\pi\tau^2}\,d\theta&=\frac{1-\phi\left(\frac{\sqrt{2}a}{\tau}\right)}{2\sqrt{\pi}\tau}\\\end{aligned}\ ] ] so the expected loss is @xmath243 if the true prior is exponential with rate @xmath53 , and the estimated prior is exponential with rate @xmath244 , then the loss function is given by @xmath245_a^\infty\\ & = \frac{\lambda\left(\lambda-\hat{\lambda}\right)^2}{2}e^{-2\lambda a}\\\end{aligned}\ ] ] for the normal estimated by pareto , we have @xmath247 . the loss function is therefore @xmath248_a^\infty+\int_a^\infty\frac{1}{2\tau^2 } e^{-\frac{\theta^2}{\tau^2}}\,d\theta\right)\\ & = \frac{1}{2\pi\tau^2}\left((\hat{\alpha}+1)^2\int_a^\infty \frac{1}{\theta^2}e^{-\frac{\theta^2}{\tau^2}}\,d\theta-\left(2\hat{\alpha}+\frac{3}{2}\right)\frac{\sqrt{\pi}}{\tau}\left(1-\phi\left(\frac{\sqrt{2}a}{\tau}\right)\right ) + \frac{a}{2\tau^2 } e^{-\frac{a^2}{\tau^2}}\right)\\\end{aligned}\ ] ] if the true prior is pareto with minimum @xmath250 and index @xmath59 , and the estimated prior is pareto with minimum @xmath250 and index @xmath251 , then the loss function is given by @xmath252_a^\infty\\ & = \left(\alpha-\hat{\alpha}\right)^2\frac{\alpha^2\eta^{2\alpha}}{(2\alpha+3)a^{2\alpha+3}}\\\end{aligned}\ ] ] for a normal true prior , we have that the loss from using the mle ranking is @xmath253_a^\infty+\int_a^\infty e^{-\frac{\theta^2}{\tau^2}}\,d\theta\right)=\frac{1}{4\pi\tau^4}\left(a e^{-\frac{a^2}{\tau^2}}+\sqrt{\pi}\tau\left(1-\phi\left(\frac{\sqrt{2}a}{\tau}\right)\right)\right)\\\end{aligned}\ ] ] for the pareto true prior , the expected additional loss from using the mle ranking is @xmath257_a^\infty=\frac{\alpha^2(\alpha+1)^2\eta^{2\alpha}}{(2\alpha+3)a^{2\alpha+3}}\\\end{aligned}\ ] ] substituting @xmath90 , @xmath87 and @xmath258 , we get the loss is @xmath259 so the best choice of @xmath244 for the exponential estimating prior is @xmath270 for the simulation setting @xmath88 , @xmath254 , this is @xmath271 and the expected loss for our simulation is 0.000622064 . integrating by parts gives @xmath282_a^\infty-\int_a^\infty \frac{1}{2\lambda\theta^2}e^{-2\lambda\theta}\,d\theta = \frac{e^{-2\lambda a}}{2\lambda a}-\frac{1}{2\lambda}\int_a^\infty \frac{1}{\theta^2}e^{-2\lambda\theta}\,d\theta\\\end{aligned}\ ] ] we will assume that @xmath29 is given for each simulation , and that our objective is to estimate the parameters from the data for each estimating prior so that the distribution fits the data well on the tail . we will use maximum likelihood for this purpose . we have already seen that the loss function is different from the kullback - leibler divergence that the mle estimate attempts to optimise , so the mle is not optimal in terms of minimising our expected loss function , and further work could go into devising better estimation methods for the misspecified prior case . for the mle estimation , the details in each case are presented here : we have @xmath79 samples which we model as having mean @xmath2 following a normal distribution with mean 0 and variance @xmath60 , and each observation @xmath3 following a normal distribution with mean @xmath2 and variance @xmath4 . we want to maximise the log - likelihood of all the data points with @xmath287 for some cuttoff @xmath29 . to simplify this procedure , we will maximise the log - likelihood of all data points for which @xmath288 . the log - likelihood is then written @xmath289 ( the last term is because we must take the conditional log - likelihood conditional on @xmath288 . ) setting the derivative with respect to @xmath265 to zero , we get @xmath290 we can solve this numerically using newton s method . we can use the following method to obtain a good starting value . since @xmath29 is reasonably large compared to @xmath265 , we can approximate @xmath291 [ note : this is a poor approximation . using it gives fairly bad approximations for @xmath234 . the approximations for other estimating priors later are better . ] so that the final term in the derivative of the log - likelihood is approximately @xmath292 we have assumed that @xmath5 is small compared to @xmath265 , so we can set we solve for when this is equal to zero using the quadratic formula to get : @xmath295 which should give an approximation to the true value of @xmath265 . if we further make the approximation that @xmath296 is small , then we have @xmath297 which gives us @xmath298 we can compare this approximate mle estimate of @xmath60 to the theoretically best estimate for the exponential and pareto cases . if we assume that @xmath5 are all small , then the term @xmath299 is approximately @xmath300 , which is a weighted mean of the @xmath4 . therefore the expected value is the expected value of @xmath4 , so we have @xmath301 since @xmath3 is normally distributed with mean @xmath2 and variance @xmath4 , we have that @xmath302 therefore we have @xmath303 for the exponential true prior , we have that conditional on @xmath287 , we have @xmath304 follows an exponential distribution with @xmath89 and @xmath256 . therefore @xmath305 therefore , for a large sample @xmath306 this is quite far from the optimal estimate of @xmath307 . for the pareto true prior , the variance is infinite ( since @xmath308 ) , so the distribution of the mle @xmath229 has infinite mean . this means we can not apply the law of large numbers to assert that for large sample size @xmath229 will converge in distribution to a constant . more specifically , @xmath309 follows a pareto distribution with @xmath310 and @xmath311 . the sum of pareto distributions with small @xmath59 is approximately equal to the maximum value , which has distribution function @xmath312 we also have @xmath313 we are interested in @xmath314 , because this is the value that is important for our posterior mean estimate . the survival function of @xmath314 is @xmath315 that is , @xmath314 approximately follows an exponential distribution with parameter @xmath316 . this can be quite different from the optimal @xmath317 . indeed we get @xmath318 since @xmath288 and @xmath5 is small , we can approximate @xmath321 so the log - likelihood is approximately @xmath322 however , we want the conditional log - likelihood given @xmath288 . since @xmath5 is small , we will set this approximately equal to the likelihood conditional on @xmath287 , which is @xmath323 setting the derivative with respect to @xmath53 to zero gives @xmath324 so the log - likelihood is maximised by @xmath325 ( the other zero is because the approximation @xmath326 does not hold for @xmath327 ) since @xmath5 is small , we can approximate @xmath328 which gives @xmath329 when the true prior is normal , we see that @xmath330 is the mean of a truncated normal distribution , and is given by @xmath331 substituting @xmath88 and @xmath332 , we get that @xmath333 therefore , for a large sample , our estimate @xmath244 will converge to @xmath334 . for the pareto true prior , we have @xmath335 . despite the fact that the variance is infinite , the law of large numbers still ensures that the sample mean of the @xmath3 does converge to @xmath336 as sample size tends to infinity . we can therefore substitute @xmath336 for this sum in the expression to get @xmath337 for the pareto estimating prior , the likelihood of @xmath2 is @xmath338 and the probability of a value exceeding @xmath29 is @xmath339 . the likelihood of @xmath319 is therefore @xmath340 letting @xmath341 , this integral becomes @xmath342 so the conditional likelihood of @xmath3 given that @xmath287 is approximately @xmath343 the conditional log - likelihood is therefore @xmath344 setting the derivative with respect to @xmath59 to zero gives @xmath345 @xmath346 setting the derivative with respect to @xmath59 to zero gives @xmath347 which has solution @xmath348 assuming that @xmath349 is small . we have the approximation @xmath350 which gives the mle @xmath351 for the normal distribution , if we let @xmath352 , then @xmath353 is normally distributed with mean @xmath354 and variance @xmath355 . we have @xmath356 . for @xmath288 , we have @xmath357 , and if @xmath358 is reasonably large , then nearly all values of @xmath353 will be less than @xmath359 . this means that @xmath360 can be well approximated by the taylor expansion @xmath361 we therefore have @xmath362 for the truncated normal with mean @xmath363 , variance 1 , truncated at @xmath364 , we have @xmath365 so letting @xmath366 and @xmath367 we see that @xmath368 follows a truncated normal distribution with mean @xmath363 , variance @xmath369 and lower limit @xmath364 , @xmath370 for our specific case , the normal true prior has @xmath88 and @xmath371 . empirically , for these parameters , @xmath372 , so @xmath373 , and @xmath374 . therefore , the expected value of @xmath251 is @xmath375 since @xmath5 follows an exponential distribution with @xmath376 , so @xmath377 , which means that @xmath378 we have @xmath382 . this gives @xmath383_0^\infty-\int_0^\infty \left(\log(1+x)+\frac{x}{1+x}\right)e^{-\lambda x}\,dx\\ & = -\int_0^\infty \frac{x}{1+x}e^{-\lambda x}\,dx\\ & = -\int_0^\infty \left(1-\frac{1}{1+x}\right)e^{-\lambda x}\,dx\\ & = \int_0^\infty \frac{e^{-\lambda x}}{1+x}\,dx-\frac{1}{\lambda}\\\end{aligned}\ ] ] on the other hand , integration by parts gives @xmath384_0^\infty+\int_0^\infty \frac{1}{1+x}e^{-\lambda x}\,dx\\ & = \int_0^\infty \frac{1}{1+x}e^{-\lambda x}\,dx\\\end{aligned}\ ] ] substituting this into the previous equation gives @xmath381 we have @xmath387 . this gives @xmath388 on the other hand , integration by parts gives @xmath389_0^\infty-\int_0^\infty \frac{e^{-\lambda x}}{(1+x)^2}\,dx\\ & = 1-\frac{g(\lambda)}{\lambda}\\\end{aligned}\ ] ] this gives us @xmath390 this means that for an exponential with parameter @xmath89 and cut - off @xmath256 , @xmath391 follows an exponential distribution with rate @xmath29 , so @xmath392 , so its expected value is @xmath393 , where @xmath394 is the solution to @xmath381 similarly , @xmath395 , so @xmath396 . the expected value of @xmath251 is then @xmath397 numerically , we find @xmath398 and @xmath399 . substituting these values into the equation gives @xmath400 henderson , n. c. and newton , m. a. ( 2015 ) , making the cut : improved ranking and selection for large - scale inference . journal of the royal statistical society : series b ( statistical methodology ) . doi : 10.1111/rssb.12131 morris , a. p. , b. f. voight , t. m. teslovich , t. ferreira , a. v. segre , v. steinthorsdottir , r. j. strawbridge , h. khan , h. grallert , a. mahajan , et al . large - scale association analysis provides insights into the genetic architecture and pathophysiology of type 2 diabetes . _ nature genetics 44 , _ 981990 . west , m. , blanchette , c. , dressman , h. , huang , e. , ishida , s. , spang , r. , zuzan , h , olson , j. a. jr . , marks , j. r. and nevins , j. r. ( 2001 ) , predicting the clinical status of human breast cancer by using gene expression profiles , _ proceedings of the national academy of sciences ( 98 ) _ , 1146211467

we study the-discriminant of toric varieties . we reduce its computation to the case of irreducible configurations and describe its behavior under specialization of some of the variables to zero . we give characterizations of dual defect toric varieties in terms of their gale dual and classify dual defect toric varieties of codimension less than or equal to four . sparse discriminant , dual defect varieties . + ams subject classification : primary 14m25 , secondary 13p05 .

math0510615

in this paper we will study properties of the _ sparse _ or @xmath0-discriminant . given a configuration @xmath1 of @xmath2 points in @xmath3 we may construct an ideal @xmath4 $ ] and , if @xmath5 is homogeneous , a projective toric variety @xmath6 . the dual variety @xmath7 is , by definition , the zariski closure of the locus of hyperplanes in @xmath8 which are tangent to @xmath9 at a smooth point . generically , @xmath7 is a hypersurface and its defining equation @xmath10 , suitably normalized , is called the @xmath0-discriminant . if @xmath7 has codimension greater than one then @xmath9 is called a dual defect variety and we define @xmath11 . the @xmath0-discriminant generalizes the classical notion of the discriminant of univariate polynomials . it was introduced by gelfand , kapranov , and zelevinsky ( their book @xcite serves as the basic reference of our work ) and it arises naturally in a variety of contexts including the study of hypergeometric functions @xcite and in some recent formulations of mirror duality @xcite . when studying the @xmath0-discriminant it is often convenient to consider a gale dual of @xmath0 . this is a configuration @xmath12 , where @xmath13 is the codimension of @xmath9 in @xmath14 . the configuration @xmath15 , and by extension @xmath0 , is said to be irreducible if no two vectors in @xmath15 lie on the same line . equivalently , if the matroid @xmath16 defined by the family , @xmath17 , of linearly independent subsets of @xmath15 is _ simple_. in theorem [ resformula ] , we prove a univariate resultant formula which reduces the computation of the @xmath0-discriminant to the case of irreducible configurations . this implies , in particular , that the newton polytope of the discriminant is unchanged , up to affine isomorphism , if we replace @xmath15 by the configuration obtained by adding up all subsets of collinear vectors . this generalizes a result of @xcite for codimension - two configurations . we point out that , in their case , this is a consequence of a complete description of the newton polytope of the discriminant . in the study of rational hypergeometric functions , one is interested in understanding the behavior of the @xmath0-discriminant when specializing a variable @xmath18 to zero and its relation to the discriminant of the configuration obtained by removing the corresponding point @xmath19 from @xmath0 . theorem [ specialization ] generalizes the known results in this direction ( ( * ? ? ? * lemma 3.2 ) ; ( * ? ? ? * lemma 3.2 ) ) . this specialization result was first proved by the first author in his phd dissertation @xcite , using the theory of coherent polyhedral subdivisions . we give a greatly simplified proof in 4 , where we derive the specialization theorem as a corollary of our resultant formula . using tropical geometry methods , dickenstein , feitchner , and sturmfels have been able to compute the dimension of the dual of a projective toric variety @xmath9 and this , in particular , makes it possible to decide if a given toric variety is dual defect , i.e. if the dual variety has codimension greater than one . their formula ( * ? ? ? * corollary 4.5 ) involves the configuration @xmath0 and the geometric lattice , @xmath20 , whose elements are the supports , ordered by inclusion , of the vectors in @xmath21 . the information contained in @xmath20 is essentially the same as that contained in a family of flats in @xmath22 , for a gale dual configuration @xmath15 of @xmath0 . thus , one could say that the formula by dickenstein , feitchner , and sturmfels involves both @xmath0 and @xmath15 information . in theorem [ goodflags ] , we use theorem [ specialization ] to show that we can decide whether a configuration is dual defect purely in terms of certain _ non - splitting _ flags of flats in the matroid @xmath22 . in theorem [ th : decomposition ] we obtain a decomposition of the gale dual configuration of a toric variety and give , in terms of this decomposition , a sufficient condition for the variety to be dual defect . although we believe this condition to also be necessary , we are not able to prove it at this point . dual defect varieties have been extensively studied : @xcite . in particular , dickenstein and sturmfels have classified codimension - two dual defect varieties @xcite and , by completely different methods , @xcite has classified dual defect projective embeddings of smooth toric varieties in terms of their associated polytopes . we give a complete classification of of dual defect toric varieties of codimension less than or equal to four in terms of the gale duals . this implies , in particular , that in these cases the condition in theorem [ th : decomposition ] is necessary and sufficient . we conclude 5 by comparing di rocco s list , for codimension less than or equal to four , with our classification . we begin by setting up the notation to be used throughout . we will denote by @xmath0 a @xmath23 integer matrix or , equivalently , the configuration @xmath1 of @xmath2 points in @xmath3 defined by the columns of @xmath0 . we will always assume that @xmath0 has rank @xmath24 and set @xmath25 , the _ codimension _ of @xmath0 . viewing @xmath0 as a map @xmath26 we denote by @xmath27 the kernel of @xmath0 . @xmath28 is a lattice of rank @xmath13 . for any @xmath29 we write @xmath30 , where @xmath31 have disjoint support . let @xmath4 $ ] be the lattice ideal defined by @xmath28 , that is the ideal in @xmath32 $ ] generated by all binomials of the form : @xmath33 , where @xmath34 . note that for any vector @xmath35 in the @xmath36-rowspan of @xmath0 we have @xmath37 for all @xmath38 and , hence , @xmath5 is @xmath39-weighted homogeneous . we will say that @xmath0 is _ homogeneous _ or nonconfluent if the vector @xmath40 is in the @xmath36-rowspan of @xmath0 . note that in terms of the configuration in @xmath3 , @xmath0 is homogeneous if and only if all the points lie in a rational hyperplane not containing the origin . throughout this paper we will be interested in properties of homogeneous configurations @xmath0 which depend only on the @xmath36-rowspan of @xmath0 . thus , in those cases we may assume without loss of generality that the first row of @xmath0 is @xmath40 . we shall then say that @xmath0 is in _ standard form_. given a homogeneous configuration @xmath0 , let @xmath41 be the projective ( though not necessarily normal ) variety defined by the homogeneous ideal @xmath5 . the map @xmath42 defines a torus embedding which makes @xmath9 into a toric variety of dimension @xmath43 . generically , its _ dual variety _ @xmath7 is an irreducible hypersurface defined over @xmath44 . its normalized defining polynomial @xmath45 is called the sparse or @xmath0-discriminant . it is well - defined up to sign . if the dual variety @xmath7 has codimension greater than one , then we define @xmath11 and refer to @xmath9 as a _ dual defect variety _ and to @xmath0 as a _ dual defect configuration_. note that @xmath9 , and consequently @xmath7 , depend only on the rowspan of @xmath0 . indeed , it is shown in ( * ? ? ? * proposition 1.2 , chapter 5 ) that @xmath9 depends only on the affine geometry of the set @xmath46 . alternatively , given a configuration @xmath47 we consider the generic laurent polynomial supported on @xmath0 : @xmath48 which , for a choice of coefficients @xmath49 , we view as a regular funcion on the torus @xmath50 . then , the discriminant is an irreducible polynomial in @xmath32 $ ] which vanishes whenever the specialization of @xmath51 has a multiple root in the torus ; i.e. @xmath51 and all its derivatives @xmath52 vanishing simultaneously at some point in @xmath53 . note that when @xmath0 is in standard form : @xmath54 and , consequently , @xmath51 and @xmath55 have the same zeroes on @xmath50 . let @xmath56[t^{\pm 1}]$ ] be the ring of laurent polynomials in @xmath57 whose coefficients are polynomials in @xmath58 , and denote by @xmath59 the ideal in @xmath60 generated by @xmath51 and its partial derivatives with respect to the @xmath57 variables . set @xmath61 . let @xmath62 be the zariski closure of the projection of @xmath63 in @xmath64 , then if @xmath62 is a hypersurface , @xmath65 . if @xmath0 is homogeneous and @xmath9 is not dual defect then @xmath62 is the cone over @xmath7 . we recall that if @xmath66 are a @xmath44-basis of @xmath28 , then the @xmath67 matrix @xmath15 , whose columns are @xmath68 is called a _ gale dual _ of @xmath0 . the same name is used to denote the configuration @xmath69 of row vectors of @xmath15 . gale duals are defined up to @xmath70-action . we will also consider @xmath67 integer matrices @xmath71 , whose columns @xmath72 are a @xmath36-basis of @xmath73 . in that case we will say that @xmath71 is a @xmath36-dual of @xmath0 . for any @xmath67 integer matrix @xmath71 of rank @xmath13 we will denote by @xmath74 the greatest common divisor of all maximal minors of @xmath71 and call it the _ index _ of @xmath71 . indeed , @xmath74 is the index of the lattice generated by the row vectors of @xmath71 , @xmath75 , in @xmath76 . an @xmath77 integer matrix @xmath0 of rank @xmath24 is said to be a dual configuration of @xmath71 if @xmath78 . note that @xmath71 is a gale dual of @xmath0 if and only if it has index @xmath79 and that , if @xmath0 is dual to @xmath71 , then @xmath0 is homogeneous if and only if the row vectors of @xmath71 add up to zero . such a configuration @xmath71 will also be called homogeneous . if @xmath80 for some @xmath81 , then any dual configuration @xmath0 is a _ pyramid _ , i.e. all the vectors @xmath82 are contained in a hyperplane . it is easy to check that in that case @xmath9 is dual defect . given an @xmath67 integer matrix @xmath71 of rank @xmath13 we will denote by @xmath83 the sublattice of @xmath84 generated by the columns of @xmath71 and by @xmath85 $ ] the lattice ideal defined by @xmath83 . if @xmath71 is a gale dual of @xmath0 , then @xmath86 and @xmath87 is a prime ideal . in any case , if @xmath88 are the columns of @xmath71 and we denote by @xmath89 the ideal @xmath90 then the lattice ideal @xmath91 is the saturation @xmath92 if @xmath71 is homogeneous of index @xmath74 then the variety @xmath93 has @xmath74 irreducible components and they are all torus translates of @xmath94 , where @xmath0 is a dual of @xmath71 . similarly , the dual variety @xmath95 is a union of finitely many torus translates of @xmath7 . in particular if one of them is a hypersurface so is the other . in that case , we denote by @xmath96 $ ] the defining equation suitably normalized . moreover , there exist @xmath97 such that @xmath98 where @xmath99 denotes component - wise multiplication . we will say that @xmath71 is dual defect if and only if @xmath0 is dual defect . the computation of the @xmath0-discriminant is well - known in the case of codimension - one homogeneous configurations . let @xmath100 , @xmath101 , be a gale dual of @xmath0 . reordering the columns of @xmath0 , if necessary , we may assume without loss of generality that @xmath102 for @xmath103 and @xmath104 for @xmath105 . set @xmath106 then , up to an integer factor @xmath107 we recall the notion of _ horn uniformization _ from ( * ? ? ? * chapter 9 ) . although in @xcite this is done only in the case of saturated lattice ideals , the generalization to arbitrary lattice ideals is straightforward . let @xmath108 be an integer matrix whose rows add up to zero , the _ horn map _ @xmath109 is defined by the formula @xmath110 where @xmath111 we also define @xmath112 by @xmath113 , where @xmath88 are the column vectors of @xmath71 , and set @xmath114 the following result is proved in ( * ? ? ? * chapter 9 , theorem 3.3a ) for the case of gale duals . its extension to @xmath36-duals is straightforward . [ th : gkz3.3 ] let @xmath115 be a homogeneous configuration and @xmath116 a @xmath36-dual of @xmath0 . then if @xmath7 is a hypersurface , so is @xmath117 . moreover , @xmath118 in this section we will study the effect on the @xmath0-discriminant of removing from the gale dual configuration @xmath15 a set of collinear vectors which add up to zero . we will show that this operation preserves the dual defect property and the newton polytope of the discriminant . moreover , there is a resultant formula relating the two discriminants . we shall assume throughout this section that our configurations are homogeneous . [ splittinglines ] let @xmath0 be a configuration in @xmath84 which is not a pyramid , and @xmath119 a gale dual . suppose we can decompose @xmath15 as @xmath120 where @xmath121 and @xmath122 are homogeneous configurations , @xmath121 is of rank @xmath13 , and @xmath122 is of rank @xmath79 . let @xmath123 be a dual of @xmath121 . then @xmath124 in particular , @xmath0 is dual defect if and only if @xmath123 is dual defect . let @xmath125 be a dual of @xmath122 . we may assume without loss of generality that @xmath123 and @xmath125 are in standard form . we may also assume that @xmath126 and @xmath127 . since the vectors in @xmath121 span @xmath76 over @xmath36 , there is a @xmath44-relation @xmath128 it is then easy to check that the matrix @xmath129 is dual to @xmath15 and , consequently , we may assume that @xmath0 agrees with the matrix ( [ specialform ] ) . we can write @xmath130 , where : @xmath131 and @xmath132 and view @xmath123 , @xmath125 as configurations in @xmath133 , @xmath134 , respectively . we let @xmath135 , @xmath136 , @xmath137 , and @xmath138 . given @xmath139 , we let @xmath140 . we define @xmath141 in an analogous way . if @xmath0 is as in ( [ specialform ] ) , @xmath142 and , therefore , @xmath143 in particular , we get a map @xmath144 given by @xmath145 . we also define @xmath146 by @xmath147 let @xmath148 , and let @xmath149 denote the natural projection . finally , define @xmath150 by @xmath151 . then the diagram @xmath152 commutes . we note that @xmath153 . indeed , the natural projection @xmath154 has finite fibers since , for any @xmath155 , @xmath156 . but @xmath125 is a codimension - one configuration and therefore its discriminant is given by ( [ codim - one ] ) . hence , @xmath157 must satisfy an equation of the form @xmath158 , for some @xmath159 , @xmath160 , and @xmath161 . we now claim that the conclusion of theorem [ splittinglines ] will follow from lemma [ gensurj ] , proved below , which asserts that @xmath162 is generically surjective with fibers of dimension @xmath163 . indeed , we have @xmath164 and , consequently , @xmath165 before proving the statements on generic surjectivity and fiber dimension , we prove an auxiliary lemma . [ divisor ] let @xmath0 be a @xmath166 integer matrix of rank @xmath24 with gale dual @xmath167 . let @xmath168 . suppose that for some @xmath169 , @xmath170 then @xmath171 let @xmath172 . then , the set @xmath173 agrees with the conormal space of @xmath9 at the point @xmath174\in x_a \subset { { \ensuremath{\mathbb{p}}}}^{n-1}$ ] . for each such @xmath175 we have , by assumption @xmath176 hence @xmath177 lies in the tangent space to @xmath9 at the point @xmath174 $ ] . since this tangent space equals the row span of @xmath0 , the result follows . [ gensurj ] under the hypotheses ( [ specialform ] ) , the map @xmath178 is generically surjective with fibers of dimension @xmath179 . to prove the first statement we show that @xmath180 is generically surjective . let @xmath181 and choose @xmath182 such that @xmath183 as noted above , for any choice of @xmath184 there are finitely many possible choices of @xmath157 satisfying ( [ fiber ] ) . we next choose @xmath185 such that @xmath186 . note that the assumption that @xmath125 is in standard form implies that if @xmath187 then so does @xmath188 , where @xmath189 for the given choice of @xmath157 , let @xmath58 be defined by @xmath190 therefore , @xmath191 . thus , it suffices to show that we can choose @xmath192 such that @xmath193 satifies @xmath194 but clearly @xmath195 where @xmath196 , and similarly for @xmath197 . lemma [ divisor ] and ( [ relation ] ) imply that we may assume without loss of generality that @xmath198 have been chosen so that @xmath199 . thus , if @xmath200 are so that @xmath201 then we can certainly choose @xmath192 so that ( [ lasteq ] ) holds and , consequently , @xmath202 is surjective outside the zero locus of @xmath203 . appealing once again to lemma [ divisor ] and ( [ relation ] ) , it follows that this zero locus does not contain @xmath204 which completes the proof of the first assertion . finally , we note that the remark after ( [ fiber ] ) implies the statement about the fiber dimension of @xmath162 . suppose now that we are under the same assumptions as in theorem [ splittinglines ] . that is , @xmath0 is a configuration in @xmath84 which is not a pyramid . @xmath119 is a gale dual of @xmath0 which may be decomposed as @xmath205 , where @xmath121 and @xmath122 are homogeneous configurations . @xmath121 is of rank @xmath13 , and @xmath122 is of rank @xmath79 . moreover , let @xmath123 be a dual of @xmath121 . we then have [ splittinglines : np ] if @xmath121 has index @xmath74 , then the newton polytope @xmath206 is affinely isomorphic to @xmath207 . by theorem [ splittinglines ] , @xmath208 if and only if @xmath209 , thus we may assume @xmath210 . let @xmath211 and suppose that that @xmath126 . we will then show that the projection @xmath212 on the first @xmath213 coordinates maps @xmath206 to @xmath207 . since both of these polytopes have the same dimension the result follows . note that since the vectors @xmath214 are all collinear and @xmath215 , we have , for all @xmath216 , that the product @xmath217 is a constant @xmath218 . hence , the defining equations @xmath219 of @xmath220 are related through @xmath221 by ( [ pullback ] ) , substituting @xmath222 by @xmath223 , @xmath224 , where @xmath225 is the the @xmath81-th column vector of @xmath15 , into @xmath226 gives the discriminant @xmath10 up to a laurent monomial factor . on the other hand , this same substitution in the right hand side of ( [ eq : np ] ) yields a polynomial in @xmath227 $ ] whose support equals that of @xmath228 . hence @xmath229 since , on the other hand , ( [ discc ] ) implies that @xmath230 , the result follows . [ btilde ] a configuration @xmath231 is called _ irreducible _ if any two vectors in @xmath15 are linearly independent . if @xmath0 is dual to an irreducible configuration @xmath15 , we shall also call @xmath0 irreducible . given a configuration @xmath15 we will denote by @xmath232 the irreducible configuration obtained by removing all vectors lying on splitting lines and replacing non - splitting subsets of collinear vectors in @xmath15 by their sum . [ matroid ] @xmath16 be the matroid defined by the family , @xmath17 , of linearly independent subsets of @xmath15 . then @xmath15 is irreducible if and only if @xmath22 is _ simple_. [ def : degenerate ] let @xmath233 be a configuration and @xmath234 a gale dual . @xmath15 is said to be _ degenerate _ if and only if @xmath235 . the following corollary may be viewed as a generalization of the results in @xcite . [ sumline ] let @xmath0 be a @xmath236 integer matrix of rank @xmath24 defining a homogeneous configuration . let @xmath237 be a gale dual of @xmath0 . let @xmath232 be as above . then @xmath238 and @xmath239 are affinely isomorphic . let @xmath240 denote the set of lines in @xmath241 containing vectors in @xmath15 . for each @xmath242 , let @xmath243 consider the configuration @xmath244 repeated applications of theorem [ splittinglines : np ] gives that @xmath245 . on the other hand we may also view @xmath71 as @xmath246 where @xmath247 . theorem [ splittinglines : np ] then implies that @xmath248 . we next show that , with the notation and assumptions of theorem [ splittinglines ] , there is a univariate resultant formula relating the discriminants @xmath249 and @xmath250 . [ resformula ] let @xmath0 , @xmath15 , @xmath123 , @xmath121 , and @xmath122 be as in theorem [ splittinglines ] and let @xmath125 be a dual of @xmath122 . assume moreover that @xmath121 consists of the first @xmath213 vectors in @xmath15 . then , there exist integers @xmath251 such that @xmath252 where @xmath253 , @xmath254 , @xmath255 , and @xmath99 denotes componentwise multiplication with @xmath256 and @xmath257 . if @xmath258 then @xmath259 by theorem [ splittinglines ] and ( [ eq : resformula ] ) is clearly true . suppose @xmath260 . let @xmath74 be the index of @xmath121 and let @xmath39 be a @xmath44-generator of the one - dimensional lattice @xmath261 . since @xmath15 has index @xmath79 , @xmath74 is the smallest positive integer such that @xmath262 we can find integers @xmath263 such that @xmath264 we may then assume that @xmath0 is as in ( [ specialform ] ) and therefore , since both @xmath123 and @xmath125 are in standard form , it follows from ( [ standard ] ) that if @xmath265 then the discriminants @xmath266 and @xmath267 vanish simultaneously for some @xmath139 . let @xmath268 be such that @xmath269 and @xmath270 are polynomials in @xmath157 with non - zero constant term . then there exists a polynomial @xmath271 such that @xmath272 the proof of theorem [ splittinglines : np ] implies that the degree of @xmath10 in the variables @xmath273 equals @xmath274 . on the other hand , the degree of the left - hand side of ( [ reseq ] ) is the @xmath157-degree of @xmath275 times @xmath276 . by definition of @xmath39 , we can write @xmath277 , @xmath278 , @xmath279 , and therefore @xmath280 but then it follows from the expression ( [ codim - one ] ) for the discriminant of a codimension - one configuration that @xmath281 hence both sides of ( [ eq : resformula ] ) have the same degree in the variables @xmath273 and , consequently , @xmath271 depends only on @xmath282 . suppose @xmath283 is not constant . we can write @xmath284 @xmath285 choose @xmath286 with @xmath287 then @xmath288 for all @xmath289 this means equations ( [ aux ] ) and ( [ aux2 ] ) are solvable in @xmath157 for all @xmath290 there are at most @xmath74 possible values for @xmath157 which solve ( [ aux2 ] ) , which means that ( [ aux ] ) must be the zero polynomial which is a contradiction since the monomials appearing in @xmath291 are distinct monomials of @xmath292 thus @xmath271 is a constant @xmath293 . [ coefficients ] we note that there are many possible choices for @xmath294 in theorem [ resformula ] . indeed , it suffices that @xmath295 and @xmath296 satisfy ( [ choiceofintegers ] ) and that @xmath297 and @xmath298 be chosen so that the products @xmath269 and @xmath270 be polynomials in @xmath157 with non - zero constant term . in fact , if we replace ( [ choiceofintegers ] ) by @xmath299 where @xmath300 , with @xmath301 a positive integer , then its effect is to make a change of variable @xmath302 in the resultant and therefore we would have : @xmath303 for suitable integers @xmath304 , @xmath305 . the following corollary which will be needed in the next section describes the effect on the discriminant of adding to the @xmath15 configuration a vector and its negative . [ plusminus ] let @xmath306 be a homogeneous configuration and let @xmath307 , @xmath308 , be a gale dual . let @xmath309 be a non - zero vector and let @xmath310 let @xmath311 be dual to @xmath312 . let @xmath175 and @xmath313 $ ] , @xmath314 $ ] , the discriminants associated with @xmath0 and @xmath311 , respectively . then @xmath315 since @xmath15 has index @xmath79 , we can write @xmath316 and setting @xmath317 , @xmath318 , we can apply ( [ eq : resformula ] ) and obtain @xmath319 for a suitable integer @xmath297 . we may specialize this resultant to @xmath320 , @xmath321 since that does not change the @xmath157-degrees of the polynomials involved and obtain : @xmath322 we end this section with a simple example to illustrate how we can use theorem [ resformula ] and corollary [ plusminus ] to reduce the computation of discriminants to that of irreducible configurations and univariate resultants . * example . * we work directly on the @xmath15 side and consider a configuration @xmath15 consisting of seven vectors @xmath323 , where @xmath324 @xmath325 the last @xmath326 vectors lie on a line @xmath327 and @xmath328 . as before , we set @xmath329 where @xmath330 and @xmath331 . we let @xmath332 denote variables ssociated with @xmath323 , respectively , and let @xmath333 , @xmath334 be associated with @xmath335 and @xmath336 . we note that @xmath121 and @xmath122 are homogeneous configurations satisfying the assumptions in theorem [ resformula ] and @xmath337 . following the notation of theorem [ resformula ] we have @xmath338 and therefore @xmath339 on the other hand , using singular @xcite we compute @xmath340 while clearly @xmath341 thus , given ( [ rel1 ] ) , we may apply theorem [ resformula ] with @xmath342 , @xmath343 and obtain @xmath344 according to corollary [ plusminus ] setting @xmath345 yields @xmath346 . since @xmath334 appears only raised to even powers , the expression for @xmath346 is obtained from that for @xmath347 erasing @xmath333 and @xmath334 . finally , note that if , instead of ( [ rel1 ] ) , we use the relation : @xmath348 then , as noted in remark [ coefficients ] @xmath349 the main result of this section is a specialization theorem for the @xmath0-discriminant generalizing lemma 3.2 in @xcite and lemma 3.2 in @xcite . in these references , the lemmata in question play an important role in the study of rational hypergeometric functions . we begin with a general result on the variable grouping in the @xmath0-discriminant . [ monomialgrouping ] let @xmath0 be a @xmath236 integer matrix of rank @xmath24 and @xmath350 a gale dual of @xmath0 . let @xmath10 , @xmath175 , be the sparse discriminant . then , if @xmath351 and @xmath352 , @xmath353 , are positive multiples of each other , @xmath354 define @xmath355 by @xmath356 , @xmath357 . it is clear that the initial form @xmath358 of @xmath249 relative to the weight @xmath359 agrees with the restriction @xmath360 . thus , it suffices to show that @xmath361 we recall ( * ? ? ? * chapter 10 , theorem 1.4 a ) that the secondary fan @xmath362 is the normal fan to the newton polytope @xmath363 of the principal @xmath364determinant ( we refer to ( * ? ? ? * chapter 10 ) for the definition and main properties of the principal @xmath364determinant ) . then @xmath365 if and only if @xmath359 and @xmath366 are in the same relatively open cone of @xmath362 . on the other hand , it follows from ( * ? ? ? * lemma 4.2 ) , that the linear map @xmath367 defines an isomorphism of fans between the secondary fan , @xmath362 , and its image , a polytopal fan @xmath368 defined on @xmath369 hence , ( [ initial2 ] ) holds if and only if @xmath370 and @xmath371 are in the same relatively open cone of @xmath372 but @xmath373 is a positive multiple of @xmath374 by assumption , so they must be in the same relatively open cone of @xmath372 since @xmath249 is a factor of @xmath375 by ( * ? ? ? * chapter 10 , theorem 1.2 ) , the normal fan of @xmath375 refines that of @xmath249 . then , any two vectors giving the same initial form on @xmath375 give the same initial form on @xmath376 this proves equation ( [ initial ] ) and concludes the proof of the proposition . as before , let @xmath47 be a homogeneous configuration in @xmath3 which is not a pyramid . for any index set @xmath377 , we denote by @xmath378 the subconfiguration of @xmath0 consisting of @xmath379 . let @xmath380 be a gale dual of @xmath0 . given a line @xmath381 , let @xmath382 and @xmath383 ( ) : = _ jj _ b_j . @xmath383 if @xmath384 is a non - splitting line , let @xmath39 be the @xmath44-generator of @xmath385 in the same direction as @xmath386 and , for @xmath387 write @xmath388 . we set @xmath389 and define @xmath390 accordingly . we may now prove the main result of this section [ specialization ] let @xmath0 be a homogeneous , @xmath166 integer matrix of rank @xmath24 , and let @xmath384 be a non - splitting line . then , for any @xmath391 , @xmath392 we may assume that @xmath393 , and let us denote by @xmath254 , @xmath282 . let @xmath394 and @xmath311 a dual of @xmath312 . as we have done before , let us denote by @xmath333 , respectively @xmath334 , the variable associated with @xmath386 , respectively @xmath395 . by corollary [ plusminus ] @xmath315 on the other hand , we can write @xmath396 , where @xmath397 let @xmath39 be a generator of @xmath261 so that @xmath398 with @xmath399 a positive integer . let @xmath74 be the index of @xmath121 . then we may write : @xmath400 thus , it follows from ( [ modifiedresformula ] ) that , up to constant , @xmath401 since we can choose @xmath402 . consequently @xmath403 on the other hand , let @xmath404 , @xmath278 , @xmath279 . then , since @xmath405 , @xmath406 where @xmath407 and @xmath408 are integers . it then follows that we may specialize @xmath409 , @xmath410 , in the resultant since that does not change the leading term of @xmath411 . hence , up to constants and monomials : @xmath412 but , since @xmath386 is the unique vector in the line @xmath384 in the configuration @xmath121 , it follows that @xmath413 is a non - facial circuit in @xmath123 and therefore by ( * ? ? ? * lemma 3.2 ) , @xmath414 divides @xmath415 and the result follows . in this section we apply the specialization theorem [ specialization ] and recent results in @xcite to prove , in theorem [ goodflags ] , a gale dual characterization of dual defect toric varieties . this leads to a classification of dual defect toric varieties of codimension less than or equal to four . motivated by this classification , we prove that the gale dual of a configuration may be decomposed as a disjoint union of non dual - defect configurations which are maximal in an appropriate sense . using this decomposition we give a sufficient condition for a configuration to be dual defect . we believe that this condition is necessary as well . indeed , it follows from theorems [ three - classif ] and [ four - classif ] , that this is the case for codimension less than or equal to four . throughout this section , we let @xmath0 be a homogeneous configuration of @xmath2 points in @xmath3 which is not a pyramid . we assume moreover that the elements of @xmath0 span the lattice @xmath3 . as always , if convenient , we will view @xmath0 as a @xmath23 integer matrix of rank @xmath24 . let @xmath9 denote the associated projective toric variety and @xmath416 its dual variety . let @xmath20 denote the geometric lattice whose elements are the supports , ordered by inclusion , of the vectors in @xmath21 . the following result is proved in @xcite using tropical geometry methods . let @xmath234 , @xmath419 , be a gale dual of @xmath0 and let @xmath16 be the matroid defined by the family , @xmath17 , of linearly independent subsets of @xmath15 . given a subset @xmath420 , the _ rank _ of @xmath421 is defined as the cardinality of the maximal element of @xmath17 completely contained in @xmath421 . a subset @xmath422 is called a @xmath301-_flat _ if it is a maximal , rank-@xmath301 subset of @xmath15 . clearly every subset @xmath423 spans a subspace @xmath424 whose dimension equals the rank of @xmath421 . a subspace @xmath425 is said to be @xmath15-spanned if @xmath426 . given a flat @xmath422 we denote note that @xmath435 and @xmath436 , so we will usually drop it from the notation . if @xmath430 is a non - splitting flag then , for all @xmath434 , @xmath437 is a @xmath15-spanned subspace and @xmath438 . moreover , @xmath439 projects to a non - splitting line in @xmath440 . clearly , the projection of a non - splitting @xmath301-flag @xmath430 to @xmath441 is a non - splitting @xmath442-flag in the configuration defined by the projection of @xmath15 . we prove the if direction by induction on the codimension @xmath13 . the result is obviously true for @xmath444 . assuming it to be true for configurations of codimension @xmath445 , let @xmath15 be a codimension @xmath13 configuration with a non - splitting @xmath443-flag @xmath446 . let @xmath447 denote the projection onto a rank @xmath445 lattice complementary to @xmath448 and let @xmath449 . clearly , @xmath450 is a non - splitting @xmath451-flag for @xmath452 , where @xmath453 . we recall that @xmath452 is a gale dual for the configuration @xmath454 . by induction hypothesis , @xmath123 is not dual defect and , by theorem [ specialization ] , the discriminant @xmath250 must divide an appropriate specialization of @xmath249 . hence @xmath0 is not dual defect . we also prove the converse by induction on the codimension @xmath13 . once again , the case @xmath444 is clear . we begin by considering the special case of a configuration @xmath0 with an irreducible gale dual @xmath15 . if @xmath0 is not dual defect , by theorem [ tropical ] , there exists a proper maximal chain in @xmath20 , @xmath418 , such that the matrix @xmath455 has rank @xmath456 . after reordering the columns of @xmath0 , and consequently the entries of @xmath457 , we may assume that @xmath458 with @xmath459 we claim that there exists an index @xmath460 , @xmath461 such that the matrix @xmath462 , obtained by removing the @xmath460-th row and the last column of @xmath463 , has rank @xmath464 . indeed , if the columns of @xmath462 are linearly dependent then , since the corresponding columns of @xmath463 are independent , it follows that the basis vector @xmath465 may be written as a linear combination of the first @xmath464 columns of @xmath463 . if this were true for every @xmath460 , @xmath461 , we could write the vector @xmath466 as a linear combination of the first @xmath464 columns of @xmath463 , a contradiction . we fix now an index @xmath460 , as above , such that @xmath467 . let @xmath468 be configuration obtained by removing the @xmath460-th column of @xmath0 . notice that the vectors @xmath469 obtained , also , by removing the zero in the @xmath460-th entry from the corresponding @xmath457 , define a proper maximal chain in @xmath470 . we then have , by theorem [ tropical ] , that @xmath468 is not dual defect and , therefore any gale dual @xmath421 of @xmath468 must contain a non - splitting @xmath451-flag @xmath471 . now , since @xmath15 is irreducible , @xmath421 agrees up to @xmath36-linear isomorphism with the projection of @xmath15 onto @xmath472 . then , denoting by @xmath473 the lifting of @xmath474 to @xmath241 , @xmath475 , and setting @xmath476 @xmath477 , we have that @xmath478 is a non - splitting flag of flats in @xmath15 . finally , consider the general case . that is , let @xmath0 be a non dual - defect configuration whose gale dual @xmath15 is not necessarily irreducible . as before , let @xmath232 be the irreducible configuration obtained from @xmath15 by replacing all subsets of collinear vectors in @xmath15 by their sum . note that @xmath232 need not have index one , but we may still consider a dual @xmath123 of @xmath232 . it follows from corollary [ sumline ] that @xmath479 . moreover , a gale dual @xmath480 of @xmath123 , being @xmath36-linearly isomorphic to @xmath232 , is irreducible . therefore @xmath480 has a non - splitting @xmath443-flag . but then so do @xmath232 and @xmath15 . note that , by theorem [ goodflags ] , if @xmath0 is not a pyramid and @xmath483 , then @xmath484 if and only if a gale dual @xmath15 has no non - splitting one - flags , i.e. if and only if every line is splitting or , equivalently , if @xmath485 . this classification of codimension - two dual defect toric varieties is contained in corollary 4.5 of @xcite . this observation may be generalized to the codimension - three case : [ three - classif ] let @xmath486 be a homogeneous configuration of codimension three , which is not a pyramid . let @xmath487 be a gale dual of @xmath0 . then @xmath208 if and only if @xmath15 is degenerate . by the above corollary and theorem [ goodflags ] it suffices to show that if @xmath15 is an irreducible configuration of rank three , then @xmath15 has a non - splitting two - flag . let @xmath488 and @xmath489 be distinct elements in @xmath15 and set @xmath490 be the two - flat containing @xmath491 . if , @xmath492 then we may assume @xmath493 and @xmath494 is a non - splitting two - flag . on the other hand , suppose every @xmath15-spanned plane @xmath495 satisfies @xmath496 . then , fixing an element @xmath497 , and denoting by @xmath498 the distinct @xmath15-spanned planes containing @xmath488 we would have that @xmath499 . but , we have assumed @xmath500 for all @xmath103 . hence @xmath501 , and this implies that @xmath502 , a contradiction . [ four - classif ] let @xmath486 be a homogeneous configuration of codimension four , which is not a pyramid . let @xmath503 be a gale dual of @xmath0 . then @xmath208 if and only if either @xmath15 is degenerate , or there exist planes @xmath504 , such that @xmath505 , and every non - splitting line lies either in @xmath506 or in @xmath507 . let @xmath0 be such that @xmath11 and suppose @xmath15 is non - degenerate . let @xmath232 be the irreducible configuration as in definition [ btilde ] . since @xmath15 is non - degenerate the vectors in @xmath232 span @xmath508 and , by corollary [ sumline ] , @xmath11 if and only if @xmath509 . thus , we may assume without loss of generality that @xmath15 is irreducible . we note that if @xmath510 , where @xmath121 and @xmath122 are homogeneous configurations contained in complementary planes @xmath506 and @xmath507 , respectively , then @xmath15 may not contain any non - splitting three - flags and , therefore , @xmath0 is dual defect . [ newniceflag ] let @xmath15 be a homogeneous configuration of rank @xmath13 and let @xmath511 be a line . suppose @xmath15 has a non - splitting flag of rank @xmath301 . then , @xmath15 has a non - splitting flag @xmath512 of rank @xmath301 such that @xmath513 . we prove the result by induction on @xmath301 , @xmath514 . the result is obvious for @xmath515 since @xmath516 and @xmath15 is homogeneous which means that the number of non - splitting one - flats in @xmath15 is either zero or at least three . assume it to be true for non - splitting flags of rank less than @xmath301 , and let @xmath430 be a non - splitting flag @xmath430 in @xmath15 of rank @xmath517 . we can assume that @xmath518 . consider the projection @xmath519 to @xmath520 . @xmath384 projects to a line @xmath521 in @xmath522 . moreover , the projection of @xmath430 defines a non - splitting flag of rank @xmath523 in @xmath519 . by inductive hypothesis there exists a non - splitting flag @xmath524 in @xmath519 of rank @xmath523 such that @xmath525 . let @xmath526 be the unique subspace of dimension @xmath527 containing @xmath528 and projecting onto @xmath529 , @xmath530 . notice that by construction @xmath531 but , since @xmath532 we have @xmath533 . setting @xmath534 , @xmath535 for @xmath536 , we get the desired non - splitting @xmath301-flag in @xmath15 . [ existence ] let @xmath233 be a homogeneous configuration of codimension @xmath13 and @xmath15 a gale dual . if @xmath15 is non - degenerate , then there exists a flat @xmath537 of rank @xmath445 such that @xmath538 . moreover , if we denote by @xmath539 the homogeneous configuration in @xmath540 defined by @xmath541 , then , if @xmath539 is non dual - defect , @xmath15 is not dual defect . if every flat of rank @xmath445 is homogeneous , let @xmath542 be the maximal rank of a non - homogeneous flat @xmath543 in @xmath15 . we have @xmath544 since @xmath15 is non - degenerate . choose a flat @xmath545 of rank @xmath546 with @xmath547 and let @xmath548 be the rank @xmath549 flats which contain @xmath545 . by assumption , @xmath550 for all @xmath103 . then , @xmath551 hence @xmath501 and therefore @xmath15 has rank @xmath549 . since @xmath552 this implies that @xmath15 is degenerate , a contradiction . suppose now that @xmath553 is not dual defect . by theorem [ goodflags ] , @xmath553 has a non - splitting flag @xmath512 of rank @xmath554 and , by lemma [ newniceflag ] , we may assume that @xmath555 . but then , @xmath556 is a non - splitting flag of rank @xmath445 in @xmath15 . applying theorem [ goodflags ] again we deduce that @xmath15 is not dual defect . [ distline2 ] let @xmath233 be a homogeneous configuration of codimension four and suppose a gale dual @xmath557 of @xmath0 is irreducible . suppose @xmath15 does not have any non - splitting three - flags and let @xmath543 be a rank - three flat with @xmath558 . then @xmath559 and the elements @xmath560 span a plane @xmath561 , with @xmath562 . let @xmath539 be as in lemma [ existence ] . since @xmath15 is dual defect so is @xmath553 and hence , by theorem [ three - classif ] , @xmath539 must be degenerate . since @xmath543 has rank three and @xmath15 is irreducible , this can only happen if @xmath559 , so that @xmath563 define a splitting line . the second assertion is then clear by theorem [ three - classif ] . we now return to the proof of theorem [ four - classif ] . because of corollary [ sumline ] and theorem [ goodflags ] , it suffices to prove that if @xmath564 is an irreducible , non - degenerate configuration which does not have any non - splitting three - flags , then @xmath510 , where @xmath121 and @xmath122 are homogeneous , rank - two configurations . let @xmath422 be a rank three flat with @xmath558 . by corollary [ distline2 ] , @xmath565 and @xmath121 is a rank - two flat with @xmath566 . let @xmath567 . we claim that @xmath122 does not have any non - splitting two - flags . indeed , suppose @xmath568 is a non - splitting two - flag . let @xmath569 . such @xmath488 exists since @xmath570 . then , letting @xmath571 be the smallest three - flat containing @xmath572 , we would have that @xmath573 would be a non - splitting three - flag in @xmath15 , contradicting our assumption . but , it is easy to see that the argument used in the proof of theorem [ three - classif ] implies that since @xmath122 is irreducible and has no non - splitting two - flags , it must have rank two and @xmath574 . since @xmath15 has rank four , the planes @xmath575 and @xmath576 must be complementary . [ th : decomposition ] let @xmath15 be a homogeneous , irreducible configuration of rank @xmath13 . then , we can write @xmath577 where the @xmath578 s are homogeneous , disjoint , non dual - defect subconfigurations of @xmath15 . moreover , @xmath578 is a flat in @xmath579 and the @xmath578 s are maximal with these properties . moreover , the rank of a non - splitting flag in @xmath15 is bounded by @xmath580 hence if @xmath581 , @xmath15 is dual defect . it follows from theorems [ three - classif ] and [ four - classif ] that the condition @xmath582 is a necessary and sufficient condition for a configuration @xmath15 , of rank at most four , to be dual defect . we expect this to be the case in general . this would give a complete classification of dual defect toric varieties in terms of their gale configuration . [ niceflag2 ] let @xmath583 be a homogeneous non dual - defect configuration of rank @xmath13 . suppose @xmath584 is a @xmath301-dimensional subspace , @xmath585 . then , @xmath15 has a non - splitting flag @xmath430 of rank @xmath445 such that @xmath586 . we proceed by induction on @xmath13 . the result is clear for @xmath587 . assume our statement holds for configurations of rank @xmath445 . let @xmath512 be a non - splitting flag of rank @xmath445 in @xmath15 . if @xmath588 we are done . assume then that @xmath589 and consider the projection @xmath519 to @xmath590 . then , @xmath519 is not dual defect and , by inductive hypothesis , there exists a non - splitting @xmath451-flag @xmath591 in @xmath519 such that @xmath592 . let @xmath593 be the unique subspace of @xmath594 containing @xmath595 and projecting to @xmath596 and set @xmath597 . note that @xmath598 since @xmath591 is non - splitting . now @xmath599 implies that @xmath600 . now , since @xmath490 is spanned by non - splitting one - flats , there exists a one - flat @xmath601 , with @xmath602 , and such that @xmath603 . the flag @xmath604 is a non - splitting flag in @xmath15 with @xmath605 . [ splitflat ] let @xmath15 be an irreducible , homogeneous , dual defect configuration and let @xmath384 a line in @xmath594 . then there exists a homogeneous , non dual - defect flat @xmath428 of rank @xmath301 , @xmath606 , such that @xmath607 . we proceed by induction on @xmath608 . if @xmath609 then , by theorem [ three - classif ] , there are no irreducible , non dual - defect configurations . so assume that @xmath610 and that the result holds for configurations of rank less than @xmath13 . let @xmath611 be the largest rank of a non - splitting flag in @xmath15 . we may assume that @xmath517 . otherwise , given any one - flat @xmath612 in @xmath15 , every two - flat containing it must be homogeneous , but this is impossible since @xmath15 is irreducible . moreover , by lemma [ newniceflag ] , we may assume that @xmath15 has a non - splitting @xmath301-flag @xmath430 such that @xmath613 . let @xmath614 be the distinct @xmath615-flats in @xmath15 containing @xmath616 . since @xmath617 , @xmath618 , and at most one @xmath615-flat may contain both @xmath619 and @xmath620 . hence may assume @xmath621 for @xmath622 . if @xmath623 for some @xmath622 , then we can take @xmath624 and we are done . if not , let let @xmath625 and @xmath626 . then @xmath627 is a homogeneous configuration of rank @xmath628 , which may or may not be irreducible . let @xmath629 be as in definition [ btilde ] . suppose @xmath630 . then , since @xmath15 is irreducible , @xmath631 is a homogeneous @xmath15-flat of rank @xmath301 which , we claim , is not dual defect . indeed , let @xmath81 be such that @xmath632 , we can define a non - splitting flag @xmath633 of rank @xmath523 in @xmath71 , by @xmath634 for @xmath635 and @xmath636 for @xmath637 . if , on the other hand , @xmath638 , then note that @xmath629 is dual defect . indeed , suppose @xmath629 has a non - splitting @xmath301-flag @xmath639 . then , by lemma [ newniceflag ] , we may assume without loss of generality that @xmath640 . but then @xmath641 would be a non - splitting flag of rank @xmath628 in @xmath15 , a contradiction . hence , by inductive hypothesis , @xmath629 has a homogeneous , non dual - defect flat @xmath71 of rank at least two and such that @xmath642 . therefore , @xmath71 is a flat in @xmath15 as well and the proof is complete . we return now to the proof of theorem [ th : decomposition ] . we prove the existence of ( [ eq : decomposition ] ) by induction on the rank @xmath13 . if @xmath587 then , being irreducible , @xmath15 is not dual defect and we may take @xmath643 . suppose the theorem holds for configurations of rank less than @xmath13 and let @xmath15 be an irreducible , dual defect configuration of rank @xmath13 . by lemma [ splitflat ] , there exists a homogeneous , non dual - defect , @xmath15-flat @xmath644 . we may assume that @xmath121 is not contained in any larger , homogeneous , non dual - defect @xmath15-flat and @xmath645 . let @xmath646 . clearly , @xmath480 is homogeneous and irreducible . if @xmath480 is not dual defect then taking @xmath647 we are done . on the other hand , if @xmath480 is dual defect and of rank less than @xmath13 , then we may apply the inductive hypothesis to write @xmath648 where the @xmath649 are maximal , homogeneous , disjoint , non dual - defect subconfigurations of @xmath480 and , for @xmath650 , @xmath578 is a flat in @xmath651 . finally , if @xmath652 , we repeat the argument and write @xmath480 as a disjoint union @xmath653 , where @xmath122 is a homogeneous non dual - defect @xmath480 flat . since at each step the cardinality of the remaining homogeneous configuration @xmath654 strictly decreases , it is clear that this process terminates . clearly , this would imply the result since the distinct flats among the @xmath660 , @xmath661 would define a non - splitting flag in @xmath578 whose rank would , therefore , be bounded by @xmath662 . to prove the claim we proceed by induction on @xmath663 . if @xmath664 , then we may assume @xmath665 and it suffices to choose @xmath666 and @xmath667 for @xmath668 . suppose now that we have constructed @xmath669 , @xmath670 and set @xmath671 . then @xmath672 is the disjoint union of the @xmath578-flats @xmath673 , for @xmath670 . let @xmath674 be the first index such that @xmath675 . such an index exists since @xmath430 is a non - splitting flag . since @xmath675 , there exists a @xmath676-flat @xmath677 such that @xmath678 and @xmath679 . set @xmath659 for @xmath680 . note that since @xmath681 @xmath682 must be strictly larger than @xmath683 . hence @xmath684 and , for dimensional reasons , this must be a direct sum . we have shown in theorem [ four - classif ] that if @xmath233 is a dual defect homogeneous configuration of codimension four , which is not a pyramid , and @xmath15 is a gale dual then either @xmath15 is degenerate or @xmath685 , and @xmath575 and @xmath576 are complementary planes . in this case , if @xmath686 is a dual of @xmath232 then @xmath686 is a union of homogeneous , codimension - two configurations lying in complementary subspaces of @xmath3 . similarly , if @xmath15 is a degenerate configuration consisting of vectors in a splitting line and in a complementary three - dimensional space , then @xmath0 is a union of two homogeneous configurations , of codimension one and three respectively , lying in complementary subspaces of @xmath3 . in either case , the projective toric variety @xmath9 is obtained from a join of two varieties by attaching codimension - one configurations according to ( [ specialform ] ) . more generally , if @xmath15 is decomposed as in ( [ eq : decomposition ] ) and @xmath0 is a dual of @xmath15 , then @xmath0 will be a cayley configuration of @xmath546 configurations @xmath687 in @xmath688 , where @xmath689 , in the following sense : in the special case when @xmath695 , as in theorem [ th : decomposition ] , is an irreducible configuration such that @xmath696 then , if @xmath233 is dual to @xmath15 , the toric variety @xmath9 is a join of varieties @xmath697 lying in disjoint linear subspaces and the dual variety @xmath698 has codimension @xmath546 . however , as the following example shows , for codimension greater than four , it is no longer true that every dual defect toric variety is obtained from a join by attaching codimension - one configurations according to ( [ specialform ] ) . * example . * let @xmath0 be the cayley configuration in @xmath699 , @xmath700 the variety @xmath9 is a smooth three - fold in @xmath701 . it is easy to show that a gale dual @xmath702 may be decomposed as @xmath703 , where @xmath578 is an irreducible , homogeneous , codimension - two configuration and , therefore , non dual - defect . let @xmath704 be as in ( [ eq : decomposition2 ] ) . then @xmath705 and , by theorem [ th : decomposition ] , @xmath15 is dual defect . in fact using theorem [ tropical ] one can show that @xmath698 is a six - dimensional subvariety of @xmath701 . di rocco has obtained a classification of dual defect projective embeddings of smooth toric varieties in terms of their associated polytopes @xcite . recall that a homogeneous configuration @xmath0 is said to be _ saturated _ if @xmath47 consists of all the integer points of a @xmath43 dimensional polytope with integer vertices , @xmath506 , lying on a hyperplane off the origin . moreover , the projective toric variety @xmath9 is smooth , if and only if the polytope @xmath506 is _ delzant _ , that is , for each vertex @xmath706 of @xmath506 , there exist @xmath707 such that @xmath708 is a lattice basis of @xmath3 , and @xmath709 near @xmath706 . it is well known that projective embeddings of smooth toric varieties are in one - to - one correspondence with delzant polytopes . [ dir - classif ] let @xmath0 be a saturated , homogeneous , configuration in @xmath3 which is not a pyramid and such that @xmath710 is delzant . then @xmath0 is dual defect if and only if @xmath711 where @xmath301 is such that @xmath712 , @xmath694 are saturated and the polytopes @xmath713 are all delzant polytopes of the same combinatorial type . thus , we see that the smoothness condition puts very strong conditions on the type of cayley configuration we may consider . to illustrate this , we will list all smooth dual defect projective toric varieties of codimension at most four . we note first of all that in these cases , the configurations @xmath714 in theorem [ dir - classif ] must be one - dimensional . in fact , let @xmath0 be a dual defect , saturated , homogeneous , configuration in @xmath3 which is not a pyramid and such that @xmath710 is delzant , and write @xmath715 as in theorem [ dir - classif ] . then , if @xmath716 , each polytope @xmath717 must be one - dimensional . indeed , let us consider the simplest case when the polytopes @xmath717 are two - dimensional . then @xmath718 and since by assumption @xmath719 , we must have @xmath720 . the fewest number of integral points in a delzant polytope in @xmath721 is three . hence @xmath722 and @xmath723 . let @xmath724 $ ] denote the configuration @xmath725 . an easy counting argument now shows that the smooth dual defect toric varieties of codimension less than or equal to four are the ones associated with the cayley configurations listed below : the gale duals of the configurations in the above list are easily computed . indeed , it is easy to see that each cayley factor @xmath727 $ ] contributes a splitting line containing two vectors from @xmath15 , and this vectors are primitive relative to the lattice @xmath76 . similarly , each factor @xmath728 $ ] contributes a homogeneous subconfiguration @xmath649 of rank @xmath301 and containing exactly @xmath628 primitive vectors in @xmath15 . thus , for example , in the codimension four case , the configuration @xmath726,[2],[2])$ ] has a gale dual @xmath15 whose reduced configuration @xmath232 decomposes as @xmath729 , where @xmath578 are homogeneous configurations of rank two , lying in complementary planes , and consisting of three primitive vectors each .

we consider a cosmological scenario in which a scale - invariant spectrum of curvature perturbations is generated by a rapidly - evolving equation of state on a slowly expanding background . this scenario generalizes the adiabatic ekpyrotic " mechanism proposed recently in arxiv:0910.2230 . whereas the original proposal assumed a slowly contracting background , the present work shows that the mechanism works equally well on an expanding background . this greatly expands the realm of broader cosmological scenarios in which this mechanism can be embedded . we present a phase space analysis and show that both the expanding and contracting versions of the scenario are dynamical attractors , with the expanding branch having a broader basin of attraction . in both cases , a finite range of scale invariant modes can be generated within the regime of validity of perturbation theory .

1104.4347

there is mounting observational evidence that the large scale structure originated from a nearly scale invariant and nearly gaussian spectrum of primordial density perturbations . while these statistical properties are consistent with the simplest inflationary models @xcite , a critical question for early universe cosmology is whether inflation is unique in making these predictions . this has motivated the quest for alternative scenarios , from the pre - big bang scenario @xcite , to string gas cosmology @xcite@xmath0@xcite , to ekpyrotic theory @xcite@xmath0@xcite . ( see @xcite for reviews of ekpyrotic cosmology . ) a zeroth - order benchmark for a successful theory of the early universe is to explain the observed homogeneity , isotropy and flatness of our universe . there are only two known cosmological phases that make the universe increasingly homogeneous , isotropic and flat . the first is of course inflation , characterized by accelerated expansion after the big bang . this requires a scalar field ( or fluid ) with negative equation of state @xmath1 . a second possibility is ekpyrosis , a phase of slow contraction before the big bang @xcite . this corresponds to a scalar field with stiff equation of state @xmath2 . the smoothing power of ekpyrotic contraction was recently confirmed using numerical relativity simulations @xcite . another benchmark is to generate a nearly scale invariant and gaussian primordial spectrum . but even this is not sufficient . inflation not only generates perturbations with the desired properties , but it does so within a cosmological background that is a dynamical attractor . indeed , on super - horizon scales the curvature perturbation on uniform - density slices @xcite@xmath0@xcite , @xmath3 , measures differences in the expansion history of distant hubble patches @xcite . since @xmath4 at long wavelengths in ( single - field ) inflation , the background is an attractor @xcite . achieving both scale invariance and dynamical attraction in alternative scenarios has proven challenging . the mode function equation for @xmath5 in a contracting , matter - dominated universe takes an identical form as in inflation @xcite ; but @xmath5 grows outside the horizon in this case , indicating an instability . this is not surprising , since an anisotropic stress component , for instance , will blueshift faster than dust . similarly , contracting mechanisms that rely on a time - dependent sound speed are inevitably unstable @xcite . the contracting phase in the original ekpyrotic scenario @xcite , with @xmath6 , is an attractor @xcite , as mentioned above , and correspondingly the growing mode for @xmath5 is a constant in the long wavelength limit . however , the resulting spectrum for @xmath5 is strongly blue @xcite , rather than scale invariant . a scale invariant spectrum can be obtained through entropy perturbations @xcite , as in the new ekpyrotic scenario @xcite , but this requires two scalar fields . even in this case , the entropy direction is generically tachyonically unstable @xcite . the recently - proposed _ adiabatic ekpyrotic _ mechanism @xcite , based on a single scalar field with canonical kinetic term , offers a counterexample . the mechanism can be realized with fairly simple potentials , such as the lifted exponential " v ( ) = v_0 ( 1-e^-c / m_pl ) , [ scalarpotential ] with @xmath7 and @xmath8 . this potential takes the form of a plateau that has been lifted to positive energy at large positive @xmath9 with a steep waterfall around @xmath10 . see fig . [ potentialcartoon ] . the key difference compared to earlier renditions of ekpyrotic cosmology is that the potential assumes _ positive _ values for part of the evolution . scale invariant adiabatic perturbations are generated during a transient phase in which the equation of state parameter , - = ( 1+w ) , grows rapidly from @xmath11 , where the constant term dominates , to @xmath12 , where the negative exponential term dominates . during this transition , the scale factor is nearly constant , while the equation of state parameter varies rapidly as @xmath13 , where @xmath14 is conformal time . this is referred to as the _ transition phase_. the quantity @xmath15 , which determines the evolution of @xmath5 , therefore scales as @xmath16 exactly as in inflation , where @xmath17 and @xmath18 ! the two - point function is , therefore , identical to inflation . ( another counterexample proposed recently relies on a rapidly - varying , superluminal sound speed @xmath19 @xcite . see @xcite for earlier related work . ) . at large field values the potential is nearly constant , and there is a steep waterfall around @xmath20.,width=384 ] more generally , it has recently been shown that in fact there are only three single - field cosmologies with unit sound speed capable of generating a scale invariant spectrum for @xmath5 on an attractor background @xcite : @xmath21 inflation , with @xmath22 and @xmath23 ; @xmath24 the adiabatic ekpyrotic mechanism @xcite mentioned above , with @xmath25 on a slowly contracting background ; and @xmath26 a novel version of the adiabatic ekpyrotic mechanism , in which the background first slowly expands , then slowly contracts . the analysis has been generalized to the case of a time - dependent sound speed @xcite . see @xcite for related work . although all three cosmologies yield identical two - point function , the degeneracy is broken by the three - point function . in contrast with the extremely gaussian spectrum of the simplest inflationary models , the rapid variation of @xmath27 in the adiabatic ekpyrotic mechanisms leads to strongly scale - dependent non - gaussianities , which peak on small scales @xcite . for the lifted exponential potential ( [ scalarpotential ] ) , this results in a breakdown of perturbation theory on small scales , both classically and quantum mechanically @xcite . as shown in @xcite , however , these pathologies , and related issues raised in @xcite , can be avoided by considering more general forms of the potential , such as v()=v_0(1-e^-c()/m_pl ) , [ genpot ] where the exponent decreases from @xmath28 to a much smaller value @xmath29 after an acceptable range of scale invariant modes has been generated . the resulting spectrum is perturbative on all scales , but , because of the modified evolution , is now scale invariant over a finite range of modes , spanning a factor of @xmath30 in @xmath31 space , or a dozen e - folds . while limited , this range is sufficient to account for observations of the cosmic microwave background and the large scale structure . earlier analyses of the adiabatic ekpyrotic mechanism have focused on the case where the universe is contracting throughout . in this paper , we instead explore the phenomenology and implications of the expanding case , consisting of an expanding transition phase " , followed by a contracting ekpyrotic scaling phase . in doing so , we are motivated by the issue of embedding this mechanism in broader cosmological scenarios . specifically , one attractive feature of the expanding branch is that we expect its basin of attraction to be much broader than in the original contracting version . indeed , while the contracting transition phase in the original adiabatic ekpyrotic mechanism is a dynamical attractor , the evolution prior to the transition phase is generally not . at sufficiently early times , the field lies on a flat plateau of its potential , and , because the universe is contracting , any additional amount of kinetic energy is blueshifted and threatens to dominate the energy density . in the initially expanding version studied here , any additional kinetic energy will instead redshift away , and we therefore expect the solution to be stable for all times . after reviewing the contracting version of the adiabatic ekpyrotic mechanism in sec . [ contractadiabatic ] , we show in sec . [ expanding ] that this mechanism works equally well on a slowly expanding background and generates a scale invariant spectrum of curvature perturbations . since scale invariance relies on a rapidly - varying equation of state parameter while the scale factor is nearly constant , the density perturbation spectrum is to a good approximation insensitive to whether the background is expanding or contracting during mode production . we check this explicitly in sec . [ numver ] by numerically integrating the mode function equation in both the expanding and contracting cases . the resulting spectra are indistinguishable ( see fig . [ analyticzeta ] for a preview ) . section [ rest ] revisits the analytic calculation of the background evolution , allowing for more general initial conditions . we quantify the extent to which the scale invariant phase shortens as a function of initial scalar field kinetic energy . in sec . [ phasespace ] , we study the issue of stability globally by performing a phase space analysis , including a wide range of initial conditions , both for the expanding and the contracting branch . this confirms that the transition phase is an attractor in both cases , as indicated at the perturbative level by @xmath5 having a constant growing mode as @xmath32 . more broadly , this analysis also shows that the basin of attraction is broader in the expanding case , as anticipated in the previous paragraph . in sec . [ fusion ] we study the connection to an inflationary precursor to the transition phase , which follows in the expanding case from trusting ( [ scalarpotential ] ) at sufficiently large field values . we briefly summarize our main results and discuss prospects for future directions in sec . [ conclu ] . the expanding adiabatic ekpyrotic mechanism studied here suffers from the same strong coupling issues as the original contracting version . in particular , the calculation of the three - point amplitude in @xcite neglected the time - dependence of the scale factor , and hence should apply to the expanding case as well . ( one technical difference in our case is that @xmath27 momentarily blows up at the end of the transition phase , because @xmath33 at that time . as the numerical analysis of sec . [ numver ] clearly shows , however , the two - point function is insensitive to this momentary divergence . we expect that the same is true for the three - point function . ) in particular , just like in the contracting branch , strong coupling can be avoided by considering more general forms of the potential , such as ( [ genpot ] ) . we will not repeat the discussion of non - gaussianities in this paper , and we refer the interested reader to @xcite for more details . it is instructive to review the mechanism presented by @xcite in which scale - invariant curvature perturbations are generated by a fast - rolling scalar field during a slowly contracting phase . consider a canonical scalar field coupled to gravity s = d^4 x(-g^__-v ( ) ) , [ scalaraction ] where @xmath34 . for concreteness , we focus on the lifted exponential " potential ( [ scalarpotential ] ) . this form of the potential is simplest for analytical calculations but is not meant to represent a realistic scenario . indeed , to avoid strong coupling issues @xcite , realistic models must consider a more general potential of the form @xmath35 , where @xmath36 decreases to a smaller value after a suitable range of scale invariant modes has been generated . see @xcite for details . since the production of these scale - invariant modes occurs as the field traverses a small range @xmath37 , there is a reasonable amount of freedom in specifying the global form of the potential . the evolution of the scalar field on a cosmological background is governed by the equation of motion + 3h+v_,=0 . [ scalareom ] during the phase of interest , the background evolves extremely slowly so we may ignore hubble friction . as shown in @xcite , this approximation can be rigorously justified _ a posteriori _ by explicitly computing the first order correction and verifying that it is indeed small . intuitively , the field falls down a steep waterfall , hence the transition occurs within a hubble time . fast roll " approximation leads to the simplified equation -v_,=-v_0e^-c , [ scalareomapprox ] which is solved by ( t)(c|t| ) , [ phisol ] where @xmath38 . in writing down this solution , we have assumed that the field initially has negligible kinetic energy at early times , hence the total energy is @xmath39 . in sec . [ rest ] , we will consider departures from this choice and the impact on the spectrum of perturbations . to solve for the evolution of the background , we substitute ( [ phisol ] ) into @xmath40 to obtain h(t)=-h_0 . at sufficiently early times , the hubble parameter is approximately constant , @xmath41 , where the integration constant @xmath42 is fixed by the friedmann equation , 3m_pl^2h^2_0 ^2 + v ( ) v_0 . [ fried ] the authors of @xcite focused on the contracting branch , h_0= . indeed , since @xmath43 and @xmath44 is negative , @xmath45 is manifestly negative , corresponding to a contracting universe . ( note that the present sign convention differs from @xcite , where @xmath46 and @xmath42 is a negative quantity . ) the phase of interest is the regime where the @xmath42 term dominates the hubble parameter , @xmath47 , in which case @xmath48 . additionally , the assumption that the background remains nearly static implies @xmath49 . the quantity @xmath15 , which determines the evolution of @xmath5 , therefore satisfies za~ , exactly as in inflation , where @xmath17 and @xmath18 . moreover , as in inflation , the growing mode for @xmath5 goes to a constant , indicating that the background is a dynamical attractor @xcite . the two cosmologies yield identical power spectra , and therefore can be considered dual " to one another at the level of the two - point function . the transition phase ends when @xmath50 is no longer a good approximation . subsequently , the hubble parameter tends to an ekpyrotic scaling h_ek(t ) , [ hek ] while the scale factor slowly decreases as a_ek(t ) ~(-t)^2/c^2 . [ aek ] this ekpyrotic scaling phase is a necessary component of the story . because the hubble horizon is nearly static during the transition phase , the scale invariant modes created in this phase remain inside the hubble radius . a subsequent phase is therefore necessary to push these modes outside the hubble horizon . the scaling ekpyrotic phase fills this role since the universe is slowly contracting , modes are gently pushed outside the hubble horizon without disturbing their spectrum . ( eventually , the ekpyrotic phase must itself terminate before the big crunch . we envision that it is followed by a bounce to an expanding , radiation - dominated phase . at the level of a four - dimensional effective theory , a stable non - singular bounce can be achieved either through a phase of ghost condensation @xcite , as in the new ekpyrotic scenario , or through a phase of galileon domination @xcite . see @xcite for recent supersymmetric extensions of these theories . ) the key element of the adiabatic ekpyrotic mechanism is the rapid evolution of the equation of state parameter @xmath27 ; this growth is responsible for the generation of a spectrum of adiabatic modes . interestingly , the scale factor remains nearly constant in the process , and hence plays no essential role . as a result , we expect that the adiabatic ekpyrotic mechanism can be generalized to a case where the scale factor is slowly _ increasing_. in this section we show that this is indeed possible . though this may seem like an academic exercise , there are important implications in extending the original set - up , especially for embedding this mechanism in broader cosmological scenarios . as reviewed above , in the contracting adiabatic mechanism the transition phase is a dynamical attractor @xcite . however , the cosmological evolution prior to the onset of the transition phase is not necessarily an attractor solution . in fact , we know that at the earliest times when the field lies on the potential plateau , the background cosmology must be unstable any additional kinetic energy in the field will blueshift and can lead to kinetic domination . in an initially expanding universe , the kinetic energy will instead redshift . in this way , we expect an expanding transition phase solution to be stable for all time . the construction of the solution is quite similar to the contracting case . as before , we may ignore the background evolution to first order so the scalar field equation of motion reduces to ( [ scalareomapprox ] ) , with a solution given again by ( t)(c|t| ) , [ phisol2 ] where @xmath38 . since the cosmological background is to first order irrelevant for the scalar field evolution , it is natural that the approximate solution ( [ phisol ] ) applies irrespective of whether the universe is contracting or expanding . in the same way as before , we can integrate the @xmath51 equation to obtain h(t)=+h_0 . [ h(t ) ] the friedmann equation ( [ fried ] ) constrains the magnitude of the integration constant @xmath42 as before , h_0= . [ h0 ] note this time we have chosen the positive - definite quantity @xmath42 to appear as @xmath52 in ( [ h(t ) ] ) . hence , at sufficiently early times , @xmath53 is approximately constant and positive , corresponding to an expanding de sitter universe . the expanding transition solution will be similar to the contracting case as long as @xmath53 , we will have @xmath54 , and scale invariant perturbations will be generated . however since hubble is positive during this time , perturbations are generated on a slowly _ expanding _ background . the transition phase is once again followed by an ekpyrotic scaling phase , with @xmath55 . during this phase , the hubble radius shrinks slowly , and the scale invariant modes are gently pushed outside the horizon . there is , however , a key difference : in the expanding case , @xmath45 changes sign as the universe evolves from the ( expanding ) transition phase to the ( contracting ) ekpyrotic scaling phase . by definition , the transition phase proceeds as long as the hubble parameter is nearly constant and the hubble friction term can be neglected in the scalar field evolution . these conditions are only satisfied for a finite time . first note that the constant term dominates in the expression for @xmath45 until t_end - tran = t_beg - ek = - , [ tend ] which corresponds to the time when @xmath56 vanishes , and the universe transitions from expansion to contraction . as the subscripts indicate , this marks the end of the transition phase and the onset of the ekpyrotic scaling phase . likewise , the transition phase is also finite in the past . the solution ( [ phisol2 ] ) for @xmath57 neglected gravity , which is a poor approximation for sufficiently large positive @xmath9 where the potential is flat and hubble damping is important . more precisely , it is straightforward to show that the approximation @xmath58 used in ( [ scalareomapprox ] ) is consistent as long as @xmath59 , where t_beg - tran = - . [ tbeg ] summarizing , the transition phase solution is valid for -= t_beg - tran < t < t_end - tran = - . interestingly , the length of the transition phase lasts less than a hubble time , and thus represents a small amount of the total cosmological evolution . we can calculate the equation of state parameter @xmath27 using our expression for @xmath56 in ( [ h(t ) ] ) : = - = = . [ epsexpand ] at early times , @xmath60 , this expression reduces to @xmath61 , characteristic of the transition phase . hence we expect that this will lead to the generation of scale invariant modes . at late times , @xmath62 , we have @xmath63 , characteristic of the ekpyrotic scaling phase . note that the equation of state parameter diverges at @xmath64 the hubble parameter vanishes at that time while @xmath51 remains finite . at first sight this may seem worrisome , since the momentary divergence of @xmath27 could potentially disrupt the scale invariance of the perturbations . this turns out to be an unfounded fear . we will show in sec . [ numver ] by numerically integrating the perturbation equation that the spectrum for @xmath5 is unaffected by the momentary blow - up in the equation of state . intuitively , this is because the time scale over which the equation of state diverges is very short , _ i.e. _ much less than a hubble time . for completeness , we can integrate ( [ h(t ) ] ) to obtain an expression for the scale factor a(t)(-t)^2/c^2e^h_0t1+(-t)+h_0t+ , [ aexpand ] where the taylor expansion is a good approximation during the transition phase . the scale factor is therefore nearly constant during the transition phase , as in the contracting example , and remains finite throughout the evolution . it is worth emphasizing that while the background is expanding during the transition phase , this is decidedly not an inflationary scenario . during the expanding transition phase , the equation of state evolves rapidly in time ( @xmath65 ) , while the scale factor is nearly constant @xmath66 , in stark contrast to slow - roll inflation . + we are now in a position to check that a slowly - expanding transition phase leads to a scale invariant spectrum of density perturbations . we begin with an analytic calculation , before verifying the results numerically in sec . [ numver ] . for this purpose , we work in comoving gauge , where the spatial slices are constant density ( @xmath67 ) hypersurfaces , and the spatial metric is given by @xmath68 . in this gauge , the variable @xmath5 represents the curvature perturbation on spatial slices . its action at quadratic order is given by s_2=d^3xd z^2 , where @xmath69 as before , and primes denote derivatives with respect to conformal time @xmath14 . the resulting equation of motion for the fourier modes @xmath70 is _ k+2_k + k^2_k=0 . [ zetaeom ] it is convenient to instead work in terms of the canonically normalized variable @xmath71 : v_k+(k^2-)v_k=0 . [ modefunction ] if @xmath72 , as in inflation , then this equation is well - known to yield a scale - invariant spectrum for @xmath5 at long wavelengths . furthermore , in this case the growing mode of @xmath5 goes to a constant as @xmath32 . in this limit , @xmath5 may be interpreted as a homogeneous perturbation to the scale factor and may be locally absorbed by a spatial diffeomorphism @xcite . the background is therefore a dynamical attractor . in our case , it follows from ( [ epsexpand ] ) and ( [ aexpand ] ) that z(t)=(-t)^2/c^2e^h_0 t . [ zee ] thus @xmath73 inherits the singular behavior of @xmath27 at @xmath64 , though as mentioned earlier this will not pose a problem . one way to see this is to note that @xmath74 is a regular singular point of ( [ zetaeom ] ) . during the transition phase , the scale factor is nearly constant , hence conformal time and cosmological time are approximately the same : @xmath75 . deep in the transition phase , @xmath60 , ( [ zee ] ) therefore implies . assuming the usual adiabatic vacuum , the solution is v_k=(1- ) , which gives a scale invariant power spectrum for the curvature perturbation , @xmath76 , in the long wavelength limit ( @xmath77 ) : p_k^3|_k|^2= . [ power ] moreover , in this limit @xmath5 approaches a constant , indicating that the expanding transition phase is also an attractor . since modes freeze out when @xmath78 , the range of scale invariant modes is set by the duration of the transition phase . from ( [ tend ] ) and ( [ tbeg ] ) , we deduce that = . [ length ] equations ( [ power ] ) and ( [ length ] ) agree exactly with the contracting case @xcite . of course , this is not surprising , since away from the singular point @xmath64 , @xmath73 does not know the difference between expanding and contracting solutions . we now check such claims through numerical integration . in this section we check the analytical results derived above by numerically integrating the evolution of @xmath5 . in the process we will reassure ourselves that the momentary singularity in @xmath27 does not spoil the scale invariance of the modes created during the transition phase . our starting point is the evolution equation ( [ zetaeom ] ) for the mode functions @xmath70 . this equation is better behaved than that for @xmath79 , since , as mentioned earlier , the singularity at @xmath64 is a regular singular point in this case . as initial conditions we impose the adiabatic vacuum choice _ k(_i)=e^ik_i , with @xmath80 chosen so that @xmath81 for all modes of interest . to proceed , we need to substitute an expression for @xmath82 into ( [ zetaeom ] ) . we will do this in two ways . first , following a quasi - analytic approach , in sec . [ quasianalytic ] we use the analytic form given by ( [ zee ] ) . this expression is of course not exact , since various approximations went into deriving it . second , in sec . [ exact ] we will redo the calculation more precisely using an exact form for @xmath82 , which will be obtained by numerically integrating the background equations of motion . let us start with the quasi - analytic approach , using the approximate form for @xmath82 given by ( [ zee ] ) . as it stands , ( [ zee ] ) gives @xmath73 as a function of cosmological time @xmath44 , whereas we need it in terms of conformal time @xmath14 . note that the factors of @xmath83 and @xmath84 are important in the transition phase and the ensuing contracting ekpyrotic phase , respectively . during both of these phases , cosmological time and conformal time are nearly the same , so a good approximation is to replace @xmath44 with @xmath14 in these factors . the factor of @xmath85 , however , is important at early times and makes the scale factor differ from unity conformal time and proper time are therefore much different in this case . approximating @xmath86 at early times , we can integrate @xmath87 to obtain e^h_0t== . thus we obtain the following analytic expression for @xmath82 z()= . [ ztau ] using this analytic expression for @xmath82 we are now in a position to integrate the equation of motion for @xmath5 . in fact , we will do the integration for both the expanding and the contracting case , where the contracting case corresponds to letting @xmath88 in ( [ ztau ] ) . in both cases , we take @xmath89 , @xmath90 , and integrate ( [ zetaeom ] ) over the range of modes @xmath91 and over the time interval @xmath92 . these parameters are chosen so that the shortest wavelength modes will have just left the hubble horizon by the end of the evolution , while the background solution will be firmly within the scaling ekpyrotic phase . the resulting power spectra are shown in fig . [ analyticzeta ] . we find that there is a range of scale - invariant modes spanning roughly @xmath93 decades in @xmath31-space , which is in order - of - magnitude agreement with our analytical result ( [ length ] ) . note that the range of scale invariant modes is slightly shorter than that found in the numerical analysis of @xcite , which is due to our including the extra factor @xmath94 in @xmath82 . we see that the power spectrum for @xmath5 is indistinguishable in each case , which confirms that the adiabatic ekpyrotic generation mechanism works equally well on an expanding background . in particular , this confirms that the divergence in @xmath27 at @xmath74 , corresponding to the transition from expansion to contraction , has no effect on the perturbation spectrum . the above analysis relied on certain approximations in deriving ( [ ztau ] ) . we neglected hubble friction in the evolution for @xmath9 , but this must eventually break down at early times when the background is approximately expanding de sitter space . to check that such corrections to ( [ ztau ] ) do not spoil scale invariance , in this section we numerically solve for the background evolution to obtain an exact result for @xmath82 . this can be done using the hamilton - jacobi formalism . using the chain rule , the @xmath95 equation implies h_,=- , [ hphi ] hence the friedmann equation can be rewritten as 3m_pl^2h^2=^2+v()=2m_pl^2 h_,^2+v ( ) . [ hj ] in this way , @xmath9 is thought of as the clock tracking the background evolution . we first numerically solve this equation to find @xmath96 , and in turn obtain @xmath57 by integrating ( [ hphi ] ) . at the end of the day , this gives us the hubble parameter as a function of time @xmath97 , from which we can extract @xmath98 . in practice , we perform the integration over a sufficiently broad range of field values and correspondingly large time interval . specifically , we have solved ( [ hj ] ) over the field range @xmath99 , setting @xmath100 . meanwhile , in integrating ( [ hphi ] ) to obtain @xmath57 we fix the integration constant corresponding to a shift in time by demanding that @xmath9 matches the analytic solution ( [ phisol2 ] ) . substituting everything into @xmath69 , we can numerically integrate ( [ zetaeom ] ) to obtain the power spectrum . the result is shown in fig . [ numericalzeta]a . for comparison , [ numericalzeta]b shows the quasi - analytic result . computational constraints forced us to use slightly different integration parameters from those used in sec . [ quasianalytic ] . figure [ numericalzeta ] was obtained using @xmath101 and @xmath102 , with integration ranging over @xmath103 and @xmath104 . we see that the plots show very good agreement . the analysis of sec . [ expanding ] assumed a scalar field starting from rest ( @xmath105 ) in the asymptotic past . in this section we explore a broader range of initial conditions , corresponding to non - zero initial kinetic energy for the scalar field , and the impact on the range of scale invariant modes . specifically , our goal is to derive an expression for the length of time spent in the transition phase ( and thus the range of scale invariant @xmath31-modes generated ) as a function of the initial energy . we start with the equation of motion for the scalar field , assuming as before that the hubble damping term is negligible + v_,0 . this equation admits a first integral of motion , ^2+v()=e , [ integralofmotion ] where @xmath106 is of course the total energy of the field . in sec . [ expanding ] , as well as in earlier work @xcite , @xmath106 was taken to be equal to @xmath107 at the onset of the transition phase , while remaining agnostic about the prior evolution . here we want to consider the more general case @xmath108 . in other words , @xmath106 corresponds to the value of the potential from which the scalar field starts at rest . equation ( [ integralofmotion ] ) implies = - , where we have chosen the negative branch of the square root , corresponding to the field rolling downhill . substituting our potential ( [ scalarpotential ] ) , this can be integrated explicitly -t= = ( e^c/2m_pl ) , where the last step assumes @xmath109 . this can be inverted to obtain a solution for @xmath57 , valid as long as @xmath110 : ( t)= . [ fullphi ] this expression generalizes ( [ phisol2 ] ) to span a range of initial field energy , or equivalently initial position on the potential where the field is released from rest . as a check , this reduces to ( [ phisol2 ] ) in the limit @xmath111 . moreover , at late times , when @xmath112 becomes sufficiently small , ( [ fullphi ] ) also approaches ( [ phisol2 ] ) , confirming that this is an attractor . it is straightforward to calculate @xmath51 h=-^2=-^2(c|t| ) . [ dothgen ] similarly , substituting the above solution for @xmath57 in the friedmann equation , @xmath113 , with @xmath114 , we obtain @xmath115 . the equation of state parameter is therefore given by = -=^2(c|t| ) . this generalizes ( [ epsexpand ] ) to a broader range of initial conditions , and matches ( [ epsexpand ] ) in the limit @xmath111 , as well as at late times @xmath116 . clearly the range of scale invariant modes will depend on the initial conditions , and we can already expect a more restricted range as we move away from the case @xmath117 studied earlier . the relevant quantity to assess the shape of the power spectrum is the time - dependent mass term , @xmath118 , appearing in the mode function equation ( [ modefunction ] ) . since @xmath98 is nearly constant during the phase of interest , we have = ^2(c|t| ) . modes will be scale invariant provided they freeze out when @xmath119 . recalling the taylor expansion @xmath120 , this will be the case whenever @xmath121 . this condition is approximately satisfied for @xmath122 , where t_beg - tran- . [ beggen ] in other words , @xmath123 marks the onset of scale invariant mode production . although this expression naively diverges for @xmath117 , we of course can only make @xmath123 as large as @xmath124 at earlier times , the fast - roll approximation assumed here breaks down . more carefully , we have t_beg - tran(-,- ) . the scale invariant phase has finite duration . much like the transition phase studied earlier , it comes to an end when the approximation @xmath125 breaks down . integrating ( [ dothgen ] ) in the limit @xmath121 gives h(t)= + + , where the ellipses include terms that become increasingly small as @xmath126 . it follows that the scale invariant or transition phase concludes at t_end - tran- . [ endgen ] combining ( [ beggen ] ) and ( [ endgen ] ) , the range of scale invariant modes is thus given by < . [ lengthnew ] here we have assumed the earliest possible onset of the transition phase . comparing to our previous answer ( [ length ] ) , @xmath127 , we see that solutions with @xmath128 lead to a narrower range of scale invariant modes , as expected . note that ( [ lengthnew ] ) agrees with ( [ length ] ) in the limit @xmath129 , as it should . in sec . [ exppert ] , as well as in earlier work @xcite , the adiabatic ekpyrotic evolution has been argued to be a dynamical attractor because the growing mode for @xmath5 goes to a constant in the long wavelength limit @xcite . this argument , while true , only applies at the perturbative level , _ i.e. _ for sufficiently small deviations from the background solution . in this section , we study the issue of stability more broadly , by performing a phase space analysis for a wide range of initial conditions . this will allow us to determine the breadth of the basin of attractor both in the expanding and in the contracting branches of the transition phase . for this purpose , we consider curves in the @xmath130 phase plane parameterized by @xmath131 . the friedmann constraint and the scalar field equation of motion imply the autonomous system : @xmath132 where the hubble parameter is understood as a function of @xmath9 and @xmath133 : h=. the choice of sign for the square root corresponds to the choice of expanding or contracting branch . cosmological solutions are given by the integral curves of the vector field @xmath134 . let us be precise about what we mean by attractor behavior . as curves approach each other in the @xmath130 plane , their respective values of @xmath135 and @xmath136 get closer and closer together , up to a constant relative shift in @xmath137 . we have the gauge freedom to fix the initial values of @xmath137 such that the solutions coincide . in this sense , an attractor solution is identified by neighboring curves flowing towards it . we will see that this is the case for the transition phase and the subsequent ekpyrotic scaling phase . it is worth pointing out that this focusing of trajectories occurs in the @xmath130 plane , but not in the physical phase space . minisuperspace models are hamiltonian systems , and consequently liouville s theorem forbids reduction of phase space volume . the resolution of this apparent discrepancy is of course that @xmath138 , rather than @xmath139 , is the momentum conjugate to @xmath9 in the canonical formalism . this can be seen more generally by considering a @xmath140 dimensional phase space . minisuperspace is a constrained hamiltonian system , with the friedmann equation defining a @xmath141 dimensional constraint hypersurface on which the hamiltonian vanishes . this hypersurface is foliated by the gauge orbits corresponding to the hamiltonian flow . the @xmath142 dimensional hypersurface which is transverse to this hamiltonian flow is the space of classical trajectories ; this is the symplectic reduction or marsden - weinstein quotient @xmath143 , where @xmath144 is the locus where the hamiltonian vanishes @xcite . there is a well - defined pullback of the symplectic form in the total space to this reduced phase space . darboux s theorem tells us that there are local coordinates @xmath145 such that the symplectic 2-form is = _ i=1^n dp_iq^i . for a model with action ( [ scalaraction ] ) in the adm formalism we can choose coordinates to write this symplectic form as = dt + _ i=1^n-1 dp_iq^i . in this language , the friedmann equation is the constraint that the hamiltonian vanishes @xmath146 . the total phase space carries a natural invariant measure constructed from @xmath147 . we may pull back this symplectic form to the reduced phase space @xmath148 and construct a natural measure @xmath149 . this is the gibbons hawking stewart measure @xcite . + + note that on the reduced phase space , this measure is conserved . moreover , in the particular case of a spatially - flat ( @xmath150 ) universe we can write the equations of motion in such a way that they are independent of @xmath151 , as we did in the single - field autonomous system ( [ phidotphivector ] ) . focusing on this single - field case for concreteness , we may then choose to parameterize the reduced phase space by @xmath9 and @xmath139 @xcite . choosing to treat these as euclidean coordinates introduces the measure = d , [ adhocmeasure ] but this of course need not be conserved by the evolution . as mentioned earlier , this non - conservation can be traced to the fact that @xmath139 is not the momentum canonically conjugate to @xmath9 . using ( [ adhocmeasure ] ) is nevertheless a sensible thing to do as mentioned above , we are interested in trajectories with the same values of @xmath9 and @xmath139 . figure [ globalview ] shows the vector field @xmath134 given by ( [ phidotphivector ] ) for the expanding and contracting cases , respectively , along with some numerically integrated curves . the black curve in each case corresponds to the analytic solution , with the transition phase occurring between the dotted lines . the chosen parameters are @xmath152 and @xmath153 . figure [ transitionview ] zooms in on the transition phase . as is clear from fig . [ transitionview ] , the transition solution is an attractor , both in the expanding and contracting cases . this confirms our earlier perturbative claims based on the long - wavelength behavior of @xmath5 . the expanding solution has a slightly larger basin of attraction , as expected . in particular , note that in the contracting case curves with @xmath139 sufficiently greater than zero are forced away from the transition solution while in the expanding case , such curves are driven toward the slow - roll solution and then follow the transition solution . more globally , fig . [ globalview ] shows that in the contracting case curves are repelled from the ` slow - roll ' solution at large @xmath9 . this is due to the susceptibility of contracting de sitter to kinetic domination . in the expanding case however , we see the opposite behavior trajectories with significant initial kinetic energy are driven toward the slow - roll de sitter solution before undergoing the transition phase evolution . to summarize , both the expanding and contracting transition solutions are attractors for some range of initial conditions , but the expanding solution has a larger basin of attraction . one of our motivations for considering an initially expanding cosmology was the instability of the contracting solution at sufficiently early times . the adiabatic ekpyrotic mechanism requires that the transition phase starts out with a sufficiently small equation of state parameter , @xmath154 , which implies that prior evolution will generically be unstable to kinetic domination in the contracting case . as shown in sec . [ phasespace ] , this is true , in particular , for the simplest potential ( [ scalarpotential ] ) the pre - transition phase evolution corresponds to a contracting de sitter universe , which is clearly unstable . the expanding branch , on the other hand , is better behaved , as it extrapolates backwards in time to an expanding de sitter space . thus , in choosing an initially expanding universe , we have greatly expanded the basis of attraction . ( it is worth emphasizing that even in the initially expanding case , the universe is driven towards a contracting ekpyrotic scaling phase . ) as mentioned in the introduction , our fiducial potential ( [ scalarpotential ] ) serves as the simplest illustration of our mechanism . more generally , this form need only hold approximately during the transition phase , corresponding to a small range in field space , @xmath155 , and there is ample freedom in choosing the global form of the potential . that said , if we take the potential ( [ scalarpotential ] ) at face value , then at sufficiently early times the evolution in the expanding case should correspond to an inflating space - time . in this section , we take this possibility seriously and study the transition between the initial de sitter phase to the transition phase . we will see that @xmath57 evolves smoothly between the two regimes . to see this , let us split the evolution into two regimes : the ` transition ' regime , @xmath156 , where hubble friction is negligible and the scalar field equation of motion reduces to ( [ scalareomapprox ] ) : + v_,=0 . [ eomtran ] the ` slow - roll ' regime , @xmath157 , where the equation of motion is approximately given by 3h+v_,=0 , [ eomslow ] with @xmath158 . the field value delineating these two regions , @xmath159 , can be estimated as the value of @xmath9 at the onset of the transition phase . combining ( [ phisol2 ] ) , ( [ h0 ] ) and ( [ tbeg ] ) , we obtain _ t = ( t_beg - tran ) = ( c)c , where in the last step we have used @xmath8 . we are now in a position to argue that the approximate solutions in regions i and ii match each other smoothly at @xmath160 . since the potential is certainly continuous , this amounts to showing that the kinetic energy also matches continuously . the kinetic energy of the transition solution is easy to write down , as we have done it several times already , and is just the first integral of ( [ eomtran ] ) : ^2_tranv_0e^-c / m_pl . [ ketran ] likewise , the kinetic energy in the slow - roll regime follows from ( [ eomslow ] ) : ^2_slow - rollv_0c^2e^-2c / m_pl . [ kesr ] it immediately follows that ( [ ketran ] ) and ( [ kesr ] ) are equal at @xmath160 , as we wanted to show . this result tells us that trajectories starting in the slow roll region , @xmath161 , are quickly driven to the attractor slow - roll solution and reach @xmath162 with precisely the correct kinetic energy as assumed by the transition solution . ( in the notation of sec . [ rest ] , solutions that emerge from the slow regime generically reach the transition phase with @xmath163 . ) note that the spectrum of fluctuations also matches smoothly between the two regimes . the power spectrum of modes generated during the inflationary epoch is given by p_^inf = , which agrees up to an order unity factor with ( [ power ] ) . the adiabatic ekpyrotic mechanism is the unique , non - inflationary single - field mechanism with unit sound speed that generates a scale invariant spectrum for @xmath5 on an attractor background . while originally proposed assuming a contracting universe , in this paper we have shown that the mechanism works equally well on an initially expanding background . the evolution consists of an expanding transition phase " , followed by a contracting ekpyrotic scaling phase . we have shown , both through analytical arguments and exact numerical integration , that the power spectrum for @xmath5 is indistinguishable from the contracting version of the mechanism . this confirms that the perturbation spectrum is to a good approximation insensitive to whether the background is expanding or contracting during mode production scale invariance relies on a rapidly - varying equation of state parameter while the scale factor is nearly constant . by performing a phase space analysis for both the expanding and contracting branches , we have verified that the transition phase and subsequent ekpyrotic scaling phase are attractors . the basin of attraction is broader in the expanding case , since intuitively any additional kinetic energy present at early times gets redshifted , instead of blueshifted , in this case . for the simplest potentials considered here , the evolution is an expanding de sitter space - time asymptotically in the past . taking this precursor inflating phase seriously , we have shown that the transition from inflation to the adiabatic ekpyrotic phase is smooth , with the scale invariant spectra generated in each phase matching at the transition , up to order unity coefficient . as mentioned in the introduction , the degeneracy with inflation is broken at the three - point level . unlike the nearly gaussian spectrum of inflation , the rapidly - varying equation of state @xmath13 characteristic of the adiabatic ekpyrotic phase leads to large non - gaussianities on small scales . for the simplest lifted exponential potential ( [ scalarpotential ] ) , this results in a breakdown of perturbation theory , both at the classical and quantum mechanical levels . as shown in detail in @xcite in the contracting case , however , this perturbative breakdown can be avoided for more general potentials of the form ( [ genpot ] ) , but this comes at a cost the range of scale invariant perturbations is now limited , spanning a factor of @xmath30 in @xmath31 space . these considerations should carry over to the expanding case . we leave a careful exploration of this issue to future work . the expanding version of adiabatic ekpyrosis presented here greatly expands the realm of larger cosmological scenarios in which this mechanism can be embedded . because the basin of attraction is broader compared to its contracting counterpart , the expanding mechanism is less sensitive to the details of the prior evolution . more importantly for model - building , the universe is expanding during mode production and is contracting later on , which naturally suggests embedding our mechanism in a cyclic scenario @xcite . it is conceivable that tying our mechanism to the present phase of cosmic acceleration may also explain why the @xmath30 scale invariant modes fall in the observable window . we are currently investigating this possibility . _ acknowledgments : _ it is our pleasure to thank daniel baumann , adam brown , kurt hinterbichler , mitchell lerner , godfrey miller , leonardo senatore , paul steinhardt , daniel wesley and matias zaldarriaga for helpful discussions . this work was supported in part by funds from the university of pennsylvania , the alfred p. sloan foundation ( jk ) and the ella n. pawling fellowship from the university of pennsylvania ( aj ) . a. h. guth , `` the inflationary universe : a possible solution to the horizon and flatness problems , '' phys . d * 23 * , 347 ( 1981 ) ; a. albrecht and p. j. steinhardt , `` cosmology for grand unified theories with radiatively induced symmetry breaking , '' phys . * 48 * , 1220 ( 1982 ) ; a. d. linde , `` a new inflationary universe scenario : a possible solution of the horizon , flatness , homogeneity , isotropy and primordial monopole problems , '' phys . lett . b * 108 * , 389 ( 1982 ) . m. gasperini and g. veneziano , `` pre - big bang in string cosmology , '' astropart . * 1 * , 317 ( 1993 ) [ arxiv : hep - th/9211021 ] . a. nayeri , r. h. brandenberger and c. vafa , `` producing a scale - invariant spectrum of perturbations in a hagedorn phase of string cosmology , '' phys . lett . * 97 * , 021302 ( 2006 ) [ arxiv : hep - th/0511140 ] . r. h. brandenberger , a. nayeri , s. p. patil and c. vafa , `` tensor modes from a primordial hagedorn phase of string cosmology , '' phys . lett . * 98 * , 231302 ( 2007 ) [ arxiv : hep - th/0604126 ] . r. h. brandenberger , a. nayeri , s. p. patil and c. vafa , `` string gas cosmology and structure formation , '' int . j. mod . a * 22 * , 3621 ( 2007 ) [ arxiv : hep - th/0608121 ] . r. h. brandenberger _ et al . _ , `` more on the spectrum of perturbations in string gas cosmology , '' jcap * 0611 * , 009 ( 2006 ) [ arxiv : hep - th/0608186 ] . t. battefeld and s. watson , `` string gas cosmology , '' rev . phys . * 78 * , 435 ( 2006 ) [ arxiv : hep - th/0510022 ] . j. khoury , b. a. ovrut , p. j. steinhardt and n. turok , `` the ekpyrotic universe : colliding branes and the origin of the hot big bang , '' phys . d * 64 * , 123522 ( 2001 ) [ arxiv : hep - th/0103239 ] . r. y. donagi , j. khoury , b. a. ovrut , p. j. steinhardt and n. turok , `` visible branes with negative tension in heterotic m - theory , '' jhep * 0111 * ( 2001 ) 041 [ arxiv : hep - th/0105199 ] . j. khoury , b. a. ovrut , n. seiberg , p. j. steinhardt and n. turok , `` from big crunch to big bang , '' phys . d * 65 * , 086007 ( 2002 ) [ arxiv : hep - th/0108187 ] . j. khoury , b. a. ovrut , p. j. steinhardt and n. turok , `` density perturbations in the ekpyrotic scenario , '' phys . d * 66 * , 046005 ( 2002 ) [ arxiv : hep - th/0109050 ] . d. h. lyth , `` the primordial curvature perturbation in the ekpyrotic universe , '' phys . b * 524 * , 1 ( 2002 ) [ arxiv : hep - ph/0106153 ] . r. brandenberger and f. finelli , `` on the spectrum of fluctuations in an effective field theory of the ekpyrotic universe , '' jhep * 0111 * , 056 ( 2001 ) [ arxiv : hep - th/0109004 ] . j. steinhardt and n. turok , `` cosmic evolution in a cyclic universe , '' phys . d * 65 * , 126003 ( 2002 ) [ arxiv : hep - th/0111098 ] . a. notari and a. riotto , `` isocurvature perturbations in the ekpyrotic universe , '' nucl . b * 644 * , 371 ( 2002 ) [ arxiv : hep - th/0205019 ] . f. finelli , `` assisted contraction , '' phys . b * 545 * , 1 ( 2002 ) [ arxiv : hep - th/0206112 ] . s. tsujikawa , r. brandenberger and f. finelli , `` on the construction of nonsingular pre - big - bang and ekpyrotic cosmologies and the resulting density perturbations , '' phys . d * 66 * , 083513 ( 2002 ) [ arxiv : hep - th/0207228 ] . s. gratton , j. khoury , p. j. steinhardt and n. turok , `` conditions for generating scale - invariant density perturbations , '' phys . d * 69 * , 103505 ( 2004 ) [ arxiv : astro - ph/0301395 ] . a. j. tolley , n. turok and p. j. steinhardt , `` cosmological perturbations in a big crunch / big bang space - time , '' phys . d * 69 * , 106005 ( 2004 ) [ arxiv : hep - th/0306109 ] . b. craps and b. a. ovrut , `` global fluctuation spectra in big crunch / big bang string vacua , '' phys . d * 69 * , 066001 ( 2004 ) [ arxiv : hep - th/0308057 ] . j. khoury , p. j. steinhardt and n. turok , `` great expectations : inflation versus cyclic predictions for spectral tilt , '' phys . lett . * 91 * , 161301 ( 2003 ) [ arxiv : astro - ph/0302012 ] . j. khoury , p. j. steinhardt and n. turok , `` designing cyclic universe models , '' phys . lett . * 92 * , 031302 ( 2004 ) [ arxiv : hep - th/0307132 ] . j. khoury , `` a briefing on the ekpyrotic / cyclic universe , '' arxiv : astro - ph/0401579 . p. creminelli , a. nicolis and m. zaldarriaga , `` perturbations in bouncing cosmologies : dynamical attractor vs scale invariance , '' phys . d * 71 * , 063505 ( 2005 ) [ arxiv : hep - th/0411270 ] . j. l. lehners , p. mcfadden , n. turok and p. j. steinhardt , `` generating ekpyrotic curvature perturbations before the big bang , '' phys . rev . d * 76 * , 103501 ( 2007 ) [ arxiv : hep - th/0702153 ] . e. i. buchbinder , j. khoury and b. a. ovrut , `` new ekpyrotic cosmology , '' phys . d * 76 * , 123503 ( 2007 ) [ arxiv : hep - th/0702154 ] . e. i. buchbinder , j. khoury and b. a. ovrut , `` on the initial conditions in new ekpyrotic cosmology , '' jhep * 0711 * , 076 ( 2007 ) [ arxiv:0706.3903 [ hep - th ] ] . e. i. buchbinder , j. khoury and b. a. ovrut , `` non - gaussianities in new ekpyrotic cosmology , '' phys . lett . * 100 * , 171302 ( 2008 ) [ arxiv:0710.5172 [ hep - th ] ] . p. creminelli and l. senatore , `` a smooth bouncing cosmology with scale invariant spectrum , '' jcap * 0711 * , 010 ( 2007 ) [ arxiv : hep - th/0702165 ] . k. koyama and d. wands , `` ekpyrotic collapse with multiple fields , '' jcap * 0704 * , 008 ( 2007 ) [ arxiv : hep - th/0703040 ] . k. koyama , s. mizuno and d. wands , `` curvature perturbations from ekpyrotic collapse with multiple fields , '' class . * 24 * , 3919 ( 2007 ) [ arxiv:0704.1152 [ hep - th ] ] . j. l. lehners and p. j. steinhardt , `` non - gaussian density fluctuations from entropically generated curvature perturbations in ekpyrotic models , '' phys . d * 77 * , 063533 ( 2008 ) [ erratum - ibid . d * 79 * , 129903 ( 2009 ) ] [ arxiv:0712.3779 [ hep - th ] ] . j. l. lehners and p. j. steinhardt , `` intuitive understanding of non - gaussianity in ekpyrotic and cyclic models , '' phys . d * 78 * , 023506 ( 2008 ) [ erratum - ibid . d * 79 * , 129902 ( 2009 ) ] [ arxiv:0804.1293 [ hep - th ] ] . j. l. lehners and p. j. steinhardt , `` non - gaussianity generated by the entropic mechanism in bouncing cosmologies made simple , '' phys . rev . d * 80 * , 103520 ( 2009 ) [ arxiv:0909.2558 [ hep - th ] ] . j. khoury and p. j. steinhardt , `` adiabatic ekpyrosis : scale - invariant curvature perturbations from a single scalar field in a contracting universe , '' phys . lett . * 104 * , 091301 ( 2010 ) [ arxiv:0910.2230 [ hep - th ] ] . a. linde , v. mukhanov and a. vikman , `` on adiabatic perturbations in the ekpyrotic scenario , '' jcap * 1002 * , 006 ( 2010 ) [ arxiv:0912.0944 [ hep - th ] ] . j. khoury and p. j. steinhardt , `` generating scale - invariant perturbations from rapidly - evolving equation of state , '' arxiv:1101.3548 [ hep - th ] . j. l. lehners , `` ekpyrotic and cyclic cosmology , '' phys . rept . * 465 * , 223 ( 2008 ) [ arxiv:0806.1245 [ astro - ph ] ] . j. l. lehners , `` ekpyrotic non - gaussianity a review , '' adv . astron . * 2010 * , 903907 ( 2010 ) [ arxiv:1001.3125 [ hep - th ] ] . j. k. erickson , d. h. wesley , p. j. steinhardt and n. turok , `` kasner and mixmaster behavior in universes with equation of state @xmath164 , '' phys . d * 69 * , 063514 ( 2004 ) [ arxiv : hep - th/0312009 ] . d. garfinkle , w. c. lim , f. pretorius and p. j. steinhardt , `` evolution to a smooth universe in an ekpyrotic contracting phase with @xmath165 , '' phys . d * 78 * , 083537 ( 2008 ) [ arxiv:0808.0542 [ hep - th ] ] . j. m. bardeen , p. j. steinhardt and m. s. turner , `` spontaneous creation of almost scale - free density perturbations in an inflationary universe , '' phys . d * 28 * , 679 ( 1983 ) . v. f. mukhanov , h. a. feldman and r. h. brandenberger , `` theory of cosmological perturbations . classical perturbations . quantum theory of perturbations . extensions , '' phys . rept . * 215 * , 203 ( 1992 ) . f. finelli and r. brandenberger , `` on the generation of a scale - invariant spectrum of adiabatic fluctuations in cosmological models with a contracting phase , '' phys . d * 65 * , 103522 ( 2002 ) [ arxiv : hep - th/0112249 ] ; d. wands , `` duality invariance of cosmological perturbation spectra , '' phys . rev . d * 60 * , 023507 ( 1999 ) [ arxiv : gr - qc/9809062 ] . l. a. boyle , p. j. steinhardt and n. turok , `` a new duality relating density perturbations in expanding and contracting friedmann cosmologies , '' phys . d * 70 * , 023504 ( 2004 ) [ arxiv : hep - th/0403026 ] . c. lin , r. h. brandenberger and l. p. levasseur , `` a matter bounce by means of ghost condensation , '' arxiv:1007.2654 [ hep - th ] ; y. f. cai , r. brandenberger and x. zhang , `` the matter bounce curvaton scenario , '' arxiv:1101.0822 [ hep - th ] . j. khoury and f. piazza , `` rapidly - varying speed of sound , scale invariance and non - gaussian signatures , '' jcap * 0907 * , 026 ( 2009 ) [ arxiv:0811.3633 [ hep - th ] ] . a. j. tolley and d. h. wesley , `` scale - invariance in expanding and contracting universes from two - field models , '' jcap * 0705 * , 006 ( 2007 ) [ arxiv : hep - th/0703101 ] . y. -s . piao , e. zhou , `` nearly scale invariant spectrum of adiabatic fluctuations may be from a very slowly expanding phase of the universe , '' phys . rev . * d68 * , 083515 ( 2003 ) . [ hep - th/0308080 ] . j. khoury , g. e. j. miller , `` towards a cosmological dual to inflation , '' [ arxiv:1012.0846 [ hep - th ] ] . d. baumann , l. senatore , m. zaldarriaga , `` scale - invariance and the strong coupling problem , '' [ arxiv:1101.3320 [ hep - th ] ] . c. armendariz - picon and e. a. lim , scale invariance without inflation ? , " jcap * 0312 * , 002 ( 2003 ) ; c. armendariz - picon , near scale invariance with modified dispersion relations , " jcap * 0610 * , 010 ( 2006 ) . p. creminelli , m. a. luty , a. nicolis and l. senatore , `` starting the universe : stable violation of the null energy condition and non - standard cosmologies , '' jhep * 0612 * , 080 ( 2006 ) [ arxiv : hep - th/0606090 ] . a. nicolis , r. rattazzi and e. trincherini , `` energy s and amplitudes positivity , '' jhep * 1005 * , 095 ( 2010 ) [ arxiv:0912.4258 [ hep - th ] ] . j. khoury , j. l. lehners and b. ovrut , `` supersymmetric p(x , phi ) and the ghost condensate , '' arxiv:1012.3748 [ hep - th ] . j. khoury , j. l. lehners and b. a. ovrut , `` supersymmetric galileons , '' arxiv:1103.0003 [ hep - th ] . g. w. gibbons , s. w. hawking and j. m. stewart , a natural measure on the set of all universes , " nucl . b * 281 * , 736 ( 1986 ) .

we have analyzed different correlation functions in a realistic spin - orbital model for half - doped manganites . using a finite - temperature diagonalization technique the ce phase was found in the charge - ordered phase in the case of small antiferromagnetic interactions between electrons . it is shown that a key ingredient responsible for stabilization of the ce - type spin and orbital - ordered state is the cooperative jahn - teller ( jt ) interaction between next - nearest mn neighbors mediated by the breathing mode distortion of mn octahedra and displacements of mn ions . the topological phase factor in the mn - mn hopping leading to gap formation in one - dimensional models for the ce phase as well as the nearest neighbor jt coupling are not able to produce the zigzag chains typical for the ce phase in our model .

cond-mat0306532

half - doped perovskite manganites re@xmath3a@xmath3mno@xmath4 ( re - rare earth , a - alkaline earth ) exhibit very specific properties quite different from manganites with other @xmath5 electron concentrations.@xcite generally , for large size re and a cations or when external pressure is applied@xcite these compounds are in the a - type antiferromagnetic ( af ) , metallic state at low temperatures showing no or only weak sign of charge ordering , and the occupied @xmath5 electron states are predominantly @xmath6 like . recently it has been reported that the a - phase may develop an intrinsic charge - stripe modulation , which controls the transport properties.@xcite with decreasing size of the cations , which implies a decrease of the bandwidth , charge ordering is observed which at lower temperatures is accompanied by the formation of peculiar ferromagnetic ( fm ) zigzag chains@xcite that are staggered antiferromagnetically . this ce spin order , a notation introduced in the pioneering work of wollan and koehler@xcite and goodenough@xcite , is accompanied by a checkerboard charge order ( co ) and directed occupied orbitals on the mn@xmath7 sites amplifying double exchange ( de ) along one lattice diagonal . the ce - structure is a particularly fascinating manifestation of the control of magnetic order due to the orbital degree of freedom@xcite in combination with charge ordering . although a number of approaches have been followed to explain this structure , the mechanism for its stability is still not clear and hotly debated with emphasis put on the role of the de,@xcite the coulomb interaction,@xcite or the coupling to the lattice degrees of freedom.@xcite the ce phase has been observed experimentally both in cubic ( nd , pr)@xmath3(sr , ca)@xmath3mno@xmath4@xcite and in layered lasr@xmath8mn@xmath8o@xmath9@xcite and la@xmath3sr@xmath10mno@xmath11@xcite manganites . in some cases it can coexist with the a - type af spin ordering.@xcite regarding co in the ce phase , some experiments are interpreted in terms of almost perfect co @xcite while others@xcite are considered to be consistent with a smooth charge density wave ( cdw ) of the @xmath5 electron density . although , it is plausible that the complex spin and orbital arrangement in the charge - ordered manganites can result from the competition between de@xcite and af superexchange interactions ( @xmath12 ) , the role played by the co and jahn - teller ( jt ) interactions@xcite in the system is still unclear . finally in systems like e.g. pr@xmath3ca@xmath3mno@xmath4 co and ce - type orbital correlations are found to develop well above the neel temperature@xcite or even coexist with fm spin state,@xcite which indicates that these orbital correlations are not primarily driven by magnetic interactions . this work focuses on a microscopic understanding of the stability of the ce - phase which appears to be the ground state of narrow band manganites at and near quarter - filling . we argue here that the key interaction , apart from the nearest - neighbor coulomb repulsion @xmath13 which supports the charge ordering,@xcite is the next - nearest neighbor jt interaction . this interaction between orbitals at next - nearest neighbor mn@xmath7 ions emerges from the jt - distortions of the mn@xmath7 octahedra which is amplified by the breathing distortion of the intermediate mn@xmath14 octahedra . we show that this interaction generates a narrow regime in the @xmath13-@xmath15 phase diagram where the ce - phase ( and sometimes the c - phase ) is stable . otherwise one encounters a homogeneous fm phase at small @xmath15 fully controlled by de and a conventional nearest - neighbor af phase at larger @xmath15 . an important feature of further neighbor jt interactions are the displacements of the mn@xmath14 ions . we are only aware of work by radaelli _ _ et al.__@xcite where detailed lattice coordinates for the ce - phase in la@xmath3ca@xmath3mno@xmath4 are reported . in particular they observed that the 4 planar o(2 ) move towards the mn@xmath14 ion and the mn@xmath14 ions are found to be displaced from their regular perovskite positions , while the mn@xmath7 ions are not shifted . this observation confirms the importance of the breathing distortion and also shows the mechanism of the further neighbor jt - distortion in the ce - phase . as a consequence the appearance of new bragg reflections in the ce - phase would be associated not primarily with quadrupolar electronic orbital order but rather with the displacements of ions associated with the ce orbital order.@xcite an alternative long - ranged jt based interaction was recently proposed by khomskii and kugel@xcite and by calderon , millis and ahn.@xcite they employ elasticity theory to derive the orbital interaction between jt - distorted mn@xmath7o@xmath16 octahedra for the perfectly charge ordered case . we have also investigated the consequences of this interaction . our numerical simulations confirm the estimate presented in ref . that for perfect charge ordering the af ce phase is stable . yet in this model this requires large nearest neighbor coulomb repulsion @xmath13 , while for moderate values of @xmath13 the ce phase gets unstable . the finite temperature diagonalization@xcite used in our study allows to include the full interplay of spin and orbital states , and to monitor the onset of different orders as function of temperature . this gives us the ability for an unbiased investigation of the formation of different spin - orbital orderings emerging from all multi - electron configurations in the cluster . in this paper we concentrate on the stability and temperature dependence of spin , orbital , and charge ordered states in the two - dimensional ( 2d ) realistic model for half - doped manganites . in our microscopic spin - orbital model co develops as a result of nearest - neighbor coulomb repulsion @xmath13 and as a consequence of further neighbor jt interactions , while the spin - orbital order results from the competition between de ( i.e. kinetic energy ) , af superexchange and further neighbor jt interactions . as shown in sec . [ sec : nnnjt ] the latter interaction can lead to a co state at low temperatures even in the absence of inter - site coulomb repulsion . furthermore we argue that a topological sign in the hopping amplitudes of @xmath5 orbitals@xcite invoked in one - dimensional ( 1d ) models for the ce phase is not sufficient to explain the formation of ce spin - orbital structure in the 2d model studied by us . further motivations to put effort in a better understanding of the interplay of spin , charge , orbitals and lattice in the formation of the ce - phase are the following : it has been argued that the intrinsic mechanism that leads to colossal magnetoresistance are tendencies towards ce - type charge and orbital ordering in the vicinity of @xmath17.@xcite furthermore a ce - insulator to fm - metal transition has been observed in relatively small magnetic field , which is accompanied by a major change of the optical conductivity on a surprisingly large energy scale ( @xmath18 ev).@xcite such dramatic changes in relatively weak magnetic fields are a clear manifestation of the subtle interplay between spin , orbital and co. the present study provides a basis for further investigations of the temperature dependence of e.g. the optical conductivity and the study of spin and orbital excitations in the ce - phase , as well as the effect of doping into the quarter filled state . the paper is organized as follows . in sec . [ sec : model ] we present the model hamiltonian , and describe in sec . [ sec : lanczos ] the finite - temperature lanczos method . [ sec : numerics ] contains numerical results for different two - site correlation functions characterizing the charge , orbital and spin structure . these correlation functions are evaluated as functions of temperature and collected in a form of semi - quantitative phase diagrams in three different regimes : ( i ) including nearest - neighbor effective orbital - orbital ( oo ) interactions ( sec . [ sec : nnjt ] ) ; ( ii ) assuming the elastic form of oo coupling at different distances ( sec . [ sec : elastic ] ) ; ( iii ) considering the local form of next- and nearest - neighbor oo coupling ( sec . [ sec : nnnjt ] ) , while in sec . [ sec : stacking ] the effect of charge stacking is discussed . the paper is summarized in sec . [ sec : summary ] . in the appendix the stability of the ce versus the c phase is analysed for a 1d band model with lifted orbital degeneracy . we consider the generic fm kondo lattice model for manganites@xcite with the mn @xmath5 electrons coupled to the core @xmath0 spins via the hund coupling @xmath19 . the model is augmented by inter - site coulomb repulsion @xmath13 and cooperative jt interactions and treated in the limit of infinite on - site coulomb repulsion between two @xmath5 electrons on the same site ( @xmath20 ) : @xmath21 the first term describes the motion of the @xmath5 electrons , @xmath22 with a constraint that allows only for mn@xmath14 and mn@xmath7 configurations , @xmath23 , where all other alternative @xmath5 states are projected out . here and in the following the index @xmath24 ( @xmath25 ) denotes the @xmath5 orbital ( spin ) orthogonal to the @xmath26 ( @xmath27 ) one , respectively . the hopping matrix elements depend on the basis chosen . for the basis @xmath28 with @xmath29 and @xmath30 they are given by , @xmath31=\frac{t}{4}\left ( \begin{array}{cc } 3 & \mp\sqrt{3}\\ \mp\sqrt{3 } & 1 \end{array}\right ) , \ ] ] where , @xmath32(@xmath33 ) refer to the @xmath34(@xmath35 ) direction , respectively . the second term of hamiltonian ( [ eq : h_total ] ) stands for the hund interaction between @xmath5 electrons spin and the @xmath36 @xmath0 core spin @xmath37 , @xmath38 resulting in the parallel alignment of spins between @xmath0 and @xmath5 electrons at each site in the limit @xmath39 which corresponds to the realistic situation in manganites . the @xmath40 part of the total hamiltonian alone represents the de mechanism and yields a fm ground state with @xmath41 orbitals being occupied which is favored by the kinetic energy in 2d model . such a fully spin polarized case at finite @xmath42 leads to _ the orbital @xmath43-@xmath44 model _ as considered in refs . . here , we are interested in the more complex case where both spin and orbital degrees of freedom play an active role . therefore , an af ( @xmath45 ) coupling between nearest - neighbor @xmath0 spins is also incorporated , @xmath46 in our numerical studies we assume @xmath47 for the core spins which decreases the resulting hilbert space considerably . we have tested that this assumption gives similar results as in the @xmath36 case when @xmath19 is renormalized@xcite by the factor @xmath48 . moreover , to promote the charge ordering as observed experimentally in half - doped manganites we include the coulomb repulsion ( @xmath49 ) between @xmath5 electrons on neighboring sites,@xcite @xmath50 with @xmath51 . this term reduces the probability of occupancy of neighboring sites by @xmath5 electrons . the form of the inter - site oo interactions included in the @xmath52 term of the hamiltonian will be specified in the following section . to investigate the possible spin and orbital orderings in our spin - orbital model ( [ eq : h_total ] ) we have calculated the temperature dependence of different two - site correlation functions given by the operator expectation value : @xmath53 of two - site operators , @xmath54 with @xmath55 , @xmath56 , @xmath57 , or @xmath58 with @xmath59.@xcite here , @xmath60 , @xmath61 is the number of basis states @xmath62 that span @xmath63 , @xmath64 number of sites in the cluster and @xmath65 is the partition function . the calculation of ( [ eq : a_r_exact ] ) would , however , require a complete diagonalization of a @xmath66 matrix of the hamiltonian . therefore , we use for the calculation of @xmath67 a generalization of the exact diagonalization technique developed by jakli and prelovek.@xcite in this approach the trace of the thermodynamic expectation value is performed by a monte carlo sampling leading to the following approximate expression : @xmath68 where the first sum is performed over a restricted number @xmath69 of random states @xmath70 being initial states in the lanczos algorithm to generate a set @xmath71 of eigenfunctions of @xmath63 with respective eigenvalues @xmath72 . @xmath73 is the number of lanczos functions in the expansion of a given state @xmath70 . in the same manner the partition function is approximated by , @xmath74 it has been shown@xcite that the results get very accurate already for @xmath69 , @xmath75 . even though we exploit the translational symmetry the calculations are restricted to small clusters ( @xmath76 is used in this paper ) as only @xmath77 subspaces can be treated separately , while a similar symmetry in the orbital sector does not exist . although a @xmath78 cluster is large enough to capture the ce - type correlations , boundary conditions can influence further neighbor interactions . to assess such finite size effects quantitatively , a full diagonalization on larger clusters would be necessary . in the case of the co ce phase this would involve at least a @xmath79 site cluster , which is far beyond numerical realization . in the numerical studies we assume @xmath47 for the core spins and the limit of strong hund s coupling @xmath80 . the numerical results are almost identical for all @xmath81 in the whole range of temperatures considered by us . the coulomb repulsion @xmath82 and the af - exchange @xmath83 are free parameters , and the data is summerized in form of a phase diagram in the @xmath84 plane . the hopping parameter @xmath43 is chosen as energy unit . we focus in this study on co states and thus assume finite positive values for @xmath13 in all the calculations . in perovskite manganites the effective hopping @xmath43 strongly depends on the lattice parameters and distortions _ e.g. _ increases with contraction of the mn@xmath33o bond length and also depends on the mn@xmath33o@xmath33mn bond angle.@xcite thus , its value can be only roughly estimated as @xmath85ev.@xcite a major difficulty when searching for spin structures like those indicated in fig . [ fig : spins ] , and the same applies to charge and orbital order , is that in a cluster with periodic boundary conditions translational ( and rotational ) symmetries are not broken . therefore we have to investigate correlation functions and compare their features with the different spin configurations ( presented in fig . [ fig : spins ] ) . schematic view of different spin patterns found in the 2d model in presence of checkerboard co ( half - doped case ) . thick ( thin ) arrows represent states with total spin @xmath86 ( @xmath36 ) with ( without ) an @xmath5 electron present at a given site , respectively . the shaded lines in c and ce phases indicate the direction of de carrier propagation . the preferred directional @xmath5 orbitals occupied at sites @xmath86 are parallel to the shaded paths.,width=226 ] we shall consider three different forms for the orbital interaction term @xmath52 and show that one has to go beyond the nearest - neighbor interactions to obtain the ce - type correlations . we begin the description of our calculations with a simplified model ( without @xmath87 term ) but including the crystal - field splitting of @xmath5 orbitals @xmath88 , @xmath89 where @xmath90 is the @xmath91 component of the pseudospin operator . this orbital splitting accounts for the elongation of octahedra along the c - axis in layered compounds , which favors @xmath92 over @xmath41 orbital occupation , and also counterbalances the trend towards occupation of @xmath41 orbitals , which is strongly favored by the kinetic energy in the 2d - model . although , the model without @xmath87 interactions can lead to the stability of the ce - type orbital pattern,@xcite the magnetic structure was fm . in a wide range of parameters @xmath93 , @xmath94 , @xmath95 no stabilization of the af ce spin phase was found , indicating that the inter - chain coupling is quite strong and can not be neglected in realistic models for half - doped co manganites . such a conclusion agrees this recent density - functional calculations showing a considerable band dispersion normal to the chains.@xcite thus , we argue that the _ topological effect _ considered in 1d model approximations , which gave a simple explanation of the existence of the ce spin - orbital phase at half - doping , is not likely to be the decisive mechanism in a more realistic 2d case investigated here . our simulations with purely electronic interactions do not provide evidence for 1d fm chains that are coupled antiferromagnetically to neighbor chains , which is the basic assumption that could justify such 1d models . therefore , in the following sections we concentrate on the electron - lattice coupling@xcite and present a mechanism stabilizing the ce spin / orbital state in manganites . phase diagram indicating the character of the dominant spin correlations found at small temperatures @xmath96 . the calculations include the nearest - neighbor oo effective interactions ( @xmath97 ) . @xmath98 indicates representative points for which the correlation functions are presented below.,width=132 ] the first aim of our study of the spin - orbital model ( [ eq : h_total ] ) in half - doped manganites is to investigate the role of nearest - neighbor oo coupling played in co state . there are two different mechanisms contributing to this interaction : ( i ) the cooperative jt effect@xcite and ( ii ) superexchange interactions.@xcite neglecting more complex spin - orbital terms@xcite both effects can be described by the following simple term derived before for undoped lamno@xmath4 compound,@xcite @xmath99 where the two - site orbital operator , @xmath100 , between nearest - neighbor mn sites in the @xmath28 basis , @xmath101 is described in terms of pseudospin operators : @xmath102 , and @xmath103 . the prefactor of the mixed term @xmath104 is negative in the @xmath34 direction and positive in the @xmath35 direction . the nearest - neighbor oo term can lead to _ attraction _ of electrons on neighboring sites with different orbital orientations but having the coulomb repulsion ( [ eq : h_v ] ) included in the model we avoid phase separation . temperature dependence of the charge ( a ) , orbital ( b ) , and spin ( c ) two - site correlation functions calculated for neighbors at different distances @xmath105 for the phase with fm spin correlations . parameters : @xmath106 , @xmath97 , @xmath107 , @xmath108.,width=377 ] temperature dependence of the correlations as in fig . but for the phase with af spin correlations . parameters : @xmath109 , @xmath97 , @xmath110 , @xmath108.,width=377 ] temperature dependence of the charge ( a ) , orbital ( b ) , and spin - charge ( c ) correlation functions calculated for the phase with c spin correlations . parameters : @xmath106 , @xmath97 , @xmath111 , @xmath108.,width=377 ] as shown in fig . [ fig : oonn_phase ] for large values of the superexchange interaction between core spins ( @xmath112 ) an af spin ground state is realized while with decreasing @xmath15 the kinetic energy of the @xmath5 electrons ( @xmath113 ) starts to play an active role via the de mechanism@xcite breaking some of the af bonds which leads to 1d fm chains which form straight lines ( c phase ) . further decrease of @xmath15 yields fm correlations in both @xmath34 and @xmath35 directions . as the kinetic energy is controlled by the inter - site coulomb repulsion @xmath13 and suppressed in the limit of @xmath114 we find the af region increasing in size with increasing @xmath82 . we discuss now in more detail the temperature dependence of the different correlation functions . starting from the fm region ( see fig . [ fig : corr_fm_nn ] ) we find the evolution of weak co for @xmath115 with nearest - neighbor @xmath116 charge correlations smaller than the @xmath117 and @xmath118 next - neighbor ones which is consistent with the alternating order of mn@xmath1 and mn@xmath2 ions . the orbital correlations are positive between all neighbors indicating the preference of @xmath6 orbital occupancy due to the kinetic energy which is quite large in the fm state . although all spin correlation functions are positive , they remain highly anisotropic as @xmath119 as a result of the strong competition between the de mechanism and af superexchange interaction between core spins . next , we consider the af spin state in the case of large @xmath15 and @xmath13 ( see fig . [ fig : corr_af_nn ] ) . increasing coulomb repulsion ( @xmath109 ) we find more distinct and almost isotropic alternation of the @xmath5 charge which now sets in already at @xmath120 . here , the orbital correlations are correlated with the onset of the af spin order at @xmath121 . for @xmath122 the spin - spin correlations are strong and represent an almost isotropic af spin state . for intermediate values of @xmath15 and moderate coulomb repulsion we find the phase with predominantly c spin correlations at @xmath122 ( see fig . [ fig : oonn_phase ] ) . as presented in fig . [ fig : corr_c_nn ] for @xmath123 and @xmath124 the charge correlations are similar to those obtained for the fm state shown in fig . [ fig : corr_fm_nn ] ( a ) while the spin ordering changes dramatically . moreover for @xmath122 the orbital correlations are weaker than those found in the fm state discussed above indicating the orbital tendency towards in - plane directional - type order . a convenient way to distinguish between phases with c or ce and other spin correlations is to evaluate a _ combined spin - charge correlation function _ being restricted to sites occupied by @xmath5 electrons , @xmath125 . analyzing the localized limit presented in fig . [ fig : spins ] ( or its by @xmath126 rotated version ) one can easily see that , for the c phase one should find @xmath127 for @xmath128 and @xmath129 for @xmath130 while for the ce phase only @xmath129 for @xmath128 with any other spin - charge correlations remaining small . in the case presented in fig . [ fig : corr_c_nn ] @xmath125 functions are negative ( positive ) for nearest - neighbors along diagonal ( next - nearest - neighbors along @xmath34/@xmath35 direction ) , respectively , in agreement with the arguments presented above . furthermore , nearest - neighbor spin - charge correlations are small as a result of averaging over different cluster orientations ( rotated by angles @xmath131 , @xmath132 ) . it is straightforward to see why the c - structure is favored by the nearest - neighbor jt interaction . the c - phase can be viewed as a realization of the alternating orbital structure of lamno@xmath4 but with a modulated charge density . hence this structure is compatible with the nearest - neighbor jt interaction , whereas in the ce - structure this interaction is frustrated . the role of local lattice distortions due to the jt effect , which creates an anisotropic strain field decaying as @xmath133 and thereby promotes long - range orbital ordering , was considered by khomskii ( see ref . ) . following these arguments , we treat the oo term as an elastic interaction which is active also between @xmath5 electrons on next - nearest - neighbor sites along @xmath34 or @xmath35 direction , @xmath134 where @xmath135 refers to next nearest - neighbors along @xmath34 and @xmath35 direction with @xmath136 which follows from the @xmath137 decay of elastic interactions.@xcite the existence of longer - range oo interactions is justified by its cooperative lattice - driven nature where an orbital flip on a single mn site distorts not only the neighboring oxygens but also positions of other manganese ions around it . here , the jt coupling preferring the tilting of the orbitals from the uniform orbital state leads to an alternating orbital order ( see fig . [ fig : ce_orb ] ) and in this way compensates for the kinetic energy gain in uniform @xmath41 orbital state favored in the 2d model . here we neglect non - central elastic interactions between neighboring mn sites in @xmath138 direction as their strength does not scale with @xmath139 ( see ref . ) and thus would involve introduction of an additional parameter in the model . schematic configuration of the charge - orbital ordered ce phase . the sites * a * and * b * indicate two orbital mn@xmath7 sublattices while empty circles stand for the mn@xmath14 ions . the dashed - line box indicates a typical orbitals orientation around a given mn@xmath14 ion.,width=226 ] phase diagram as in fig . including the nearest- and next - neighbor elastic interactions ( @xmath97 , @xmath136 ) . @xmath98 indicates a representative point in the ce region for which the correlation functions are presented here.,width=132 ] temperature dependence of the two - site correlations as in fig . but for the phase with ce spin and orbital correlations promoted by elastic interactions ( @xmath97 , @xmath136 ) . other parameters : @xmath140 , @xmath141 , @xmath108 . the inset in ( b ) shows the orbital correlations in the rotated basis : @xmath142.,width=377 ] the ce phase is favored as compared to the c phase by the oo coupling between second neighbors ( @xmath143 ) along @xmath34/@xmath35 direction . however , this interaction is only active ( not screened ) when strong charge ordering takes place ( large @xmath82 case ) and here the ce phase can persist even for small values of @xmath15 as long as the @xmath5 kinetic energy is sufficiently suppressed by a large @xmath13 ( see fig . [ fig : elastic_phase ] ) . thus , one has to assume large inter - site coulomb repulsion ( where almost perfect mn@xmath1/mn@xmath2 charge ordering@xcite is realized ) and small superexchange interaction [ see an example for @xmath144 and @xmath145 in fig . [ fig : corr_ce_el ] ] . in the @xmath28 basis , which we use in most of the numerical evaluations , the spin and orbital correlation functions with @xmath146 become large as @xmath119 while all the others remain very small in the whole range of temperatures . such small orbital correlations imply either that the respective orbitals are uncorrelated or that they order in such a way , that the occupied orbitals are linear combinations with almost equal amplitude expressed in the basis used for the measurement of correlation functions.@xcite this difficulty is not present in su(2 ) symmetric models , and appears here as consequence of the cubic symmetry in the orbital sector . therefore , the same functions were evaluated also in the @xmath142 basis as presented in the inset of fig . [ fig : corr_ce_el ] ( b ) . in the new orbital basis the function @xmath147 is more pronounced and saturates for @xmath148 at @xmath149 close to its maximum value @xmath150 indicating almost perfect ce orbital structure ( see fig . [ fig : ce_orb ] ) . shorter range @xmath151 correlations should become small when averaged over differently oriented clusters ( rotated by angles @xmath131 ) [ see fig . [ fig : spins ] ( d ) ] . from fig . [ fig : spins ] ( d ) one can see that a similar pattern as found for the two - site orbital correlations should be found for the @xmath125 functions . in our cluster in ideal ce state only electrons at distance @xmath128 can contribute to the charge - spin correlations ( being negative in this case ) which do not vanish when averaged over different cluster orientations . as shown in fig . [ fig : corr_ce_el ] ( c ) the @xmath125 correlations obtained here are consistent with the above characterization of the ce pattern with a strong negative signal dominating at distance @xmath128 . the existence of charge - ordered insulating state concomitant with orbital and spin order of the ce - type@xcite was observed in different half - doped manganites with small one - electron bandwidth.@xcite the average change in the orbital occupation can be studied calculating @xmath152 which changes between @xmath153 and @xmath154 for the uniform @xmath41 and @xmath92 orbital states , respectively , for the case with one @xmath5 electron per two mn sites . furthermore , when the orbital state involves @xmath41 and @xmath92 orbitals in the same proportion one finds @xmath155 . in fig . [ fig : tz_susc_el ] ( a ) we see that increasing the coulomb interaction from @xmath106 to @xmath109 the orbital ordering hardly changes for @xmath122 with @xmath152 only slightly decreasing to @xmath156 for @xmath109 . this indicates that as long as the system is in the fm state ( see fig . [ fig : elastic_phase ] ) the ground state is dominated by the @xmath6 orbital promoted by the kinetic energy term . further increase in the coulomb inter - site repulsion to @xmath140 leads to a drastic change in the orbital order when system enters the ce phase region ( see fig . [ fig : elastic_phase ] ) . here , @xmath157 for @xmath122 indicating dominating role played by the @xmath158 orbitals . this fact is consistent with our results for the two - site orbital correlation functions [ see inset in fig . [ fig : corr_ce_el](b ) ] . temperature dependence of the average value of the @xmath159 orbital operator ( a ) and the inverse of the staggered charge susceptibility @xmath160 ( b ) obtained for different values of the inter - site repulsion @xmath82 in the presence of elastic interactions ( @xmath97 , @xmath136 ) . other parameters : @xmath141 , @xmath108.,width=226 ] a change in the inter - site coulomb interaction is directly connected with the robustness of the charge ordering as seen in figs . [ fig : corr_fm_nn ] - [ fig : corr_c_nn ] ( a ) and fig . [ fig : corr_ce_el ] ( a ) . although the characteristic temperature of the co ( @xmath161 ) can be roughly estimated from the charge - charge correlation functions it can be more precisely extracted from the staggered charge susceptibility,@xcite @xmath162 : @xmath163 which at high temperatures follows a curie - weiss law @xmath164 . from fig . [ fig : tz_susc_el ] ( b ) one can easily estimate @xmath165 , @xmath166 , and @xmath167 for @xmath168 , @xmath169 , and @xmath170 , respectively , in agreement with the mean - field solution@xcite which for @xmath114 predicts , @xmath171 . although the model considered in the previous section with longer - range oo elastic coupling leads to the ce - type spin and orbital correlations , this phase is found only in a very extreme case of almost perfect mn@xmath14/mn@xmath7 co ( that is for strong coulomb repulsion ) which may not be very realistic in the light of some experimental facts indicating the cdw character of charge redistribution between different manganese ions.@xcite therefore in this section we derive the further neighbor oo interactions from a microscopic model including apart from the jt - distortion of the mn@xmath7 octahedra the breathing distortion around the mn@xmath14 ions as well . we then show that this interaction is capable to produce the zigzag ( spin and orbital ) phase in a region of small coulomb repulsion changing the previous phase diagrams in a drastic way . to assess the gain in lattice energy connected with the formation of the ce long - range orbital order we follow the work of millis for lamno@xmath4 by treating the lattice distortions classically.@xcite the effect of lattice displacements is described by assuming arbitrary values of four different lattice distances @xmath172 ( @xmath173 ) ( see fig . [ fig : distortions ] ) which are expressed by @xmath174 , @xmath175 uniform deformation along the @xmath34 and @xmath35 direction of the mn lattice , and @xmath176 displacements of o ions along the mn o mn bonds within the @xmath177 planes : @xmath178 where @xmath35 is the lattice constant of the ideal perovskite.@xcite moreover , we assume @xmath5 charge stacking in the @xmath179 direction with the respective mn mn distance being @xmath180 . schematic representation of the lattice distortions contributing to the second neighbor jt - interaction . the corresponding ce orbital order for the 5 atom structure above is indicated in fig . by dashed lines . different mn@xmath7 , mn@xmath14 , and o ions are presented by full , shaded and empty circles , respectively . the distances between the mn@xmath14 ion and the mn@xmath7 ions in @xmath181 [ @xmath182 and @xmath183 [ @xmath184 directions , respectively , are identical . the arrow indicates the displacement of the central mn@xmath14 ion when @xmath185.,width=283 ] phase diagram as in fig . including the local nearest- and next - neighbor oo interactions with @xmath97 and @xmath186 ( spring constants @xmath187 ) . @xmath98 indicate representative points in the ce region for which the correlation functions are presented here . the dashed arrow presents the direction of the transition from the localized ce to the itinerant electron a - phase at low temperatures.,width=132 ] in order to define orbital sublattices ( a and b ) we introduce a transformation which describes the tilting of orbitals by making two different transformations at both sublattices,@xcite @xmath188 = \left [ \begin{array}{cc } \cos\left(\frac{\pi}{4}\pm\phi\right ) & \sin\left(\frac{\pi}{4}\pm\phi\right ) \\ -\sin\left(\frac{\pi}{4}\pm\phi\right ) & \cos\left(\frac{\pi}{4}\pm\phi\right ) \end{array}\right ] \left [ \begin{array}{c } \end{array}\right ] , \ ] ] where @xmath33 ( @xmath32 ) refers to @xmath189 ( @xmath190 ) sublattice , respectively , and @xmath191 . here , @xmath192 and @xmath193 stands for local basis orbitals : @xmath41 and @xmath92 at site @xmath194 , respectively . in the rotated basis @xmath195 we assume for the transformed operators : @xmath196 ( @xmath197 ) for @xmath189 ( @xmath198 ) sublattice , respectively . at intermediate values of @xmath199 one can reach , e.g. , @xmath200/@xmath201 ( @xmath202/@xmath203)-type orbital ordering at @xmath204 ( @xmath205 ) , respectively . the rotation ( [ eq : rotation ] ) does not include complex orbital order which was proposed as a possible orbital order in manganites at smaller doping.@xcite the classical energy of the distorted lattice @xmath206 ( per mn site ) is given by,@xcite @xmath207 + k_1\left(u_1 ^ 2 + u_2 ^ 2\right),\ ] ] normalized per mn ion , where @xmath208 ( @xmath209 ) are the nearest - neighbor mn o ( mn mn ) spring constant , respectively . here the force constants include the factor @xmath210 , and hence have the dimension of an energy . next , we have to add the energy of the breathing mode ( bm ) describing the attraction between oxygen ions and the unoccupied mn@xmath2 site : @xmath211 here , all oxygens are equally attracted towards the central manganese ion . the bm induces charge alternation on neighboring mn sites and contributes to the effective inter - site charge - charge repulsion ( @xmath13 ) . the coupling constant to the breathing distortion is written as @xmath212 , where @xmath213 denotes the coupling to the jt distortion . the term describing the jt coupling depends on the character of the occupied orbital at a mn@xmath1 site . the jt interaction describes the coupling between the occupied @xmath5 orbital at a particular mn@xmath7 ion and the distortions of the surrounding oxygen ions , as introduced by millis.@xcite in the orbitally rotated state ( [ eq : rotation ] ) this energy has the following form , @xmath214\sin(2\phi)\right . \nonumber\\ & - & \left.\sqrt{3}\cos(2\phi)\left[\frac{1}{2}\left(\delta_x - \delta'_x\right ) + u_2 - u_1\right]\right\},\end{aligned}\ ] ] and depends on the rotation angle @xmath199 and on the deformation of the lattice . the optimal values of the ionic displacements for a given angle @xmath199 are found by minimization of the total energy , @xmath215 , which for ( [ eq : e_l])-([eq : jt ] ) and neglecting higher - order nonlinear terms gives , [ eq : dis ] @xmath216 } { 2\left(k_1 + 2k_2\right ) } , \\ \delta'_x & = & \frac{-\lambda\left[2\beta -\sin(2\phi)+\sqrt{3}\cos(2\phi)\right ] } { 2\left(k_1 + 2k_2\right ) } , \\ \delta_z & = & \frac{-\lambda\left[\beta+\sin(2\phi)\right]}{k_1 + 2k_2 } , \\ u_1 & = & \frac{-\lambda\left[2\beta+\sin(2\phi)+\sqrt{3}\cos(2\phi)\right ] } { 4k_1 } , \\ u_2 & = & \frac{-\lambda\left[2\beta+\sin(2\phi)-\sqrt{3}\cos(2\phi)\right ] } { 4k_1 } .\end{aligned}\ ] ] this yields for the total energy of the distorted lattice , @xmath217 + 2k_1+k_2\right\ } .\nonumber \\\end{aligned}\ ] ] this implies a change of the volume of the lattice with respect to the ideal perovskite ( @xmath218 ) @xmath219 . hence the volume decreases with increasing bm coupling @xmath220 , but does not depend on the character of the orbital order . the relative strength of the effective next- and nearest - neighbor jt interactions can be obtained by comparing the lattice stiffnesses of the orbital modulations in the undoped ( @xmath221 ) and the quarter - filled ( @xmath222 ) case , respectively . for _ undoped _ lamno@xmath4 the lattice energy was found to be [ see eq . ( 2.14 ) in ref . ] , @xmath223\right\ } , \label{eq : e^0}\ ] ] in the half - doped case the orbital alternation at mn@xmath7 neighbors is driven by the mn@xmath14 ion shift . the energy ( 21 ) can be decomposed @xmath224 , where @xmath225 is the gain of energy due to the mn@xmath14 displacement : @xmath226 . \label{eq : de}\ ] ] this cooperative energy does not depend on the breathing mode coupling @xmath220 directly , though there is an indirect dependence via @xmath227 influencing the optimal value for @xmath199 . contrary to the undoped case ( [ eq : e^0 ] ) , where the mn - mn potential ( @xmath228 ) is necessary to obtain finite orbital stiffness , in the half - doped case ( [ eq : de ] ) the mn - o interaction alone can lead to cooperative jt effect stabilizing the alternating orbital order . assuming @xmath229ev , @xmath230 ev , and @xmath231 ( see ref . ) one finds for the @xmath232 ( @xmath233 ) orbital order the ce energy contribution per mn ion @xmath234 mev . in table [ tab : ene ] we present the ratio @xmath235 , characterizing the stability of the ce orbital order , as function of the bm coupling strength @xmath220 and the ratio of spring constants @xmath236 . the largest relative gain in energy due to the mn@xmath14 ions shift ( @xmath237 ) is obtained for the @xmath238 orbital order ( @xmath233 ) which for small bm coupling ( @xmath239 ) can reach up to @xmath240 of the total lattice energy , whereas for larger values of @xmath220 , although @xmath225 is not changed , the total lattice energy increases rapidly as @xmath241 being dominated by the distortions of the mno@xmath16 octahedra rather then by their displacements . furthermore , with increasing mn lattice stiffness ( @xmath242 ) both the @xmath235 ratio and the shift of unoccupied mn@xmath14 ions [ @xmath243 decrease . .[tab : ene ] the relative energy gain , @xmath235 , associated with the mn@xmath14 ion displacement ( see fig . ) for the orbital orderings given by @xmath233 and @xmath244 presented as function of @xmath220 and @xmath236 . [ cols="<,^,^,^,^",options="header " , ] correlation functions in the case of a ce ground state ( as in fig . ) but obtained for the local oo interaction ( 26 ) with spring constants @xmath187 ( @xmath245 ) . other parameters : @xmath106 , @xmath97 , @xmath141 , @xmath108.,width=377 ] charge correlations in the case of a fm ground state as obtained for the local oo interaction with spring constants @xmath187 ( @xmath245 ) . other parameters : @xmath246 , @xmath97 , @xmath247 , @xmath108.,width=132 ] comparing the cooperative part [ @xmath248 of both jt energy contributions [ ( [ eq : e^0 ] ) and ( [ eq : de ] ) ] we end up with a simple relation between different oo interaction strengths , @xmath249 for @xmath250 the @xmath133 scaling of elastic interactions@xcite is recovered . in the above estimate we have neglected the superexchange interaction which can also contribute to the effective @xmath251 [ see _ e.g. _ eq . ( 2.7 ) in ref . ] . the relative strength of both spring constants in the co case can be assessed from the relative length of different bonds in the half - doped mno@xmath8 plane . as measured for low - temperature la@xmath252ca@xmath252mno@xmath4 superstructure@xcite the length of all four in - plane mn@xmath14-o bonds is very similar ( @xmath253 ) . neglecting the effect of mn - o - mn bond bending and assuming @xmath254 ( see fig . [ fig : distortions ] ) one obtains the relation , @xmath255 by using eqs . ( [ eq : dis ] ) . the above assessment together with ( [ eq : kappa_ratio ] ) underlines the importance of mn@xmath7mn@xmath7 orbital coupling ( @xmath222 ) in co manganites . the structural data of radaelli _ et al._@xcite may be used in combination with eqs . ( [ eq : dis ] a - e ) to infer the values for @xmath213 and @xmath220 . we have not attempted to determine the force constants as well , but assume @xcite @xmath230 ev . using basically data for the jt - distortion of mn@xmath7o in - plane bonds ( @xmath256 and @xmath257 ) and the mn@xmath14 displacement , we obtain for @xmath258 @xmath259 and @xmath260 ev , respectively . we note that the ratio @xmath261 is rather close to the values determined by millis@xcite for lamno@xmath4 . this also implies that the energy gain due to the breathing distortion is substantial [ see eq . ( [ eq : e_tot ] ) ] . we return to the many - body theory for a translation invariant system . as the oo interaction between further mn@xmath7 neighbors is mediated by an mn@xmath14 ion we introduce the following effective three - site interaction , @xmath262 where the operator @xmath263 is defined as in eq . ( [ eq : tij ] ) , and @xmath264 denotes three neighboring sites along @xmath34 or @xmath35 direction . the reasoning here is similar as for the nearest - neighbor jt interaction in eq . ( [ eq : h_oo ] ) , i.e. , the expectation value of the operator @xmath263 in an orbital ordered state depends on the orbital angle precisely as in eq . ( [ eq : de ] ) , namely @xmath265 $ ] . the @xmath266 factor in eq . ( [ eq:3site ] ) reflects the charge state of mn@xmath14 and favors co in the quarter - filled system , while the contribution of the interaction ( [ eq:3site ] ) would vanish in the undoped compound . investigating the model ( [ eq : h_total ] ) with both nearest - neighbor ( [ eq : h_oo ] ) and three - site ( [ eq:3site ] ) oo couplings we find a strong change in the low - temperature phase diagram ( see fig . [ fig : oonnn_phase ] ) as compared with the previous ones ( figs . [ fig : oonn_phase ] and [ fig : elastic_phase ] ) . assuming @xmath245 ( which corresponds to @xmath267 ) , the ce - like correlations set in already at rather small inter - site repulsion ( @xmath268 ) while for stronger coulomb interactions ( @xmath269 ) the spin correlations have predominantly af character . furthermore , the fm region is quite small and can be found only for @xmath270 and @xmath271 . when decreasing the ratio @xmath236 ( and keeping the value of @xmath139 constant ) the fm region shrinks at the expense of the ce region . next , we present the two - site correlations for one representative ( @xmath106 , @xmath141 ) point in the ce region in fig . [ fig : oonnn_phase ] . from fig . [ fig : corr_ce_nnn ] ( a ) one sees that the ce state can exist for less pronounced co than in the elastic strain model considered in sec . [ sec : elastic ] , where almost perfect mn@xmath7/mn@xmath14 co was necessary to stabilize the zigzag state ( see fig . [ fig : corr_ce_el ] ) . moreover , the evolution of ce - type orbital order with decreasing temperature is here correlated with the onset of co [ compare fig . [ fig : corr_ce_nnn ] ( a)-(b ) with [ fig : corr_ce_el ] ( a)-(b ) ] . regarding @xmath125 correlations , the ce - type spin order develops at lower temperatures than the charge and orbital order but at higher temperature than previously found for the strain induced interactions . the three - site term ( [ eq:3site ] ) which is active for a mn@xmath7mn@xmath14mn@xmath7 sequence along the @xmath34 and @xmath35 direction , respectively , promotes not only orbital alternation on neighboring mn@xmath7 sites but also strengthens co. this is clearly seen in the charge correlations which are now stronger than those _ e.g. _ in fig . [ fig : corr_fm_nn ] , although the same coulomb repulsion ( @xmath106 ) was assumed in both cases . going from the ce to the fm phase ( as indicated by an arrow in fig . [ fig : oonnn_phase ] ) one finds that the cdw is now reduced at low temperatures in the fm state ( fig . [ fig:<nn > ce ] ) , while the charge modulation is still rather strong at higher temperatures ( @xmath121 ) above the nel temperature where the electron kinetic energy is quenched due to the random @xmath0 spin orientations . this behavior is a clear manifestation of the de mechanism . we note that the ce - type orbital correlations are still present in the fm phase in fig . [ fig : oonnn_phase ] for @xmath272 , whereas for smaller values for @xmath13 @xmath273-@xmath274 orbital correlations dominate , which are characteristic for the metallic a - phase . finally , we consider possible mechanisms stabilizing the charge stacking of ce - ordered planes @xcite as observed in most charge - ordered manganites at doping @xmath275 . any interaction stabilizing such a stacking must overcome the inter - site charge - charge repulsion . there are two interactions which contribute to such a repulsion : ( i ) the electron - electron coulomb force @xmath13 ; and ( ii ) the bm coupling . both interactions will effectively tend to maximize the number of mn@xmath7mn@xmath14 bonds . the bm coupling contribution to the inter - site charge - charge repulsion ( @xmath276 ) can be estimated comparing the lattice energies with charge alternating and stacked along the @xmath179 direction which leads to the estimate @xmath277 ^ 2 /(2k_1)$ ] . assuming typical values for @xmath278 and @xmath279 ( see ref . ) together with our previous estimate @xmath280 one can obtain _ @xmath281 for the @xmath142 ce orbital order . in a multi - layer system the oo interaction along the @xmath179 direction is given by the nearest - neighbor coupling @xmath221 , which in the @xmath28 basis has the form,@xcite @xmath282 energy gained by @xmath5 electron stacked in the @xmath179 direction as function of charge modulation @xmath283 obtained for @xmath246 , @xmath186 and different values of @xmath251.,width=264 ] this interaction together with the @xmath222 coupling , which controls the ce - orbital order in the @xmath177 plane , in a charge - ordered case yields an approximate orbital model of the form , @xmath284 where @xmath56 ( @xmath285 ) are defined with respect to the @xmath142 orbital basis . defining @xmath286 and assuming that in the @xmath177 planes of a bilayer system @xmath287 and @xmath288 orbitals are occupied on two neighboring sites in the @xmath179 direction , the @xmath221 coupling [ see eq . ( [ eq : stacking ] ) ] rotates orbitals towards the @xmath289 configuration and leads to binding of two electrons with energy @xmath290 . for charge disproportion with mn@xmath291 ions alternating in the @xmath177 planes and assuming that the @xmath251 ( @xmath139 ) two ( three)-site interaction scales as @xmath292 with @xmath293 ( @xmath48 ) , respectively , one finds for the energy gain ( per mn ion ) due to charge stacking along @xmath179 direction for a 3d lattice : @xmath294 -2\delta^2v .\ ] ] the last term stands for the effective coulomb repulsion at given charge modulation . with charge modulation sufficiently reduced @xmath295 ( see fig . [ fig : stacking ] ) the oo inter - plane coupling can stabilize charge - stacked phase ( @xmath296 ) . such stabilization is strengthened ( weakened ) by increasing @xmath251 ( @xmath139 ) oo interaction , respectively . for smaller charge disproportion ( @xmath297 ) and considerable charge on mn@xmath298 sites the @xmath221 in - plane coupling [ neglected in eq . ( [ eq : stacking ] ) ] can also play an important role in the inter - plane charge modulation . however , when the co is strongly reduced the orbital ground state is too complex to be analyzed analytically . a further mechanism stabilizing the observed @xmath5 charge stacking along the @xmath179 direction results from the af inter - plane superexchange interaction.@xcite yet , the magnetic interaction ( @xmath299 ) results mainly from the @xmath300 superexchange which is of the order of only a few mev.@xcite thus , the antiferromagnetism in the @xmath179 direction observed in most of the ce structures appears to be a _ result _ rather than the _ origin _ of the stacking pattern of the charge / orbital order . we have investigated the potential of different electron - lattice mechanisms to stabilize the ce spin - orbital ordered phase in charge ordered half - doped manganites ( filled with one @xmath5 electron per two mn ions ) . our study is based on a generalized de model containing the @xmath5 orbital degrees of freedom , and also including inter - site coulomb repulsion ( @xmath13 ) and af superexchange ( @xmath15 ) between core spins . the effect of three types of orbital interactions derived from the cooperative jt effect have been studied : ( i ) nearest neighbor jt - coupling , which is responsible for the alternating orbital order in undoped lamno@xmath4 , ( ii ) in addition second neighbor jt - interaction with a strength as expected for elastic strain , and ( iii ) a 2nd neighbor jt interaction which accounts for the breathing distortion of the mn@xmath14 octahedra and is in general stronger than ( ii ) . a particular feature of this interaction , which has been proposed based on the classical model considerations in sec . [ sec : nnnjt ] , are the displacements of the mn@xmath14 ions in the ce - phase . we have studied these models employing finite temperature diagonalization on a @xmath78 cluster with periodic boundary conditions , thereby keeping the translational invariance . the evolution of the different phases with temperature is monitored by the calculation of correlation functions for charge , orbital and spin , respectively . starting out from the generalized fm kondo lattice model ( [ eq : h_total ] ) without any orbital interaction ( @xmath301 ) we find the orbital correlations dominated by the @xmath6 orbitals which are favored by the kinetic energy.@xcite this orbital state is characteristic for the metallic a - phase in wide - band manganites at half doping . for small af superexchange interactions ( @xmath302 ) between @xmath0 core spins the fm state is found in the spin sector , while for larger values of @xmath15 the ground state is that of a regular 2 sublattice antiferromagnet . even for large coulomb repulsion @xmath13 we found no evidence for c or ce structures for some intermediate values of the parameter @xmath15 . the inclusion of the nearest - neighbor jt - interaction ( i ) , however , was found to stabilize the af - c phase in a wide parameter range in the @xmath303 phase diagram . to obtain the more complex ce orbital and spin state not only the kinetic energy must be reduced by increasing the coulomb repulsion @xmath13 but also a further neighbor _ antiferro - orbital _ interaction must be included in the model . this interaction results from the electron - lattice coupling and is a generalization of the jt interaction which was found to play an important role in the parent lamno@xmath4 compound.@xcite in co compounds at quarter - filling the effect of the oo coupling will be dominated by interactions between further distant mn neighbors . we have shown that such interactions can produce the ce structure in both _ spin and orbital _ sectors . in the case of the three - site effective interaction ( @xmath143 ) the zigzag chains are found to be stable not only for modest inter - site repulsion ( @xmath268 ) but even in the limit of @xmath304 and provided @xmath251 is small , where the co is induced entirely by the three - site orbital coupling ( [ eq:3site ] ) . this effective next - nearest - neighbor oo coupling was derived taking into account the cooperative nature of both mn@xmath7o and mn@xmath7mn@xmath14 bonds length deformation . the free relaxation of the latter length leads in presence of co to orbital alternation in planar directions , and the pattern of directed orbitals typical for the orbital ce structure appears . consequently , with the help of the orbital topology,@xcite this gives rise to the ce af spin ordering for realistic values for @xmath15 . an important aspect of the second neighbor jt - interaction ( iii ) is its 3-site structure which involves a coupling to the charge at the mn@xmath14 ion . this implies that the evolution of the ce orbital correlations with temperature are linked to the evolution of co , while the magnetic ce correlations develop at lower temperature similar to pr@xmath3ca@xmath3mno@xmath4@xcite . this is different from ( ii ) , where first co develops , then oo and finally at lowest temperature af spin order . as discussed in sec . [ sec : parameters ] we did not find the ce spin and/or orbital phase considering a pure electronic model , i.e. , neglecting the electron - lattice coupling . although , the topological effect combined with the de mechanism leads to the preference of the ce over c phase when all @xmath5 ( bridge and corner ) orbitals are degenerate,@xcite the situation is more complex when such degeneracy is lifted by jt lattice distortions . as described in the appendix , one finds band insulators in both cases while the stability of the ce versus the c phase strongly depends on the crystal - field splittings ( @xmath305 , @xmath306 ) . an important issue is the degree of charge ordering in the ce phase . the ce spin - orbital structures obtained in our simulations appear in the regime of relatively strong co , as infered from our calculated charge correlation functions . our calculations also suggest that the second neighbor jt coupling even for small or vanishing coulomb repulsion can lead to a pronounced charge modulation . these results are consistent with the simple physical picture , that de perpendicular to the fm chains must be strongly suppressed to allow af superexchange to trigger the ce structure , which may not be the case for weak co. on the other hand , a recent study by mahadevan , terakura and sarma@xcite based on the band structure approach arrived at the conclusion that the charge difference between the two mn species in the ce phase is basically negligible , which would imply the absence of a significant breathing distortion of mn@xmath14 octahedra . further experimental information on the degree of lattice distortion in the ce phase will certainly help to test the proposed jt based mechanism for the stability of the ce phase . the charge modulations determined in our calculations may be influenced by numerical limitations which overestimate quantum fluctuations : ( i ) spins @xmath47 are used to describe @xmath0 electrons ; ( ii ) the calculations are made using a 2d cluster ; ( iii ) the oxygen ions are not present explicitly in the cluster and thus only _ dynamical _ interactions are included ( @xmath307 ) , which are mediated by oxygen displacements , while possible _ static _ lattice effects@xcite are neglected . certainly each of these approximations weakens the charge ordering and the spin - orbital order . it is , however , not straightforward to decide whether the region of the ce - phase in the phase diagram may grow or shrink , if quantum fluctuations are suppressed . finally we note that further neighbor jt interactions can also play an important role in the charge stripe formation at other commensurate carrier concentrations . for @xmath308 ( @xmath309 ) the dominant coupling arises from the nearest- ( next nearest- ) neighbor oo interactions , respectively , with a pronounced electron - hole asymmetry present . in the latter case with small inter - site coulomb repulsion ( @xmath13 ) the bi - stripe state@xcite would be preferred while larger values of @xmath13 would favor a `` wigner crystal '' charge arrangement.@xcite we would like to thank g. khaliullin , m. mayr and a. m. ole for stimulating discussions and valuable comments . j. b. acknowledges the support of the mpi fr festkrperforschung , stuttgart . the financial support by the polish state committee of scientific research ( kbn ) of poland , project no . 5 p03b 055 20 is also acknowledged ( j. b. ) . a quite appealing explanation for the stability of the ce as compared to the c phase was suggested in refs . by considering a 1d model , where in the ce phase a gap opens as a consequence of a topological sign in the hopping matrix - elements . however , the intra - site coulomb repulsion @xmath42 ( leading to some redistribution of the @xmath5 charge ) can destabilize the ce phase.@xcite here we investigate how these arguments are modified when orbital degeneracy is lifted due to jt distortions . we consider the hamiltonian : @xmath310 which contains the kinetic energy @xmath311 of an electron hopping along a fm chain@xcite with possible @xmath5 level shift at the bridge sites ( @xmath312 ) and level splitting at corner sites ( @xmath313 ) considered in two different orbital basis @xmath28 and @xmath314/@xmath315 orbitals are located between the split energy levels at the corners assuming @xmath316 . both kinds of level shifts can result from the lattice distortions . at the bridge site only the orbital @xmath317 ( @xmath315 ) is considered while the orthogonal one @xmath318 ( @xmath319 ) , respectively , is decoupled in chain direction . in momentum space this leads to a @xmath320 matrix problem : @xmath321 \left(\begin{array}{c } d_{b , k}\\ d_{x , k}\\ d_{z , k}\end{array } \right ) , \ ] ] where @xmath322 ( @xmath323 ) and @xmath324 ( @xmath325 ) are the hopping elements for the ce ( c ) phase , respectively , in the @xmath28 orbital basis . electron dispersion for the zigzag chain of the ce phase ( solid lines ) and for the linear chain of the c phase ( dashed lines ) calculated for different crystal - field splittings in the @xmath28 orbital basis at corner sites ( @xmath305 ) . the orbitals at the bridge sites are degenerate ( assuming @xmath326).,width=188 ] phase diagram of the chain model illustrating the stability of the ce against the c phase as function of the crystal - field splittings defined with respect to the orbital basis ( a ) @xmath28 and ( b ) @xmath327 , respectively.,width=188 ] calculating the band energies ( see fig . [ fig : bands ] ) one can easily determine the character of the stable phase . for the degenerate band model ( @xmath328 ) the c phase is metallic while in the ce phase a gap opens due to the _ topological sign_.@xcite the opening of the gap stabilizes the ce phase . when the degeneracy between @xmath41 and @xmath92 states at the corner sites ( @xmath329 ) is lifted , the c phase becomes insulating as well and for sufficiently large @xmath5 level splittings it can have a _ lower _ energy than the ce phase when the chain is doped with one electron per two mn sites [ see fig . [ fig : bands](b ) ] . the region of stability of the ce phase increases for positive ( negative ) @xmath305 when @xmath330(@xmath331 ) , respectively [ see fig . [ fig:1d_phase](a ) ] . generally , the ce phase is stabilized with respect to c , when the occupied bridge state approaches the lower energy level of the corner states . in a real half - doped crystal where in the ce phase mn@xmath33o bond lengths shrink ( expand ) in the @xmath179 direction [ @xmath177 plane ] , respectively , one would expect an energetic preference , due to the jt effect , for orbitals with stronger @xmath6 than @xmath332 character . thus , in the 1d model the relevant part of the phase diagram ( fig . [ fig:1d_phase ] ) is the region with @xmath333 and @xmath334 , i.e. , where the ce structure dominates . to relate our mean - field considerations to the model from sec . [ sec : nnjt ] the stability of the ce phase versus c was also considered with level splitting at the corner sites defined with respect to the orbital basis @xmath327 . such a splitting , directly related to the nearest - neighbor jt interaction ( [ eq : h_oo ] ) when @xmath335 , is operative in the c phase but vanishes by symmetry in the ce orbital phase . as shown in fig . [ fig:1d_phase](b ) the ce phase is stable for small splittings ( @xmath336 ) and in the @xmath335 , @xmath330 region . the latter is the relevant parameter range in the case of nearest neighbor jt interactions , and hence the model suggests the ce phase as stable phase . this , however , is in conflict with our numerical study of the 2d model , where only the c phase was found to be stable . as in 2d the nearest neighbor jt interaction stabilizes the c structure , whereas this interaction is frustrated in the ce phase . this effect is not included in the 1d model . a. e. bocquet , t. mizokawa , t. saitoh , h. namatame , and a. fujimori , , 3771 ( 1992 ) ; t. saitoh , a. e. bocquet , t. mizokawa , h. namatame , a. fujimori , m. abbate , y. takeda , and m. takano , _ ibid . _ * 51 * , 13 942 ( 1995 ) ; s. satpathy , zoran s. popovic and filip r. vukajlovic , , 960 ( 1996 ) .

let denote free random variables . for two linear forms and with real coefficients and we shall describe all distributions of such that and are free . for identically distributed free random variables with distribution we establish necessary and sufficient conditions on the coefficients such that the statements : + is a centered semicircular distribution ; and and are identically distributed ( ) ; are equivalent . in the proof we give a complete characterization of all sequences of free cumulants of measures with compact support and with a finite number of non zero entries . the characterization is based on topological properties of regions defined by means of the voiculescu transform of such sequences .

1110.1527

the intensive research on the asymptotic behaviour of random matrices induced more research on their infinitely dimensional limiting models as well . free convolution of probability measures , introduced by d. voiculescu , may be regarded as such a model @xcite , @xcite . the key concept of this definition is the notion of freeness , which can be interpreted as a kind of independence for non - commutative random variables . as in the classical probability the concept of independence gives rise to the classical convolution , the concept of freeness leads to a binary operation on the probability measures on the real line , the free convolution . many classical results in the theory of addition of independent random variables have their counterpart in this theory , such as the law of large numbers , the central limit theorem , the lvy - khintchine formula and others . we refer to voiculescu , dykema and nica @xcite , hiai and petz @xcite , and nica and speicher @xcite for an introduction to these topics . in many problems of mathematical statistics , conclusions are based on the fact that certain special distributions have important properties which permit the reduction of the original problem to a substantially simpler one , for instance via the notion of sufficiency . the simplest type of statistics of independent observations , admitting a fairly complete description of the mutual independence and identical distribution , are linear statistics . consider independent scalar random variables @xmath13 ( not necessary identically distributed ) and two linear statistics @xmath14 where @xmath15 are real constant coefficients . it turns out that the independence of the two linear statistics @xmath5 and @xmath6 essentially characterizes the normality of the variables @xmath16 . to be precise , the following assertion , due to darmois @xcite and skitovich @xcite , @xcite , holds . let @xmath5 and @xmath6 given by ( [ 1.1 ] ) be independent . then the random variables @xmath16 such that @xmath17 , i.e. , which enter in both @xmath5 and @xmath6 , have normal distributions . note that the converse proposition holds in the following form : if @xmath18 and all @xmath16 such that @xmath17 are normal , then @xmath5 and @xmath6 are independent . polya @xcite was the first who established that only the normal distribution leads to identically distributed linear statistics @xmath19 and @xmath20 , where @xmath19 and @xmath21 are independent and identically distributed . marcinkiewicz @xcite proved that distributions having moments of all orders and admitting the existence of a nontrivial pair of identically distributed linear statistics based on a random sample are normal . yu . v. linnik @xcite , @xcite described the class of symmetric distributions admitting identically distributed linear statistics and studied in detail the problem of characterizing the normal distribution via properties of such statistics . in this paper we give a complete description of those free random variables @xmath22 such that the linear statistics @xmath23 and @xmath24 are free . in addition we prove an analogue of yu . v. linnik s results @xcite , @xcite , @xcite , and give the solution of the problem of characterization of the semicircular distribution via identical distribution of linear statistics @xmath23 and @xmath24 , where @xmath0 are identically distributed free random variables . assume that @xmath25 is a finite von neumann algebra with normal faithful trace state @xmath26 . the pair @xmath27 will be called a _ tracial _ @xmath28-_probability space_. assume that @xmath25 is acting on a hilbert space @xmath29 . we will denote by @xmath30 the set of all operators on @xmath29 which are affiliated with @xmath25 and by @xmath31 the set of selfadjoint operators affiliated with @xmath25 . recall that a ( generally unbounded ) selfadjoint operator @xmath32 on @xmath29 is called _ affiliated _ with @xmath25 if all the spectral projections of @xmath32 belong to @xmath25 . the elements of @xmath31 will be regarded as ( possibly ) unbounded random variables . let @xmath33 . the distribution @xmath34 of @xmath35 is the unique probability measure on @xmath36 satisfying the equality @xmath37 for every bounded borel function @xmath38 on @xmath39 . recall that a family @xmath40 of elements of @xmath33 is said to be free if for all bounded continuous functions @xmath41 on @xmath39 we have @xmath42 whenever @xmath43 , and all alternating sequences @xmath44 of @xmath45 s , @xmath46 s , and @xmath47 s , i.e. , @xmath48 . bercovici and voiculescu @xcite proved that if @xmath49 are free random variables for @xmath50 , and @xmath51 is a selfadjoint polynomial in @xmath52 non - commuting variables , then the distribution of the random variable @xmath53 depends only on the distributions of @xmath54 . if @xmath55 , then the distribution of @xmath56 only depends on the @xmath57 and is called the _ additive free convolution _ of @xmath58 . denote this distribution by @xmath59 . let @xmath0 denote free random variables with distributions @xmath60 , respectively . consider two linear statistics @xmath61 with real coefficients @xmath3 and @xmath4 . in the sequel we assume without loss of generality that @xmath62 and @xmath63 for all @xmath50 . denote by @xmath64 the _ standard _ semicircular measure , i.e. , the measure with the density @xmath65 , where @xmath66 . we shall call measures with densities @xmath67 with some @xmath68 and @xmath69 semicircular . nica @xcite established that the stability of freeness under rotations characterizes semicircular random variables . lehner @xcite proved that there are free random variables @xmath70 which are not semicircular and such that @xmath71 and @xmath72 are free . hence the analogue of the darmois skitovich theorem fails in the free case if there are at least three random variables involved . we can nevertheless describe all free random variables @xmath0 for which the linear statistics @xmath5 and @xmath6 in ( [ 2.1 ] ) are free . our result extends the results of nica and lehner considerably . in order to formulate our first result we need the following notation . let @xmath73 be a given random variable with distribution @xmath7 such that @xmath74 for some @xmath75 and let @xmath76 , be its free copies . let @xmath77 be @xmath52-th primitive root of unity ( e.g. , @xmath78 ) and set @xmath79 following lehner @xcite , we define the @xmath52th free cumulant of the random variable @xmath35 to be @xmath80 in a short way . for a detailed definition of free cumulants see the monograph of nica and speicher @xcite as well . since the free cumulant @xmath81 depends on @xmath52 and the distribution @xmath7 of @xmath35 only , we will denote this cumulant by @xmath82 as well . [ 2.1ath ] consider free random variables @xmath83 , and let @xmath84 be real numbers such that @xmath85 and @xmath86 for @xmath87 , where @xmath88 , and @xmath89 for @xmath90 . the linear statistics @xmath5 and @xmath6 are free if and only if the distributions @xmath91 have compact supports and the free cumulants @xmath92 , satisfy the relations : @xmath93 for all @xmath94 and @xmath95 such that @xmath96 , and @xmath97 for @xmath98 . the following result describes all distributions @xmath91 in the previous theorem . let @xmath99 be real numbers . introduce the function @xmath100 denote by @xmath101 the component of @xmath102 which contains @xmath103 . [ 2.2ath ] a sequence @xmath104 of real numbers such that @xmath105 for @xmath106 , is a sequence of free cumulants of some probability measure with compact support if and only if every jordan curve , contained in @xmath107 and connecting @xmath108 and @xmath103 , contains a point of the boundary of @xmath101 . [ 2.2arem ] consider the set @xmath109 of free cumulant sequences of the form @xmath110 . then the set @xmath111 is closed in the space @xmath112 . we easily obtain from theorem [ 2.2ath ] the following consequence . [ 2.3acor ] a sequence @xmath104 of real numbers such that @xmath113 , @xmath105 for @xmath106 , and @xmath114 , with sufficiently small @xmath115 , is a sequence of free cumulants of some probability measure with compact support . note that bercovici and voiculescu @xcite proved a more general result than corollary [ 2.3acor ] and showed the failure of the well - known cramr and marcinkiewicz theorems in free probability theory . to illustrate these results in a low dimensional example we consider the case @xmath116 . [ 2.2acor ] a sequence @xmath117 is a free cumulant sequence of some probability measure if and only if @xmath118 , where @xmath119 with @xmath120 @xmath121 . ] we see from corollary [ 2.2acor ] that a sequence @xmath122 is a free cumulant sequence of some probability measure if and only if @xmath123 . this assertion was obtained by lehner ( oral communication ) by other means . now we consider the problem of the description of identically distributed free random variables @xmath0 with distribution @xmath7 such that the statistics @xmath5 and @xmath6 are identically distributed as well @xmath124 . following linnik @xcite , we introduce two entire functions of a complex variable @xmath125 : @xmath126 and @xmath127 where @xmath128 and @xmath129 are restricted as in ( [ 2.1 ] ) . it is easy to see that all zeros of the functions @xmath130 lie in a strip @xmath131 with some @xmath132 . we prove the following characterization of semicircular measures which is an analogue of linnik s result @xcite about a characterization of gaussian probability measures . recall that a probability measure @xmath7 is called degenerate if @xmath133 , where @xmath134 is the dirac measure concentrated at the point @xmath135 . [ 2.1th ] let @xmath136 . in order that , for some non - degenerate probability measure @xmath7 , the statement @xmath137 @xmath138 implies the statement @xmath139 @xmath7 is a semicircular measure , it is necessary and sufficient that the following conditions are satisfied : @xmath140@xmath141 @xmath46 is a simple and unique positive zero of the function @xmath130 , @xmath140@xmath142 @xmath143 for all @xmath144 . note that @xmath139 implies @xmath137 if @xmath145 . let @xmath7 be a semicircular measure with mean zero , then @xmath139 implies @xmath137 iff @xmath146 . from theorem [ 2.1th ] we obviously obtain the following consequences . [ 2.1^**co ] let @xmath147 and let @xmath148 . in order that , for some non - degenerate probability measure @xmath7 , the statements @xmath137 and @xmath139 are equivalent , it is necessary and sufficient that the conditions @xmath141 and @xmath142 are satisfied . [ 2.1^*co ] let @xmath136 . in order that , for some non - degenerate probability measure @xmath7 with a median equal to @xmath108 , the statements @xmath137 and @xmath139 are equivalent , it is necessary and sufficient that the conditions @xmath141 and @xmath142 are satisfied . moreover theorem [ 2.1th ] implies the following assertion . [ 2.1co ] assume that @xmath149 , and @xmath150 . furthermore , assume @xmath138 . then either @xmath7 is a semicircular measure or @xmath7 is a degenerate probability measure . corollary [ 2.1co ] is an analogue of a result by polya @xcite . for analogues of the polya result in non - commutative probability theory see lehner @xcite . we prove as well the following result for symmetric probability measures . [ 2.1th * ] let @xmath136 . in order that , for some non - degenerate probability measure @xmath7 , which is symmetric with respect to @xmath108 , the statements @xmath137 and @xmath139 are equivalent , it is necessary and sufficient that the condition @xmath141 is satisfied . the following result for probability measures @xmath7 with moments of finite order is an analogue of a result by linnik @xcite in classical probability theory . [ 2.2th ] assume that @xmath136 and that @xmath130 has zeros in @xmath151 . let @xmath152 denote the maximum of the real parts of such zeros . in order that , for some non - degenerate probability measure @xmath7 such that @xmath153 with @xmath154 $ ] , the statement @xmath137 implies the statement @xmath139 it is necessary and sufficient that @xmath155 and @xmath156 for all positive integers @xmath157 . we prove in lemma [ 4.1lem ] that if @xmath138 , then the function @xmath130 has a real root @xmath152 such that @xmath158 . in the case where all moments of @xmath7 exist we obtain from theorem [ 2.2th ] the following result . [ 2.3th ] assume that @xmath136 . in order that , for some non - degenerate probability measure @xmath7 such that @xmath159 for all @xmath160 , the statement @xmath137 implies the statement @xmath139 it is necessary and sufficient that @xmath155 and @xmath156 for all positive integers @xmath157 . theorem [ 2.3th ] is an analogue of a result by marcinkiewicz @xcite in classical probability theory . denote by @xmath161 the family of all borel probability measures defined on the real line @xmath36 . on the set @xmath161 define two associative composition laws denoted @xmath162 and @xmath163 . let @xmath164 . the measure @xmath165 will denote the classical convolution of @xmath166 and @xmath167 . in probabilistic terms , @xmath165 is the probability distribution of @xmath168 , where @xmath32 and @xmath169 are ( commuting ) independent random variables with distributions @xmath166 and @xmath167 respectively . the measure @xmath170 is the free ( additive ) convolution of @xmath166 and @xmath167 introduced by voiculescu @xcite for compactly supported measures . the free convolution was extended by maassen @xcite to measures with finite variance and by bercovici and voiculescu @xcite to the whole class @xmath161 . thus , @xmath170 is the distribution of @xmath168 , where @xmath32 and @xmath169 are free random variables with distributions @xmath166 and @xmath167 , respectively . let @xmath171 denote the open upper ( lower ) half of the complex plane . if @xmath172 , denote its cauchy transform by @xmath173 following maassen @xcite and bercovici and voiculescu @xcite , we introduce the _ reciprocal cauchy transform _ @xmath174 the corresponding class of reciprocal cauchy transforms of all @xmath172 we denote by @xmath175 . this class admits a simple description . recall that the nevanlinna class @xmath176 is the class of analytic functions @xmath177 the class @xmath175 is the subclass of nevanlinna s functions @xmath178 such that @xmath179 as @xmath180 non - tangentially to @xmath39 ( i.e. , such that @xmath181 stays bounded ) , and this implies that @xmath178 has certain invertibility properties . ( for details see akhiezer and glazman @xcite , akhiezer @xcite ) . to be precise , for two numbers @xmath182 we set @xmath183 then for every @xmath184 there exists @xmath185 such that @xmath178 has the right inverse @xmath186 defined on @xmath187 . the function @xmath188 will be called the voiculescu transform of @xmath7 . it is not hard to show that @xmath189 for @xmath190 where @xmath191 is defined . note that @xmath192 as @xmath193 , @xmath194 . in the sequel we will denote @xmath195 by @xmath196 for a random variable @xmath35 with a distribution @xmath7 as well . it is easy to verify that @xmath197 for fixed @xmath198 and @xmath199 , where @xmath200 and @xmath201 are defined . in the domain @xmath187 , where the functions @xmath202 , @xmath203 , and @xmath204 are defined , we have @xmath205 this characterization for the distribution @xmath170 of @xmath168 , where @xmath32 and @xmath169 are free random variables , is due to voiculescu @xcite . he considered compactly supported measures @xmath7 . the result was extended by maassen @xcite to measures with finite variance ; the general case was proved by bercovici and voiculescu @xcite . we need the following auxiliary results . the following proposition is proved in @xcite . [ 3.4pr ] for every probability measure @xmath7 we have @xmath206 where @xmath207 as @xmath208 . the following lemma is well - known , see @xcite . [ 3.4abl ] let @xmath7 be a probability measure such that @xmath209 then the following relation holds @xmath210 uniformly in the angle @xmath211 , where @xmath212 . conversely , if for some function @xmath213 the relation @xmath214 holds with real numbers @xmath215 for @xmath216 , then @xmath217 admits the representation @xmath218 , where @xmath7 is a probability measure with moments @xmath219 . by lemma [ 3.4abl ] and the cartier good formula for free random variables ( see lehner @xcite ) , we easily obtain the expansion of the function @xmath195 . see @xcite as well . [ 3.5pr ] for every probability measure @xmath7 such that @xmath220 for a non - negative integer @xmath52 we have @xmath221 where @xmath222 , are the free cumulants of the probability measure @xmath7 . conversely , if for some function @xmath195 the relation @xmath223 holds with real coefficients @xmath224 , then @xmath7 has a finite moment @xmath220 and @xmath225 . we also need the following well - known result ( see for example @xcite ) . [ 3.4abcpr ] in order that a probability measure @xmath7 has a compact support it is necessary and sufficient that the sequence @xmath226 of free cumulants of this measure satisfies the inequality @xmath227 with some constant @xmath228 . we introduced the definition of the free cumulants in section 2 . let us recall the definition of mixed free cumulants as well . let @xmath229 be random variables and let @xmath230 , denote free copies . set @xmath231 as in ( [ 2.2 ] ) . following lehner @xcite , the nth mixed cumulant may be defined via @xmath232 in a short way . we will use the following known results , see @xcite and @xcite . [ 3c.1th ] consider a non - commutative probability space @xmath27 and let @xmath233 be the corresponding free cumulant functionals . consider random variables @xmath234 in @xmath31 . then the following two statements are equivalent . \(i ) @xmath234 are freely independent . \(ii ) we have @xmath235 for all @xmath236 and @xmath237 such that at least two of these @xmath52 indices @xmath238 are different . [ 3c.1pr ] let @xmath239 , be an affine transformation of @xmath0 , then we have , for @xmath240 , @xmath241 [ 3c.2pr ] mixed cumulants vanish . that is , if there is a nontrivial subset @xmath242 $ ] @xmath243i.e . , @xmath244 and @xmath245 $ ] , @xmath246:=\{1,\dots , n\})$ ] such that @xmath247 and @xmath248\setminus i}$ ] are free , then @xmath249 . now we prove an analogue of a known lemma of linnik for the characteristic functions ( see @xcite , ch . 1 , 6 ) . recall that a probability measure @xmath7 is symmetric if @xmath250 for any real borel set @xmath109 . it is not difficult to verify ( see @xcite ) that @xmath7 is symmetric if and only if @xmath251 takes imaginary values for @xmath252 , where @xmath251 is defined . [ 3.5l ] let @xmath7 be a symmetric probability measure and @xmath253 be a sequence of positive numbers such that @xmath254 . if , for all @xmath47 , @xmath255 , where @xmath256 is a symmetric probability measure with compact support , then @xmath257 . we shall show that @xmath7 has moments @xmath258 of all orders and that @xmath259 for all @xmath260 . the proof proceeds by induction for even @xmath52 . from the assumptions of the lemma we see @xmath261 where @xmath262 as @xmath263 , and @xmath264 is sufficiently large positive integer . by ( [ 3.5l1 ] ) , we obtain the following equation , using the symmetry of the measures @xmath7 and @xmath256 , @xmath265 we shall prove by induction that @xmath266 for all @xmath267 . letting @xmath268 , we conclude from ( [ 3.5l2 ] ) that @xmath269 and @xmath270 . now suppose that , for all @xmath271 , @xmath272 exists and @xmath273 . using ( [ 3.5l1 ] ) and the formula @xmath274 we arrive at the relation @xmath275 letting here @xmath268 , we obtain @xmath220 and @xmath276 that was to be proved . it remains to note that since @xmath266 for all @xmath267 and the measure @xmath256 has compact support , we have @xmath257 . hence the lemma is proved . voiculescu @xcite , maassen @xcite and in the general case biane @xcite proved that there exist unique functions @xmath277 and @xmath277 from the class @xmath175 such that @xmath278 in addition @xmath279 . the relation ( [ 3a.1a ] ) was proved by purely analytic methods by chistyakov and gtze @xcite and belinschi and bercovici @xcite . introduce the class @xmath280 $ ] in the following way . a function @xmath281 is in class @xmath280 $ ] if \1 ) @xmath281 is in class @xmath176 , and \2 ) @xmath281 is holomorphic and positive in the interval @xmath282 , and holomorphic and negative in the interval @xmath283 . the following theorem is due to krein @xcite . [ 3.5th ] a function @xmath281 is in class @xmath280 $ ] if and only if it admits a representation @xmath284}\frac{\sigma(dt)}{t - z},\ ] ] where @xmath285 is a finite non - negative measure . we now prove a free analogue of result by wintner , see @xcite , ch . 3 , 2 . [ 3.6lem ] assume that @xmath286 , where @xmath7 has compact support . then @xmath166 and @xmath167 have compact support as well . it suffices to prove the lemma for the measure @xmath166 . the proof for the measure @xmath167 is similar . by ( [ 3a.1a ] ) , there exists @xmath287 such that @xmath288 . hence we obtain the relation @xmath289}\frac{\mu(du)}{z - u}= \int\limits_{\bbb r}\frac{\mu_1(du)}{z(z)-u},\quad z\in\bbb c^+,\ ] ] with @xmath290 . since @xmath291 as @xmath208 non - tangentially to @xmath39 and @xmath292 for @xmath293 , then @xmath294 and we may write @xmath295 where @xmath296 is a probability measure for every @xmath297 and @xmath298 is a measurable function for every borel @xmath109 set in @xmath36 . using this representation we deduce from ( [ 3.6lem1 ] ) @xmath299 for every borel @xmath109 set in @xmath36 let @xmath300 . we see from ( [ 3.6lem3 ] ) that @xmath301 , where @xmath302 . therefore , for every point @xmath303 , the measure @xmath304 has support contained in @xmath305 $ ] . it remains to show that @xmath166 has a bounded support . return to ( [ 3.6lem2 ] ) with @xmath306 . by theorem [ 3.5th ] , the function @xmath307 is holomorphic and real for @xmath308 and for @xmath309 . since @xmath310 admits the representation @xmath311 where @xmath312 and @xmath256 is a finite nonnegative measure , it follows from the stieltjes perron inversion formula that the measure @xmath256 has bounded support contained in @xmath305 $ ] . thus @xmath313}\frac{(1+t^2)\,\nu(dt)}{t - z},\ ] ] where @xmath314 . the parameter @xmath152 and the measure @xmath256 depend on @xmath315 only and do not depend on @xmath316 . let @xmath317 and be sufficiently large , i.e. , @xmath318 . then , by ( [ 3.6lem5 ] ) , @xmath319 for @xmath320 , a contradiction with ( [ 3.6lem2 ] ) for @xmath321 and @xmath322 . an analogous argument holds for @xmath323 . hence there exists @xmath324 such that the points @xmath303 satisfy the inequality @xmath325 . the lemma is proved . in this section we first describe some results ( see kagan , linnik , rao @xcite ) on continuous solutions of special equations which were used to characterize distributions via independence of linear statistics and identical distribution of linear statistics . [ 3.1aal ] consider the following equation , for @xmath326 , @xmath327 where @xmath328 is a polynomial of degree @xmath47 ; @xmath329 and @xmath330 are complex valued functions of two real variables @xmath38 and @xmath331 . we assume that @xmath9 the numbers @xmath4 are all distinct @xmath10 the functions @xmath332 , and @xmath333 are continuous . then , in some neighborhood of the origin , all the functions @xmath332 , and @xmath333 are polynomials of degree at most @xmath334 . consider the equation @xmath335 for a bounded continuous function @xmath336 defined on @xmath337 . here @xmath338 and @xmath339 are nondecreasing functions satisfying the condition @xmath340 for some @xmath341 . we assume that the relation ( [ 3.1 ] ) is nondegenerate , i.e. , @xmath342 . applying a mellin transform to ( [ 3.1 ] ) we easily obtain , for @xmath343 , @xmath344 in view of ( [ 3.2 ] ) we deduce from ( [ 3.3 ] ) , for @xmath345 , @xmath346 where @xmath347 the functions @xmath348 and @xmath349 are analytic in the half - plane @xmath350 , and the function @xmath351 is analytic in the half - plane @xmath352 . we use the relation ( [ 3.4 ] ) for analytic continuation of @xmath351 into the half - plane @xmath353 as a meromorphic function . keeping the same notation , we have @xmath354 the singularities of @xmath351 in the half - plane @xmath353 happen to be poles distributed among the zeros of @xmath348 . taking an arbitrary @xmath355 , the inversion formula for the mellin transform yields @xmath356 let @xmath357 be some zero of @xmath348 of multiplicity @xmath358 in the half - plane @xmath353 . we see that @xmath359 where @xmath360 is a polynomial of degree at most @xmath361 . these residues may depend on @xmath362 , but it easily seen that the degree of the polynomial @xmath360 does not depend on @xmath362 . if @xmath363 , we call the number @xmath364 an _ active exponent _ of the solution @xmath336 and the number @xmath365 will be called the _ multiplicity _ of the active exponent @xmath366 . the leading coefficient of @xmath360 will be denoted @xmath367 and the degree by @xmath368 . all the active exponents of @xmath336 are located in the half - plane @xmath369 . if the number of the active exponents @xmath370 of @xmath336 is finite , then , by jordan s lemma on residues , it follows from ( [ 3.6a ] ) that @xmath371 we shall introduce the notation @xmath372 we need the following results on active exponents and differentiable solutions @xmath336 . [ 3.1lem ] if @xmath373 is a continuous solution of @xmath374 such that @xmath375 as to @xmath376 , then @xmath377 is an active exponent and @xmath378 [ 3.3lem ] if under the hypothesis of lemma @xmath379 the function @xmath336 has a continuous derivative @xmath380 for some @xmath381 and for all @xmath382 and the following limit exists and is finite @xmath383 then all the active exponents @xmath384 of @xmath336 which are not simultaneously integers and simple active exponents satisfy the condition @xmath385 . in this section we prove theorem [ 2.1ath ] . in order to prove theorem [ 2.1ath ] we follow the proof of the classical darmois - skitovich theorem ( see @xcite ) . _ necessity_. assume that free random variables @xmath0 with distributions @xmath386 , @xmath387 , respectively , are such that the linear statistics @xmath5 and @xmath6 ( see ( [ 2.1 ] ) ) are free and the coefficients @xmath388 of these statistics satisfy the assumptions of theorem [ 2.1ath ] . then for every pair of real numbers @xmath389 the linear statistics @xmath390 and @xmath391 are free and we have the relation @xmath392 using ( [ 2.3 * ] ) , we deduce from ( [ ds.1 ] ) that @xmath393 for @xmath199 with some @xmath184 and @xmath394 , where all functions @xmath395 , and @xmath396 are defined . hence ( [ ds.2 ] ) holds for @xmath397 and for @xmath398 and @xmath399 with sufficiently small @xmath400 . note that the functions @xmath401 @xmath402 depend on @xmath38 or @xmath331 only . consider the functions @xmath403 , for @xmath404 and @xmath405 with sufficiently small @xmath406 . since @xmath407 , and @xmath408 as @xmath409 , we see that ( [ ds.2 ] ) with @xmath397 has the form @xmath410 where @xmath411 , and @xmath412 are complex - valued continuous functions , and @xmath413 are all distinct . then , by lemma [ 3.1aal ] , the functions @xmath411 , are polynomials of degree @xmath414 . therefore we have the representation @xmath415 for @xmath416 with some @xmath417 and @xmath418 , where @xmath419 are complex valued coefficients . since @xmath420 as @xmath421 , all @xmath422 . in view of the relations @xmath423 we see that @xmath424}(-1)^l\frac { d_{2l , j}}{y^{2l-1}}\ ] ] and @xmath425}(-1)^l\frac { d_{2l+1,j}}{y^{2l}}\ ] ] for @xmath426 . we easily deduce from the last two relations that the coefficients @xmath419 are real - valued . then , by proposition [ 3.5pr ] , all moments @xmath427 exist and @xmath428 and @xmath429 for @xmath430 . by proposition [ 3.4abcpr ] , the measures @xmath431 , have compact supports . we now return to the relation ( [ ds3 ] ) . by ( [ ds4 ] ) , the functions on both sides of ( [ ds3 ] ) are differentiable . differentiating sequentially both sides of ( [ ds3 ] ) with respect to @xmath38 and @xmath331 we obtain a relation from which ( [ 2.4 ] ) follows immediately . _ sufficiency_. assume that free random variables @xmath0 satisfy the assumptions of theorem [ 2.1ath ] . consider mixed cumulants @xmath432 such that @xmath433 indices @xmath238 are equal 1 and @xmath434 indices @xmath238 are equal 2 . then , by propositions [ 3c.1pr ] , [ 3c.2pr ] , and by ( [ 2.4 ] ) , @xmath5 and @xmath6 have vanishing mixed cumulants @xmath435 for @xmath94 and @xmath436 . in addition we clearly have @xmath437 for all @xmath98 and @xmath436 . hence , by theorem [ 3c.1th ] , the linear forms @xmath5 and @xmath6 are free independent . the theorem is completely proved . first we shall prove theorem [ 2.2ath ] . _ sufficiency_. we conclude from the definition of the domain @xmath101 that its boundary is a jordan curve consisting of a finite number of curves parametrized by algebraic functions . moreover , by the assumptions of the theorem , there exist real numbers @xmath438 and @xmath341 such that @xmath439 , where @xmath440 is the half - line @xmath441 with @xmath442 , @xmath443 is is a jordan curve laying in @xmath444 and connecting the point @xmath445 and @xmath446 , @xmath447 is the half - line @xmath441 with @xmath448 . note that the function @xmath449 is analytic such that @xmath450 we shall show that the function @xmath451 defined via @xmath452 takes every value in @xmath444 precisely once . the inverse @xmath453 thus defined is in the class @xmath454 . let @xmath455 be a sufficiently large positive number . for every fixed @xmath456 we consider a closed rectifiable curve @xmath457 consisting of a curve @xmath458 , which is a part of the curve @xmath152 , connecting @xmath459 to @xmath455 , and the circular arc @xmath460 with @xmath461 connecting @xmath455 to @xmath459 . the curve @xmath462 depends on @xmath455 . we see from the construction of the curve @xmath458 that if @xmath125 runs through @xmath458 the image @xmath463 lies on the interval @xmath464 $ ] , where @xmath465 and @xmath466 . here @xmath467 as @xmath468 . we note as well that if @xmath125 runs through @xmath469 the image @xmath463 lies in the domain @xmath470 . hence @xmath471 winds around @xmath472 once when @xmath125 runs through @xmath457 , and it follows from the argument principle that there is a unique point @xmath357 in the interior of the curve @xmath457 such that @xmath473 . since this relation holds for all all curves @xmath474 with sufficiently large @xmath475 , we deduce that the point @xmath357 is unique in @xmath101 . therefore the inverse function @xmath476 exists and is analytic on @xmath477 . by condition ( [ ch.c.2 ] ) , @xmath478 as @xmath180 non - tangentially to @xmath39 and therefore @xmath479 . hence there exists a probability measure @xmath7 such that @xmath480 and the sufficiency of the assumptions of the theorem is proved . _ necessity_. let there exist a jordan curve @xmath152 , laying in @xmath101 and connecting @xmath108 and @xmath103 . we shall show that in this case it does not exist a probability measure @xmath7 such that @xmath480 . assume to the contrary that there exists a probability measure @xmath7 such that @xmath480 for @xmath199 with some positive @xmath481 and @xmath482 . then @xmath480 for @xmath483 and @xmath484 with a sufficiently large constant @xmath228 . by proposition [ 3.4abcpr ] , @xmath7 has a compact support . it is obvious that the relation @xmath485 holds for @xmath486 and therefore it holds for @xmath487 . since @xmath488 as @xmath489 , and @xmath490 as @xmath208 , the relation ( [ ch.c.3 ] ) with @xmath491 and @xmath489 leads to a contradiction . this proves the necessity of the assumptions of the theorem and completely proves the theorem . * proof of remark [ 2.2arem]*. since probability measures corresponding to the sequences of free cumulants of the set @xmath109 have the uniformly bounded support , the conclusion of the remark follows immediately from the fact that @xmath109 is conditionally compact . @xmath492 * proof of corollary [ 2.2acor]*. let @xmath493 . without loss of generality we assume that @xmath494 . it follows from theorem [ 2.2ath ] that if the sequence @xmath495 is a sequence of free cumulants of some probability measure then the polynomial @xmath496 has at the least one positive root for every fixed @xmath497 $ ] . denote @xmath498 the maximum of such roots . assume that @xmath499 $ ] . in this case @xmath500 , has one positive root @xmath501 and this root is a continuous function on @xmath502 $ ] . let @xmath503 . then @xmath504 has two positive roots only , say @xmath505 and @xmath506 ( @xmath507 , and @xmath508 as @xmath509 . these roots are continuous functions on @xmath510 and @xmath511 . in addition @xmath512 for @xmath513 and for @xmath514 . consider the function @xmath515 for @xmath516 . note that @xmath517 for @xmath518 and @xmath519 for @xmath520 . it is easy to verify that the function @xmath521 , is strictly monotone for @xmath522 and for @xmath523 and has a unique minimum at the point @xmath524 . note that @xmath525 if @xmath526 . in this case @xmath527 . note as well that @xmath528 for @xmath518 and @xmath529 for @xmath520 . now let us show that if @xmath117 is a free cumulant sequence , then @xmath530 let to the contrary @xmath531 for some @xmath532 . then we have the inequality @xmath533 fix a parameter @xmath534 with @xmath535 . the line @xmath536 is a tangent to the curve @xmath537 at the point @xmath538 iff @xmath539 . let @xmath540 . then @xmath504 has a positive root iff @xmath541 . assume that @xmath542 then , by ( [ 2.2acor.1 ] ) , this inequality holds for @xmath543 and there exists @xmath544 such that @xmath512 for @xmath545 , a contradiction with the assumptions of theorem [ 2.2ath ] . hence @xmath546 . it is easy to see that , for every fixed @xmath540 , @xmath504 has a positive root iff @xmath504 has two positive roots only , say @xmath547 and @xmath548 ( @xmath549 , and @xmath550 , are continuous functions on @xmath551 $ ] . in addition @xmath552 and @xmath512 for @xmath553 and for @xmath554 . hence we can conclude that @xmath498 is a continuous function for @xmath555 $ ] . now assume that @xmath556 . let @xmath557 . it is easy to see in this case , in view of ( [ 2.2acor.1 ] ) , that @xmath504 has three positive roots , say @xmath558 . these functions are continuous for @xmath559 $ ] and @xmath560 , @xmath561 . in addition @xmath498 is a continuous function for @xmath562 $ ] . by ( [ 2.2acor.1 ] ) , we see that , for @xmath563 , @xmath504 has only one positive root @xmath564 . since three positive roots of @xmath500 , with fixed @xmath565 coincide at the point @xmath566 only , we note that @xmath567 and @xmath568 . then there exists a jordan curve in @xmath444 containing @xmath108 and @xmath103 on which @xmath569 , a contradiction . thus @xmath570 and ( [ 2.2acor.1a ] ) is proved . we can rewrite this assumption in the following form . if @xmath571 , then @xmath572 and we have the estimate @xmath573 . if @xmath520 , then @xmath574 and @xmath575 . in this case we have the estimate @xmath576 . now we assume that ( [ 2.2acor.1a ] ) holds . let us show that @xmath117 is a free cumulant sequence . repeating the previous arguments we conclude that @xmath498 is a continuous function for @xmath577 $ ] . finally note that if @xmath578 $ ] and @xmath574 , then , for such fixed @xmath579 the polynomial @xmath504 has one positive root @xmath580 only . the function @xmath580 is continuous on @xmath578 $ ] and such that @xmath581 and @xmath582 . therefore @xmath498 is a continuous function for all @xmath497 $ ] and @xmath583 . hence the assumption ( [ 2.2acor.1a ] ) is necessary and sufficient in order that the sequence @xmath495 is a sequence of free cumulants of some probability measure in the case @xmath584 . assume that @xmath585 . then @xmath586 , and , by theorem [ 2.2ath ] , the sequence @xmath495 is not a sequence of free cumulants of a probability measure . we now assume that @xmath587 . note that @xmath588 has a positive root iff @xmath573 . but it is easy to see that under this condition @xmath588 has a positive root for every @xmath497 $ ] and @xmath498 is a continuous function on @xmath589 $ ] . hence the condition @xmath573 is necessary and sufficient in order that the sequence @xmath495 is a sequence of free cumulants of some probability measure . assume that @xmath590 and @xmath591 . in this case @xmath592 , if @xmath593 and @xmath594 , if @xmath595 . assume that @xmath596 . we easily conclude from the previous arguments that the condition @xmath123 is necessary and sufficient in order that the sequence @xmath597 is a sequence of free cumulants of some probability measure . it remains finally to note that the assertion of the corollary for @xmath598 follows immediately from remark [ 2.2arem ] . in order to prove theorem 2.1 we need the following results . the first of them is a description of @xmath163-stable distributions ( see @xcite and @xcite ) . [ 5.1lem ] every @xmath163-stable probability measure is equivalent to a unique probability measure whose voiculescu transform is given by one of the following @xmath137 @xmath599 ; @xmath139 @xmath600 with @xmath601 ; @xmath602 @xmath603 , @xmath604 with @xmath605 ; @xmath606 @xmath607 with @xmath608 . here and in the sequel we choose the principal branch of the functions @xmath609 and @xmath610 . the stability index of a @xmath163-stable probability measure is equal to @xmath46 in case ( 1 ) , to @xmath481 in cases ( 2 ) and ( 4 ) , and to 1 in case ( 3 ) . the parameter @xmath611 which appears in cases ( 2 ) , ( 3 ) and ( 4 ) will be called the _ asymmetry coefficient _ , and one can see that the measure corresponding to the parameters @xmath612 is the image of the measure with parameters @xmath613 by the map @xmath614 on @xmath36 . the next two lemmas are an analog of a result by linnik ( see @xcite , @xcite ) . [ 5.2lem ] let @xmath615 and @xmath616 , where @xmath160 . the function @xmath617 with sufficiently small parameter @xmath115 is the voiculescu transform of some symmetric probability measure . define the following region : @xmath618 where @xmath619 . the region @xmath620 is a jordan domain with boundary curve @xmath152 which is given by the equation @xmath621 , where @xmath622 is defined by the equation : @xmath623 we see that ( [ 5.1 ] ) has an unique solution @xmath622 which is greater than 1 for @xmath624 such that @xmath625 . if @xmath626 , ( [ 5.1 ] ) has two solutions @xmath627 . we choose the larger of them . if @xmath628 , then @xmath629 . note that the function @xmath630 is analytic with @xmath631 we shall now show that the function @xmath632 defined via @xmath633 takes every value in @xmath477 precisely once . the inverse @xmath453 thus defined is in the class @xmath454 . denote by @xmath634 , a point of an intersection of the curve @xmath152 with the circle @xmath635 with sufficiently large @xmath636 . for every fixed @xmath472 we consider a closed rectifiable curve @xmath440 consisting of some smooth curve @xmath637 , which is a part of the curve @xmath152 , connecting @xmath638 to @xmath639 , the arc @xmath640 of the semicircle @xmath641 , connecting @xmath639 to @xmath638 . the curve @xmath440 depend on @xmath455 . we see from the construction of the curve @xmath637 that if @xmath125 runs through @xmath637 the image @xmath463 lies on the interval @xmath642 $ ] , where @xmath643 and @xmath644 . here @xmath645 as @xmath468 . we note as well that if @xmath125 runs through @xmath640 the image @xmath463 lies in the domain @xmath646 . hence @xmath471 winds around @xmath647 once when @xmath125 runs through @xmath440 , and it follows from the argument principle that inside the curve @xmath440 there is a unique point @xmath357 such that @xmath473 . since this relation holds for all sufficiently large @xmath475 , we deduce that the point @xmath357 is unique in @xmath620 . hence the inverse function @xmath648 exists and is analytic in @xmath477 . by condition ( [ 5.2 ] ) , @xmath478 as @xmath649 , non - tangentially to @xmath39 and therefore @xmath650 . this proves our assertion . [ 5.3lem ] the function @xmath651 with sufficiently small parameter @xmath115 is the voiculescu transform of some symmetric probability measure . denote @xmath652 , and consider the function @xmath653 we see from this formula that @xmath654 for @xmath655 . in addition , for every fixed @xmath656 , @xmath657 as @xmath658 . hence , for every fixed @xmath656 , there exist points @xmath659 such that @xmath660 . denote by @xmath622 maximum of them . note that @xmath661 as @xmath662 or @xmath663 . introduce the curve @xmath152 by the equation @xmath664 . denote by @xmath665 the domain @xmath666 . note that the function @xmath667 is analytic with @xmath668 we shall now show that the function @xmath669 defined via @xmath633 takes every value in @xmath444 precisely once . the inverse @xmath670 thus defined is in the class @xmath454 . we define a closed rectifiable curve @xmath440 in the same way as in the proof of lemma [ 5.2lem ] . repeating the argument of the proof of lemma [ 5.2lem ] , we conclude that if @xmath125 runs through @xmath440 the image @xmath463 winds around every fixed point @xmath472 once , and it follows from the argument principle that inside the curve @xmath440 there is a unique point @xmath357 such that @xmath473 . since this relation holds for all sufficiently large @xmath475 , we deduce that the point @xmath357 is unique in @xmath665 . hence the inverse function @xmath671 exists and is analytic in @xmath444 . by condition ( [ 5.3 ] ) , @xmath478 as @xmath208 non - tangentially to @xmath39 and therefore @xmath650 . this proves the lemma . using similar arguments as those in the proof of lemma @xmath672 one can prove a more general result . let @xmath673 be a positive integer . the function @xmath674 with sufficiently small parameter @xmath115 is the voiculescu transform of some symmetric probability measure . in this section we shall prove theorem [ 2.1th ] , [ 2.1th * ] , [ 2.2th ] and corollary [ 2.1co ] . we use in the proof of these theorems some ideas of the papers @xcite , @xcite and @xcite . _ sufficiency_. let @xmath7 be the non - degenerate distribution of free random variables @xmath0 which satisfy the relation @xmath138 . the voiculescu transform @xmath195 of the probability measure @xmath7 is defined in a domain @xmath187 with some @xmath184 and @xmath394 . since one can extend the function @xmath676 on @xmath677 assuming @xmath678 , we can extend the voiculescu transform @xmath195 on the domain @xmath679 assuming @xmath680 . now we note that the voiculescu transform @xmath195 satisfies the following equation @xmath681 without loss of generality we assume that @xmath682 . as shown in section 3 , @xmath189 for @xmath683 . denote @xmath684 note that the function @xmath685 , is infinitely differentiable and @xmath375 as @xmath686 . moreover @xmath687 for real @xmath445 and @xmath688 , therefore it follows from ( [ 4.1 ] ) that @xmath689 in the sequel we consider special solutions of these equations . we shall apply the auxiliary results of section 4 which describe solutions of the equation ( [ 4.2 ] ) in the case @xmath690 , where @xmath338 and @xmath339 are distribution functions of the measures @xmath691 and @xmath692 , respectively , and @xmath693 . first we shall prove the following lemma . by lemma [ 3.1lem ] , we only need to prove the inequality @xmath697 . let us assume to the contrary that @xmath698 . by the definition of @xmath377 we see that the function @xmath351 is analytic for @xmath699 . since @xmath700 , we conclude by lvy s and raikov s theorem ( see @xcite , ch . 2 , theorem 2.2.1 ) that @xmath701 it follows from this relation that there exists a sequence @xmath702 such that @xmath703 as @xmath704 and for which @xmath705 by proposition [ 3.4pr ] , @xmath706 where @xmath707 as @xmath208 non - tangentially to @xmath39 . denote by @xmath708 the probability measure such that @xmath709 for any borel set @xmath109 . it is easy to see that @xmath710 for @xmath711 . in addition the measure @xmath712 is symmetric . therefore it easily follows from ( [ 4.3 ] ) that the relation @xmath713 holds , where @xmath714 as @xmath421 . we conclude from ( [ 4.3a ] ) and ( [ 4.4 ] ) that @xmath715 for @xmath716 . this relation implies @xmath717 and therefore the measure @xmath718 . since @xmath719 for @xmath720 with some @xmath721 , and @xmath189 and @xmath722 for such @xmath125 , we easily conclude that @xmath723 , and @xmath724 , a contradiction . the lemma is proved . from the definition of the active exponent @xmath377 ( see ( [ 3.7 ] ) , where @xmath349 and @xmath348 are defined in ( [ 3.5 ] ) with @xmath690 , where @xmath338 and @xmath339 are distribution functions of the measures @xmath691 and @xmath692 , respectively , and @xmath693 ) we conclude that @xmath725 , is a root of the function @xmath130 . by the assumptions of the theorem and lemma [ 3.1lem ] it follows that @xmath726 . consider the function @xmath727 , where we have chosen @xmath728 . the coefficient @xmath729 and the parameter @xmath362 were chosen in section 4 . it is clear that @xmath730 is a solution of equation ( [ 4.2 ] ) . moreover @xmath731 therefore @xmath732 [ 3.1lem ] , there exists a positive active exponent of the solution @xmath730 , a contradiction . hence , we may consider the second case only . in this case it is easy to see that the function @xmath739 satisfies the assumptions of lemma [ 3.5l ] and we obtain that @xmath712 is a semicuclar measure . [ 4.2lem ] assume that the function @xmath740 satisfies the condition : @xmath741 for all @xmath742 . let the statistics @xmath5 and @xmath6 be identically distributed and let @xmath712 be a semicircular measure . then @xmath7 is a semicircular measure as well . by lemma [ 3.6lem ] , the measure @xmath7 has a compact support . hence , the voiculescu transform @xmath195 is an analytic function in the domain @xmath743 with some parameter @xmath475 and it admits in this domain the following laurent expansion @xmath744 here @xmath745 . since @xmath712 is a semicircular measure , we have , using ( [ 2.3 ] ) , @xmath746 where @xmath341 . from this formula we deduce that @xmath747 and @xmath748 for @xmath749 . since the function @xmath195 satisfies the equation ( [ 4.1 ] ) , we obtain the relation @xmath750 by the assumptions of the lemma @xmath751 for @xmath749 , we conclude that @xmath752 for @xmath749 . we shall first assume that the function @xmath130 has a root @xmath440 such that @xmath753 . let @xmath754 or @xmath755 . by lemma [ 5.1lem ] , there exist a symmetric probability measure @xmath7 whose the voiculescu transform has the form @xmath756 . we conclude for this function that @xmath757 let @xmath758 . by lemma [ 5.1lem ] , there exist symmetric probability measure @xmath7 whose the voiculescu transform has the form @xmath759 . we obtain for this function @xmath760 we shall now assume that @xmath761 and @xmath46 is not a simple root of the function @xmath130 . by lemma [ 5.3lem ] , there exist a symmetric probability measure @xmath7 whose the voiculescu transform has the form @xmath762 with sufficiently small parameter @xmath115 . it is easy to see that @xmath763 assume that @xmath764 and @xmath440 is not even . by lemma [ 5.2lem ] , there exist a symmetric probability measure @xmath7 whose the voiculescu transform @xmath195 has the form @xmath765 with sufficiently small parameter @xmath115 . we deduce as above that @xmath766 for @xmath483 . we shall now show that if there exists a positive integer @xmath157 such that @xmath767 , then the statement ( 1 ) of the theorem does not imply the statement ( 2 ) . using corollary [ 2.3acor ] ( see @xcite as well ) , consider a probability measure @xmath7 with the voiculescu transform @xmath768 where @xmath769 and is sufficiently small by modulus . we easily see that the function @xmath195 satisfies the equation ( [ 4.1 ] ) . moreover , the probability measure has a compact support . thus , we have established that if 2 is not unique simple positive zero of the function @xmath130 or there exist odd positive numbers @xmath770 such that @xmath771 the statement ( 1 ) does not imply the statement ( 2 ) of the theorem . * proof of theorem @xmath773*. we keep all previous notations . we assume that the statistics @xmath5 and @xmath6 are identically distributed . by the assumptions of the theorem , @xmath774 with @xmath775 $ ] , where @xmath152 is maximum of the real parts of zeros of the function @xmath130 . by proposition [ 3.5pr ] and ( [ 4.1 * ] ) , we have @xmath776 therefore @xmath777 . we now conclude from lemmas [ 3.1lem ] and [ 3.3lem ] that all active exponents of @xmath336 are positive integers and simultaneously simple exponents . since the number of active exponents of @xmath336 is finite we can use the formula ( [ 3.7a ] ) . using this identity we easily obtain the relation @xmath778 where @xmath779 , are real coefficients . we deduce from ( [ 4.6 ] ) that @xmath780 , and @xmath781 for @xmath782 . the function @xmath195 satisfies the equation ( [ 4.1 ] ) . therefore , using ( [ 4.6 ] ) , we get @xmath783 we conclude from ( [ 4.7 ] ) that @xmath784 for @xmath785 . thus , @xmath786 is a semicircular measure . * proof of corollary @xmath787*. since @xmath788 , and @xmath146 , we note that @xmath789 for all positive @xmath790 . in addition , it is easy to see that @xmath791 . by the inequality @xmath792 . by theorem [ 2.1th ] , if @xmath7 is a non - degenerate probability measure , then @xmath7 is a semicircular measure . if @xmath133 with some @xmath135 , then ( [ 4.1 ] ) holds if @xmath793 . if @xmath794 , then ( [ 4.1 ] ) holds for @xmath795 only .

we examine whether the spectral energy distribution of uv continuum emission of active galactic nuclei ( agns ) changes during flux variation . we used multi - epoch photometric data of qsos in the stripe 82 observed by the sloan digital sky survey ( sdss ) legacy survey and selected 10 bright qsos observed with high photometric accuracies , in the redshift range of where strong broad emission lines such as ly and do not contaminate sdss filters , to examine spectral variation of the uv continuum emission with broad - band photometries . all target qsos showed clear flux variations during the monitoring period 1998 , and the multi - epoch flux data in two different bands obtained on the same night showed a linear flux - to - flux relationship for all target qsos . assigning the flux in the longer wavelength to the x - axis in the flux - to - flux diagram , the x - intercept of the best - fit linear regression line was positive for most targets , which means that their colors in the observing bands become bluer as they become brighter . then , the host galaxy flux was estimated on the basis of the correlation between the stellar mass of the bulge of the host galaxy and the central black hole mass ; the latter was estimated on the basis of the luminosity scaling relations for or emission lines and their line width . we found that the longer wavelength flux of the host galaxy was systematically smaller than that of the fainter extension of the best - fit regression line at the same shorter wavelength flux for most targets . this result strongly indicates that the spectral shape of the continuum emission of qsos in the uv region ( in rest - frame wavelength ) usually becomes bluer as it becomes brighter . the multi - epoch flux data in the flux - to - flux diagram were found to be consistent with the wavelength - dependent amplitude of variation presented in , which showed a larger amplitude of variation in shorter wavelengths . we also found that the multi - epoch flux - to - flux plots could be fitted well with the standard accretion disk model changing the mass accretion rate with a constant black hole mass for most targets . this finding strongly supports the standard accretion disk model for uv continuum emission of qsos .

1103.3619

it is generally accepted that the vast amount of continuum emission of type 1 active galactic nuclei ( agns ) in uv and optical wavelengths originates in the accretion disk surrounding a supermassive black hole , and the uv - optical variability found at the beginning of agn studies has been considered a powerful tool for understanding the nature of the agn central engine . however , the mechanism of this variability is still under discussion . many models for variability have been proposed , including accretion disk instabilities , x - ray reprocessing ( krolik et al . 1991 ; kawaguchi in prep . ) , star collisions ( courvoisier , paltani , & walter 1996 ; torricelli - ciamponi et al . 2000 ) , and gravitational microlensing ( hawkins 1993 ) ; however , none of these successfully explained more than a few properties of uv - optical variability @xcite . the spectral variability of the uv - optical continuum emission during flux variation is a key property for understanding the central engines and their variability mechanisms in agns . for example , since the spectral energy distribution ( sed ) of a hot spot or a flare in the accretion disk caused by local enhancement of mass accretion or disk instabilities would be different from that of the entire disk , the spectral shape of continuum emission is expected to vary with the flux variation when caused by those mechanisms . on the other hand , a change of the global mass accretion rate of the disk and a certain reprocessing model @xcite would not change the temperature distribution of the accretion disk at larger radii ; thus , the absence of the spectral variation of optical continuum emission during flux variation suggests those variation mechanisms . sakata et al . ( 2010 ; hereafter paper i ) addressed the spectral variability of optical continuum emission of agns . paper i examined the long - term multi - band monitoring data of 11 nearby agns and precisely estimated the contaminated flux of the host galaxy and the narrow emission lines . then , it was found that the multi - epoch optical flux data in any two different bands obtained on the same night showed a very tight linear flux - to - flux relationship for all target agns and that the non - variable component of the host galaxy plus narrow lines was located on the fainter extension of the linear regression line of multi - epoch flux - to - flux plots . from these results , paper i concluded that the spectral shape of agn continuum emission in the optical region ( @xmath4 ) does not systematically change during flux variation and that the trend of spectral hardening in which the optical continuum emission becomes bluer as it becomes brighter , which has been reported by many studies @xcite , is caused by the contamination of the non - variable component of the host galaxy plus narrow emission lines , which is usually redder than agn continuum emission . in contrast to optical continuum emission , two opposite claims have not been resolved for the spectral variability of uv continuum emission . @xcite statistically examined the properties of flux variations of qsos from the two - epoch photometric observations of about 25,000 qsos obtained by the sloan digital sky survey ( sdss ) and found a larger amplitude of variation at shorter wavelengths in the uv region of @xmath5 , indicating spectral hardening during flux variation of qsos . @xcite obtained a composite of the differential spectrum of two - epoch spectroscopic observations for hundreds of sdss qsos and found that the composite differential spectrum was bluer than the composite spectrum of qsos , which also indicates spectral hardening of the uv continuum emission . on the other hand , applied principal component analysis ( pca ) to the multi - epoch uv spectra of 15 nearby agns obtained by the _ international ultraviolet explorer _ ( iue ) satellite and concluded that the uv flux variation of the agns consists of a variable component with a constant spectral shape and a non - variable component . based on the decomposition of the multi - epoch spectra to a variable and a non - variable components , they further concluded that the variable component has a power - law shape of fixed spectral index and that the non - variable component can be reproduced using the sum of a steep balmer continuum and a pseudo - continuum , which corresponds to the spectral feature of the small blue bump ( sbb ) . @xcite and @xcite observed ngc 4593 and fairall 9 , respectively , using the iue satellite . they found that the multi - epoch uv flux data in two different bands obtained on the same night showed a linear flux - to - flux relationship and that the contaminated flux of the sbb was located on a fainter extension of the linear regression line . these authors concluded that the uv continuum emission retains a constant spectral shape during the agns flux variation . in this paper , we examine the spectral variability of the uv continuum emission of agns , in the same way as in paper i , on the basis of the long - term multi - epoch photometric data of 10 mid - redshift luminous qsos obtained by the sdss . in section [ s_obsdata ] , we describe the selection of target qsos , their basic properties such as the central black hole mass and accretion rate , and present their light curves . in section [ s_result ] , we examine the uv color variability of the target qsos from the analysis of the flux - to - flux plots and the contaminated fluxes of the host galaxies . in section [ s_discuss ] , we compare our results with previous observational studies about the uv color variability of agns and also with an accretion disk model . in section [ s_summary ] , we summarize our results . we assume cosmological parameters of @xmath6 throughout this paper . stripe 82 is located in the south galactic gap and was scanned multiple times by the sdss legacy survey in order to enable a deep co - addition of the data and to find variable objects . it is defined as the region spanning 8 h in right ascension ( ra ) from @xmath7 = 20@xmath8 to 4@xmath8 and @xmath9 in declination ( dec . ) from @xmath10 = @xmath11 to @xmath12 , consisting of two scan regions referred to as the north and south strips . both the north and south strips have been repeatedly imaged in @xmath13 , @xmath14 , @xmath15 , @xmath16 , and @xmath17 bands about 80 times on average by more than 300 nights of observations from 1998 to 2007 , with about 70 percent imaging runs obtained after 2005 since when the sdss - ii supernova survey started . there are about 37,000 qso candidates in the stripe 82 region @xcite of which spectroscopic data were available for about 8,300 candidates . in order to examine rest - uv continuum color variation of qsos , we first selected the spectroscopically identified qsos in stripe 82 from the qso candidate list of @xcite . then , we selected the targets using the following criteria : ( 1 ) the target redshift is @xmath18 in which strong broad lines such as ly@xmath7 and do not contaminate sdss filters and in which rest wavelengths of the @xmath13- or @xmath14-band are just longer or shorter than that of the line ( around @xmath19 or @xmath20 ) . or @xmath14 bands at a certain redshift , its flux is estimated as about @xmath21% of uv continuum flux based on the typical equivalent width for sdss qsos @xcite . ] ( 2 ) the signal to noise ratio ( s / n ) of the photometric data in two bands used for the analysis is more than 30 . the s / n values are taken from @xcite and are based on the single - epoch of observation . finally , we selected 10 luminous qsos . although wide wavelength coverage is preferable for examining the spectral variability , we basically used the @xmath16-band data in place of the @xmath17-band data as the longest wavelength observation because the s / n value of the former is much higher than that of the latter for most qsos . the targets , their redshifts , and the filter selections are listed in table [ t_filterwave ] . we then calculated the black hole mass and the mean eddington ratio of accretion rate for our targets . the black hole mass was estimated using the scaling relations for the emission line , or the emission line if the emission line was unavailable , according to @xcite . the full width at half maximum ( fwhm ) of the emission line was estimated from the standard deviation of the emission - line width taken from the sdss database , which was derived by fitting the gaussian profile to the emission line , by multiplying a factor of @xmath22 . we selected the specific luminosity at @xmath23 for the scaling relation of the emission line and at @xmath24 for that of the emission line where the effective wavelengths in the rest frame of the photometric bands are neighbored , estimated by interpolating or extrapolating the fluxes of the two photometric bands , and they were averaged over the light curves of the targets . strong absorption features in the emission line were found in three targets ( j0136@xmath20046 , j0346@xmath20042 , and j2134@xmath250048 ) , and the error of the black hole mass estimate might be larger for them . the bolometric luminosity was calculated using the equation @xmath26 @xmath27 @xcite and the eddington luminosity was calculated as @xmath28 , where @xmath29 is the gravitational constant , @xmath30 is light speed , @xmath31 is the mass of a proton , @xmath32 is the thomson scattering cross - section , and @xmath33 is the black hole mass estimated as described above . then , the mean eddington ratio of accretion rate , @xmath34 was calculated . the specific luminosities , the standard deviation of the emission lines , the black hole masses , and the eddington ratios are listed in table [ t_objbasicparm ] . the black hole mass of the targets are about @xmath35 , and the mean eddington ratio is between @xmath36 . we also examined the radio flux of the targets . as listed in table [ t_objbasicparm ] , 8 of the 10 targets were not detected by the very large array ( vla ) first survey ( becker , white , & helfand 1995 ) , and the remaining two targets are not covered by the data of the first survey . from these estimations of the black hole mass and the eddington ratio , and the examination of the radio flux , the targets can be characterized as radio quiet qsos with very massive black holes . we used the psf magnitude corrected with galactic extinction obtained from the sdss database for the photometry of the target qsos because the photometric error of the psf magnitude was smaller than that of the aperture magnitude . then , we further examined those data in order to improve the photometric accuracy of the light curves of the targets . first , we evaluated the atmospheric condition for photometry using the qa value and selected the data after 2004 , when the qa value was available . it was provided by the sdss database and gives @xmath37 fluctuations by the millimag in the recalibrated g - band camcol 3 zero point for fields in the run . if this number is zero , then the night was photometric or no recalibration was done , and if this number is non - zero , a large number corresponds to more variable clouds and worse photometric calibration can be expected in general . then , we adopted only the data whose qa values were less than 50 , that is , @xmath38 mag error of the photometric calibration , if the qa value was available . in addition , we examined the atmospheric condition using the flux of reference stars located near the target , which should be constant . when the flux of the reference star for a target qso at some epoch was more or less than @xmath39 from its mean flux for all available data , we did not use the photometric data of the target at that epoch . even though the qa value was unavailable for the data before 2004 , the selection procedure based on the flux of the reference stars efficiently rejected the data obtained in bad atmospheric condition . finally , the magnitude of the target qso was measured relative to the nearby reference stars , then the magnitudes of the reference stars are added . one or two reference stars were selected , which locate in the same field of the target and are brighter than the target by more than 2 mag . this relative photometry would reduce the fluctuation of the flux caused by changing atmospheric conditions during the observation and would improve the photometric accuracy of the light curves of the targets . the light curves of the target qsos are presented in figures [ f_lc1 ] and [ f_lc2 ] , and the parameters of the light curves are listed in table [ t_filterwave ] . we found that all targets showed significant flux variations during 7 years of observation , that is , a time span of @xmath40 years in the rest frame of the target qsos . in order to examine the color variability of the uv continuum emission of qsos with flux variation , we applied the flux - to - flux plot analysis as applied to the color variability of optical continuum emission of seyfert galaxies in paper i , which was originally proposed by @xcite . we plotted the flux data in two different bands observed on the same night in the flux - to - flux diagram , where the flux in longer wavelength is assigned to the @xmath41-axis and the flux in shorter wavelength is assigned to the @xmath42-axis . the flux - to - flux plots for all target qsos are presented in figures [ f_ff1 ] and [ f_ff2 ] , and the band pair and its effective wavelength in the rest frame for each target are listed in table [ t_filterwave ] . the rest wavelength of the shorter-@xmath43 filter is either @xmath44 or @xmath45 , while that of the longer-@xmath43 is between @xmath46 . as shown in figures [ f_ff1 ] and [ f_ff2 ] , the rest - frame uv flux data in the two different bands are distributed linearly for all targets , and especially for those with large flux variation such a correlation seems very tight . in order to examine the linear relationship between the uv fluxes in the two different bands , both straight - line fitting and power - law fitting to the data in the flux - to - flux diagram were carried out using the fitting code made by @xcite following the most generalized multivariate least square process given by jefferys ( 1980 , 1981 ) . the fitting functions adopted for the straight - line fitting and the power - law fitting are @xmath47 and @xmath48 , respectively , because larger contamination of the host galaxy flux is expected in the longer - wavelength flux ( @xmath49 ) than in the shorter - wavelength flux ( @xmath50 ) . the best - fit parameter values and the reduced @xmath51 values of the straight - line fitting and the power - law fitting are listed in table [ t_fffit ] . as listed in table [ t_fffit ] , the residuals of the straight - line fitting of the flux - to - flux plots in uv wavelengths are so small that the reduced @xmath51 is near unity , except for j2045@xmath20051 . indeed , according to the @xmath51-test for the straight - line fitting , the linear relationship is not rejected for 6 of the 10 targets ( j0312@xmath20113 , j0346@xmath20042 , j2111@xmath250024 , j2119@xmath250032 , j2123@xmath20050 , and j2134@xmath250048 ) at a 1% level of significance . although the @xmath51-test rejects the linear relationship at a 1% level of significance for the remaining four targets , no clear curvature can be seen in their flux - to - flux plots . indeed , no significant improvement of @xmath51 values applying the power - law fitting instead of the straight - line fitting can be found for two of the four targets ( j0136@xmath20046 and j2045@xmath20051 ) . the larger @xmath51 values of the straight - line fitting of the four targets are probably caused by another variable component of fluxes that is not synchronized to the uv continuum emission , or by underestimation of the photometric errors . in fact , as will be described in section [ s_effectsbp ] , the reduced @xmath51 values of the straight - line fitting are significantly decreased for three of the four targets ( j0105@xmath20050 , j0105@xmath250040 , and j2045@xmath20051 ) , and the linear relationship is not rejected at a 1% level of significance for two targets ( j0105@xmath20050 and j0105@xmath250040 ) when the variable flux component of the sbb is considered . based on these considerations , we conclude that the uv flux of all of the target qsos in the two different bands shows a linear correlation during the flux variations . the x - intercept of the best - fit linear regression line , @xmath52 , is positive for 9 of the 10 targets , as listed in table [ t_fffit ] ( with the exception of j2111@xmath250024 , for which @xmath52 is consistent to be zero within @xmath53 error ) . hence , their rest - frame uv colors of the fluxes observed in the two different bands become bluer as they become brighter . however , as presented in the discussion in paper i , if the sum of the non - variable flux components , such as that of the host galaxy , is located on the fainter extension of the best - fit regression lines of the flux - to - flux plots , the rest - frame uv color is almost constant during the flux variation . in this section , we estimate the host galaxy fluxes of the target qsos and examine whether the uv colors of the targets show the trend of spectral hardening or remain almost constant during the flux variations . since the host galaxies of the target qsos are not spatially resolved in the sdss images , we first estimated the stellar mass of the host galaxy from the black hole mass estimated in section [ s_targetselec ] and the magorrian relation @xcite , a tight correlation between the central black hole mass and the stellar mass of the bulge of the host galaxy . next , the host galaxy fluxes in observing bands were converted from the bulge mass using a mass - to - light ratio and an sed of stellar population of the host galaxy . we assumed bulge luminosity as the whole luminosity of the host galaxy because the qsos at @xmath54 are mostly hosted by the bulge - dominated early - type galaxies ( e.g. , bahcall et al . 1997 ; hutchings et al . 2002 ; dunlop et al . 2003 ; zakamska et al . 2006 ) we used the magorrian relation presented in @xcite , @xmath55 to estimate the stellar mass of the host galaxy from the black hole mass . the scatter of the correlation was estimated as 0.21 dex by them . including the scatter of the black hole mass estimation based on the scaling relation for against uv luminosity 0.32 dex presented in @xcite we estimated the scatter of the stellar mass of the host galaxy as 0.38 dex , that is , a factor of 2.4 for @xmath56 error and 5.8 for @xmath53 error . although the redshift evolution of the magorrian relation is still uncertain , @xcite suggested that the @xmath57 mass ratio increases a factor of @xmath58 at @xmath59 and @xcite also showed that it increases a factor of @xmath60 at @xmath61 compared to that of the local universe . therefore , the stellar mass , thus the luminosity of the host galaxy of the targets estimated here , might be systematically overestimated by a factor of @xmath62 . contrary to the morphology , recent studies show that the optical colors of agn host galaxies at @xmath63 are slightly bluer than those of quiescent early - type galaxies , indicating a component of newly formed stellar population ( e.g. , jahnke , kuhlbrodt , & wisotzki 2004b ; snchez et al . 2004 ; nandra et al . 2007 ; silverman et al . in addition , at higher redshift , @xcite showed that the rest - frame uv colors of host galaxies of qsos at @xmath64 are bluer than that expected from an old stellar population with a formation epoch at @xmath65 , suggesting a young stellar population of a few percent of the total mass . schramm , wisotzki , & jahnke ( 2008 ) showed that the rest - frame @xmath66 colors of the host galaxies of three qsos at @xmath67 are close to zero , indicating a substantial contribution from young stars , and a stellar mass - to - light ratio below 1 . since kiuchi , ohta , & akiyama ( 2009 ) demonstrated that most optical seds of host galaxies of type-2 agns at @xmath68 follow composite seds of e to sbc galaxies of the local universe , and since @xcite showed that the obs - frame @xmath69 colors of host galaxies of type-2 qsos at @xmath70 are widely scattered between the k - corrected colors of composite seds of local e and i m galaxies , we adopt a local sbc galaxy as a typical model for estimating the mass - to - light ratio and the sed of host galaxies of the target qsos ; we use local e and i m galaxies for the comparison . we use mass - to - light ratios of 20 , 2 , and 0.5 for e , sbc , and i m , respectively @xcite , and calculate the fluxes of the host galaxies in the rest - frame wavelength coverage of the observing filters using an uv - optical spectrum template of nearby galaxies for e , sbc , and i m presented in @xcite . figures [ f_ff1 ] and [ f_ff2 ] present the location of the host galaxy flux of the target qsos in the flux - to - flux diagrams . the estimated host galaxy fluxes , as described , are listed in table [ t_host ] . figures [ f_ff1 ] and [ f_ff2 ] clearly show that the location of the host galaxy flux is systematically on the left side of the best - fit regression line . for a typical case of the sbc - type galaxy , the host galaxy fluxes of 9 of the 10 targets are different from the fainter extension of the best - fit regression line by more than @xmath53 error of the host galaxy fluxes , or a factor of 5.8 . the only target for which the host galaxy flux is located on the extension of the best - fit regression line within @xmath53 error , j2115@xmath250024 , is in fact @xmath71 consistent to be zero , and the host galaxy flux is estimated to be considerably small . instead , if we assume the im - type galaxy for the host galaxy to model recent star - forming activity , as reported for high - redshift qsos by schramm et al . ( 2008 ) , the host - galaxy fluxes of 5 of the 10 targets ( j0105@xmath20050 , j0105@xmath250040 , j2045@xmath20051 , j2111@xmath250024 , and j2134@xmath250048 ) are located on the extension of the best - fit regression line within @xmath53 error . however , if the estimated black hole mass using the magorrian relation at the local universe is systematically smaller by a factor of @xmath62 caused by the redshift evolution , the location of the host galaxy flux is on the left side of the best - fit regression line for four of those five targets ( except the only target , again j2115@xmath250024 ) . we conclude that most of the target qsos show spectral hardening in uv wavelengths , that is , the uv continuum emission becomes bluer as it becomes brighter , in spite of maintaining the linear correlations between the qsos two - band flux data . these results suggest that the uv continuum emission of qsos usually shows spectral hardening , which is consistent with the results reported by @xcite and @xcite from the colors of the differential fluxes or the composite differential spectrum of two - epoch observations for sdss qsos . the best - fit parameters and the reduced @xmath72 values of the straight - line fitting are listed in table [ t_fffit ] . as described in section [ s_ffplotfit ] , the reduced @xmath72 value is so large that the linear relationship is rejected at a 1% level of significance for 4 of the 10 targets ( j0105@xmath20050 , j0105@xmath250040 , j0136@xmath20046 , and j2045@xmath20051 ) . these larger @xmath72 values suggest contamination from another variable component of fluxes that is not synchronized to the uv continuum emission . the sbb consisting of emission lines and balmer continuum emission , and the @xmath73 emission line , which more or less contaminate our observing bands of the longer - wavelengths , are thought to be the most promising sources for such additional variable components . they are considered to originate in the broad emission - line region ( blr ) , and moreover , reverberation mapping observations have found the lag of the sbb flux variation behind that of the uv continuum emission for ngc 5548 @xcite and the lag of the emission - line flux variation for ngc 4151 @xcite . since the size of the blr of the target qsos is estimated to be more than 200 light days for h@xmath74 emission line by the luminosity scaling relation @xcite , the flux variation of the sbb emission and the emission line is not at all synchronized with that of the continuum emission and would provide a scatter around the linear correlation between the fluxes of the uv continuum emission in the two different bands . we estimated the scatter of the linear correlation in the flux - to - flux plots caused by the contaminations from the sbb and emissions and re - examined the linear relationship of the uv continuum emission in the two different bands . we first fit the target spectrum obtained from the sdss database by a power - law model with the spectral windows of @xmath75@xmath76 , @xmath77 , and @xmath78 , where the spectrum is free from any strong emission lines . the synthesized flux of the spectrum in the photometric band was estimated by convolving the spectrum with the filter transmission curve , and then , the flux of the spectrum was recalibrated by scaling the synthesized flux by the photometric data at the observing date of the spectroscopy that was estimated by interpolating the light curve . the contaminated flux of the sbb and emissions in the broad - band filter was estimated by subtracting the synthesized flux of the best - fit power - law spectrum from that of the original spectrum . the contaminated fluxes of the sbb and emissions in the longer wavelength filter are listed in table [ t_sbb ] . their uncertainties were derived from the error of the power - law fitting . next , the @xmath56 scatter of the flux variation of the sbb and emission - line component , @xmath79 , was estimated as @xmath80 , where @xmath81 , @xmath82 , and @xmath83 are the contaminated flux of the sbb and emissions , the average flux , and the @xmath56 scatter over the broad - band light curve , respectively . the @xmath79 is also listed in table [ t_sbb ] . finally , assuming that the flux variation of the sbb and emission line is totally uncorrelated with that of the continuum emission in time , we recalculated the reduced @xmath51 of the straight - line fitting for the flux - to - flux plots by replacing @xmath84 by @xmath85 , where @xmath84 is the observational error of the photometry . the new reduced @xmath51 values of the straight - line fitting in which the scatter caused by the contaminations of the sbb and emissions is considered , are listed in table [ t_sbb ] . we found that the reduced @xmath51 values of the straight - line fitting were significantly decreased for three targets ( j0105@xmath20050 , j0105@xmath250040 , and j2045@xmath20051 ) out of the four targets , for which the linear relationship was rejected by the @xmath72 analysis in section [ s_ffplotfit ] , and the linear relationship is now not rejected at a 1% level of significance for two targets ( j0105@xmath20050 and j0105@xmath250040 ) . these results strongly indicate that a part of the scatter of the flux - to - flux plot from the best - fit regression line is caused by the contaminations from the sbb and emission line , and the uv continuum emissions in two different bands of the target qsos show a tight linear correlation during the flux variations . @xcite examined the relationship between variability amplitude and the rest - frame wavelength of the uv - optical region from sdss two - epoch multi - band observations of about 25,000 qsos , and found a larger amplitude of variation at shorter wavelengths in the uv region of @xmath5 . in this section , the tracks of the flux - to - flux plot of the target qsos obtained from the photometric monitoring data , which show the hardening trend in the uv region , as presented in the previous sections , are compared with the hardening trend of the amplitude of variation presented in @xcite . assuming that the wavelength - dependent amplitude of variation always holds for the flux variation at all times for individual qsos , a power - law function of @xmath86 in a flux - to - flux plot is obtained , where @xmath87 is derived from the wavelength - dependent amplitude of variation @xmath88 , presented in equation 11 of @xcite . then , we fit the power - law function , @xmath89 , to the flux - to - flux plot data of the target qsos , fixing the parameter @xmath90 as @xmath91 , where @xmath92 and @xmath93 are the effective wavelengths of the observing filters in the rest frame . figures [ f_ff1 ] and [ f_ff2 ] present the fitting of the power - law function to the flux - to - flux plots , and the best - fit parameters and the reduced @xmath72 values are listed in table [ t_vanden ] . the reduced @xmath51 value of the power - law fit is comparable to that of the straight - line fit for most of the target qsos . the slope of the power - law fit seems to be slightly gentler than that of the flux - to - flux plots for j2123@xmath20050 , and in fact , the reduced @xmath72 of the power - law fit is slightly larger than that of the straight - line fit even if it is still as small as @xmath94 . when we refit the power - law function , @xmath89 , with both @xmath95 and @xmath90 freed , @xmath96 is derived from the new power - law fit with the reduced @xmath97 . the @xmath90 value is slightly different from @xmath98 but would be still consistent with @xcite within the fitting error and the diversity of qso properties , and the reduced @xmath72 becomes comparable to that of the straight - line fit . from these results , we conclude that our results of the hardening trend during the flux variation obtained from the flux - to - flux plots in the rest - frame uv wavelengths are consistent with the hardening trend of the variability amplitude in the uv region presented in @xcite . [ [ s_paltaniwalter ] ] observed the uv spectra of 15 nearby agns at different epochs using the iue satellite . applying pca to the multi - epoch spectra and the decomposition to a variable and a non - variable components , they explained the uv variation in almost all their targets as the sum of a variable component of a constant power - law spectral shape and a non - variable component of the sbb , which means that the spectral shape of the uv continuum emission is almost constant during the flux variation . this finding is not consistent with our results that the uv continuum emission becomes bluer as it becomes brighter for most of the target qsos . in fact , did not estimate the sbb fluxes of the target agns by fitting a spectral model of a power - law continuum plux the sbb to their uv spectrum in order to examine the consistency of that method with the conclusion from the decomposition to a variable and a non - variable components . moreover , as discussed in section [ s_effectsbp ] , the sbb is thought to originate in the blr and its flux was found to be variable . it is suspected that a sufficient amount of the sbb flux is constant , which explains the constant spectral shape of the uv continuum emission during flux variation . on the other hand , paltani & walter s finding that one eigenvalue dominated all the others in the pca of the uv flux variation is consistent with the linear relationship of the fluxes in the two different bands found in the flux - to - flux plots of our target qsos . the models of the accretion disk predict that the spectral shape of the uv - optical continuum emission depends on the mass accretion rate and the black hole mass ; thus , the spectral properties of flux variation would also depend on those agn properties . therefore , while it is beyond the scope of this paper , it is very interesting to re - examine the multi - epoch uv spectra of nearby agns of using the flux - to - flux plot analysis with accurate estimation of non - variable components such as the host galaxy and also to discuss the luminosity or the black hole mass dependency of the spectral variability of the uv continuum emission . we showed that the spectral shape of the optical continuum emission is almost constant during the flux variation for 11 seyfert galaxies in paper i , which suggests that the radial temperature profile of an optical emitting region of an accretion disk does not change during the flux variation . on the other hand , we here find that the uv continuum emission becomes bluer as it brightens for most of the target qsos , while the uv fluxes in the two different bands show a tight linear correlation . this suggests that the temperature structure of the uv emitting region of an accretion disk shows systematic change . when the mass accretion rate changes in a standard accretion disk model , the spectral shape of the continuum emission from the accretion disk changes in uv wavelengths while it remains almost constant in optical wavelengths for agns whose central black hole mass is larger than @xmath99 ( e.g. , kawaguchi , shimura , & mineshige 2001 ) . that is because the variation of the mass accretion rate changes the maximum temperature of the accretion disk , which affects the spectral shape of uv continuum emission , while it does not change the radial profile of outer accretion disk of @xmath100 where the optical continuum emission originates , which would result in an almost constant optical continuum spectral shape . @xcite presented that a standard accretion disk model in which the mass accretion rate changed from one epoch to the next could be successfully fitted to the composite differential spectrum of two epochs of observations for hundreds of sdss qsos , which is bluer than the composite spectrum of qsos indicating the spectral hardening of uv continuum emission @xcite . however , the composite spectrum represents the average characteristics of qsos ; moreover , the differential spectrum corresponds only to the slope of the linear relationship between fluxes in two different bands in the flux - to - flux plots . here , we fit a standard accretion disk model with a varying mass accretion rate to the flux - to - flux plots for individual target qsos , and examine whether the spectral variation of the uv continuum emission during flux variations at various epochs can be described by the standard accretion disk model with various mass accretion rates and a constant black hole mass . the inner radius of the standard accretion disk model is set as @xmath101 , where @xmath102 is the schwarzschild radius . the outer radius is set as @xmath103 , but it is not important for the uv spectrum . a face - on view is assumed to calculate the flux from the accretion disk . then , a trajectory of the accretion disk model in the flux - to - flux diagram can be traced by changing the mass accretion rate with a constant black hole mass . we fit the trajectory of the accretion disk model to the flux - to - flux plots of the individual target qsos with a free parameter of black hole mass . since the observed flux includes the contaminating fluxes of the host galaxy , the sbb and emission line in addition to the continuum emission , the sbc - type host galaxy flux estimated in section [ s_effectotagncont ] , and the sbb and contaminating flux estimated in section [ s_effectsbp ] are added to the trajectory of the accretion disk model before fitting . figures [ f_ffsim1 ] and [ f_ffsim2 ] present the best - fit accretion disk model for the flux - to - flux plots of the target qsos , and the best - fit black hole mass and the reduced @xmath72 are listed in table [ t_simstat ] . the error of the black hole mass in the table is estimated from the statistical uncertainty of @xmath51-fitting of the model . as shown in figures [ f_ffsim1 ] and [ f_ffsim2 ] , the flux - to - flux plots can be fitted well by the standard accretion disk model , changing the mass accretion rate with a constant black hole mass . indeed , the reduced @xmath51 values of the standard accretion disk model are comparable to those of the straight - line fitting for 9 of the 10 target qsos , and even for the exception , j2123@xmath20050 , the reduced @xmath72 is as small as @xmath94 . in addition , as shown in figure [ f_bhmass ] , the best - fit black hole mass of the standard accretion disk model is consistent with the black hole mass estimated from the emission - line width and the luminosity scaling relation in section [ s_targetselec ] , within 0.76 dex for @xmath53 error of the latter black hole mass estimation . we also calculate the bolometric luminosity of the best - fit standard accretion disk model whose mass accretion rate corresponds to the average flux of the light curve . the ratio of the bolometric luminosity to the specific luminosity in uv , the bolometric correction @xmath104 , is listed in table [ t_simstat ] . we find that the bolometric correction is consistent with those obtained for hundreds of type 1 qsos from @xcite , which are distributed between 4 to 9 and whose average value is 5.62 . the black hole mass and the luminosity of the best - fit standard accretion disk model fitted to the flux - to - flux plots in uv wavebands are reasonable , and it could be a new technique for estimating a black hole mass of luminous qsos although the systematic uncertainties should be examined . as presented in paper i , the optical color of the best - fit regression line in the flux - to - flux diagram can be regarded as that of the agn optical continuum , because the optical spectral shape remains almost constant during the flux variation and the colors derived from the flux - to - flux plot analysis of nearby seyfert galaxies were consistent with those of the standard accretion disk model , @xmath105 . although it has long been known that the standard accretion disk model apparently shows a contradiction with observations that the composite spectrum of qsos is redder than that of the standard accretion disk model @xcite , the detailed analyses of the spectral shape of the uv - optical continuum emission of agns using flux variation presented here and in paper i strongly support the standard accretion disk . a possible explanation for the composite spectra being redder is that host galaxy light contaminates the spectra preferentially at longer wavelengths . these results are consistent with the recent analyses of the spectral shape of the optical to near - infrared continuum emission , which comes from the accretion disk by flux variations @xcite and polarizations @xcite . however , it has a critical difficulty of considering the variation of global accretion rate of a standard accretion disk as a primary source of flux variation of qsos . the timescale of changes of the global mass accretion rate is believed to correspond to the viscous timescale ( e.g. , pringle 1981 ; frank et al . 2002 ) , which is about @xmath106 years for our target qsos and much longer than the timescale of the flux variation we observed . clearly , further study of the flux variation mechanism is desired , which can explain the observed properties of spectral variations of the uv - optical continuum emission of agns , suggesting a variation of the characteristic temperature of an accretion disk . we examined the spectral variability of uv continuum emission of agns based on the multi - epoch photometric data of 10 sdss qsos in the stripe 82 . the target redshift was @xmath107 in which strong broad lines such as ly@xmath7 and do not contaminate sdss filters . the target luminosity was about @xmath108 erg s@xmath109 , the black hole mass was estimated to be about @xmath35 based on the luminosity scaling relations for or emission lines and their line width , and the mean eddington ratio was between @xmath36 . all target qsos showed significant flux variation within 7 years of observation , or @xmath40 years in the rest - frame . we plotted the flux data in two different bands ( shorter-@xmath110 or @xmath111 , longer-@xmath112 @xmath113 ) observed on the same night in the flux - to - flux diagram and found a strong linear correlation between the fluxes in the two different bands during the flux variations for all target qsos . the linear relationship was not rejected for 6 of the 10 targets at a 1@xmath114 level of significance based on the @xmath51-test for straight - line fitting . even for the remaining four targets , no clear curvature could be seen in their flux - to - flux plots ; no significant improvement of @xmath51 values by the power - law fitting than those of the straight - line fitting could be found for two of the four targets , and the reduced @xmath51 values of the straight - line fitting were significantly decreased for three of the four targets when the contamination of the sbb was considered . we then examined the location of the host galaxy flux in the flux - to - flux diagram . the stellar mass of the host galaxy was estimated from the black hole mass on the basis of the magorrian relation @xcite ; then , the host galaxy flux was estimated from the stellar mass using the mass - to - light ratio and the sed of a local sbc galaxy @xcite . we found that the location of the host galaxy flux was systematically on the left side of the best - fit regression line for 9 of the 10 targets ; that is , the uv continuum emission becomes bluer as it brightens for most of the target qsos in spite of holding their linear correlations between the two - band flux data . this trend of spectral hardening in uv wavelengths is consistent with the findings of @xcite . indeed , we found that the flux - to - flux plots of the target qsos could be fitted well by a power - law function assuming the wavelength - dependent variability amplitude of qsos . we presented in paper i that the spectral shape of the optical continuum emission of agns remains almost constant and that its color is consistent with that of the standard accretion disk model , @xmath115 . then , in the present paper , we fit a standard accretion disk model with a changing mass accretion rate and a constant black hole mass to the uv flux - to - flux plots for the individual target qsos , and found that the spectral variation of the uv continuum emission during flux variations at various epochs could be described by the standard accretion disk model with varied mass accretion rates . the black hole mass and the bolometric luminosity from the best - fit standard accretion disk model were reasonable . although it has long been known that the composite spectrum of qsos is redder than that of the standard accretion disk model , the detailed analyses of the spectral shape of the uv - optical continuum emission of agns using flux variation strongly support the standard accretion disk model . however , the variation timescale of the global mass accretion rate was considered to be too large to explain the flux variations within a few years , and further study of the flux variation mechanism is desired . we thank toshihiro kawaguchi for useful discussion and comments . this research has been partly supported by the grant - in - aids of scientific research ( 10041110 , 10304014 , 12640233 , 14047206 , 14253001 , and 14540223 ) and coe research ( 07ce2002 ) of the ministry of education , science , culture and sports of japan . tomoki morokuma has been supported by the jsps ( japan society for the promotion of science ) research fellowship for young scientists . assef , r. j. , et al . 2008 , , 676 , 286 bahcall , j. n. , kirhakos , s. , saxe , d. h. , & schneider , d. p. 1997 , , 479 , 642 barger , a. j. , cowie , l. l. , capak , p. , alexander , d. m. , bauer , f. e. , fernandez , e. , brandt , w. n. , garmire , g. p. , & hornschemeier , a. e. 2003 , , 126 , 632 becker , r. h. , white , r. l. , & helfand , d. j. 1995 , , 450 , 559 bell , e. f. , & de jong , r. s. 2001 , , 550 , 212 choloniewski , j. 1981 , acta astronomica , 31 , 293 collier , s. , horne , k. , wanders , i. , & peterson , b. m. 1999 , , 302 , l24 courvoisier , t. j .- l . , paltani , s. , & walter , r. 1996 , , 308 , l17 decarli , r. , falomo , r. , treves , a. , labita , m. , kotilainen , j. k. , & scarpa , r. 2010 , , 402 , 2453 dunlop , j. s. , mclure , r. j. , kukula , m. j. , baum , s. a. , odea , c. p. , & hughes , d. h. 2003 , , 340 , 1095 faber , s. m. , & jackson , r. e. 1976 , , 204 , 668 francis , p. j. , hewett , p. c. , foltz , c. b. , chaffee , f. h. , weymann , r. j. , & morris , s. l. 1991 , , 373 , 465 frank , j. , king , a. , & raine , d. j. 2002 , accretion power in astrophysics , by juhan frank and andrew king and derek raine , pp . 398 . isbn 0521620538 . cambridge , uk : cambridge university press , february 2002 . , giveon , u. , maoz , d. , kaspi , s. , netzer , h. , & smith , p. s. 1999 , , 306 , 637 hawkins , m. r. s. 1993 , , 366 , 242 hutchings , j. b. , frenette , d. , hanisch , r. , mo , j. , dumont , p. j. , redding , d. c. , & neff , s. g. 2002 , , 123 , 2936 jahnke et al.2004 , , 614 , 568 jahnke , k. , kuhlbrodt , b. , & wisotzki , l. 2004 , , 352 , 399 jefferys , w. h. 1980 , , 85 , 177 jefferys , w. h. 1981 , , 86 , 149 kaspi , s. , maoz , d. , netzer , h. , peterson , b. m. , vestergaard , m. , & jannuzi , b. t. 2005 , , 629 , 61 kawaguchi , t. , mineshige , s. , umemura , m. , & turner , e. l. 1998 , , 504 , 671 kawaguchi , t. , shimura , t. , & mineshige , s. 2001 , , 546 , 966 kishimoto , m. , antonucci , r. , blaes , o. , lawrence , a. , boisson , c. , albrecht , m. , & leipski , c. 2008 , nature , 454 , 492 kiuchi , g. , ohta , k. , & akiyama , m. 2009 , , 696 , 1051 krolik , j. h. , horne , k. , kallman , t. r. , malkan , m. a. , edelson , r. a. , & kriss , g. a. 1991 , , 371 , 541 magorrian , j. , et al . 1998 , , 115 , 2285 maoz , d. , et al . 1993 , , 404 , 576 marconi , a. , & hunt , l. k. 2003 , , 589 , l21 metzroth , k. g. , onken , c. a. , & peterson , b. m. 2006 , , 647 , 901 nandra , k. , georgakakis , a. , willmer , c. n. a. , cooper , m. c. , croton , d. j. , davis , m. , faber , s. m. , koo , d. c. , laird , e. s. , & newman , j. a. 2007 , 660 , l11 paltani , s. , & walter , r. 1996 , , 312 , 55 peng , c. y. , impey , c. d. , rix , h .- w . , kochanek , c. s. , keeton , c. r. falco , e. e. , leher , j. , & mcleod , b. a. 2006 , , 649 , 616 pereyra , n. a. , vanden berk , d. e. , turnshek , d. a. , hillier , d. j. , wilhite , b. c. , kron , r. g. , schneider , d. p. , & brinkmann , j. 2006 , , 642 , 87 pringle , j. e. 1981 , , 19 , 137 rees , m. j. 1984 , , 22 , 471 richards , g. t. , et al . 2006 , , 166 , 470 richards , g. t. , et al . 2009 , , 180 , 67 rodriguez - pascual , p. m. , et al . 1997 , , 110 , 9 sakata , y. , minezaki , t. , yoshii , y. , kobayashi , y. , koshida , s. , aoki , t. , enya , k. , tomita , h. , suganuma , m. , katsuno u. y. , & sugawara , s.2010 , , 711 , 461 ( paper i ) snchez et al . 2004 , , 614 , 586 santos - lleo , m. , clavel , j. , barr , p. , glass , i. s. , pelat , d. , peterson , b. m. , & reichert , g. 1995 , , 274 , 1 schramm , m. , wisotzki , l. , & jahnke , k. 2008 , , 478 , 311 shakura , n. i. , & syunyaev , r. a. 1973 , , 24 , 337 silverman , j. d. et al . 2008 , 675 , 1025 tomita , h.2005 , phd thesis , department of astronomy , graduate school of science , univ . of tokyo tomita , h. , yoshii , y. , kobayashi , y. , minezaki , t. , enya , k. , suganuma , m. , aoki , t. , koshida , s. , & yamauchi , m. 2006 , , 652 , l13 torricelli - ciamponi , g. , foellmi , c. , courvoisier , t. j .- l . , & paltani , s. 2000 , , 358 , 57 trump , j. r. , et al . 2009 , , 700 , 49 vanden berk , d. e. , et al . 2001 , , 122 , 549 vanden berk , d. e. , et al . 2004 , , 601 , 692 vestergaard , m. , & peterson , b. m. 2006 , , 641 , 689 wamsteker , w. , et al . 1990 , , 354 , 446 webb , w. , & malkan , m. 2000 , , 540 , 652 wilhite , b. c. , vanden berk , d. e. , kron , r. g. , schneider , d. p. , pereyra , n. , brunner , r. j. , richards , g. t. , & brinkmann , j. v. 2005 , , 633 , 638 wilhite , b. c. , brunner , r. j. , grier , c. j. , schneider , d. p. , & vanden berk , d. e. 2008 , , 383 , 1232 wold , m. , brotherton , m. s. , & shang , z. 2007 , , 375 , 989 xu , y. , bian , w .- h . , yuan , q .- r . , & huang , k .- 2008 , , 389 , 1703 zakamska , n. l. , strauss , m. a. , krolik , j. h. , ridgway , s. e. schmidt , g. d. , smith , p. s. , heckman , t. m. ; schneider , d. p. , hao , l. , & brinkmann , j. , 132 , 1496 sdss [email protected] & 1.585 & u & i & 1374 & 2894 & 68 & 0.146 & 0.017 & 0.251 & 0.017 + sdss [email protected] & 1.074 & u & i & 1712 & 3607 & 57 & 0.238 & 0.056 & 0.361 & 0.059 + sdss [email protected] & 1.716 & g & z & 1725 & 3288 & 39 & 0.169 & 0.015 & 0.263 & 0.015 + sdss [email protected] & 2.311 & g & i & 1415 & 2215 & 67 & 0.140 & 0.019 & 0.161 & 0.016 + sdss [email protected] & 1.535 & u & i & 1400 & 2951 & 33 & 0.156 & 0.020 & 0.436 & 0.030 + sdss [email protected] & 1.590 & u & i & 1371 & 2888 & 57 & 0.133 & 0.026 & 0.246 & 0.029 + sdss [email protected] & 2.325 & g & i & 1409 & 2250 & 52 & 0.216 & 0.012 & 0.245 & 0.012 + sdss [email protected] & 2.327 & g & i & 1408 & 2249 & 52 & 0.106 & 0.008 & 0.136 & 0.007 + sdss [email protected] & 2.261 & g & i & 1437 & 2294 & 62 & 0.808 & 0.058 & 1.064 & 0.043 + sdss [email protected] & 2.336 & g & i & 1405 & 2243 & 64 & 0.087 & 0.011 & 0.100 & 0.010 + j0105@xmath20050 & @xmath116 & 46.30 & 46.16 & 2027 & 1720 & 9.29 & 9.21 & 0.33 + j0105@xmath250040 & @xmath116 & & 45.94 & & 1848 & & 9.19 & 0.27 + j0136@xmath20046 & @xmath116 & 46.31 & 46.26 & * 4465 & 1340 & * 9.93 & 8.99 & 0.74 + j0312@xmath20113 & @xmath116 & 46.47 & 46.40 & 1498 & & 9.10 & & 0.94 + j0346@xmath20042 & not covered & 46.32 & 46.35 & * 1358 & 1746 & * 8.92 & 9.31 & 0.49 + j2045@xmath20051 & not covered & 46.26 & 46.16 & 2066 & 1622 & 9.20 & 9.09 & 0.34 + j2111@xmath250024 & @xmath116 & 46.70 & 46.56 & 2392 & & 9.64 & & 0.37 + j2119@xmath250032 & @xmath116 & 46.37 & 46.34 & 1323 & & 8.94 & & 1.13 + j2123@xmath20050 & @xmath116 & 47.24 & 47.20 & 2541 & & 9.99 & & 0.78 + j2134@xmath250048 & @xmath116 & 46.24 & 46.17 & * 3731 & & * 9.74 & & 0.12 + j0105@xmath20050 & 1.057 @xmath117 0.060 & 0.113 @xmath117 0.008 & 1.70 & & 0.310 @xmath117 0.023 & 0.228 @xmath117 0.004 & 0.194 @xmath117 0.029 & 1.23 + j0105@xmath250040 & 0.910 @xmath117 0.025 & 0.090 @xmath117 0.007 & 1.86 & & 0.589 @xmath117 0.037 & 0.267 @xmath117 0.012 & 0.350 @xmath117 0.049 & 1.47 + j0136@xmath20046 & 1.082 @xmath117 0.132 & 0.107 @xmath117 0.019 & 1.81 & & 1.683 @xmath117 3.412 & @xmath20.145 @xmath117 2.779 & 2.569 @xmath117 17.297 & 1.84 + j0312@xmath20113 & 1.155 @xmath117 0.047 & 0.040 @xmath117 0.005 & 0.74 & & 0.466 @xmath117 22.985 & @xmath20.680 @xmath117 7.208 & 6.960 @xmath117 59.681 & 0.72 + j0346@xmath20042 & 0.647 @xmath117 0.039 & 0.193 @xmath117 0.015 & 1.26 & & 0.657 @xmath117 0.660 & @xmath20.016 @xmath117 1.050 & 1.810 @xmath117 4.048 & 1.29 + j2045@xmath20051 & 0.865 @xmath117 0.022 & 0.090 @xmath117 0.004 & 2.71 & & 0.903 @xmath117 0.307 & 0.083 @xmath117 0.055 & 1.048 @xmath117 0.385 & 2.76 + j2111@xmath250024 & 0.992 @xmath117 0.074 & 0.028 @xmath117 0.016 & 0.57 & & 0.427 @xmath117 0.121 & 0.199 @xmath117 0.030 & 0.221 @xmath117 0.141 & 0.53 + j2119@xmath250032 & 1.112 @xmath117 0.069 & 0.041 @xmath117 0.006 & 0.83 & & 4.180 @xmath117 16.912 & @xmath20.127 @xmath117 1.163 & 2.754 @xmath117 12.147 & 0.80 + j2123@xmath20050 & 1.453 @xmath117 0.116 & 0.507 @xmath117 0.045 & 0.69 & & 1.565 @xmath117 0.167 & 0.731 @xmath117 0.559 & 0.598 @xmath117 1.005 & 0.70 + j2134@xmath250048 & 1.131 @xmath117 0.039 & 0.023 @xmath117 0.003 & 0.97 & & 0.905 @xmath117 0.435 & 0.032 @xmath117 0.021 & 0.868 @xmath117 0.279 & 0.99 + j0105@xmath20050 & 6.7e@xmath207 & 4.1e@xmath205 & 6.4e@xmath204 & 4.4e@xmath203 & 3.5e@xmath203 & 2.1e@xmath202 + j0105@xmath250040 & 1.8e@xmath206 & 7.0e@xmath204 & 1.4e@xmath203 & 1.3e@xmath202 & 9.0e@xmath203 & 7.8e@xmath202 + j0136@xmath20046 & 4.6e@xmath207 & 9.3e@xmath205 & 3.5e@xmath204 & 2.4e@xmath203 & 2.3e@xmath203 & 1.2e@xmath202 + j0312@xmath20113 & 2.3e@xmath207 & 1.0e@xmath206 & 2.2e@xmath204 & 6.0e@xmath204 & 1.2e@xmath203 & 3.0e@xmath203 + j0346@xmath20042 & 7.5e@xmath207 & 6.5e@xmath205 & 7.2e@xmath204 & 5.0e@xmath203 & 3.9e@xmath203 & 2.4e@xmath202 + j2045@xmath20051 & 5.4e@xmath207 & 3.3e@xmath205 & 5.1e@xmath204 & 3.4e@xmath203 & 2.8e@xmath203 & 1.6e@xmath202 + j2111@xmath250024 & 7.9e@xmath207 & 3.7e@xmath206 & 7.6e@xmath204 & 2.2e@xmath203 & 4.2e@xmath203 & 1.1e@xmath202 + j2119@xmath250032 & 1.5e@xmath207 & 6.8e@xmath207 & 1.4e@xmath204 & 4.0e@xmath204 & 7.6e@xmath204 & 2.0e@xmath203 + j2123@xmath20050 & 2.0e@xmath206 & 9.2e@xmath206 & 1.9e@xmath203 & 5.4e@xmath203 & 1.1e@xmath202 & 2.7e@xmath202 + j2134@xmath250048 & 9.9e@xmath207 & 4.7e@xmath206 & 9.5e@xmath204 & 2.7e@xmath203 & 5.2e@xmath203 & 1.4e@xmath202 + j0105@xmath20050 & 4.78e@xmath202 @xmath117 3.15e@xmath203 & 3.19e@xmath203 & 1.41 & 1.70 + j0105@xmath250040 & 6.46e@xmath202 @xmath117 3.82e@xmath203 & 1.04e@xmath202 & 0.98 & 1.86 + j0136@xmath20046 & 1.24e@xmath202 @xmath117 2.87e@xmath203 & 7.07e@xmath204 & 1.80 & 1.81 + j0312@xmath20113 & 3.15e@xmath203 @xmath117 1.17e@xmath203 & 3.10e@xmath204 & 0.73 & 0.74 + j0346@xmath20042 & 3.58e@xmath202 @xmath117 7.85e@xmath203 & 2.50e@xmath203 & 1.17 & 1.26 + j2045@xmath20051 & 3.88e@xmath202 @xmath117 3.73e@xmath203 & 4.50e@xmath203 & 1.69 & 2.71 + j2111@xmath250024 & 1.33e@xmath202 @xmath117 1.88e@xmath203 & 6.68e@xmath204 & 0.56 & 0.57 + j2119@xmath250032 & 6.50e@xmath205 @xmath117 1.45e@xmath203 & 3.56e@xmath206 & 0.83 & 0.83 + j2123@xmath20050 & 5.55e@xmath202 @xmath117 7.60e@xmath203 & 2.26e@xmath203 & 0.69 & 0.69 + j2134@xmath250048 & 5.05e@xmath203 @xmath117 7.64e@xmath204 & 4.88e@xmath204 & 0.96 & 0.97 + j0105@xmath20050 & 1.164 @xmath117 0.007 & 1.505 & 1.89 & 1.70 + j0105@xmath250040 & 1.040 @xmath117 0.006 & 1.419 & 2.11 & 1.86 + j0136@xmath20046 & 1.056 @xmath117 0.008 & 1.375 & 1.82 & 1.81 + j0312@xmath20113 & 1.539 @xmath117 0.008 & 1.315 & 0.72 & 0.74 + j0346@xmath20042 & 0.544 @xmath117 0.004 & 1.499 & 1.43 & 1.26 + j2045@xmath20051 & 1.107 @xmath117 0.005 & 1.506 & 2.82 & 2.71 + j2111@xmath250024 & 1.369 @xmath117 0.006 & 1.315 & 0.62 & 0.57 + j2119@xmath250032 & 1.460 @xmath117 0.007 & 1.315 & 0.83 & 0.83 + j2123@xmath20050 & 0.745 @xmath117 0.003 & 1.314 & 1.07 & 0.69 + j2134@xmath250048 & 1.805 @xmath117 0.008 & 1.316 & 0.98 & 0.97 + j0105@xmath20050 & 9.38@xmath118 & 1.86 & 1.70 & 6.04 + j0105@xmath250040 & 9.18@xmath119 & 2.03 & 1.86 & 7.00 + j0136@xmath20046 & 9.70@xmath118 & 1.92 & 1.81 & 5.50 + j0312@xmath20113 & 9.26@xmath118 & 0.72 & 0.74 & 7.30 + j0346@xmath20042 & 9.90@xmath118 & 1.25 & 1.26 & 4.80 + j2045@xmath20051 & 9.46@xmath118 & 2.71 & 2.71 & 5.73 + j2111@xmath250024 & 9.24@xmath120 & 0.55 & 0.57 & 8.63 + j2119@xmath250032 & 9.43@xmath118 & 0.93 & 0.83 & 5.95 + j2123@xmath20050 & 9.84@xmath118 & 1.11 & 0.69 & 5.84 + j2134@xmath250048 & 9.02@xmath120 & 1.06 & 0.97 & 8.88 +

in this paper , the optimal sampling strategies ( uniform or nonuniform ) and distortion tradeoffs for stationary gaussian bandlimited periodic signals with additive white gaussian noise are studied . unlike the previous works that commonly consider the average distortion as the performance criterion , we justify and use both the average and variance of distortion as the performance criteria . to compute the optimal distortion , one needs to find the optimal sampling locations , as well as the optimal pre - sampling filter . a complete characterization of optimal distortion for the rates lower than half the landau rate is provided . it is shown that nonuniform sampling outperforms uniform sampling . in addition , this nonuniform sampling is robust with respect to missing sampling values . next , for the rates higher than half the landau rate , we find bounds that are shown to be tight for some special cases . an extension of the results for random discrete periodic signals is discussed , with simulation results indicating that the intuitions from the continuous domain carry over to the discrete domain . sparse signals are also considered where it is shown that uniform sampling is optimal above the nyquist rate . finally , we consider a sampling / quantization scheme for compressing the signal . here , we show that the total distortion can be expressed as the sum of sampling and quantization distortions . this implies a lower bound on the distortion via shannon s rate distortion theorem .

1405.3980

shannon s rate distortion theory finds the optimal compression rate for a given distortion for an i.i.d . discrete signal . for continuous signals , shannon assumes that the signal is sampled at the nyquist rate . thus , in shannon s protocol , distortion occurs at the quantization phase ; no information about the signal is discarded in the sampling phase . but since only the end - to - end distortion is of importance , one can accept some distortion at the sampling phase , by sampling the signal below the nyquist rate . in other words , we can discard information about the signal in both the sampling and quantization phases . this has the benefit of saving in the sampling rate , which can improve power and computational efficiency of the system @xcite . furthermore , choosing the sampling locations ( _ nonuniform sampling _ ) , one can further reduce the sampling rate . nonuniform sampling works by measuring signal values at a set of arbitrary locations in time . this can occur when we are either unable to uniformly sample the signal due to some physical constraints , or when we lose some of the samples after uniform sampling . furthermore , nonuniform sampling is helpful in dealing with aliasing @xcite , even though characterizing the effective bandwidth " under nonuniform sampling is difficult . similar to uniform sampling , it has been shown that for reliable recovery , the average sampling rate must be at least twice the bandwidth of the signal @xcite . see @xcite for an overview of nonuniform sampling . consider the problem of finding the best locations for sampling a continuous signal to minimize the reconstruction distortion . to find the optimal sampling points , one should utilize any available prior information about the structure of the signal . furthermore , sampling noise should be taken into account . due to its importance , nonuniform sampling and its stability analysis has been the subject of numerous works . however , there has been relatively less work for a fully bayesian model of the signal and sampling noise , where one is interested in the statistical _ average _ distortion of signal reconstruction over a class of signals . in this paper , we study this problem for stationary periodic bandlimited signals . * previous works : * while there has been some previous work on the tradeoffs between sampling rate , quantization of samples and reconstruction distortion @xcite-@xcite ; there has been relatively fewer papers on finding the best locations of nonuniform sampling for minimizing reconstruction distortion @xcite-@xcite . the best locations of nonuniform sampling via an _ adaptive sampling _ method is considered in @xcite for a first order markov source . they propose a new framework based on dynamic programming and greedy methods to compute the sampling time increments using the @xmath0 most recent sampling times and values . a tradeoff among sampling rate , information rate and distortion of uniform sampling is derived in @xcite for gaussian sources corrupted by noise . one major difference between the assumptions in @xcite and ours is that they looks at uniform sampling while we consider both uniform and nonuniform sampling . the authors in @xcite look at choosing the sampling points for a stationary gaussian signal with auto - correlation function @xmath1 . it is shown that uniform sampling is optimal in this case . matthews @xcite considers a wide - sense stationary process , and computes the linear mmse estimate from sub - nyquist uniform samples . the authors in @xcite compute the optimal pre- and post - filters for a signal corrupted by noise , when uniform sampling is used . the problem is formulated in such a way that it is applicable to different types of modulation systems such as pam , pcm and dpcm . the authors in @xcite derive a rate distortion function for lossy reconstruction of a multi - component signal , when only a subset of the signal components are sampled . there has been some work in compressed sensing on the tradeoffs between the number of samples and the reconstruction error . here the samples are linear measurements , and the goal is to recover the sparsity pattern with a given distortion level . in @xcite , a sampling rate - distortion region is found ; it is shown that arbitrarily small but constant distortion is achievable with a constant measurement rate and per - sample signal - to - noise ratio . the authors in @xcite define a sampling - rate distortion function for an i.i.d . source , and derive lower bounds on the achievable performance . application of the sampling - rate distortion framework to compressed imaging with a multi - resolution statistical image model are also considered . computing optimal sampling points when limited information about the signal is available , has also been considered in the literature . universal _ set of sampling points for a class of signals is one that allows reliable reconstruction , uniformly for all signals in this class . in @xcite , a class of discrete signals with a limited number of non - zero frequency components are considered , and universal sampling locations are specified . finally , the problem of finding optimal sampling points is a task of general importance and has been studied in miscellaneous applications , e.g. @xcite-@xcite . * system model : * in this paper , we consider a continuous stationary periodic signal . the signal is assumed to be bandlimited , with at most @xmath2 non - zero fourier series coefficients from frequency @xmath3 to @xmath4 where @xmath5 is the fundamental frequency . since there are @xmath6 free variables , the signal can be uniquely reconstructed from @xmath7 noiseless samples ( landau rate ) . but as we consider taking @xmath8 _ noisy _ samples ( at time instances that we choose ) , unique reconstruction is not feasible , and distortion is inevitable . we also consider a pre - sampling filter in our model . see fig . [ fig1 ] for a description of our problem . the fourier series coefficients and the sampling noise are all random variables . therefore , both the input signal and its reconstruction from its samples are random signals , and so is the reconstruction distortion . one measure of fidelity is the expected value of distortion . another measure is the probability that the distortion exceeds a certain threshold ; this probability will be small if both the expectation and variance of the reconstruction distortion are small . in this paper , we consider minimizing both the expected value and variance of distortion , by choosing the best sampling points and the pre - sampling filter . we denote the minimum of the expected value and variance of distortion by @xmath9 and @xmath10 , respectively . a small @xmath9 guarantees a good _ average _ performance over all instances of the random signal , while a small @xmath10 guarantees that with high probability , we are close to the promised average distortion for a given random signal ( see also appendix [ appendixa ] for further justification ) . in essence , our goal is to find the tradeoffs between the number of samples ( @xmath8 ) , the pre - sampling filter , the variance of noise ( @xmath11 ) and the optimal expected value and the variance of distortion ( @xmath9 and @xmath10 ) . the literature in sampling theory looks only at minimizing the average distortion and ignores the variance of distortion . * overview of main contributions : * we provide tight results or bounds on the tradeoffs among various parameters such as distortion , sampling rate and sampling noise . when we are below half the landau rate , @xmath12 , we find the optimal average distortion ( @xmath9 ) and variance ( @xmath10 ) . interestingly , we show that the minimum of both @xmath9 and @xmath10 are obtained at the same sampling locations and the same pre - sampling filter . these two minima are obtained when we use no pre - sampling filter on the signal bandwidth , and use a nonuniform sampling procedure as given below : any arbitrary @xmath8 points from the set @xmath13 , which are uniform sampling points at the landau rate . observe that for @xmath14 , this is a non - uniform set of sampling points . it is worth to note that the sampling locations only depend on the bandwidth of the signal ; they are optimal for all values of the noise variance ( @xmath11 ) and @xmath15 . moreover , the sampling points are robust with respect to missing samples . note that the optimal sampling points are any arbitrary @xmath8 points from the set @xmath13 . thus , if we sample at these positions and we miss some samples , _ i.e. , _ getting @xmath16 samples instead of @xmath8 samples , the set of @xmath17 sample points is still a subset of @xmath13 , and hence optimal . therefore , if we knew in advance that we get @xmath17 samples , this knowledge would not have helped us . finally , complete characterization of distortion in terms of @xmath8 and @xmath18 allows us to answer the question of whether it is better to collect a few accurate samples from a signal , or collect many inaccurate samples from the same signal ( a similar question has been raised in @xcite ) . for rates above half the landau rate , we provide lower and upper bounds that are shown to be tight in some cases . when @xmath19 , _ i.e. , _ between half the landau rate and the landau rate , we find optimal average distortion when @xmath20 , using a non - uniform set of sampling locations . in addition , when @xmath21 , uniform sampling is shown to be optimal under certain constraints . whenever we find @xmath9 and @xmath10 explicitly , the minima are achieved simultaneously at the same optimal sampling points and the same pre - sampling filter . as discussed earlier , there are two stages in compression of a continuous signal : sampling phase and quantization phase . the overall reconstruction distortion depends on the choices that we make in both phases . we show that the optimal distortion can be written as the sum of two terms : one for sampling distortion and one for quantization distortion . therefore , the problem essentially decomposes into two independent subproblems . furthermore , we show that the asymptotic shannon s rate distortion theory yields a lower bound on the compression distortion term . * structure of the paper : * this paper is organized as follows : the preliminaries are given in section [ preliminaries ] . in section [ sec : problemdef ] , the problem is formally defined . in section [ sec : main - results ] , the main results of the paper are presented ; the proofs are relegated to appendix [ proofs ] . in section [ simulation ] , numerical simulation results are provided . in section [ future ] , the future work is stated and finally the proofs and technical details are given in the appendices . [ cols="^,^",options="header " , ] in this section , some theorems and lemmas that are used in the proofs of the main results are provided . @xcite[peierl ] ( peierl s inequality ) + let @xmath22 be an @xmath23 hermitian matrix , and @xmath24 be any convex function on @xmath25 . let @xmath26 be any orthonormal base of @xmath27 . then , @xmath28\geq\sum_{j=1}^{n}f(\langle u_j , au_j\rangle),\ ] ] where @xmath29 is the inner product between @xmath30 and @xmath31 . furthermore , for a strictly convex function @xmath24 , equality holds if and only if each @xmath30 is an eigenvector of @xmath22 . [ t2]@xcite every @xmath32 circulant matrix @xmath33 $ ] has eigenvectors @xmath34,\ ] ] for @xmath35 and corresponding eigenvalues @xmath36and can be expressed in the form @xmath37 , where @xmath38 has the eigenvectors as columns in order and @xmath39 is a diagonal matrix with diagonal elements @xmath40 . [ t3]@xcite let @xmath41 $ ] and @xmath42 $ ] be two @xmath32 circulant matrices with eigenvalues @xmath43 respectively . then matrices @xmath44 and @xmath45 commute and @xmath46 where @xmath47 and @xmath48 is also a circulant matrix . [ t4]@xcite let @xmath22 be an @xmath49 matrix with singular values @xmath50 let @xmath44 be a @xmath51 submatrix of @xmath22 ( intersection of any @xmath52 rows and any @xmath53 columns of @xmath22 ) with singular values @xmath54 then , + @xmath55 [ fig1 ] we consider a continuous stationary bandlimited periodic signal defined as follows : @xmath56,\label{eqdefs}\end{aligned}\ ] ] where @xmath5 is the fundamental frequency . the summation is from @xmath57 to @xmath58 , indicating that the signal is bandlimited . we assume that @xmath59 and @xmath60 for @xmath61 are mutually independent gaussian rv s , distributed according to @xmath62 for some @xmath63 . thus , the signal power is @xmath64 where @xmath65 . the discrete version of the signal is @xmath66=\sum_{\ell = n_1}^{n_2}[a_\ell\cos(\ell\omega_0 n)+b_\ell\sin(\ell\omega_0 n)]\label{eqdefs - discrete}\end{aligned}\ ] ] where @xmath5 for some integer @xmath67 . our model is depicted in figure [ fig1 ] . the signal @xmath68 , given in , is passed through a pre - sampling filter , @xmath69 , to produce @xmath70 . the signal @xmath70 is sampled at time instances @xmath71 , where @xmath72 . these samples are then corrupted by @xmath73 , an i.i.d . zero - mean gaussian noise denoted as @xmath74 . thus , our observations are @xmath75 . an estimator uses the noisy samples @xmath76 to reconstruct the original signal , denoted by @xmath77 . the incurred sampling distortion given by mmse criterion is equal to @xmath78 this distortion is a random variable . our goal is to compute the minimum of the expected value and minimum of the variance of this random variable , _ i.e. , _ to minimize @xmath79 we are free to choose @xmath69 and sampling locations @xmath80 s , @xmath81 . therefore , the optimal average and variance of distortion are defined as follows : @xmath82 where the minimization is taken over the sampling locations and the pre - sampling filter . we assume that @xmath69 is a real lti filter such that @xmath83 meaning that the frequency gain of the signal is at most one ; in other words , we assume that the filter is passive and hence can not increase the signal energy in each frequency . in particular , all pass filters @xmath84 satisfy this assumption . the reason for introducing this assumption is that we can always normalize the power gain of the filter by scaling the sampling noise power . more specifically , replacing some given @xmath69 with @xmath85 for some @xmath86 , changes @xmath87 to @xmath88 . hence , @xmath89 would change to @xmath90 . but reconstruction from @xmath91 s yields the same distortion as the reconstruction from @xmath92 , which is equivalent to dividing the noise power by @xmath86 . in this section , we first state two general lower bounds on the average and variance of distortion and then , using these lower bounds , we provide the main results of the paper . note that @xmath93 and @xmath94 is proportional to the bandwidth of the signal . remember that @xmath8 is the number of noisy samples . the proofs of the main results are given appendix [ proofs ] . we did not provide the formal proofs below the statement of the theorems because we first need to represent the problem in a matrix form , as given in appendix [ sec : matrix - form ] . [ lemma : lower:1 ] for any noise variance @xmath95 and any choice of the pre - sampling filter , the following lower bounds hold for all values of @xmath8 : @xmath96 when we do not use a pre - sampling filter on the signal bandwidth ( _ i.e. , _ @xmath97 for @xmath98 ) , equality in the above equations hold if and only if for any time instances @xmath80 and @xmath99 one of the following two equations hold : @xmath100 for a proof . in the next lemma , alternative lower bounds on @xmath101 and @xmath10 are given , which are tighter than the ones given in lemma [ lemma : lower:1 ] for @xmath102 . [ lemma : lower:2 ] for any @xmath95 and any choice of the pre - sampling filter , the following lower bounds hold for all values of @xmath8 : @xmath103 furthermore , equality in the above equations hold if and only if @xmath104 for @xmath105 and the sampling time instances satisfy the following equation : @xmath106 see appendix [ lower2proof ] for a proof . here , the sampling - rate distortion tradeoffs are studied when the sampling rate is lower than the landau rate . the optimal sampling strategy and thus the optimal distortion and variance are given in the following theorems . [ t1a ] for @xmath12 , the optimal average and variance of distortion for any given pre - sampling filter are @xmath107 both the minimal average distortion and its variance are obtained with the pre - sampling filter @xmath97 for @xmath98 and choosing @xmath8 distinct time instances arbitrarily from the following set of @xmath108 samples samples , unless @xmath109 . uniform sampling for @xmath8 points takes the samples at times @xmath110 @xmath111 the optimal interpolation formula for this set of sampling points is given by @xmath112 where @xmath91 is the noisy sample of the signal at @xmath113 . see appendix [ proof:4:2:1 ] for a proof . _ consider . observe that for fixed values of @xmath114 and @xmath115 , the minimum distortion is linear in @xmath8 . but it is not linear in @xmath108 ( as snr in the denominator is @xmath116 ) , except when @xmath117 goes to infinity . in the case of noiseless samples , @xmath118 , the snr will be infinity and the minimal distortion will be @xmath119 to intuitively understand this equation , observe that there are @xmath120 free variables and we can recover @xmath8 of them using the samples . therefore , we will have @xmath121 free variables , giving a total distortion of @xmath122 as the power of each sinusoidal function is @xmath123 . moreover , the maximum distortion is @xmath124 , which is obtained when @xmath125 ( @xmath126 ) or @xmath127 . finally , observe that when @xmath117 is large , @xmath128 which is linear in both @xmath114 and @xmath11 . observe that @xmath9 is increasing in both @xmath114 and @xmath11 , since @xmath129 for @xmath12 . furthermore , @xmath130 when @xmath12 , implying that the coefficient of @xmath114 is greater than or equal to that of @xmath11 . figure [ fig3 ] depicts @xmath9 as a function of @xmath114 , @xmath11 , @xmath8 and @xmath108 . _ [ fig3 ] [ thm : unique ] if we do not use a pre - sampling filter , the optimal sampling times @xmath131 given in the statement of theorem [ t1a ] are essentially unique if @xmath108 does not divide @xmath132 . more specifically , sampling times @xmath131 are optimal if there exists some constant @xmath133 such that for at least @xmath134 of the sampling points @xmath80 we have @xmath135 more generally , a set of sampling points @xmath131 are optimal if and only if @xmath80 s are the solutions of the following equation for @xmath136 and @xmath137 @xmath138 see appendix [ proof:4:2:3 ] for a proof . when @xmath12 , we have aliasing and can not reconstruct the signal even when the sampling noise is zero . assume that the number of samples , @xmath8 , is even and consider a low - pass filter of bandwidth @xmath139 as follows : @xmath140 for some @xmath141 $ ] . this filter makes all of the coefficients @xmath142 and @xmath60 equal to zero , except for @xmath8 variables @xmath143 when @xmath144 . therefore , the number of remaining non - zero coefficients will be equal to the number of samples , @xmath8 , and we will not have aliasing at the time of sampling . therefore , there will be no interference from other signal coefficients . however , we incur a distortion when we filter out some of the coefficients @xmath59 and @xmath60 . this scheme resembles an interference cancellation strategy . on the other hand , if we use the following filter with bandwidth @xmath8 , @xmath145 we have kept more coefficients but with aliasing and interference from other coefficients . we compare the performance of these two filters under uniform sampling in the following theorem : [ thm : unifmlessn ] suppose that we insist on using uniform sampling at rate @xmath8 , _ i.e. , _ taking the points @xmath146 then , the performance of filter @xmath147 is equal to @xmath148 on the other hand , the performance of filter @xmath149 satisfies : @xmath150 which is strictly greater than @xmath151 , even when we increase the signal power by a factor of two . see appendix [ proof:4:2:2 ] for a proof . _ filter @xmath149 is like an _ interference cancellation _ scheme when we interpret aliasing as interference . but with filter @xmath147 , we can do _ interference management _ and use the interfered aliased information to recover the signal . this yields an snr gain of @xmath152 , comparing and . to convey the essential intuition , consider the following different but simpler problem : suppose @xmath153 are two independent gaussian random variables . we are interested in recovering these two variables via one observation that is corrupted by a gaussian noise of variance @xmath11 . the interference cancellation strategy corresponds to discarding @xmath154 and observing @xmath155 where @xmath156 is the noise . this yields an estimation error of @xmath157 where the @xmath114 and @xmath158 are the estimation errors for @xmath154 and @xmath159 , respectively , and @xmath117 is defined as @xmath160 . on the other hand , an interference management scheme observes @xmath161 . here both @xmath159 and @xmath154 are interference terms for each other . the total estimation error in this case is equal to @xmath162 in comparison to , the snr gain of two is attained due to the interference management . _ here , the sampling - rate distortion tradeoffs for the rates between half the landau rate and the landau rate are stated . [ t : nm2n ] for @xmath164 and any choice of the pre - sampling filter , the optimal average distortion can be bounded as follows @xmath165 in which @xmath166 and @xmath167 is equal to @xmath168 where @xmath169 is the remainder of dividing @xmath170 by @xmath171 . the lower bound given in is tight when @xmath20 for any arbitrary @xmath172 . to achieve the optimal distortion , no pre - sampling filter is necessary ; a set of optimal time instances is @xmath173 moreover , the upper bound on the average distortion given in is obtained when we do not use a filter on the signal bandwidth , and choose the @xmath8 sample points as follows : @xmath174 see appendix [ proof:4:3 ] for a proof . [ th m : unique2 ] _ if we do not use a pre - sampling filter on the signal bandwidth , the necessary and sufficient conditions for optimality of the lower bound on average distortion is the one given in . the proof is similar to that given in proposition . _ in this subsection , we develop the sampling - rate distortion tradeoffs for the rates above the landau rate . we have the following theorems : [ t1c ] for @xmath176 and any choice of the pre - sampling filter , the following lower bounds on optimal average and variance of distortion hold : @xmath177 moreover , these lower bounds are tight if and only if @xmath178 for @xmath179 and time instances @xmath80 s satisfy the following equation : @xmath180 uniform sampling , _ @xmath181 is a solution to the above equation if for each @xmath182 in the interval @xmath183 , we have @xmath184 . in particular , uniform sampling is optimal when @xmath185 . also the reconstruction formula for the optimal set of sampling points is given by @xmath186 where @xmath91 is the noisy sample of the signal at @xmath113 . see appendix [ proof:4:4:1 ] for a proof . _ when @xmath8 goes to infinity , the minimum distortion @xmath187 goes to zero , regardless of the value of sampling noise . to see this , observe that when @xmath8 is very large ( @xmath185 ) , the lower bounds given in and are tight . _ [ remarks3 ] _ unlike the case of @xmath12 where the optimal sampling points could be found without any need to know @xmath15 , observe that the constraint given in depends on @xmath15 . furthermore , the uniform sampling strategy , @xmath188 , is not in general a solution to the above equation . for instance , uniform sampling is not an optimal strategy if @xmath189 for some @xmath190 , since @xmath191 _ below are some examples of non - uniform solutions of : _ when @xmath192 , @xmath193 and @xmath194 , sampling points @xmath195 are optimal . _ _ given any @xmath15 and @xmath58 , let @xmath196 . let us show @xmath197 by @xmath198 . then the following @xmath199 non - uniform sampling points are optimal for @xmath200 $ ] : let @xmath201 where @xmath202 is the binary expansion of time index @xmath203 $ ] . we verify that holds : @xmath204 which is zero if @xmath205 , _ i.e. , _ @xmath206 for some @xmath169 . _ so far we have focused on sampling distortion . we now consider a more general model depicted in fig . [ fig2 ] , consisting of a pre - sampling filter , a sampling strategy , a compressor and a decompressor . the compressor ( or the encoder ) takes in the @xmath8 samples of the signal and maps it to @xmath207 bits , via a function @xmath208 . the decoder @xmath209 uses the @xmath207 bits to recover the signal @xmath210 . this is done via a decoding function @xmath211 where @xmath212 is the space of all periodic signals with bandwidth @xmath213 $ ] . here the compression rate @xmath214 bits per second , and the sampling rate is @xmath8 samples per @xmath67 seconds , i.e. @xmath215 samples per second . the goal is to minimize the average distortion between the original signal @xmath216 and the decoder s reconstruction @xmath217 , where @xmath218 are the @xmath8 noisy samples of @xmath216 . in other words , we would like to minimize @xmath219 the minimization is taken over the sampling points , the pre - sampling filter , and the choice of the encoder @xmath220 and the decoder @xmath209 . to minimize the total distortion over these four parameters , we fix the first two parameters ( the sampling points and the pre - sampling filter ) and minimize distortion over the encoder and decoder functions . then , one should take the minimum over the first two parameters . the following theorem addresses the first minimization over the encoding and decoding functions , when the sampling instances and the pre - sampling filter are fixed . [ thm : compression]for a fixed pre - sampling filter and sampling instances @xmath80 s , the optimal encoder works as follows : it first computes its best estimate of the original signal , @xmath221 , from the noisy samples . in other words , it first finds a signal @xmath221 as a function of samples @xmath218 that minimizes @xmath222 then , it tries to convey @xmath221 to the decoder , by compressing it to @xmath207 bits . the optimal decoder works as follows : it takes the @xmath207 bits and computes @xmath223 such that its average distortion with @xmath221 is minimized , _ @xmath224 is minimized at the decoder . this choice of encoder and decoder is optimal as it will also minimize the end - to - end distortion @xmath219 moreover , the optimal total distortion is equal to the sum of sampling distortion and compression distortions , _ i.e. , _ @xmath225 see appendix [ proof:4:5:1 ] for a proof . _ the tradeoff between sampling distortion and quantization distortion is evident from the above theorem . for instance , if we do not sample at all , the optimal reconstruction @xmath221 will be the all zero signal . in this case , we will have a high sampling distortion but the compression distortion @xmath226 will be zero . on the other hand , if we use the optimal sampling strategy for the signal , by assuming a constant total distortion , if the sampling distortion decreases , we can endure more compression distortion . unfortunately , we lack an exact closed form expression for rate distortion of a given source when we have a single copy of the source ( not many i.i.d . copies ) . therefore , solving the joint optimization seems difficult . _ [ proposition : compression ] for a given pre - sampling filter and sampling instances @xmath80 s , the value of @xmath227 in the statement of theorem [ thm : compression ] is given by @xmath228 which is given in . furthermore , by employing a known result in rate - distortion theory , given @xmath207 bits for compression , we must have @xmath229 where @xmath230 s are the eigenvalues of @xmath231 , and @xmath232 is some constant such that @xmath233 . see appendix [ proof:4:5:2 ] for a proof . it would be interesting to extend our results to the case when the coefficients @xmath59 and @xmath60 in are mutually independent zero - mean gaussian rv s with arbitrary variances @xmath234 and @xmath235 respectively for @xmath236 . the signal @xmath216 will be stationary if and only if @xmath237 for some @xmath238 . to see this , observe that @xmath239 & = \mathbb{e}\{s(t)^2\ } -\mathbb{e}^2\{s(t)\ } \\&= \mathbb{e}\{\sum_{\ell = n_1}^{n_2}[a_\ell\cos(\ell\omega_0 t)+b_\ell\sin(\ell\omega_0 t)]\}^2 \\&=\sum_{\ell = n_1}^{n_2}\left(\alpha_\ell\cos^2(\ell\omega_0 t)+\beta_\ell\sin^2(\ell\omega_0 t)\right ) \\&=\sum_{\ell = n_1}^{n_2}\left(\frac{\alpha_\ell+\beta_\ell}{2}+\frac{\alpha_\ell-\beta_\ell}{2}\cos(2\ell\omega_0 t)\right)\end{aligned}\ ] ] which is independent of @xmath240 if and only if @xmath237 for some @xmath238 . observe that the signal is _ sparse _ if @xmath238 is non - negligible for few values of @xmath241 . for the case of equal variances @xmath242 for @xmath179 , we stated two general lower bounds on the average distortion ( see subsection [ sec:4:1 ] ) . when @xmath238 values are distinct , we can generalize the second lower bound given in lemma [ lemma : lower:2 ] as follows ( we assume that the frequency support is known ) : [ proposition : sparse ] for any @xmath95 and any choice of the pre - sampling filter , the following lower bound on the average distortion holds for all values of @xmath8 : @xmath243 furthermore , equality in the above equation holds if and only if @xmath178 for @xmath179 and the sampling time instances satisfy the following equation : @xmath106 uniform sampling , _ @xmath181 is a solution to the above equation if for each @xmath182 in the interval @xmath183 , @xmath184 . in particular , for the rates above the nyquist rate , _ i.e. , _ @xmath185 , uniform sampling is optimal . see appendix [ proof:4:5:2 ] for a proof . similar to the proof of theorem [ t1c ] , one can find optimal sampling points for a sparse signal under certain conditions : uniform sampling with no filter on the signal bandwidth is optimal when we are above the nyquist rate @xmath185 . consider a real periodic discrete signal of the form @xmath66=\sum_{\ell = n_1}^{n_2}[a_\ell\cos(\ell\omega_0 n)+b_\ell\sin(\ell\omega_0 n)],\label{eqdefs - discrete}\end{aligned}\ ] ] where @xmath244 for some integer @xmath67 , and @xmath245 are natural numbers . observe that @xmath67 is the period of the discrete signal and @xmath246 is the @xmath241-th dft coefficients of the discrete signal @xmath247 $ ] . suppose that we have @xmath8 noisy samples at time instances @xmath248 . we would like to use these samples to reconstruct the discrete signal @xmath247 $ ] . if the reconstruction is @xmath249=\sum_{\ell = n_1}^{n_2}[\hat{a}_\ell\cos(\ell\omega_0 n)+\hat{b}_\ell\sin(\ell\omega_0 n)]$ ] , the distortion @xmath250-\hat{s}[n]|^2 $ ] is proportional to @xmath251 . thus , the formulation for minimizing the distortion is the same as that of the continuous signals which is given in [ sec : matrix - form ] . the only exception is that we additionally have the restriction that @xmath80 s should be integers . below , is an observation about the discrete time formulation : whenever the optimal sampling points in the continuous formulation turns out to be integer values , they are the optimal points in the discrete case . and when the optimal sampling points are not integers , simulation results with exhaustive search show that their closest integer values are either optimal or nearly optimal . for example , when the signal period is @xmath252 and @xmath253 . in this case @xmath254 and the optimal sampling points in the continuous case are @xmath255 for some @xmath256 $ ] . the rounds of these point for @xmath257 are optimal in the discrete case , _ i.e. , _ any choice of time instances from the set @xmath258 is optimal . on the other hand , when the signal period is @xmath252 and @xmath259 . here again @xmath254 and @xmath260 are the only continuous optimal points . rounding these points to their closest integers yields @xmath261 . however , the optimal points are @xmath262 resulting in @xmath263 . note that the distance between the first two optimal points is 1 , which can not be achieved by choosing any arbitrary value of @xmath264 , since the distance between any of the points in @xmath265 is 2.5 . in this section , we evaluate the optimal reconstruction interpolation formulas derived in the previous sections and also simulate an iterative method for the reconstruction of optimal non - uniform positions . we will see that the simulation results match the stated results . [ reconstructio ] illustrates the mmse reconstruction from optimal sampling points given in for high snr scenario . the figure also depicts the reconstruction result using an iterative method @xcite . its performance is near optimal in the high snr scenario . [ reconstruction2 ] uses the same parameters for generating the signal , but the variance of the noise is higher ( low snr scenario ) . figs . [ dvarm2n24 ] and [ fig : dvarm2n1n12 ] represent the average distortion , @xmath266 , and the variance of distortion , @xmath267 , for different values of @xmath8 , @xmath268 and @xmath108 . these figures show that whenever the average distortion is minimum , the variance of distortion is also minimum . this confirms the results given in theorems [ t1a ] and [ t1c ] . even though we can not prove the same result for all values of @xmath269 , simulation results for @xmath270 indicate that the two terms are minimized at the same sampling locations . in fig . [ lower12 ] , the two lower bounds on average distortion derived in lemmas [ lemma : lower:1 ] and [ lemma : lower:2 ] are plotted . the first lower bound is tight for @xmath12 , as shown in theorem [ t1a ] . the second lower bound is tight for sufficiently large @xmath21 , as discussed in theorem [ t1c ] . the maximum of these lower bounds is depicted in fig . [ lowersupper ] . in this figure , the distortion of uniform sampling without any pre - sampling filter on the signal bandwidth is also depicted . we make the following observations in this figure : the performance of the uniform sampling is close to optimal for @xmath12 ( its curve almost matches that of the lower bound in lemma [ lemma : lower:1 ] which is optimal for @xmath12 ) . while we expect that increasing the number of uniform samples should lead to a better performance , we see that near the landau rate , increasing the sampling points does not decrease the uniform sampling distortion . here , the curve for uniform sampling reaches its maximum at @xmath271 , whereas the landau rate is 16 . the curve decreases from @xmath271 onwards and matches the second lower bound at @xmath272 . thus , uniform sampling is optimal for @xmath273 . finally , the explicit upper bound on the performance of the sampling points given in in theorem [ t : nm2n ] is also depicted in this figure . it matches that of uniform sampling at @xmath109 and @xmath7 , as expected . [ lowersupper2 ] is similar to fig . [ lowersupper ] , but for different values of @xmath15 and @xmath108 . it shows that the explicit upper bound of theorem [ t : nm2n ] can be below the distortion of uniform sampling . while the above figures have been plotted for @xmath274 , our simulation results show that decreasing the variance of noise will reduce the gap between the lower and the upper bounds . finally , fig . [ lowersupper3 ] compares uniform sampling ( without a pre - sampling filter ) with the optimal non - uniform sampling strategy for @xmath12 . the performances of these two strategies are close to each other . more specifically , the figure depicts the difference of the distortion of the two strategies , divided by the distortion of optimal sampling . it shows that we can get a gain of at most 0.3 percentage of the distortion obtained by using the optimal non - uniform sampling strategy . however , a main advantage of the optimal nonuniform sampling is its _ robustness _ with respect to missing samples ( as discussed in the introduction ) . 1 are depicted . their reconstruction with @xmath275 samples are also depicted , employing the optimal mmse method and the iterative method of @xcite . the sampling points are optimally chosen . the average reconstruction distortion ( mmse ) over _ all _ random signals with the given parameters is @xmath276 . subfigure @xmath277 : the distortion of the optimal mmse method is @xmath278 and the distortion of the iterative method is @xmath279 . subfigure @xmath280 : the distortion of the optimal mmse method is @xmath281 and the distortion of the iterative method is @xmath282 . , title="fig : " ] 1 are depicted . their reconstruction with @xmath275 samples are also depicted , employing the optimal mmse method and the iterative method of @xcite . the sampling points are optimally chosen . the average reconstruction distortion ( mmse ) over _ all _ random signals with the given parameters is @xmath276 . subfigure @xmath277 : the distortion of the optimal mmse method is @xmath278 and the distortion of the iterative method is @xmath279 . subfigure @xmath280 : the distortion of the optimal mmse method is @xmath281 and the distortion of the iterative method is @xmath282 . , title="fig : " ] 1 are depicted . their reconstruction with @xmath275 samples are also depicted , employing the optimal mmse method and the iterative method of @xcite . the sampling points are optimally chosen . the average reconstruction distortion ( mmse ) over _ all _ random signals with the given parameters is @xmath283 . subfigure @xmath277 : distortion of the optimal mmse method is @xmath284 and the distortion of the iterative method is @xmath285 . subfigure @xmath280 : distortions of the optimal mmse method is @xmath286 and the distortion of the iterative method is @xmath287.,title="fig : " ] 1 are depicted . their reconstruction with @xmath275 samples are also depicted , employing the optimal mmse method and the iterative method of @xcite . the sampling points are optimally chosen . the average reconstruction distortion ( mmse ) over _ all _ random signals with the given parameters is @xmath283 . subfigure @xmath277 : distortion of the optimal mmse method is @xmath284 and the distortion of the iterative method is @xmath285 . subfigure @xmath280 : distortions of the optimal mmse method is @xmath286 and the distortion of the iterative method is @xmath287.,title="fig : " ] and @xmath108 . the signal period is @xmath288 . here @xmath192 and @xmath289 . the only degree of freedom is therefore @xmath290 , which is shown on the x - axis . ] and the signal period is @xmath288 . here @xmath192 and @xmath289 . the only degree of freedom is therefore @xmath290 , which is shown on the x - axis . ] . the signal period is @xmath288 . here @xmath192 and @xmath289 . the only degree of freedom is therefore @xmath290 , which is shown on the x - axis . ] [ lower12 ] [ future ] we studied the effect of sampling strategy and a pre - sampling filter on the reconstruction distortion of a periodic bandlimited signal of the form given in . deriving the optimal points for the discrete case in general is also of importance . moreover , it would be interesting to find the optimal quantizer . other extensions include considering a more general class of signals . furthermore , we consider only the sampling noise in this work ; one can consider adding noise to the signal itself , before sampling . it would be interesting to extend our results to the case of non - uniform power spectrum of the input signal . in , we had assumed that @xmath59 and @xmath60 are normal random variables with variance @xmath114 . more generally , one can assume coefficients with different variances . in this case @xmath291 is still diagonal but not proportional to identity matrix . for the case of @xmath292 , one can use and similar arguments to show that @xmath293 is minimized if ( @xmath294 ) is diagonal . therefore , we get the same sufficient condition for optimality , as given in proposition [ thm : unique3 ] . the case of @xmath295 is more complicated and the optimal sampling points depend on the values of variances of the coefficients . the authors would like to thank amin gohari , salman beigi and arash amini for helpful discussions . in particular , the proof of theorem [ mmse - var ] is due to amin gohari . the conventional mmse estimation problem for estimating a vector @xmath296 from vector @xmath297 asks for minimizing the expected value of @xmath298 where the estimator @xmath299 is created as a function of observation @xmath297 . however , this would only ensure that the distortion is minimized on _ average_. in practice , we get one copy of @xmath296 and @xmath297 , and we want to ensure that the distortion that we obtain is small for that one copy , not just on average . minimizing the variance of @xmath300 makes sense for the following two reasons : * if random variables @xmath300 and @xmath301 are independent and have the same distribution , then @xmath302.\ ] ] in other words , if we take two independent instances of the problem ( @xmath303 ) and ( @xmath304 ) . then the average gap between the distortion of these two problems , _ @xmath300 and @xmath301 , has average @xmath305 ; * by chebyshev s inequality , the probability of @xmath300 exceeding a threshold depends not only on the average of @xmath300 , but also on its variance . in this appendix we show that for jointly gaussian random variables , the estimator that minimizes the expected value of @xmath300 , also minimizes its variance . [ mmse - var ] suppose @xmath296 and @xmath297 are two correlated jointly gaussian vectors , having covariance matrices @xmath291 , @xmath306 and @xmath307 . let @xmath299 be a function of @xmath297 , and consider the cost @xmath308 the estimator @xmath299 that minimizes the above cost constraint is @xmath309 $ ] , and the minimum variance is equal to @xmath310 where @xmath311 we have @xmath312.\end{aligned}\ ] ] we claim that both of the terms @xmath313 and @xmath314 $ ] are minimized when @xmath309 $ ] . the second term is always non - negative and becomes zero when @xmath309 $ ] . this is because @xmath315 will be equal to @xmath316 which is a constant and does not depend on the value of @xmath297 for jointly gaussian random variables . therefore , its variance is zero if @xmath309 $ ] . to prove that the second term is minimized at @xmath309 $ ] , it suffices to show that this claim for all values of @xmath317 . we will be done if we have the following statement : for any jointly gaussian vector @xmath296 , the function @xmath318 is minimized at @xmath319 $ ] . let @xmath320 . then observe that @xmath321 . if we replace @xmath322 by @xmath323 for some unitary matrix @xmath38 , the norm @xmath324 remains invariant . therefore , without loss of generality we can assume that @xmath325 is diagonal . then @xmath326 is the sum of independent gaussian random variables . let @xmath327 be the variance of @xmath328 . therefore @xmath329 which is clearly minimized when @xmath330 , and the minimum will be equal to @xmath331 . coming back to the original minimization problem , we see that the covariance matrix of @xmath296 given any value for @xmath332 , does not depend on the value of @xmath317 , and is equal to @xmath333 , where @xmath311 therefore , the overall minimum variance is equal to @xmath334 here , we first provide a matrix representation of the problem . let @xmath296 be the vector of coefficients @xmath59 and @xmath60 ( @xmath335 ) of the original signal , @xmath216 , given in , _ i.e. , _ @xmath336^\dagger.\label{eqn : def : coef : ab}\end{aligned}\ ] ] since the vector @xmath296 consists of real numbers , @xmath337 is just the transpose operation in the above equation . after filtering @xmath216 , we get @xmath338\label{eqdefs2}\end{aligned}\ ] ] in which @xmath339 for @xmath335 , where @xmath340 and @xmath341 are the real and imaginary parts of @xmath69 , respectively . if we denote the vector of coefficients of @xmath70 as @xmath342^\dagger\ ] ] and express in matrix form , we get @xmath343 where @xmath344 and @xmath345 @xmath346 the signal @xmath70 is then sampled at time instances @xmath80 for @xmath347 . we have @xmath348.\ ] ] let @xmath349 be the vector of the samples @xmath350^\dagger.\ ] ] therefore , in matrix form , we have @xmath351 where @xmath352 is an @xmath353 matrix of the form @xmath354finally , the vector of observations denoted by @xmath297 can be writen as @xmath355 where @xmath356 is the noise vector . estimation of @xmath216 is equivalent to the estimation of its fourier coefficients . if we denote the coefficients of the reconstruction signal by @xmath357^\dagger,\ ] ] using the parseval s theorem , we will have @xmath358 our goal is to use @xmath297 to estimate the signal with minimal distortion , _ i.e. , _ from , we would like to minimize @xmath359 since all the random variables are gaussian , the linear mmse is optimal and thus we would like to use @xmath360 to find @xmath299 such that @xmath361 is minimized . the mmse estimator and the mean square error are given by @xmath362\right\}={\mathtt{tr}}(c_e ) , \label{eqn : mmsece}\end{aligned}\ ] ] where the matrix @xmath363 and the error covariance @xmath333 are of the forms : @xmath364 in the above formulas , we have used the fact that @xmath365 and @xmath366 . also we have defined @xmath367 using the alternative form for the linear mmse estimator derived in @xcite , we have @xmath368 where @xmath369 the value of @xmath300 is equal to @xmath370 from and if we use , it follows : @xmath371 where results from the fact that the trace operator is invariant under cyclic permutations . if @xmath172 , by setting @xmath118 , we get a distortion @xmath372 regardless of @xmath373 ( as long as they are distinct ) . when @xmath21 , from , one can observe that distortion becomes arbitrary small as @xmath374 . equivalently , @xmath266 can be found using as @xmath375 the main challenge is to find the minimum of @xmath376 or @xmath377 over all choices of @xmath80 s and all pre - sampling filters . + note that the matrix @xmath378 equals to @xmath379 , where the entries of the matrix @xmath380 for @xmath381 are of the following form : @xmath382 in which @xmath383 and @xmath384 for @xmath385 ( see ) . also , a direct calculation shows that the diagonal entries of matrix @xmath386 are given by : @xmath387 ; \\\qquad \qquad \qquad \qquad\text{where } \ell = k+n_1 - 1 \text { for } k = 1 , \cdots , n , \\ \\\sigma^2+{\mathsf{p}}\sum_{i = 1}^{m}\big[h_i^2(\ell\omega_0)\cos^2(\ell\omega_0 t_i)\\\quad + 2h_r(\ell\omega_0)h_i(\ell\omega_0)\cos(\ell\omega_0 t_i)\sin(\ell\omega_0t_i)+h_r^2(\ell\omega_0)\sin^2(\ell\omega_0 t_i)\big ] ; \\\qquad\qquad \qquad \qquad\text{where } \ell = k+n_1 - 1-n\text { for } k= n+1 , \cdots , 2n.\end{cases } \label{eqn : gammaentries}\end{aligned}\ ] ] * computing variance of distortion : * the variance of distortion , using the parseval s theorem given in , is of the form @xmath388 using theorem [ mmse - var ] from appendix [ appendixa ] , we have @xmath389 hence , using similar steps as the ones used in computing average distortion , from , @xmath267 is found to be @xmath390 \nonumber \\ & = { \mathsf{p}}^2 [ 2n + { \mathtt{tr}}\left(- 2\pi { \mathsf{p}}ql l^\dagger q^\dagger + ( \pi { \mathsf{p}}ql l^\dagger q^\dagger)^2\right ) ] \nonumber \\ & = { \mathsf{p}}^2 [ 2n -{\mathtt{tr}}(i_{m\times m})+ { \mathtt{tr}}\left(i_{m\times m}- 2\pi { \mathsf{p}}ql l^\dagger q^\dagger + ( \pi { \mathsf{p}}ql l^\dagger q^\dagger)^2\right ) ] \nonumber \\ & = { \mathsf{p}}^2 [ 2n - m + { \mathtt{tr } } ( ( i - \pi { \mathsf{p}}ql l^\dagger q^\dagger ) ^2 ) ] \nonumber \\ & = { \mathsf{p}}^2 [ 2n - m + { \mathtt{tr}}(\sigma^2 \pi)^2 ] \nonumber \\ & = { \mathsf{p}}^2 [ 2n - m + \sigma^4 { \mathtt{tr}}(\pi^2 ) ] \label { var1}.\end{aligned}\ ] ] alternatively , using , we get @xmath391 so far , two closed formulas for @xmath266 and @xmath267 have been derived in equations , , and . in the following subsections , the proof of the main results will be developed using the notation and formulas given in this subsection . the matrix @xmath392 is positive definite and real , since @xmath393 . in addition , the entries of matrix @xmath394 as stated in are of the form @xmath395 where @xmath383 and @xmath384 for @xmath385 . therefore , all the diagonal entries of matrix @xmath396 are ( @xmath397 ) which are less than or equal to @xmath398 for @xmath399 . using peierl s inequality ( see lemma [ peierl ] in section [ preliminaries ] ) for the convex function @xmath400 on @xmath401 , the positive matrix @xmath402 and the normal basis @xmath403 s ( @xmath404 ) , where @xmath405 is the unit vector with 1 in the @xmath406th position and 0 elsewhere , we have @xmath407 substituting into , results in a lower bound on distortion as follows @xmath408 since sampling time instances @xmath80 s and pre - sampling filter @xmath69 are arbitrary , for any value of @xmath8 , we get @xmath409 since @xmath400 is a strictly convex function for @xmath401 , equality in the above equation holds if and only if @xmath30 s are eigenvectors of @xmath378 , _ i.e. , _ if and only if @xmath378 is a diagonal matrix . in particular , assuming that that we do not use a pre - sampling filter on the signal bandwidth _ @xmath410 , equality holds if and only if @xmath411 is a diagonal matrix . therefor , we have the following equation for @xmath412 : @xmath413(i , k ) & = \sum_{\ell = n_1}^{n_2}\cos(\ell\omega_0(t_i - t_k))\\&=\mathsf{real } \ { \sum_{\ell = n_1}^{n_2 } e^{j2\pi \frac{t_i - t_k}{t } \ell } \ } \\&= \mathsf{real } \ { e^{j\pi \frac{t_i - t_k}{t } ( n_1+n_2)}. ~\frac{\sin(\pi \frac{t_i - t_k}{t } n)}{\sin(\pi \frac{t_i - t_k}{t } ) } \ } \\&=\cos(\pi \frac{t_i - t_k}{t } ( n_1+n_2 ) ) .~ \frac{\sin(\pi \frac{t_i - t_k}{t } n)}{\sin(\pi \frac{t_i - t_k}{t } ) } = 0.\end{aligned}\ ] ] consequently , the sampling time instances @xmath80 s should satisfy the following equations : @xmath414 thus , given any @xmath415 and @xmath406 , we should either have @xmath416 . then , @xmath417 hence , using , we have @xmath418 = { \mathsf{p}}^2 \left[2n - m + \frac{m}{(1 + snr)^2}\right ] .\ ] ] here again , equality holds if and only if @xmath378 is a diagonal matrix . _ from the proof , one can see that the lower bounds given in lemma [ lemma : lower:1 ] are achieved if and only if matrix @xmath419 ( defined in ) is diagonal for the optimizing @xmath80 s and the pre - sampling filter . _ [ rmk : suff - cond - lemma1 ] we use the distortion formula given in and minimize distortion subject to the sampling locations and the pre - sampling filter . let @xmath400 for @xmath401 . we have @xmath420\nonumber\\ & = { \min_{h(.),\{t_i , i = 1 , \cdots , m\}}}[\frac12 { \mathsf{p}}\sigma^2{\mathtt{tr}}(f(\gamma^{-1}))]\nonumber \\&\geq \frac12{\mathsf{p}}\sigma^2 { \min_{h(.),\{t_i , i = 1 , \cdots , m\ } } } \sum_{k = 1}^{2n } f(\gamma^{-1}(k , k ) ) \label{eq . 1 } \\&= \frac12{\mathsf{p}}\sigma^2 { \min_{h(.),\{t_i , i = 1 , \cdots , m\}}}\sum_{\ell = n_1}^{n_2 } \frac{1}{\eta_\ell}+\frac{1}{\gamma_\ell},\label{eq . 6}\end{aligned}\ ] ] where follows from the peierl s inequality ( lemma [ peierl ] ) for the covex function @xmath421 and is written by using with the following values for @xmath422 and @xmath423 : @xmath424 and @xmath425 observe that @xmath426 where @xmath427 for @xmath385 . the proof for the variance is similar . using , and the peierl s inequality for the choice of @xmath430 we have @xmath431\nonumber\\ & = { \min_{h(.),\{t_i , i = 1 , \cdots , m\}}}[\frac12 { \mathsf{p}}\sigma^2{\mathtt{tr}}(f(\gamma^{-1}))]\nonumber \\&\geq { \mathsf{p}}^2 \sigma^4 { \min_{h(.),\{t_i , i = 1 , \cdots , m\ } } } \sum_{k = 1}^{2n } f(\gamma^{-1}(k , k ) ) \nonumber \\&= { \mathsf{p}}^2 \sigma^4 { \min_{h(.),\{t_i , i = 1 , \cdots , m\}}}\sum_{\ell = n_1}^{n_2 } \frac{1}{\eta^2_\ell}+\frac{1}{\gamma^2_\ell}\nonumber,\end{aligned}\ ] ] where @xmath422 and @xmath423 are given in and . from the convexity of the function @xmath432 we have @xmath433 for nonzero values of @xmath170 and @xmath171 . therefore , @xmath434 _ necessary and sufficient conditions for tightness of the lower bounds : _ the ideas are similar to the ones given in the proof of lemma [ lemma : lower:1 ] . for pierel s inequality to be tight , we need @xmath435 defined in to be a diagonal matrix . we further need @xmath436 to have all the inequalities as equalities . we also need @xmath437 for all @xmath241 , implying that @xmath410 . therefore , the equality holds if and only if matrix ( @xmath438 ) is diagonal , and all of its diagonal entries are equal . + the off - diagonal and diagonal entries of ( @xmath438 ) are given by @xmath439 ( i , k ) = \begin{cases } { \mathsf{p}}\cdot \sum_{\ell=1}^{m}\cos((n_1+i-1)\omega_0t_\ell ) \cos((n_1+k-1)\omega_0t_\ell)~~~;~1 \leq i < k \leq n\\ { \mathsf{p}}\cdot \sum_{\ell=1}^{m}\cos((n_1+i-1)\omega_0t_\ell ) \sin((n_1+k - n)\omega_0t_\ell)~~;~1 \leq i\leq n\leq k \leq2 n\\ { \mathsf{p}}\cdot \sum_{\ell=1}^{m}\sin((n_1+i - n)\omega_0t_\ell ) \sin((n_1+k - n)\omega_0t_\ell)~;~ n\leq i \leq2 n,~ n\leq k \leq2 n\\ \end{cases},\ ] ] @xmath440 ( i , i ) = \begin{cases } \frac{m{\mathsf{p}}}{2}+\sigma^2 + \frac{{\mathsf{p}}}{2}\cdot \sum_{\ell=1}^{m}\cos(2(n_1+i-1)\omega_0t_\ell)~;~ 1\leq i\leq n\\ \frac{m{\mathsf{p}}}{2}+\sigma^2 - \frac{{\mathsf{p}}}{2}\cdot \sum_{\ell=1}^{m}\cos(2(n_1+i-1)\omega_0t_\ell)~;~ n\leq i \leq2n \end{cases},\ ] ] respectively . if we put the off - diagonal entries zero and write the above equations in a simpler form , we get @xmath441 substituting @xmath442 in the above equations , we obtain @xmath443 or in another form , we have @xmath444 these equations imply that the off - diagonal entries are all zero . but they also imply that the diagonal entries are all equal to zero , _ i.e. , _ @xmath445 here we are in the case of @xmath12 . in lemma [ lemma : lower:1 ] , the desired lower bounds on @xmath9 and @xmath10 were developed . moreover , we showed that these lower bounds are achievable if we do not use the pre - sampling filter and for any time instances @xmath80 or @xmath99 @xmath446 is optimal . to find the interpolation formula , consider the mmse estimator given in and the matrix @xmath363 in . we have @xmath447 since in the optimal case , @xmath448 and @xmath449 , the fourier coefficients of the reconstructed signal are as follows : @xmath450 which results in the desired reconstruction formula in . remember the lower bounds given in lemma [ lemma : lower:1 ] . we showed that , when we do not use a pre - sampling filter , equality in this lemma holds if and only if latexmath:[\ ] ] where @xmath600 and @xmath601 are of the forms : @xmath602 and @xmath603 observe that @xmath604 where @xmath427 for @xmath385 . similar to the proof of lemma [ lemma : lower:2 ] , using inequality we continue @xmath605 furthermore , the distortion given in is achieved under the same conditions given in lemma [ lemma : lower:2 ] that ensures the matrix @xmath606 becomes diagonal . the proof of theorem [ t1c ] can be mimicked to show that uniform sampling , _ @xmath181 is a solution to the above equation if for each @xmath182 in the interval @xmath183 , @xmath184 . in particular , for the rates above the nyquist rate , _ i.e. , _ @xmath185 , uniform sampling is optimal . a. kipnis , a. goldsmith , t. weissman , y. eldar , distortion rate function of sub - nyquist sampled gaussian sources corrupted by noise , " in proc . allerton conf . on comm . , control , and computing , 2013 , pp . 901 - 908 . d. chan and r. donaldson , optimum pre- and post - filtering of sampled signals with application to pulse modulation and data compression systems , " ieee trans . on commun . 141 - 157 , apr . 1971 . d. i. shuman , s. k. narang , p. frossard , a. ortega , and p. vandergheynst , the emerging field of signal processing on graphs : extending high - dimensional data analysis to net- works and other irregular domains , " ieee signal process . mag . 83 - 98 , 2013 . l. j. alvarez - vazquez , a. martinez , m. e. vazquez - mendez and m. a. vilar , optimal location of sampling points for river pollution control , " mathematics and computers in simulation , vol . 149 - 160 , feb . 2006 . s. rajendran and k. m. liew , optimal stress sampling points of plane triangular elements for patch recovery of nodal stresses " international j. for numerical methods in engineering , vol . 579 - 607 , apr . 2003 . r. venkataramani and y. bresler , perfect reconstruction formulas and bounds on aliasing error in sub - nyquist nonuniform sampling of multiband signals , " ieee trans . inf . theory , vol . 46 , no . 6 , pp . 2173 - 2183 , jun .

we present a study of analytic models of starless cores whose line profiles have `` infall asymmetry , '' or blue - skewed shapes indicative of contracting motions . we compare the ability of two types of analytical radiative transfer models to reproduce the line profiles and infall speeds of centrally condensed starless cores whose infall speeds are spatially constant and range between 0 and 0.2 km s . the model line profiles of hco ( ) and hco ( ) are produced by a self - consistent monte carlo radiative transfer code . the analytic models assume that the excitation temperature in the front of the cloud is either constant ( `` two - layer '' model ) or increases inward as a linear function of optical depth ( `` hill '' model ) . each analytic model is matched to the line profile by rapid least - squares fitting . the blue - asymmetric line profiles with two peaks , or with a blue shifted peak and a red shifted shoulder , can be well fit by one or both of the analytic models . for two - peak profiles is best matched by the `` hill5 '' model ( a five parameter version of the hill model ) , with an rms error of 0.01 km s , while the `` twolayer6 '' model underestimates the infall speed by a factor of . for red - shoulder profiles , the hill5 and twolayer6 fits reproduce infall speeds equally well , with an rms error of 0.04 km s . the fits are most accurate when the line has a brightness temperature greater than 3 k. our most accurate models tend to reproduce not only the line profile shape , but also match the excitation conditions along the line of sight . a better match to the line profile shape does not necessarily imply a better match to the infall speed and provide guidance on how to minimize the risk of obtaining a poor infall speed fit . a peak signal to noise ratio of at least 30 in the molecular line observations is required for performing these analytic radiative transfer fits to the line profiles . moderate amounts of depletion and beam smoothing do not adversely affect the accuracy of the infall speeds obtained from these models .

astro-ph0410748

there are four distinct stages in isolated star formation : ( 1 ) formation of a gravitationally bound core withing a molecular cloud , ( 2 ) the gravitationally driven collapse of that core , ( 3 ) the formation of a central star - like object , ( 4 ) dispersal of the remaining cloud material @xcite . the initial two stages are predominantly studied through millimeter- and submillimeter - wave - length observations . the molecular line rotational transition emission visible in these wavelength regimes allow one to probe both the physical and chemical composition of the molecular cloud cores out of which stars form . spectral analysis of the molecular line emission also allows one to investigate the kinematic motions within the molecular cloud , particularly the phase of gravitational collapse @xcite as well as the associated bipolar outflows which are ubiquitous around young stars @xcite . in a dense cloud core , commonly used molecular line tracers of moderate optical depth , such as cs ( ) , or hco@xmath1 ( ) tend to become self - absorbed . as the cloud contracts under the influence of gravity , the central regions tend to become more dense than the outer regions . this sets up a gradient in the excitation temperature of the molecular gas within the cloud such that the excitation temperatures in the central regions of the cloud are greater than those in the outer regions , even if the kinetic temperature is constant within the molecular cloud . if the cloud is spherical and static the spectral line emission from this cloud in these tracers will be symmetric about the line of sight velocity and self - absorbed . however , if there is radial motion around the molecular cloud core , a velocity gradient develops around the center of the core , precisely where the excitation temperature is the greatest . much of the emission from the rear of the cloud is not absorbed as it travels through the core , due to the high excitation temperature within the core , and continues unabsorbed through the front section of the cloud because the velocity gradient between the rear and front sections of the cloud results in a doppler shift which is significant compared to the broadening by thermal motions . emission from the front of the cloud however is absorbed by nearby molecular gas with lower excitation temperature and only a small doppler shift . in the case of inward radial motion the rear emission is blue - shifted and the front emission is red - shifted , yielding the `` classic '' blue - asymmetric infall line profile . this line profile has become the predominant tool for investigating infall motions within star - forming molecular cloud cores @xcite . although this blue - asymmetry results from infall , it is not unique to that phenomenon . the superposition of multiple clouds along the line of sight can also cause an asymmetric line profile . by observing an optically thin molecular line , which should remain symmetric around the line of sight velocity of the cloud core , it is possible to reduce the likelihood of mistaking separate clouds for an infalling cloud . asymmetric line profiles can also arise in kinematically complex regions . studies have shown that rotation and outflow can also produce asymmetric line profiles @xcite . even the asymmetric line emission of b335 , once thought to be the prototypical example of spectrally detected inside - out collapse @xcite , is now also thought to also contain an outflow which gives rise to the line wings previously modeled as part of the collapse @xcite . in spite of the multiple molecular cloud configurations that can result in blue - asymmetric line profiles , they are still the best way to study infall motions available , provided we understand the context in which they occur and make clear the necessary caveats . understanding how blue - asymmetric line profiles arise from infall motion is easy , but creating a complete model matching the molecular emission from one or more tracers is a difficult and time - consuming task . typically one constructs an infalling physical model and uses lvg , monte carlo @xcite or microturbulent @xcite radiative transfer to model the emission . by iteratively changing the model parameters one is able to match some or all of the observations reasonably well . this process can take several minutes per iteration , and becomes prohibitively slow as the number of physical parameters becomes large and as the number of molecular species increases . the derived cloud parameters are also highly dependent on the input physical model . changing the physical model can lead to an entirely different set of solutions . `` starless cores '' are useful regions for observations of inward motion because they are numerous , nearby , relatively simple in their density structure , and their motions are not confused by outflows . they have been the subject of several studies @xcite , some of which have used radiative transfer models to fit their line profiles . in this paper we present simple analytic radiative transfer models , based on the simplest available assumptions of possible excitation temperature trends within starless molecular cloud cores . these models can easily be fit to blue - asymmetric self - absorbed molecular line profiles , and produce an estimate of the infall velocity suggested by those line profiles . we also test the efficacy of the analytic radiative transfer models by comparing the infall velocities derived using those models with actual infall velocities in rigorously modeled condensing molecular clouds . our main finding is that our analytic radiative transfer models can yield good estimates for the physical conditions of the starless cores with few assumptions and little computational investment . we compare two analytic models in this paper . although it is possible to construct more analytic models using the same techniques in this paper we feel that these two models are the most appropriate for modeling infall . it is possible to analytically integrate the equation of transfer @xmath3 where @xmath4 is the brightness temperature , @xmath5 is the excitation temperature and @xmath6^{-1}$ ] . the brightness temperature is defined as @xmath7 and is directly proportional to the specific intensity @xmath8 . the general solution to the equation of transfer , assuming that optical depth increases away from the observer is @xmath9 where @xmath5 is the excitation temperature of a region which varies over the optical depth interval from 0 to @xmath10 , and @xmath11 is the incident specific intensity of radiation on that region at @xmath10 in units of brightness temperature . this equation can easily be integrated over regions of constant excitation temperature . we assume regions of constant excitation temperature along the line of sight in our first model . the first model , first discussed in @xcite , we call the `` two - layer '' model . it is perhaps unfortunate that this name has been given to this model and has stuck , as it implies that this is a model of two plane - parallel layers along the line of sight . this is not true . the model actually applies to a line of sight in which two regions with differing excitation temperatures are moving toward each other . we assume the near region has a lower excitation temperature than the far region . we assume both regions have a velocity dispersion for the observed molecule @xmath12 and a total optical depth of @xmath10 at line center . the excitation temperature of the front region is @xmath13 , while that of the rear region is @xmath14 , and the regions are approaching each other with a speed of @xmath15 . the optical depth of each region at a velocity @xmath16 is @xmath17 , \\ \tau_{r}(v ) & = & \tau_{0}\exp\left[-(v - v_{\rm lsr}+v_{\rm in})^{2}/2\sigma^{2}\right],\end{aligned}\ ] ] assuming that the average line of sight velocity of both regions is @xmath18 . figure [ models ] provides a graphical representation of the model . the brightness temperature of the spectral line , obtained from the equation of radiative transfer , is @xmath19 + j(t_{r})\left[1-e^{-\tau_{r}(v)}\right]e^{-\tau_{f}(v ) } - j(t_{b})\left[1-e^{-\tau_{r}(v)-\tau_{f}(v)}\right],\ ] ] where @xmath20 is the background temperature . the parameters @xmath10 , @xmath12 , @xmath13 , @xmath14 , @xmath21 , and @xmath22 are free parameters which can be adjusted to fit blue - asymmetric line profiles . @xcite has used this analytic model to derive infall velocity estimates for 29 starless cores . they find that infall velocities are typically on the order of a tenth of a kilometer per second , comparable to the velocity dispersions measured in these sources . equation [ xfer ] is also analytically integrable if @xmath23 is a linear function of the optical depth @xmath24 . assuming a simple linear function @xmath25\tau$ ] we can integrate the equation of transfer to obtain @xmath26 in the above equation @xmath24 is a function of the doppler velocity , so in order to calculate the intensity of emission across a line profile @xmath23 must be a linear function of @xmath24 at every frequency . it is possible to construct a series of regions along the line of sight where @xmath23 varies linearly with @xmath24 within each region and analytically calculate the brightness of radiation at a given frequency emitted from that line of sight using the above equation . we have constructed our next model using the above technique and equations which we feel approximates the excitation conditions seen along the line of sight in infalling clouds . the second model , which we call the `` hill '' model , is introduced for the first time in this paper . this model consists of a core with a peak excitation temperature @xmath27 at the center and an excitation temperature of @xmath28 at the near and far edges of the core . the @xmath23 drops linearly from @xmath29 at the center to @xmath30 at edges of the core , forming a hill in the @xmath23 profile . the optical depth of the core is @xmath31 , and its infall velocity is @xmath32 , while the systematic velocity of the system is @xmath21 . a schematic representation of this model is shown in figure [ models ] . we introduce this new type of model because we believe it may be more analogous to the excitation profile we are likely to be observing in starless cores . in figure [ excivtau ] we compare the excitation profile as a function of optical depth at line center for a simulated cloud , based on a monte carlo radiative transfer model , to the best fit hill and two - layer models of the line profile derived from that cloud . the excitation profile of our modeled starless core has a significant slope , which the hill model can replicate , but the two - layer model , due to its use of constant excitation temperature zones , can not replicate . this improvement provides motivation for introducing a new analytic radiative transfer model of infall . in order to solve the equation of radiative transfer we separate the hill model profile into two regions along the line of sight : ( 1 ) the portion of the cloud in which the excitation temperature is rising along the line of sight , whose optical depth is @xmath33 ; and ( 2 ) the portion of the cloud in which the excitation temperature is falling along the line of sight , whose optical depth is @xmath34 . the optical depth as a function of line of sight velocity for each of these regions is @xmath35 , \\ \tau_{r}(v ) & = & \tau_{\rm c}\exp\left[-(v - v_{\rm lsr}+v_{\rm c})^{2}/2\sigma^{2}\right],\end{aligned}\ ] ] assuming a velocity dispersion of @xmath12 in the entire cloud . we can now solve the equation of transfer by integrating along the line of sight through each of these two regions to derive the brightness temperature @xmath36 \nonumber \\ & & \mbox { } + \left(j(t_{0})-j(t_{b})\right)\left[1-e^{-\tau_{r}(v)-\tau_{f}(v)}\right],\end{aligned}\ ] ] where @xmath20 is the background temperature . we believe the excitation temperature gradient in this model might more closely approximate the excitation profiles in observed clouds . the six free parameters in the hill model are @xmath31 , @xmath12 , @xmath28 , @xmath27 , @xmath21 , and @xmath32 . it is also possible to add an envelope of constant excitation temperature moving with its own infall velocity . assuming this envelope has an excitation temperature of @xmath28 , the optical depth of the envelope and the velocity of the envelope adds two more free parameters , creating a maximum of eight free parameters possible in the hill models . in order to reduce the total number of free parameters we consider several variants of each model by holding certain parameters fixed . we considered two variants of the two - layer model , which we call twolayer5 and twolayer6 , as well as 4 variants of the hill model , which we call hill5 , hill6 , hill6core , and hill7 . the number is each variant name refers to the number of free parameters in that variant . the parameterization of the two - layer model variants are shown in table [ twolayervar ] , while those of the hill model variants are shown in table [ hillvar ] . lllllll & free & free & free & free & @xmath20 & free + twolayer6 & free & free & free & free & free & free + lllllllll & 0 & free & free & 0 & free & free & @xmath20 & free + hill6 & free & free & free & free & 0 & free & @xmath20 & free + hill6core & free & free & free & 0 & free & free & @xmath20 & free + hill7 & free & free & free & free & 0 & free & free & free + the analytic radiative transfer models do an excellent job of fitting blue - asymmetric line profiles @xcite . their efficacy in determining physical parameters of the molecular cloud , such as infall velocity , has not yet been studied . in this paper we study how effectively we can recover the physical parameters of modeled clouds using both the two - layer and hill analytic radiative transfer models . we then use this information to determine the point at which the systematic errors of the model dominate over random observational errors . in our first three simulations , we modeled molecular clouds with a density law of the form @xmath37 where @xmath38 is the central density , @xmath39 is the turnover radius , and @xmath40 is the power law of the density profile at large radii @xcite . this model reproduces the flat density region ( at @xmath41 ) and power law region ( at @xmath42 ) required to model starless cores . it also is in good agreement with the density profile of a bonnor - ebert sphere @xcite as well as the `` plummer - like '' density model of @xcite . the clouds modeled in this paper have peak densities ( @xmath38 ) of @xmath43 @xmath44 , a turnover radius ( @xmath39 ) of @xmath45 cm , an outer radius of @xmath46 cm , and an @xmath40 of 4 . we chose to model hco@xmath1 rotational line emission from these clouds , as it is often moderately optically thick in starless cores . we assumed that the kinetic temperature of the gas is 10 k throughout the cloud . we set the relative abundance of hco@xmath1 at @xmath47 @xcite and the hco@xmath1/h@xmath48co@xmath1 abundance ratio to 64 . depletion is not included here , however we do investigate the effect of depletion in [ depletionbeam ] . there are four velocity laws which we investigate in this paper . the first ( simulation a ) is constant infall velocity for all radii in the cloud . the second ( simulation b ) is a constant infall velocity for @xmath42 and 0 velocity for @xmath41 . the third ( simulation c ) is constant infall velocity for @xmath41 and 0 velocity for @xmath42 . these simulations are summarized in table [ sims ] . these velocity laws were chosen because they are simple velocity laws to investigate . our analytic radiative transfer models define a characteristic single infall velocity which is analogous to simulations a , b , and c. we set the velocity dispersion to 0.1 km s@xmath0 over the entire cloud and typically allow the velocities within the cloud to be up to twice the velocity dispersion . these results should all scale with the velocity dispersion and the peak line brightness . in addition to the simulations described above , we also simulated a different density profile based on the best fit bonnor - ebert model of b68 @xcite . the parameters of this simulation ( simulation d ) are discussed in [ besphere ] . llll a & tmcwc & @xmath49 & @xmath50 0.00 , 0.02 , , 0.2 km s@xmath0 + b & tmcwc & @xmath51 & @xmath50 0.00 , 0.02 , , 0.2 km s@xmath0 + c & tmcwc & @xmath52 & @xmath50 0.00 , 0.02 , , 0.2 km s@xmath0 + d & be & @xmath49 & @xmath50 0.000 , 0.024 , , 0.24 km s@xmath0 + we use these simulated starless cloud cores to synthesize hco@xmath1 ( ) and hco@xmath1 ( ) molecular line emission using the @xcite 1-d monte carlo radiative transfer model . the simulated clouds were divided into 20 spherical shells , each of uniform density , kinetic temperature , radial velocity , and with a radial size of @xmath53 cm . the level populations within the first 21 rotational levels of the hco@xmath1 molecule were calculated using the einstein - a and hco@xmath1 with h@xmath54 collisional rate coefficients from @xcite and @xcite . we then integrate along the central line of sight to find the and molecular line spectral profiles of hco@xmath1 because they are frequently observed , and often have blue - asymmetric line profiles attributed to infall @xcite . we then fit these synthesized spectra with the two - layer and hill analytic radiative transfer models to check the ability of the analytic models to recover physical parameters within the modeled cloud . several algorithms exist for minimizing multidimensional functions . there is always the danger of minimizing to a `` local '' minimum ( meaning the lowest value within some finite region of the function to be minimized ) instead of the `` global '' minimum ( meaning the lowest function value over the entire parameter space ) and it becomes more difficult to avoid local minima as your function goes to higher dimensions . there are several strategies to avoid local minima , and we have chosen a hybrid minimization algorithm to do just that . we use the differential evolution ( de ) algorithm of @xcite initially to separate the local minima from the global minimum , then we use the nelder - mead simplex method @xcite to optimize the fit . the de minimization algorithm is applicable to many astronomical problems , and is similar in scope to simulated annealing . de is an evolutionary algorithm in that several parameter sets are randomly selected and these are randomly modified over several `` generations '' until acceptable convergence has been achieved . the variability in each parameter is scaled by the variation of the solution sets in each generation , so as the algorithm converges the region of the search decreases without the need to include a decreasing `` temperature '' as in simulated annealing . de is also superior to simulated annealing in that it is more likely to achieve convergence , is a simpler algorithm , and has fewer adjustable parameters @xcite . we made four simulations of infalling clouds and used the @xcite monte carlo radiative transfer model to extract hco@xmath1 ( ) and hco@xmath1 ( ) line profiles from those simulations . in this section we will discuss how effective our analytic models are at deriving the input velocities in those simulation , quantify the relative contributions of both random and systematic errors in model fits of simulated data sets , and apply this model to an observed infall profile in a source which has been rigorously modeled using monte carlo radiative transfer techniques . simulation a , which has a constant infall velocity for all cloud radii , has asymmetric line profile typical of those seen in infalling clouds . the hco@xmath1 ( ) line profiles are shown in figure [ majo ] , along with their optically thin h@xmath48co@xmath1 ( ) counterparts . the spectra in the bottom on the left and right side are for a simulation with no infall . the middle spectra on the left and right are of a simulation with an infall velocity of 0.1 km s@xmath0 . the top spectra on the left and right are from a simulation with an infall velocity of 0.2 km s@xmath0 . overlain on the hco@xmath1 ( ) spectra on the left panel are the twolayer5 and hill5 best fit models , while on the right the twolayer6 and hill6 spectra are shown . figure [ majt ] depicts hco@xmath1 ( ) and h@xmath48co@xmath1 ( ) spectra from the same simulations along with their twolayer5 , hill5 , twolayer6 , and hill6 best fit spectral line profiles . two variants are not shown in these fits , the hill6core model best fit in most cases has a @xmath55 which is very close to 0 , yielding the same parameters and spectral line shape as the hill5 model . the hill7 model always manages a very good fit to the spectral line shape , however in some cases , especially in spectra with a small red component that is not well separated from the blue component ( such as the middle and top spectra in figure [ majt ] ) , the parameters of the hill7 model are not very well constrained . a wide range of infall velocities , optical depths , and excitation temperatures in the seven parameter hill7 model can yield the same line profile . there are several interesting features in these spectral line fits . when the infall velocity is greater than the velocity dispersion the h@xmath48co@xmath1 line profiles become double peaked . this is not due to self - absorption as the lines are optically thin . it is due to the fact that the molecular gas along the central line of sight is traveling at two velocities separated by more than the velocity dispersion , resulting in two peaks in the optically thin component . this feature is unlikely to be observed in h@xmath48co@xmath1 ( ) lines as they are too weak to be easily detected . also interesting is that while the dip between the red and blue peaks of the spectral line profile is deep , both the hill variants and the two - layer variants do a good job a reproducing the spectral line , but as the red peak drops in intensity relative to the blue peak the analytic model fits tend to produce a spectral line with a single blue peak and a red shoulder with no local maximum but only a flattening of the slope . the two - layer model variants tend to produce this type of spectrum at lower infall speeds than the hill model variants ( figure [ majo ] ) . we call the spectra with distinct blue and red peaks `` dip '' spectra and those with only a flattening of the slope , `` shoulder '' spectra . we also see that none of the models can reproduce a spectral line which has a minimal red peak superimposed on a broadly asymmetric line ( as in the top spectra of figure [ majt ] ) . these models tend to produce spectra that are similar in shape to two overlapping gaussians of equal width . if we try to fit a spectral line that differs greatly from one that can be obtained by adding two gaussians of equal width , then we tend to not get good line fits with any of the analytic models . in figure [ maf ] we show the quality of the infall velocity fit attained by the analytic model fits to the hco@xmath1 ( ) line profiles in this simulation ( left side ) as well as the model fits to the hco@xmath1 ( ) line profiles in this simulation ( right side ) . in this simulation , for the entire infall velocity range up to twice the velocity dispersion , the hill5 model tends to reproduce the modeled infall velocity with an rms systematic error of less than 0.01 km s@xmath0 in the hco@xmath1 ( ) lines and 0.02 km s@xmath0 in the hco@xmath1 ( ) lines . the twolayer5 model tends to consistently underestimate the infall velocity by over a factor of 2 in both transitions when it produces good line fits . the twolayer6 model fits the infall velocity with an rms error of 0.03 km s@xmath0 in both the hco@xmath1 ( ) lines and the hco@xmath1 ( ) lines . generally the twolayer6 model does a better job in cases where the spectrum is dominated by the blue component and the relative size or integrated intensity of the red component is small . this happens to be the true for cases with higher infall velocities in the hco@xmath1 ( ) and hco@xmath1 ( ) spectra in this simulation . the twolayer6 model also performs well when the excitation temperature of the rear layer matches the simulated excitation temperature in the cloud . the hill6 model does a good job at fitting the infall velocity for infall velocities greater than the velocity dispersion , but tends to greatly overestimate the infall velocity in cases where it is smaller than the velocity dispersion . the hill7 model also tends to reproduce the infall velocity fairly well , but not as well as the hill5 model . it should be noted that a good fit to the line profile does not always translate to a good fit to the physical parameters , such as the infall velocity , in the simulated clouds . in the case of the twolayer6 models , often the dip solution is a good fit to the line profile , but a worse fit to the infall velocity than the shoulder solution . the hill7 model is able to fit just about any line profile , but often yields worse infall velocities than hill5 model fits with larger minimized @xmath56 . in summary , the infall velocities for simulation a ( constant infall ) are most accurately reproduced by the hill5 model with a rms deviation of 0.016 km s@xmath0 over 22 cases . the twolayer6 model is significantly worse with an overall rms deviation of 0.030 km s@xmath0 over 22 cases . simulation b , which has a constant infall velocity for @xmath42 and no infall for @xmath41 , has asymmetric line profiles typical of those seen in infalling clouds . figure [ mbjo ] shows the hco@xmath1 ( ) line profiles for three different infall velocities . the self - absorbed solid line in the lowest left and right panels is the line profile for a cloud with no infall , as well as the optically thin h@xmath48co@xmath1 ( ) line profile . in the middle left and right are the hco@xmath1 and h@xmath48co@xmath1 ( ) line profiles for a cloud with an infall velocity of 0.1 km s@xmath0 , equal to the velocity dispersion in that cloud . the upper left and right panels are the hco@xmath1 and h@xmath48co@xmath1 ( ) line profiles for a cloud with an infall velocity of 0.2 km s@xmath0 . these line profiles are slightly narrower than those seen in simulation a ( constant infall ) . this is due to the fact that the cores , where the most dense gas is located , are stationary and not producing any broad line wings . as in our previous simulation we have overlaid the twolayer5 , hill5 , twolayer6 , and hill6 best fit to the hco@xmath1 spectra . figure [ mbjt ] indicates these same results for the hco@xmath1 ( ) and h@xmath48co@xmath1 ( ) spectra in simulation b ( envelope infall ) . we investigate the efficacy of our models in matching the infall velocity of the simulation b ( envelope infall ) clouds in figure [ mbf ] . the twolayer5 model once again underestimates the infall velocity in all cases and in both transitions , although it monotonically increases throughout the simulation . the twolayer6 model tends to yield the same result as the twolayer5 model over many of the infall velocities , but jumps to a higher infall velocity which in many cases approximates the correct infall velocity ( especially in the hco@xmath1 ( ) fits ) . we investigated these jumps in the twolayer6 model further and found that there are two local minima in the @xmath56 surface for many spectrum shapes . the twolayer6 minima we found correspond to `` dip '' spectra , with two distinct peaks , and `` shoulder '' spectra with one peak and a flattening of the spectrum on the red side . these two spectral line fits tend to have very different parameters for the infall velocity which we show in figure [ dipshoulder ] for the hco@xmath1 ( ) lines generated in this simulation . the letter with the larger size indicates the fit with the lowest @xmath56 , and the solid line indicates the expected value if the infall velocity modeled analytically matches the infall velocity of the cloud . in the hco@xmath1 ( ) profiles , the `` dip '' solution is a better fit to the line profile up to a modeled infall velocity of 0.12 km s@xmath0 , however the `` shoulder '' solution yields a better fit to the infall velocity for modeled infall velocities above 0.04 km s@xmath0 . a similar trend is seen in the hco@xmath1 ( ) results . the `` shoulder '' fits tend to have a higher infall velocity , and are often a much better match to the infall velocity . the `` dip '' spectra are typically the equivalent to the twolayer5 model results . the best `` dip '' spectrum fit of the twolayer6 model usually results in an excitation temperature of the front layer equal to the background temperature and an underestimate of the infall velocity . we therefore recommend that one should attempt to find the local minimum corresponding to a `` shoulder '' fit when using the twolayer6 model , and that one can verify that one has found the `` shoulder '' fit by making sure that the excitation temperature of the front layer is significantly higher than the background temperature . another trend revealed in figure [ mbf ] is that the twolayer6 model does not do a good job of fitting high velocity infall using the hco@xmath1 ( ) line profile in this modeled cloud . we can see in the upper - most line profiles of figure [ mbjt ] that the hco@xmath1 ( ) line profiles do not separate into multiple distinct components at high infall velocities in this model . as the lines begin to blend into a profile that is not easily separable into multiple components , the two - layer model begins to break down and does not accurately match the actual infall velocity . the lower panels of figure [ mbf ] show the hill model variants infall velocities compared to the infall velocities in those same clouds . once again we see that hill5 does a good job of matching most of the infall velocities using the hco@xmath1 ( ) line profile , but tends to break down at high velocities . overall the hill5 model fits to the hco@xmath1 ( match the simulated infall velocities with an rms error of 0.08 km s@xmath0 over the full range of simulated infall velocities , however for infall velocities less than 1.5 times the velocity dispersion the rms error is less than 0.01 km s@xmath0 . the hill5 infall velocity fits break down at an even lower infall velocity in the case of the hco@xmath1 ( ) line , however the overall rms error in the infall velocity fits for that transition are only 0.07 km s@xmath0 . we believe this is because the contribution of the red component is so low in these cases . when the magnitude of the red component drops hill5 does not produce a good profile fit ( see the topmost spectrum of figure [ mbjo ] ) . we also believe that the hill5 model does a poor job when the shoulder starts to disappear yielding a sloping asymmetric line such as the uppermost spectrum of figure [ mbjt ] . once again the hill6 model does a poor job at fitting low infall velocities . the hill7 does a good job at fitting all the infall velocities in both line transitions , typically within 0.03 km s@xmath0 of the modeled value . the error in the fit begins to increase , especially in the case of the hco@xmath1 ( ) fit , as the infall velocity increases . this is due to the diminishing magnitude of the red - shifted component . as the blue - shifted component begins to dominate the hill7 model fit , the seven parameters of the model begin to overdetermine the line profile , resulting in a range of solutions that fit equally well . in practice it is possible to avoid this by reducing the number of parameters and using the hill6 model , however in most observed infall profiles the magnitudes of both the red and blue components are not as large as seen in the top panel of figure [ mbjo ] ( cf . * ) . in typical line profiles , such as those seen in this spherically symmetric infalling cloud , with a uniform infall velocity in the outer layers , the hill5 does an excellent job at extracting the infall velocity , while the twolayer6 does an adequate job for lines with a small red peak or narrow shoulder component . we confine infall to the core ( @xmath41 ) , and surround the infall region with a static envelope in simulation c. the resulting hco@xmath1 and h@xmath48co@xmath1 ( ) line profiles are shown in figure [ mcjo ] . the hco@xmath1 and h@xmath48co@xmath1 ( ) line profiles are shown in figure [ mcjt ] . once again modeled infall velocity , this time in the core , is increasing upwards . the bottom spectra are for no infall velocity , the middle spectra are for an infall velocity equal to the velocity dispersion of 0.1 km s@xmath0 , and the top spectra are for an infall velocity of 0.2 km s@xmath0 . the first thing we notice is that the hco@xmath1 ( ) lines are not very blue - asymmetric . although the width increases as the infall velocity in the core increases , the infall asymmetry does not because the transition is mostly sensitive to motions in the envelope . the emission from the core is mostly absorbed by the envelope because the optical depth in the line is high . the hco@xmath1 ( ) transition , which is more optically thin , does a better job of probing the motions in the core , and shows a greater degree of asymmetry as the infall velocity increases . the low level of asymmetry in the line results in a relatively poor fit to the infall velocity for both the two - layer model variants and the hill model variants . figure [ mcf ] depicts the efficacy of the model variants at matching the simulated infall velocity in this simulation . the twolayer5 , twolayer6 , hill5 , and hill7 all significantly underestimate the infall velocity from the hco@xmath1 ( ) line profiles . this is not surprising as that transition is very thick and not very sensitive to the motions in the core , where the infall is occurring . the hill5 fit tends to be consistently a factor of @xmath57 lower than the simulated infall velocity . the situation is more encouraging when examining the thinner hco@xmath1 ( ) line profile fits . the twolayer5 model continues to significantly underestimate the infall velocity , but the twolayer6 model matches the simulated infall velocity well for velocities between 0.1 km s@xmath0 and to 0.15 km s@xmath0 . overall the rms error on the infall velocity fits of the twolayer6 model to the hco@xmath1 ( ) lines is 0.03 km s@xmath0 . the hill5 model continues to underestimate the infall velocity from the hco@xmath1 ( ) line , resulting in an rms error on the fit infall velocity of 0.07 km s@xmath0 . we believe this is due to the fact that the red peak quickly becomes a shoulder in this simulation , and the hill5 model tends to do much better when fitting a distinct red peak . the hill6 model overestimates the infall velocity , while the hill7 model tends to do about as well as the twolayer6 model at estimating the infall velocity when fitting the hco@xmath1 ( ) lines in this simulation . both the hill7 model and the twolayer6 model begin to underestimate the infall velocity for core infall velocities greater that approximately 0.14 km s@xmath0 . at this point we run into the problem that the line profile is not becoming more asymmetric , just wider , due to the increasing velocities in the core . in our fourth simulation we choose a different density profile and a uniform velocity throughout the cloud . we specifically choose the bonnor - ebert fit to the density profile of b68 obtained by using extinction measurements of the background k giant population to produce an azimuthally averaged column density profile @xcite . we calculate the bonnor - ebert density profile using the modified lane - emden equation @xmath58 where @xmath59 is the non - dimensional radial parameter , @xmath60 is the isothermal sound speed , @xmath61 is the volume density at the cloud center , and @xmath62 . we split this second order differential equation in to two first order equations by using the variable substitutions suggested by @xcite and integrate the equations numerically to derive the density profile . we use @xmath63 , the free parameter derived by @xcite to be the best fit to the b68 cloud extinction observations . @xcite find that the relative abundance of hco@xmath1 to molecular hydrogen is less than @xmath64 , which is significantly lower than the value of @xmath47 which we have been using . under such low abundances the hco@xmath1 lines we would derive from radiative transfer calculations within this model cloud would be optically thin . since optically thick lines are required for our analysis , we choose to maintain the hco@xmath1 to molecular hydrogen relative abundance at @xmath65 and only change the density profile in this simulation . we divide the spherical cloud into thirty equally spaced shells , the central shell has a molecular hydrogen density of @xmath66 @xmath44 , while the outermost shell has a molecular hydrogen density of @xmath67 @xmath44 and a maximum radius of @xmath68 cm or 12,500 au . we assume a constant kinetic temperature of 11 k throughout the cloud , as obtained by @xcite using nh@xmath69 ( 1,1 ) and ( 2,2 ) hyperfine line observations conducted at the dsn 34 m telescope . we assume a constant infall velocity throughout the cloud at all radii . we modeled 11 infall velocities ranging from 0 to twice the velocity dispersion , which we chose to be 0.12 km s@xmath0 . we base our choice of the velocity dispersion on the n@xmath54h@xmath1 ( ) line width in b68 reported by @xcite . hco@xmath1 ( ) and h@xmath48co@xmath1 ( ) spectra from this simulation are shown in figure [ mdjo ] . as in previous simulations we choose a representative sample of three spectra in the direction of the cloud center at different infall velocities . in this case the infall velocities are 0 km s@xmath0 , 0.12 km s@xmath0 , and 0.24 km s@xmath0 . on the left side of figure [ mdjo ] the best fit twolayer5 and hill5 models are shown , while on the right side the best fit twolayer6 and hill6 models are shown . the first thing to note is that the h@xmath48co@xmath1 lines are slightly asymmetric , similar to h@xmath48co@xmath1 ( ) , hc@xmath70o@xmath1 ( ) , and d@xmath48co@xmath1 ( ) observations of l1544 seen in figure 1 of @xcite . this is as a result of self - absorption , because although the h@xmath48co@xmath1 molecule has an abundance 64 times lower than that of hco@xmath1 , it is not so low that it is optically thin . it is important to be aware of that asymmetry may arise in less abundant species , however the degree of asymmetry is indeed less than the asymmetry of the more abundant hco@xmath1 molecule . as in previous simulations , the line profiles generated by the analytic models reproduce the monte carlo generated profiles very well . similar results for the hco@xmath1 ( ) and h@xmath48co@xmath1 ( ) lines are shown in figure [ mdjt ] . the h@xmath48co@xmath1 lines are more symmetric in this case as there is less optical depth along the line of sight . we compare the infall velocities in this simulation with those obtained by fitting the line profiles with our analytic models in figure [ mdf ] . as in previous sections we present the fits to the hco@xmath1 ( ) line profiles on the left panels of the figure , and those to the hco@xmath1 ( ) line profiles on the right panels of the figure . the upper panels refer to the two - layer model fits , while the lower panels refer to the hill model fits . neither of the two - layer models reproduce the simulated infall velocity across the entire range of infall velocities we tested . the rms systematic error in the twolayer6 model fits of hco@xmath1 ( ) is 0.03 km s@xmath0 , and 0.04 km s@xmath0 with respect to the hco@xmath1 ( ) fits . the hill5 model does a better job at estimating the infall velocity in the simulated clouds . the rms systematic error in the hill5 estimates of the infall velocity based on the hco@xmath1 ( ) profile fits is 0.03 km s@xmath0 , while the rms systematic error based on the hco@xmath1 ( ) profile fits is 0.01 km s@xmath0 . as we have seen in previous simulations the twolayer6 model yields a closed fit to the infall velocity at higher infall velocities . the hill5 model provides a fairly good fit for all infall velocities . these results are similar to the results obtained in simulation a ( constant infall ) , which is the other simulation in which infall was constant at all radii . this is reassuring as the central densities in these two simulations differ by a factor of 4 and they have different radial density profiles . this suggests that these analytic models may be able to predict infall velocities from optically thick line profiles for a wide range of densities and radial density profiles in star forming clouds . there are two components to the error in any fit of a model to data . the first , systematic error , is error which arise in the model s description of the data . it is due to the fact that the model may not correctly describe the nature of the observed region . the second component is random error , which arises from noise in the observation . although random error can be reduced over time or with better instrumentation , usually up to some limit , it is always a factor in any observation , and in fitting a model to those observations . in the preceding sections we examined how well our analytic radiative transfer models fit synthetic noiseless spectra . from this we found systematic errors due to the models shown in figures [ maf ] and [ mbf ] . in order to investigate the behavior of the fits in the presence of random error we decided to add noise to the spectra . @xcite have derived analytic expressions for the contribution of random error in a spectrum to the three parameters of a gaussian fit based upon the work of @xcite . they find that the random error in the line of sight velocity and in the velocity dispersion of a gaussian fit are the same and equal to @xmath71 where @xmath21 is the line of sight velocity of the gaussian , @xmath12 is the velocity dispersion of the gaussian , @xmath72 is the width of the spectral channels , @xmath73 is the peak intensity of the gaussian line profile , and @xmath74 is the rms noise level in the spectrum . unfortunately deriving expressions for the random error in the parameters of the two - layer or hill models is more difficult . we therefore use a bootstrap method to investigate the errors in the parameters of these models . we generate a noiseless synthetic spectrum from one of the above simulations , add random noise to it , and then perform the analytic radiative transfer model fits . we do this 100 times with different random noise of the same rms value in order to record the variance of each model parameter . assuming the errors are normally distributed , we find that the random error in each parameter is merely the square root of the variance in that parameter . we initially tested the method on gaussian line fits , whose random errors should follow the expression in equation [ gauss ] . figure [ gaussboot ] shows our resulting error determinations for different rms noise values ( filled circles ) versus the expected value ( solid line ) . we express the rms noise in terms of the signal to noise ratio on the lower axis , and also in terms of the integration time it would take to achieve a given rms for a telescope with a system temperature of 200 k and a main beam efficiency of 0.5 when observing a gaussian line with a peak brightness temperature of 5 k , along the top axis . we also use a channel width of 0.04 km s@xmath0 in this case , less than the input 0.1 km s@xmath0 velocity dispersion of the gaussian line and typical of current spectrometers . the bootstrap estimates are a good fit to equation [ gauss ] , with minimal scatter due to the random nature of the technique . this test indicates that the bootstrap method is a robust estimator of the random error in the parameters of the gaussian fit and should be a robust estimator of the random error in the parameters of the analytic radiative transfer model fits . we apply the bootstrap method to the hco@xmath1 ( ) line profiles of simulation a ( constant infall ) shown in figure [ majo ] . figure [ hillboot ] summarizes our findings for the random errors in the velocity dispersion , infall velocity , and line of sight velocity parameters of the hill5 model . the closed circles indicate the error in parameters as a function of integration time for a spectral resolution of 0.04 km s@xmath0 , while the open circles are those for a spectral resolution of 0.16 km s@xmath0 . the error increases with the lower spectral resolution because the self - absorption feature starts to be resolved out at that resolution . the cross in the center of each panel indicates the average systematic error obtained from the fits on figure [ maf ] . the average systematic error in the velocity dispersion is 0.008 km s@xmath0 , the average systematic error in the infall velocity is 0.009 km s@xmath0 , while the average systematic error in the line of sight velocity is 0.005 km s@xmath0 . the solid line in both the velocity dispersion panel and the line of sight velocity panel are the gaussian errors from figure [ gaussboot ] . unsurprisingly the errors in the hill model parameters are greater than those in the gaussian fit , which has fewer free parameters of the hill5 model , however they tend to follow a similar power law trend with increasing integration times . what we can see from this plot is that it is possible to be dominated by the systematic errors of the hill model after a reasonable amount of integration time on a typical millimeter radio telescope . for spectral resolutions adequate to observe the line shape , the velocity dispersion random error drops below the systematic level after less than 30 seconds of integration , while the line of sight velocity is well determined after less than two minutes . getting to a point where the systematic errors in the infall velocity dominate over the random errors takes the longest amount of time , but only about 10 minutes for a line whose brightness temperature is approximately 5 k. this means that generally the infall velocity is as well determined as it can get by the hill model when the signal to noise ratio is greater than @xmath75 . we assume , based on our experience with gaussian line fitting that maintaining a similar signal to noise ratio in lines of differing peak intensities should yield similar errors . this implies an integration time of 30 minutes for a line whose brightness temperature is approximately 3 k , and 250 minutes for a line whose brightness temperature is approximately 1 k. this means the observations required to obtain the required signal to noise can take a prohibitively long time for weak asymmetric line profiles . depletion in high density regions of a molecular cloud core is a significant effect in observed infall regions . many observations of co , cs , and hco@xmath1 in dense cores show a ring distribution thought to result from an abundance drop in the gas phase due to freeze out of these molecules onto grains @xcite . we have added depletion in some cases to perform an investigation of the efficacy of the analytic models at estimating the infall velocity as depletion increases . we adopt the formalism of @xcite and set the abundance of hco@xmath1 to vary exponentially with the density , @xmath76,\ ] ] where @xmath77 is the abundance of hco@xmath1 as a function of cloud radius , @xmath78 is the hco@xmath1 abundance in the low density limit ( @xmath47 ) , @xmath79 is the density of molecular hydrogen as a function of cloud radius , and @xmath80 is the density at which the abundance drops by a factor of @xmath81 . we included depletion in the simulation a ( constant infall ) case where @xmath82 km s@xmath0 for @xmath83 and @xmath84 @xmath44 . we found for the hill5 model the fit infall velocity changed by less that one hundredth of a kilometer per second down to an @xmath80 of @xmath85 @xmath44 for the hco@xmath1 ( ) line and @xmath86 for the hco@xmath1 ( ) line . figure [ depletion ] illustrates that in each of these cases that level of depletion corresponds to a change in the line profile from a two - peaked self - absorbed profile to a `` shoulder '' profile . severe depletion , beyond the levels reported in figure [ depletion ] tends to result in a line too weak to be observed . based on this simulation , we infer that moderate levels of depletion which does not render the line profile symmetric or unobservable has only a small effect on the hill5 model s infall velocity determinations . another important observational effect is smoothing by a telescope beam . our simulated line profiles were generated by integrating along a pencil beam through the center of a simulated cloud . we have generated 2 dimensional grids of pencil beam integrations of our simulated clouds and can convolve them to produce a simulated gaussian telescope beam . we begin by projecting the cloud on the sky plane at a distance of 140 pc , so that a beam of 60 is equal to a distance of about 0.04 pc in the cloud . we then convolved two cases from simulation a ( constant infall ) , where @xmath87 and @xmath88 km s@xmath0 , to generate beam sizes of 20 ( 0.014 pc ) , 40 ( 0.027 pc ) , and 60 ( 0.041 pc ) in the hco@xmath1 ( ) and h@xmath48co@xmath1 ( ) lines . the resulting beam - smoothed line profile seen toward the center of each simulated clouds is shown in figure [ beam ] . from the left column of figure [ beam ] we see that beam smoothing has a similar effect to depleting the core . this is not surprising as both depletion and beam smoothing tend to increase the relative contribution of emission from the envelope of the molecular cloud . our investigation indicates that the infall speed is well fit in cases where the beam width is less than or equal to the core diameter ( @xmath89 cm in our simulation ) , however this is dependent on both the density structure and velocity structure of the molecular cloud . in practical terms , a higher resolution probes a smaller region which may have a large infall velocity and therefore a more asymmetric line . as a final exercise we have applied the hill5 model to a high signal to noise line profile . we have learned from the simulations that the hill5 model is the most robust model for determining the infall velocity of a molecular cloud from the line profile for a majority of line profile shapes . in particular the hill5 model yields a good fit when the line profile has distinct blue - shifted and red - shifted peaks with a self - absorption minimum between them . for line profiles with this shape , in simulations a and d , which had the same infall speed at all radii , the rms error of the hill5 infall velocity determination is 0.01 km s@xmath0 , better than the overall rms error in these simulations of 0.02 km s@xmath0 . we recommend using the hill5 model to derive infall velocities from asymmetric molecular line profiles . in some cases there is little or no red - shifted peak , but merely a shoulder on the red side of the line profile . these cases tend to occur when the infall velocity is greater than the velocity dispersion . in simulation a ( constant infall ) and simulation c ( core infall ) red - shoulder profiles occur in the hco@xmath1 ( ) transition for simulated infall velocities greater than 0.1 km s@xmath0 , in simulation b ( envelope infall ) red - shoulder profiles occur in both the transitions for infall velocities greater than 0.1 km s@xmath0 , while in simulation d ( dense bonnor - ebert sphere ) red - shoulder profiles never occur for all the infall velocities in our simulations . in these `` shoulder '' spectra the twolayer6 model gives a comparable infall velocity estimate to the hill5 model . we note that in simulation c , where the twolayer6 model performs better than the hill5 model the red - shoulder model , lines tend to be weak ( @xmath90 k ) . when using the twolayer6 model it is important to choose the local minimum ( which should also be the global minimum in these cases ) that corresponds to the `` shoulder '' solution set . one can verify that the `` shoulder '' solution has been found by examining the best fit excitation temperature of the front layer , which should be greater than the background temperature . if the front layer excitation temperature is equal to the background temperature then it is likely that the one has found the `` dip '' solution which is usually never a good fit to the infall velocity . why is the hill5 model good at finding the infall velocity of most clouds , while the twolayer6 model is successful only on spectra with an asymmetric shoulder ? the key to understanding this lies in how well each model matches the excitation conditions along the line of sight at each doppler velocity . in figure [ excivtau ] we presented cases in which both the hill5 and twolayer6 models match the excitation profile of the simulation quite well for @xmath91 at the line center velocity . although the figure depicts only one doppler velocity , the match continues to hold over the entire line profile . the resulting infall velocities derived from these models closely match the infall velocities in the simulated contracting cloud . in cases where the infall velocities derived from the analytic models do not match the infall velocities in the simulated clouds , the modeled excitation temperature profiles as a function of optical depth along the line of sight derived from the spectral line fits do not match the simulated excitation temperature profiles . even in cases where the spectral line match is quite good , if the excitation conditions are not well matched then all the physical parameters , including the infall velocity , will not match the parameters of the simulated or observed cloud . we find that the hill5 model tends to reproduce the excitation temperature profile as a function of optical depth quite well for a wide range of infall velocities . the reason that this is true is that the excitation conditions in a cloud do tend to rise as a function of optical depth over the first couple of optical depths , until the excitation temperature reaches the kinetic temperature of the core or begins dropping again as the density drops . in all of our simulations a linear fit to the rising excitation temperature , starting with an excitation temperature equal to the background temperature , is a good approximation of the actual excitation temperature profile at all doppler velocities . as a result the hill5 model is usually sensitive to the physical conditions of the collapsing cloud and yields a good estimate of the infall velocity if the cloud radii which comprise the first few optical depths of the line profile are infalling . for this reason we recommend using the hill5 model . our simulations and error analysis suggest that the line profile should have a peak brightness temperature greater than 1 k and a signal to noise ratio greater than 30 . the line profile must also have a well defined dip or shoulder which must be resolved by the spectrometer resolution . if these conditions are met the hill5 model can be expected to provide an estimate of the infall velocity with an accuracy of 0.02 km s@xmath0 . our simulations suggest the accuracy for dip profiles alone is 0.01 km s@xmath0 . the twolayer6 model tends to reproduce the excitation temperature profile as a function of optical depth well only for higher infall velocities which tend to produce spectral lines that are very asymmetric or have well defined red shoulders . in these cases it may be tempting to use the twolayer6 method to derive the infall velocity , however there are several caveats which one must take in to account . in cases with low infall velocities or more symmetric line profiles the excitation temperature of the front layer is often low ( at or near the cosmic microwave background temperature ) while that of the rear layers is often much higher than the maximum excitation temperature in the cloud , even in cases where the twolayer6 model produces a good fit to the emerging line profiles . in these cases the infall velocities as other physical conditions derived from the twolayer6 model do not match the conditions in the simulated cloud . the reason this occurs is that there is a certain amount of degeneracy in reproducing a symmetric or nearly symmetric line profile with the twolayer6 model . a very high excitation temperature region behind a very low excitation temperature region produces the same nearly symmetric profile as a moderately high excitation temperature region behind a moderately low excitation temperature region . at higher infall velocities , when the line profile is very asymmetric or has only a shoulder , the excitation temperature of the front region can be accurately determined from the red - shifted emission of that line profile . this breaks the degeneracy and creates an excitation temperature profile along the line of sight that more closely matches the simulation and is more well determined overall . as a result the physical parameters of the simulation , including the infall velocity , tend to be well determined by the twolayer6 model in these cases . in figure [ data ] we present a high signal to noise line profiles upon which we will carry out the procedure prescribed above . the line profile is the isolated component of an n@xmath54h@xmath1 ( ) line profile in l1544 observed by bourke et al . ( in preparation ) . based on our results in the previous sections we choose to model the emission using the hill5 model . that best fit is shown by the points in figure [ data ] . we see that the fit is fairly good , though it deviates from the line profile by producing a second peak instead of a shoulder . the hill5 parameters that result in this fit are shown in table [ datahillfit ] . in this case our hill5 infall estimate of 0.099 km s@xmath0 matches well with the best fit twolayer6 estimate of 0.095 km s@xmath0 . unfortunately since we did not create this profile from a model we do not know the actual kinematic conditions in the cloud , however there are two other estimates obtained using detailed radiative transfer methods . @xcite have fit a model to n@xmath54h@xmath1 ( ) and n@xmath54h@xmath1 ( ) spectra at 4 positions around l1544 . they find an infall speed profile that accelerates inwards from the cloud edge to 0.24 km s@xmath0 . although this peak infall speed is much more than our estimate , we note that their fits to l1544 include no velocity dispersion . we speculate that the infall speed may be artificially high in order to match the line width in the absence of any turbulent velocity component . bourke et al . ( in preparation ) have observed this particular source in several molecular line transitions , include n@xmath54h@xmath1 ( ) , n@xmath54h@xmath1 ( ) , dco@xmath1 ( ) , dco@xmath1 ( ) , cs ( ) , cs ( ) , cs ( ) , and c@xmath92s ( ) . they have used the density model of this cloud proposed by @xcite and assumed infall at the same velocity over the entire cloud along with a kinetic temperature of 10 k to create radiative transfer models of l1544 and match their observed line profiles . a simultaneous match of all the components in the n@xmath54h@xmath1 spectral lines currently yields an infall velocity of 0.1 km s@xmath0 , which is in agreement with our estimate to within a hundredth of a kilometer per second . lllll 3.26 & 7.183 & 0.099 & 0.070 & 6.14 + we have shown that analytic radiative transfer models are effective at estimating the infall velocities of contracting clouds from the blue - asymmetric line profiles of species of moderate optical depth in the region of collapse . the models provide good velocity estimates with few assumptions . the greatest advantage of the analytic models is that they allow for a quick and reasonable estimate of infall velocity with a minimum of computational effort . in cases where nothing else is known about the source , they provide the only possible estimate of infall , and in regions where a more robust model can be constructed , the analytic models provide an excellent starting point for further modeling . ] the main conclusions of this work are as follows : 1 . analytic radiative transfer models can reproduce the blue - asymmetric line profiles seen in infalling starless cores . in addition to the two - layer model first presented in @xcite we have found a second class of radiative transfer models ( the hill models ) and explored 5 variants of these two classes of analytic radiative transfer models . all the model variations were able to produce blue - asymmetric line profiles similar to those we observe in starless cores . successful analytic radiative transfer models reproduce not only the line profile shape , but also match the initial excitation conditions along the line of sight up to or beyond an optical depth of 1 . when this requirement is met , the modeled infall velocity successfully fits the simulated infall velocity . those models which match the line profile shape , but do not match the line of sight excitation conditions can not reveal the infall velocity and velocity dispersion in the cloud . 3 . the hill5 analytic radiative transfer model is the preferred model when fitting a blue - asymmetric line profile . in simulations with constant velocity as at all cloud radii the infall velocities attained by fitting the hill5 analytic radiative transfer model to the simulated line profiles match the infall speed with an rms error 0.02 km s@xmath0 . the hill5 model tends to perform especially well when there are two distinct peaks such that the intensity difference between the self - absorption trough and the redshifted peak is greater than 10% of the intensity of the blueshifted peak . the rms error in the infall determination of hill5 fits to simulated spectral lines with those characteristics from simulations a and d is 0.01 km s@xmath0 . the physical parameters of simulated clouds that produce blue - asymmetric line profiles with no distinct shoulder or redshifted peak are generally not well determined by any of the model variations we have studied in this paper . we recommend not using these models to estimate infall velocities from line profiles that do not have a separate redshifted peak or a distinct redshifted shoulder . a better match to the line profile shape does not necessarily imply a better match to the excitation profile or the physical parameters , such as the infall velocity , of the observed cloud . this can be seen most clearly in figure [ dipshoulder ] where local minimum of the `` dip '' solution often provides a lower @xmath56 fit to the line profile shape , but a poorer fit to the infall velocity than the `` shoulder '' solution . following the instructions in [ method ] when fitting a blue - asymmetric line profile will minimize the risk of obtaining an unusually poor estimate of the physical parameters from the analytic radiative transfer model fit to the line profile all the analytic models lose sensitivity to the infall velocity if there is a significantly optically thick static envelope between the observer and the infall region . in these cases a more optically thin line can provide a better estimate to the infall velocity in the obscured region . conversely an optically thin tracer may underestimate the infall velocity of an infalling envelope around a static cloud core . these methods are sensitive to the velocities and conditions in the regions in which the line is becoming optically thick . a peak signal to noise ratio greater than or equal to 30 is required for the systematic error on the infall velocity estimate to be greater than the random error . we recommend achieving a peak signal to noise ratio of 30 in order to minimize the random error of a line profile fit using our analytic radiative transfer models , however nothing is gained by achieving a higher signal to noise level . achieving this signal to noise level can take a prohibitively long time for lines whose peak antenna temperature is less than or equal to 1 k. 8 . depletion and beam smoothing can greatly affect the observed line profile in infalling starless cores . we conclude that , even in the presence of these effects , if the line profile is asymmetric and a good candidate to be fit using these analytic radiative transfer models ( as described in [ method ] ) , then generally the radiative transfer models do yield an accurate estimate of the infall velocity . if the cloud is depleted to such an extent or viewed with such a large beam that the line profile becomes very weak or symmetric , then these models will not accurately probe the infall region of the starless core . we are grateful to tyler bourke for contributing his l1544 data and some preliminary results from his radiative transfer modeling for use in this paper . we also acknowledge henrik beuther for his encouraging advice and an anonymous referee for his or her valuable comments and suggestions . this research was conducted with support from nasa origins of solar systems grant nag5 - 13050 .

models of the continuum radiation from accreting hot plasmas typically assume that the plasma heating mechanism produces energetic particles distributed in energy either as a maxwellian ( the `` thermal '' models ) or as an extended power law ( the `` non - thermal '' models ) . the reality , however , is that neither description is probably accurate . in other astrophysical contexts where we have been able to observe the actual particle energy distributions , e.g. solar system plasmas , and in many particle acceleration theories , the heating mechanism supplies only some fraction of the available energy to very energetic particles . the remainder goes into producing lower energy particles which settle into a quasi - maxwellian energy distribution . here , i review the arguments for `` thermal '' versus `` non - thermal '' plasmas in accreting black hole systems and discuss the physics and emission properties of `` hybrid '' plasmas , where the particle distribution energy is approximately a maxwellian plus a power law tail . using results from a new emission code , i then show that such plasmas may be relevant to explaining recent observations , particularly those of galactic black hole candidates in their soft state . = = = 1=1=0pt = 2=2=0pt = 2=2=0pt

astro-ph9903158

most attempts at modeling the emission from accreting black hole systems typically assume that the particle energy distribution responsible for the emission is either purely `` thermal '' ( i.e. , a maxwellian ) or purely `` non - thermal '' ( i.e. , a power law ) . this is partly due to the desire for convenience and simplicity , and partly due to our continued ignorance as to when and how non - thermal particles exchange energy and thermalize with neighboring particles . despite our ignorance , however , the reality is that nature probably never makes particle distributions that are strictly of one type or the other . as a concrete example which may be very relevant to the process of black hole accretion , consider solar flares . ( many analogies have been made between magnetic reconnection and flare events in the solar corona , and those that might occur in accretion disk coronae , e.g. , galeev , rosner , & vaiana 1979 . ) the watch instrument on the granat satellite has recently collected a large sample of flare events observed at energies above @xmath0 kev ( crosby et al . 1998 ) . many of those observations were made in coincidence with the goes satellite which measures solar flare at lower @xmath1 kev . while the goes observations usually can be adequately interpreted in terms of purely thermal emission , in many instances watch sees a significant high - energy excess over the emission predicted from the goes spectral model . in other words , many solar flare events are produced by particle distributions with not only a strong thermal component , but also a significant non - thermal tail : the emitting plasma is a `` hybrid '' thermal / non - thermal plasma . if the solar corona - accretion disk corona analogy is correct , then something similar may well occur near black holes and could have important consequences for black hole emission models . the goals of this contribution are to review the arguments for and against the existence of similar particle distributions in the context of accreting black hole systems , to see what the practical consequences of having a hybrid plasma might be , and then finally to ask whether we have any concrete concrete evidence that such plasmas play a role in the observed emission . in 2 below , i examine the theoretical arguments for and against having a purely thermal electron distribution in a black hole accretion disk corona . in 3 , i examine some of the consequences of having a hybrid plasma and present model spectra calculated assuming varying amounts of thermal heating and non - thermal acceleration power are supplied to the plasma . in 4 , i argue that we may already have strong evidence for hybrid plasmas in the spectra of galactic black hole candidates , especially in spectra obtained during their soft state . i show how a simple phenomenological model based on a hybrid plasma can explain the spectra of cyg x-1 in its soft , hard , and transitional states in terms of one basic parameter , a critical radius outside of which the accretion disk is a cold sunyaev - shakura ( 1973 ) disk , and inside of which the disk is a hot , adaf - like ( narayan & yi 1995 ) corona . i summarize my conclusions concerning the importance and relevance of hybrid plasmas in 5 . consider a relativistic electron that finds itself in the middle of a low - temperature , thermal bath of background electrons . will the electron couple to the thermal electrons and share its energy with them before it does something else interesting , e.g. , radiate its energy away ? if it does , then it is valid to treat that electron and its energy as belonging to the thermal pool of electrons . before anything else happens , the electron will have exchanged its energy many times with other electrons , which is the requirement for setting up a maxwellian ( `` thermal '' ) energy distribution . hence , when considering processes that occur on time scales longer than this `` thermalization '' ( energy exchange ) time scale , one is justified in assuming that the electron energy distribution at energy @xmath2 is indeed described by a relativistic maxwellian , @xmath3 , where @xmath4 is the electron velocity , and @xmath5 is the temperature of the thermal bath electrons . if it does not , then for practical purposes , e.g. , when computing the emitted radiation spectrum , that electron is decoupled from the thermal distribution . assuming that the distribution at this energy is a maxwellian could be quite dangerous . the question of whether a plasma is best described as thermal or `` hybrid '' ( thermal with a significant non - maxwellian component ) thus comes down to how rapid the thermalization process is for energetic electrons . unfortunately , the details of thermalization and energy exchange in plasmas are in general poorly understood . witness the current controversy ( see the contribution of quataert , these proceedings ) on whether accretion disks can support proton and electron energy distributions with different temperatures ( one of the key assumptions behind adaf disk models , e.g. , narayan & yi 1995 ) . we know of at least one process that will exchange energy between protons and electrons ( and similarly between electrons and electrons ) , namely coulomb collisions / scatterings where one particle is deflected by the electric field of the other particle . under the conditions supposed to exist in an adaf ( advection dominated accretion flow ) , the rate of energy transfer due to coulomb collision is not rapid enough and if no other process intervenes , the electrons will lose most of their energy before sharing it with the protons , i.e. , the protons and electron distributions will not have the same temperature . but might another energy exchange process be important ? possibly . the electromagnetic force is a long - range one , and plasmas have so - called `` collective modes '' ( involving the coherent interactions of many particles rather than the simple interaction between two particles that occurs in a coulomb scattering ) . under certain conditions , these modes can efficiently couple electrons and/or protons of different energies even when coulomb scattering ( `` two body relaxation '' ) is not important . for an astrophysical example of such collective effects applied to the proton - electron coupling problem , see begelman and chiueh ( 1988 ) . ( see also tajima & shibata 1997 and references contained therein for a recent overview of collective and other processes in astrophysical plasmas . ) in solar system plasmas observed directly by spacecraft , we know that some effect like this must be operating since quasi - maxwellian electron energy distributions are observed even though the coulomb energy exchange time scales are enormous ( since the plasma densities are so low ) . similar processes might well be operating in a black hole accretion disk or in a corona above the disk . however , realistic calculations of collective processes are notoriously difficult from first principles , especially near an accreting black hole where we still have a relatively poor understanding of the exact physical conditions . besides collective processes , we note one other way for electrons to exchange energy , namely via the photon field . if the source is optically thick to photons emitted by an electron , the absorption of those photons by different electrons will effectively transfer energy between the electrons and produce a quasi - maxwellian distribution . this is the basic principle behind the `` synchrotron boiler '' of ghisellini , guilbert , & svensson ( 1988 ) ( see also ghisellini , haardt , & svensson 1998 ; svensson , these proceedings ) , which can lead to hot , quasi - thermal electrons distributions . if a thermal electron distribution near a black hole can be maintained only by coulomb collisions , however , we run into a situation similar to that of the inadequate electron - proton coupling in adafs . in particular , the photon energy density near an efficiently accreting black hole is so high that energetic electrons cool very rapidly due to compton scattering . to see this , let us make the standard first - order assumptions that the high - energy emission region near a black hole is roughly spherical with a characteristic radius or size @xmath6 and that inside this region , the photon and particle distributions are uniform and isotropic . for simplicity , let us also assume that the characteristic energy of photons in the source is significantly less than @xmath7 ( @xmath8 is the electron mass , and @xmath9 is the speed of light ) so that klein - nishina corrections are not important ( which is roughly true for radio quiet agn and galactic black hole candidate systems ) . in this case , the compton cooling rate for an electron of lorentz factor @xmath10 is then roughly @xmath11 where @xmath12 is the thomson cross - section , and @xmath13 is the characteristic radiation energy density inside the source . to estimate @xmath14 we note that if the source is optically thin , emitted photons will escape the source on a time scale @xmath15 the characteristic source light - crossing time . the total luminosity of the source is then @xmath16 which , for a spherical source , gives us @xmath17 a convenient way to work with the source luminosity is to re - express it in terms of the dimensionless `` compactness '' parameter , @xmath18 in terms of this parameter , the cooling rate is then simply @xmath19 which gives a characteristic compton cooling time @xmath20 using the total source luminosity and a lower limit on source size from variability consideration ( @xmath21 ) , many accreting black hole systems are inferred to have compactnesses @xmath22 significantly greater than one ( e.g. , see done & fabian 1989 for a compilation of agn compactnesses ; for the galactic black hole candidate , cyg x-1 , the best fit compactness seems to be @xmath23 , e.g. , see 4 below ) . in other words , one does not have to go to a very high lorentz factor before @xmath24 i.e. , before compton cooling becomes very rapid . note that if the source is optically thick , then the time it takes photons to leave the source is longer , @xmath13 is correspondingly higher , and the compton cooling time for an energetic electron is even shorter . to see if this rapid cooling inhibits electron thermalization , we must compare this cooling time scale with the corresponding time scale for an energetic electron to transfer its energy to a thermal bath of lower energy electrons via coulomb collisions . for purposes of making a rough estimate , we will assume that the lorentz factor of the electron is @xmath25 and the thermal electrons are relatively cold ( @xmath26 ) . then the coulomb `` cooling '' rate of the energetic electron is simply given by ( e.g. , see discussion and references in 5 of coppi & blandford 1990 , also dermer & liang 1989 ) , @xmath27 where @xmath28 is the number density of thermal bath electrons and @xmath29 is the usual coulomb logarithm , a weak function of the plasma parameters with typical value @xmath30 for accreting black hole sources . in terms of the characteristic thomson ( electron scattering ) optical depth of the source , @xmath31 this gives us a coulomb energy exchange time scale @xmath32 which grows with increasing electron energy and eventually becomes longer than the compton cooling time ( which decreases with energy ) . setting @xmath33 we obtain the lorentz factor above which electrons cool before thermalizing : @xmath34 for typical values @xmath35 and @xmath36 and @xmath37 we then have @xmath38 i.e. , if coulomb collisions are the only energy exchange mechanism , thermalization very quickly becomes ineffective and maintaining a thermal , maxwellian energy distribution with relativistic temperature @xmath39 is impossible . ( see ghisellini , haardt , & fabian 1993 for more discussion along these lines . they also point that @xmath40 can significantly exceed the variability time scale @xmath41 again calling into question validity of assuming a single temperature thermal distribution . ) for significant compactnesses @xmath42 note also that the synchrotron boiler mechanism is also quenched by the rapid compton cooling ( coppi 1992 ) unless the source is strongly magnetically dominated , i.e. , the magnetic energy density @xmath43 . this condition might hold if the particle acceleration is due to magnetic flares in an active corona , but this is far from clear yet . if , as in cyg x-1 , we observe photons with energies @xmath44 ( 511 kev ) , one must then seriously consider the possibility that they are produced by electrons in the non - thermal tail of a non - relativistic quasi - maxwellian distribution . one loophole in this argument is that if a plasma is very photon starved ( e.g. , see zdziarski , coppi & lightman 1990 ) , i.e. , if the power supplied to the electrons is much larger than the power supplied to the background photon field , most of @xmath45 will emerge at energies @xmath46 in this case , klein - nishina corrections can dramatically decrease the effectiveness of the compton cooling and thermalization might be effective to much higher energies . the typical agn and galactic black hole systems known to date appear not to be so photon - starved , however . remember also that the compton cooling rate depends critically on the source luminosity . if the accretion disk radiates very inefficiently as in the adaf model proposed for the galactic center ( and the black holes at the centers of some elliptical galaxies ) , then @xmath22 can be quite small and thermalization can be correspondingly effective . for a more detailed discussion of the electron energy distribution in the context of adafs , see mahadevan & quataert ( 1997 ) . they note that while the bulk of the electron population may thermalize ( especially at low accretion rates ) , a significant non - thermal , high - energy tail is still likely to be present due to synchrotron cooling effects . compressional heating effects may also significantly distort the shape of the electron distribution . the preceding argument and the current lack of a thermalization mechanism that clearly operates faster than coulomb collisions are not the only reasons to consider a `` hybrid '' electron energy distribution , where only the lower energy end is well - described by a maxwellian . observationally , for example , the electron energy distributions in solar system space plasmas are never exactly described by maxwellians and often show significant non - thermal tails . in fact , they sometimes show an energy distribution with two ( ! ) quasi - maxwellian peaks at different energies , i.e. , as if the energy exchange processes are effective only at coupling electrons of similar energy . also , whenever we see dissipative events , such as solar flares or reconnection events in the earth s magnetotail or shocks in the solar wind , we often see direct evidence of significant particle acceleration accompanied by non - thermal " emission . this particle acceleration , however , is almost never 100% efficient in the sense that all the dissipated power ends up in high energy ( non - thermal ) particles ( e.g. , see discussion in zdziarski , lightman , & maciolek - niedzwiecki 1993 ) . many theoretical particle acceleration models have an injection `` problem '' or condition , depending on one s point of view , where only particles above some threshold momentum , for example , are accelerated to very high energies ( e.g. , see benka & holman 1994 for a simple model of hard x - ray bursts in solar flares , or blandford & eichler 1987 for a discussion in the context of shock acceleration theories ) . only above some energy does the relevant acceleration time scale for a particle become significantly shorter than its corresponding thermalization / energy exchange time scale for that particle . particles below this energy are tightly coupled and their acceleration simply results in bulk heating of the thermal plasma component i.e. , one creates a hot , hybrid plasma . in accreting black hole systems , the problem of determining exactly what electron energy distribution results from dissipation and acceleration is further complicated by the fact that radiative cooling can be strong and that the dissipation process is highly variable ( e.g. , the `` steady state '' spectrum of cyg x-1 that one fits may really be the superposition of many shot - like events with time - varying spectra , e.g. , as explicitly illustrated by poutanen & fabian in these proceedings ) . the most heroic effort to date in terms of incorporating a particle acceleration scenario into a time - dependent code and then actually computing a spectrum is probably that of li , kusunose , & liang ( 1996 ) . see the review by li ( these proceedings ) for some of the more recent results obtained with this code as well as a detailed discussion of the particle acceleration problem . the physics incorporated into this code may ultimately turn out to be overly simple or simply not relevant ( given our large ignorance in this problem ) , but the code nonetheless provides an explicit demonstration of how a `` hybrid '' energy distribution might arise . one final complication is that the dissipation rate and non - thermal acceleration efficiency may vary considerably inside the source ( as observations by the yokoh satellite have shown to indeed be the case for solar flares ) . if the source is only moderately optically thin ( @xmath35 ) , escaping photons can traverse the source and sample several regions with different acceleration efficiencies . the effective electron distribution required to produce the observed emission will then be a spatial average over the source and may not be predictable by any ( ! ) standard one - zone particle acceleration calculation . ( the same is true if one averages the emission over time the effective electron distribution is then some time average and depends critically on the variability properties of the dissipation regions . ) the bottom line is that spectral modelers should not be surprised to find that their data requires a somewhat `` strange '' electron energy distribution . given the increasing quality of data ( uncertainties in 2 - 10 kev x - ray spectra are now down to the few percent level and are often dominated by systematics not statistics ) , it is frankly amazing that the simple models employed to date have worked as well as they have . ( in the next section , though , i will show why this may not be quite as surprising as one might first think . ) hybrid emission models have been considered for some time in the context of solar flares , e.g. , see benka & holman ( 1994 ) , and more recently benz & krucker ( 1999 ) . as the relevant processes and parameters in these plasmas are quite different from those near accreting black holes ( e.g. , the ambient radiation energy density is much higher in the black hole case ) , i will not consider these models further here . to make the discussion more concrete , i will show results from one specific incarnation of a plasma model intended for black hole applications , namely a plasma code called eqpair ( coppi , madejski , & zdziarski 1999 ) that we have recently developed . the code is an extension of the code of coppi ( 1992 ) , which makes no intrinsic assumptions about the electron spectrum other than the lowest energy electrons are in a quasi - maxwellian distribution . in terms of the basic physics considered , the code is essentially the same as other codes that have been recently applied to study hybrid plasma emission : zdziarski , lightman , & maciolek - niedzwiecki ( 1993 ) , ghisellini , haardt & fabian ( 1993 ) , li , kusunose , & liang ( 1996 ) , and ghisellini , haardt , & svensson ( 1998 ) . the source geometry assumed in these models is either a disk - corona ( slab ) geometry ( ghisellini , haardt & svensson model ) , or a spherical geometry ( other models ) . low energy ( uv or x - ray ) thermal photons from the accretion disk are assumed to be emitted uniformly inside the source region for the spherical models and enter from the base of the corona in the disk - corona model . the total luminosity of these `` soft '' photons is parameterized by a compactness parameter @xmath47 analogous to @xmath22 ( although remember that @xmath22 is defined in terms of the _ total _ photon luminosity escaping the source ) . the spectrum of these soft photons is typically assumed to be a blackbody or disk blackbody ( mitsuda et al . 1984 ) with characteristic temperature @xmath48 the main practical difference of the two geometry assumptions is that spherical models do not show the strong `` anisotropy break '' ( e.g. , stern et al . 1995 ) in the 2 - 10 kev region that the disk - corona models do . to date , strong evidence for such a break has not been seen . in the li et al . model , a turbulent wave spectrum is then specified and electrons from a cool background thermal plasma with thomson optical depth @xmath49 are accelerated by the waves to form the non - thermal tail . the total luminosity supplied to the waves and eventually the electrons is specified by the compactness parameter @xmath50 ( @xmath51 where @xmath52 is the wave luminosity ) . in the other models , the exact acceleration and heating mechanism for the electrons is left unspecified . rather , electrons and/or positrons are injected into the source with some fixed ( non - thermal ) energy spectrum @xmath53 that does not need depend on conditions in the plasma . the typical assumptions are that @xmath53 is either a power law extending from lorentz factor @xmath54 to lorentz factor @xmath55 a delta function at lorentz factor @xmath56 or an exponentially truncated power law at some energy @xmath57 the total luminosity supplied to the source via these injected non - thermal pairs is parameterized by a compactness parameter @xmath58 to mimic , for example , an acceleration mechanism with less than 100% non - thermal efficiency , additional power , parametrized by the compactness parameter @xmath59 is supplied to the electrons ( or pairs ) once they have thermalized . background electrons which participate in the thermal pool may be present , and in eqpair , pair balance is not automatically assumed . ( in electron - positron pair plasmas , pair balance means that the total pair annihilation rate in the plasma equals the pair creation rate and is the standard condition used to set the equilibrium density in the source , e.g. , see the discussion in svensson 1984 . ) hot electrons and/or positrons are allowed to cool via compton scattering , synchrotron radiation ( so far included only in the ghisellini et al . 1998 code , which , however , does not include pair processes ) , coulomb energy exchange with colder thermal pairs , and bremsstrahlung emission . the energetic photons created by these processes either escape the source , produce pairs on lower energy photons , compton scatter off more electrons , or are synchrotron self - absorbed . pairs that are injected or created in the plasma annihilate away once they have cooled . any `` excess '' electrons that were injected ( without a positron partner ) are assumed to be removed from the system once they thermalize , e.g. , via reacceleration , so that no net particles are added to the system except via pair creation . in steady state , the escaping photon luminosity must equal the sum of the various input luminosities , i.e. , @xmath60 where for the codes with no acceleration prescription , @xmath61 to summarize then , the main parameters of a hybrid model like eqpair are : ( i ) @xmath62 ( ii ) @xmath63 ( iii ) @xmath64 ( iv ) the source radius @xmath65 ( which enters into some of the rate coefficients and relates the compactness parameters to absolute source luminosities ) , ( v ) @xmath66 the optical depth of the background electron - proton plasma , ( vi ) the characteristic soft photon energy as specified by @xmath67 and ( vii ) the non - thermal electron / pair injection spectrum @xmath68 the usual spectral modeling parameters of comptonization models , e.g. , temperature of the electron distribution , @xmath69 and the total thomson scattering optical @xmath70 are all computed self - consistently . the main advantages of eqpair over other the codes is that : ( i ) all the microphysics is treated self - consistently without significant approximations ( in particular all klein - nishina corrections are included , which is crucial ) , and ( ii ) it is still fast enough to use for real data fitting ( one model iteration takes @xmath71 sec on a 300 mhz pentium ii ) . the code has been ported to xspec and incorporates ionized compton reflection ( as in the pexriv xspec model ) including smearing due to relativistic motion in the disk ( as in the diskline xspec model ) . ( note that in the current version , the reflected radiation is assumed _ not _ to pass back through the emission region as it in fact might in a slab - like disk - corona geometry such as that of haardt 1993 . ) the code will be available for general use once the description / user s manual paper is ( finally ) submitted . one drawback of eqpair and the other codes ( except for ghisellini et al . 1998 who calculate the compton upscattered spectrum from the corona by solving the radiative transfer equation in a slab geometry ) is the use of an escape probability to handle the radiation transfer . this is a potentially serious limitation when the optical depth @xmath35 and the electron temperature is high , @xmath72 kev . however , as noted in coppi ( 1992 ) , the errors are typically less than the uncertainties introduced by our ignorance of the exact source geometry , e.g. , are the soft photons actually emitted in the center of the corona or do they enter the corona from some outer , cool region of the disk ( as in poutanen , krolik , & ryde 1997 ) ? figure 7 below shows an explicit comparison of the eqpair output versus the output from a monte carlo simulation with three different soft photon injection scenarios . the optical depth in that calculation is low ( @xmath73 ) but the temperature is rather high ( @xmath74 kev ) , and the agreement is quite reasonable . one final caveat on using eqpair to fit observed spectra is that the spectrum it produces is of course a steady - state one , while the real spectra that are being fit are typically time integrations over many flares . to illustrate the effects of a hybrid electron distribution ( or , equivalently , simultaneously accelerating non - thermal particles and heating thermal particles ) , let us first consider the transition from a purely `` non - thermal '' plasma ( @xmath75 ) to a mainly thermal one with @xmath76 . this is shown in figure 1 where all plasma parameters are kept fixed except @xmath77 the initial plasma configuration has @xmath78 which means that compton cooling of pairs is strong and pair production of gamma - rays on x - rays is moderately important . ( the compactness parameter not only measures the effectiveness of compton cooling , but also the optical depth to photon - photon pair production in the source , e.g. , see guilbert , fabian & rees 1983 . ) because the non - thermal electrons were `` injected '' with a fixed lorentz factor @xmath79 the cooled electron distribution should have been @xmath80 which should have given a power law photon energy distribution with @xmath81 . the deviations from this power law are the result of photon - photon pair production which removes some of the highest energy photons and adds lower energy pairs to the electron distribution ( see , e.g. , svensson 1987 for a discussion of how pair `` cascading '' transforms spectra ) . also visible is a pair annihilation feature at @xmath1 511 kev caused by the annihilation of cooled , essentially thermal pairs . because compton cooling is so rapid , the pairs do not annihilate or thermalize until they have already lost most of their energy . ( the temperature of the cool , thermalized pairs is only @xmath82 10 kev . ) hence , the annihilation feature is narrow and has the shape expected from the annihilation of pairs with a thermal distribution . note an interesting effect . the initially non - thermal plasma has , in fact , already turned itself into a hybrid thermal / non - thermal plasma . depending on the exact plasma parameters , compton upscattering of the soft photons by the thermal component can be quite important and will produce a soft x - ray excess ( see zdziarski & coppi 1991 ) without having to invoke any additional emission component . in figure 1 , we see this soft excess emerging and becoming increasingly visible as we increase @xmath83 and make the cooled thermalized pairs hotter . note that as the soft x - ray excess increases , the pair production optical depth for gamma - rays increases correspondingly , and the flux above 511 kev drops . eventually as we keep increasing @xmath59 by about @xmath84 or @xmath85 the soft `` excess '' is no longer really an excess but in fact dominates the entire x - ray spectrum . the spectrum then is essentially that of a thermal plasma except for a pronounced gamma - ray excess and a hint of a broad annihilation line above @xmath1 100 kev . note the behavior of the annihilation line . even though the importance of pair production continually increases with @xmath83 and the annihilation flux is actually always growing , the annihilation feature eventually disappears as it is broadened and downscattered ( e.g. , see maciolek - niedzwiecki , zdziarski , & coppi 1995 ) . by @xmath86 the spectrum is very close to that expected from a purely thermal plasma although a detector with sufficient sensitivity and energy resolution above @xmath87 kev would still find an excess relative to the purely thermal model since the hybrid gamma - ray does _ not _ fall off exponentially . this gamma - ray excess will be the key in understanding the effects of having a hybrid plasma in pair balance ( see below ) . in sum , as figure 1 shows , the most unambiguous signature of a hybrid plasma would be the presence of excess emission above @xmath88 kev . unfortunately , if the hybrid plasma is thermally dominated ( @xmath89 ) , hot , and moderately optically thick ( @xmath90 and @xmath91 kev for the @xmath92 model ) , it becomes very hard to distinguish spectroscopically from a purely thermal one except at the highest energies @xmath93 kev . any firm conclusions , for example , on the nature of the plasma in cyg x-1 s hard state ( which is likely to be thermally dominated , hot , and moderately optically thick ) will have to await better detectors in this energy range like astro - e and integral , although we note that batse and comptel may already have detected such an excess ( see mcconnell et al . 1994 , and ling et al . 1997 ) . the question of how much the details of the electron energy distribution matter is unfortunately somewhat more complicated than the preceding example would suggest . the real test seems to be whether or not multiple compton scattering is important . in a typical thermal model with @xmath94 the fractional energy shift per scattering is @xmath95 in order to cool the thermal electrons and boost the input photons to sufficiently high energies that the escaping photon luminosity equals the total input luminosity , a photon must scatter many times off the electrons before it escapes . this is particularly true if @xmath96 and a significant energy boost is required to satisfy the energy conservation requirement @xmath97 . the final emergent spectrum in a thermal model is thus typically composed of many so - called `` orders '' of compton scattering . ( each order of compton scattering is calculated by computing the compton scattered photon assuming the previous order as the input photon spectrum . the initial soft photon input spectrum is the zeroth order . ) since compton scattering by hot electrons smears out spectral features ( an input photon with a given initial energy can be scattered to a range of final energies ) , the spectrum of each successive compton scattering order tends to appear smoother . the end result is that any spectral features in the first order of compton scattering ( e.g. , due to the choice of electron energy distribution ) tend to be washed out and the composite emergent spectrum is usually a rather featureless power law . as shown , e.g. , in rybicki & lightman ( 1979 ) , the slope of this power law can be derived by basically knowing only the mean photon energy change per scattering and the mean number of scatterings a photon undergoes before escaping ( i.e. , the compton @xmath98 parameter ) . if one replaces the thermal electron distribution by another one that gives the same mean photon energy shift per scattering and also insures that the thomson optical depth of the source remains constant , then to first order , nothing changes in the preceding chain of reasoning and the emergent spectrum will be the same ( ! ) . this was noted by zdziarski , coppi , & lightman ( 1990 ) in the context of photon - starved plasmas and plasmas with very steep non - thermal injection ( @xmath99 with @xmath100 ) extending to a @xmath101 close to unity , and by ghisellini , haardt , & fabian ( 1993 ) who showed that the non - thermal comptonization spectra produced in plasmas where @xmath53 goes to zero for @xmath102 are very close to thermal ones where the mean energy per scattering is the same . in other words , _ as long as most of the electrons in the source are low energy and multiple orders of compton scattering are important , it makes little difference what energy distribution the electrons have . _ if non - thermal electron acceleration near the black hole holes is not very effective , i.e. , if electrons never reach very high energies ( perhaps because the radiative cooling times are so short ) , this might help explain why objects like cyg x-1 have thermal - looking spectra in their hard state . it also explains why different hybrid plasma codes can use rather different criteria for deciding when exactly an electron has thermalized and still end up predicting similar emergent spectra . as a further illustration of how spectra from different electron distributions can be quite similar if multiple scattering is important , we show in figure 2 the spectra produced by a strictly thermal plasma , by a hybrid plasma where the non - thermal electron injection function is a delta function at @xmath103 and by a hybrid plasma where @xmath104 for all three plasmas , @xmath105 i.e. , the plasma is photon - starved and multiple scattering is important . the spectra from the thermal plasma and the hybrid , low @xmath106 plasma are rather similar ( particularly in the 1 - 100 kev range ) , even though no effort was made to tune the non - thermal electron injection ( e.g. , as in ghisellini , haardt , & fabian 1993 ) to match the mean photon energy change with that of the thermal plasma . the first key reason for this is that the equilibrium plasma parameters are determined by pair balance ( @xmath107 ) and the fact that the electron - positron distribution responsible for the multiple compton scattering is dominated by pairs created in the source , not the injected ones . if the input electron spectra are at all similar ( i.e. , have similar mean energies as is the case here ) , the different pair cascade generations initiated by these electrons tend to converge leading to similar final electron distributions . the second reason is that when multiple scattering is important , the upscattered spectrum must pivot about / start from the peak energy of the injected soft photon spectrum and then will turn down once the photon energy is comparable to the maximum average energy of the scattering electrons . if the maximum energies of the pairs created in the source are at all similar , then simple energy conservation guarantees that the slopes of the compton upscattered spectra will be correspondingly similar . for these reasons , the spectrum obtained in the hybrid plasma model with high energy non - thermal electron injection turns out to be amazingly similar ( given the radically different injection spectrum ) to the other two spectra . the agreement between cases only increases as the plasmas becomes more photon - starved . while the overall spectra from hybrid plasmas can be quite similar to those from thermal plasmas , hybrid plasmas with pairs are also systematically different from thermal ones in that the larger the excess emission they have at gamma - ray energies ( i.e. , the higher the typical injection energy of the non - thermal electrons ) , the _ lower _ the characteristic equilibrium temperature of the cooled pairs in the source . ( note the clearly separated comptonization and annihilation peaks in the dashed spectrum of fig . this is because hybrid plasmas with an energetically insignificant high energy electron tail will still produce many more pairs for a given thermalized pair temperature than will a purely thermal plasma . the gamma - ray spectrum in a thermal plasma is a wien spectrum and cuts off exponentially at photon energies @xmath108 as noted above , however , the gamma - ray spectrum in a hybrid model may fall off much more slowly with energy . this means that while the `` pair thermostat '' of svensson ( 1984 ) still operates in hybrid models ( in general , for a given @xmath109 the higher @xmath50 the higher the density of thermalized pairs and the lower their temperature ) , the exact plasma parameters it predicts depend critically on @xmath110 and the non - thermal injection spectrum . to demonstrate this last point , figure 3 shows three plots analogous to fig . 2 of ghisellini & haardt ( 1994 ) . figure 3a represents exactly the same case as their fig . 2 and shows contours of constant 210 kev spectral index and plasma temperature plotted in the plane of @xmath111 versus @xmath50 for a purely thermal model ( with no background plasma present ) . the two figures agree well given the differences in the microphysics ( eqpair includes electron - electron bremsstrahlung cooling which is important at low compactnesses . ) figure 3b shows similar contours , but now the plasma model has a small non - thermal , high - energy ( @xmath112 ) component that receives only @xmath113 of the total power provided to the electrons and pairs . while the contours look rather similar in the photon - starved , high compactness region ( @xmath114 @xmath115 ) for the reasons explained above , they behave rather differently in the rest of the diagram . figure 3c shows what happens when @xmath116 ( the plasma is mainly non - thermal ) but the injected electrons are low energy with @xmath117 ( the case in our fig . 2 ) . from the shape of the contours , we see that such a model indeed behaves much more like a purely thermal model . however , there are still considerable differences , e.g. , in the predicted spectral index , for @xmath118 the lesson here is that while there always seems to be a rough one - to - one mapping between @xmath111 and the observed spectral index ( higher @xmath111 gives harder spectra ) and between @xmath50 and the thermal electron / pair temperature ( higher @xmath50 gives lower temperatures ) , the details of the mapping are _ not _ robust . if one removes the constraint of pair balance by allowing a non - zero @xmath66 the mapping changes even more . when the time comes to extract the physical plasma parameters from the observed spectra , the possibly hybrid nature of the plasma energy distribution can make a significant difference even if the observed @xmath119 kev spectrum appears consistent with pure thermal comptonization . this having been said , figures of the type shown in figure 3 and ghisellini & haardt ( 1994 ) should still be very useful . given a particular set of assumptions about the plasma , they allow one to immediately zero in on the relevant model parameter space . in this regard , we also direct the reader to the contribution of beloborodov ( these proceedings ) where some new analytic approximations for thermal comptonization are presented . we conclude this section by showing one final example of how only the mean fractional change in photon energy per scattering matters when multiple compton scattering occurs . in figure 4 , we set the electron distribution equal to the sum of two maxwellians , one with @xmath120 kev and one with @xmath121 kev . this might at first seem a silly choice , but it is not . several authors ( e.g. , liang 1991 and moskalenko , collman , and schnfelder 1998 ) have attempted to explain the possible hard tail in cyg x-1 via a multi - zone model where thermal electrons have significantly different temperatures . the idea is that compton upscattering in the lower temperature zones explains the x - ray emission , while upscattering in the higher temperature zones explains the gamma - ray tail , i.e. , that the observed spectrum is basically the sum of two thermal comptonization spectra . while a sum of thermal comptonization spectra may indeed match the observations , there is one potential physical problem with this interpretation . in the hard state of cyg x-1 , it appears that the total optical depth of the source is at best @xmath122 thus photons are presumably free to scatter between the different thermal zones , and a uniform bi - maxwellian electron distribution is not a bad first approximation to this case . what spectrum does one obtain in this case ? as shown in figure 4 , while the overall spectrum does have an extended high enery tail , the overall spectrum is most closely approximated by a _ thermal comptonization spectrum with temperature @xmath123 ( this is the temperature of thermal plasma that gives roughly the same photon energy change as in the bi - maxwellian case . ) adding together a 10 kev and a 100 kev comptonization spectra in analogy with the way a multi - color disk black body is computed will give a very wrong answer in this case . note also that a bi - maxwellian plasma where one component is very hot has a nasty feature that may rule it out in the case of galactic black hole candidates like cyg x-1 . as we will discuss in more detail below , the fractional change in the energy of a scattered photon is not small if @xmath124 this means the multiple compton scattering approximation begins to break down , and one can see evidence of the first scattering order in figure 4 , a @xmath125 kev _ deficit _ of photons relative to a low - energy extrapolation of the 2 - 10 kev power law . in conclusion , we remark that if only one or two orders of scattering contribute significantly to the emergent spectrum ( e.g. , in a non - thermal model that is not photon - starved ) , then the shape of the spectrum is obviously extremely sensitive to the details of the underlying electron energy distribution . we do not have space to discuss the time - dependent behavior of hybrid models , e.g. , low vs. high energy lags and leads , but any such behavior will clearly depend on whether the emergent spectrum is produced in the multiple or single compton scattering regimes . in a one - zone model , if the spectrum is produced in one scattering ( e.g. , in non - thermal models ) , one expects no delay between different photons energies except perhaps for a slight soft lag due to the finite time it takes for electrons to cool and respond to changes in injection ( which can be much shorter than @xmath126 and thus hard to observe ) . if multiple compton scattering is instead important , one expects to see behavior similar to that seen in standard thermal comptonization models , e.g. , hard lags that increase logarithmically with energy . until now , we have mainly discussed hybrid plasma models in a theoretical context . is there any observational evidence that they may be important ? in active galactic nuclei ( agn ) , the situation is still unclear . radio - loud agn probably have x - ray emission that is contaminated by strongly non - thermal emission from a jet , and they will not be considered here . ( the jet environment in the emission region may be very different from that near the black hole . ) for radio - quiet agn , the composite seyfert spectra compiled by osse ( e.g. , gondek et al . 1996 ) favor a spectral cutoff at an energy @xmath127 kev . an equally good fit to the composite spectrum is obtained using either a purely thermal model or a purely non - thermal model with either steep power law injection or a low @xmath128 ( maximum electron injection energy ) , i.e. , the energy coverage and statistics of the composite spectra are not sufficient to distinguish between pure thermal , hybrid , or pure non - thermal models . it is also not clear , however , how representative the composite spectra are . for example , matt ( these proceedings ) reports on a bepposax seyfert sample that shows considerable variation in the cutoff energies , from @xmath129 kev in ngc 4151 to beyond @xmath130 kev in several objects . the strong break in ngc 4151 favors a purely thermal or hybrid model ( e.g. , zdziarski , lightman , & maciolek - niedzwiecki 1993 ) , but fits to other individual objects are inconclusive . the situation is potentially more interesting for galactic black hole candidates . power law emission extending beyond @xmath131 kev has definitely been detected by osse during the `` soft state '' of these objects ( e.g. , grove et al . 1998 , gierliski et al . 1999 ) , and in objects like cyg x-1 , we have observed several spectral state transitions from the `` soft '' to the `` hard '' state and back ( e.g. , see liang & nolan 1984 ; cui et al . 1997 ) . here , i will focus on cyg x-1 as a test case since the object always seems to be bright , and hence considerable data has been collected on it . figure 5 , taken from gierliski et al . 1999 , shows a montage of spectra obtained during the soft , hard , and intermediate ( transitional ) states of cyg x-1 . simultaneous , broad band data of this quality has had and will continue to have ( as the instruments improve ) an important impact on our understanding of this object . until simultaneous 10 - 100 kev observations were available , for example , it was impossible to constrain the contribution to the overall spectrum from a compton reflection component since the amplitude of this component depends critically on the hard tail ( @xmath72 kev ) of the spectrum . also , it was impossible to tell how the bolometric luminosity varied with the state of the source , and it was difficult to constrain the parameters of the comptonizing cloud that was supposed to be responsible for the spectrum ( at least in the hard state ) . using simultaneous ginga - osse data , however , gierliski et al . ( 1997 ) was able to show that the standard static disk - corona slab geometry was ruled out for cyg x-1 s hard state because : ( i ) no `` anisotropy break '' was seen , ( ii ) the x - ray spectrum was very hard indicating that the source was photon - starved with @xmath132 ( in the disk - corona geometry the soft reprocessed photon luminosity is comparable , i.e. , @xmath133 ) , and ( iii ) the solid angle covered by the reflecting matter substantially less than @xmath134 the value expected in the disk - corona where the corona extends uniformly over the disk . they suggested that a source geometry consistent with these would be one like that of shapiro , lightman , & eardley ( 1976 ) , with a central , hot source surrounded by a cool , thin accretion disk ( e.g. , see fig . 9 ) . relying on a composite cyg x-1 spectrum made of non - simultaneous data , dove et al . ( 1997 ) came to a similar conclusion . hard state data of this type in combination with broad band data on the transition to the soft state also led to new suggestions for the overall accretion disk / corona geometry in cyg x-1 . the data of zhang et al . ( 1997 ) showed that the bolometric luminosity in fact did not change significantly ( @xmath135 factor two ) during the recent hard - soft state transition . because the luminosity of the blackbody component in the soft state was comparable to the hard photon luminosity in the hard state , and because a thermal comptonization model for the hard tail required a low compton @xmath98 parameter ( a combination of low temperature and/or optical depth ) , poutanen , krolik , & ryde ( 1997 ) proposed that the state transition was simply a change in the state of the accretion disk : the power dissipated in the corona dropped as the inner edge of the cool disk moved inwards , correspondingly increasing the soft photon luminosity . as the cool region of the disk spread inwards , the fraction of the coronal emission intercepted by the cool disk increased ( e.g. , as the disk penetrated into the coronal region ) , causing the observed increase in the relative amplitude of the compton reflection component . esin et al . ( 1998 ) proposed a rather similar model , where the inner radius of the cool disk is interpreted as the transition radius between the sunyaev - shakura disk solution and the adaf solution . while the poutanen , krolik & ryde model is more phenomenological and the discussion in esin et al . is framed in the more physical context of adafs , both models are essentially the same , with the key free parameter controlling the `` state '' of the system being the location of the inner edge of the cool disk / the disk transition radius ( something not currently well - understood ) . both models also share the significant shortcoming that neither can simultaneously fit the low - energy ( 110 kev ) and high - energy ( @xmath136 kev ) data in the soft state . this has not been completely appreciated and is one of the strongest arguments for the presence of a hybrid plasma . the reason both models fail is that they rely on purely thermal comptonization to produce the observed spectrum . the x - ray spectrum above @xmath137 kev ( e.g. , in the rxte data ) appears to be a rather steep power law that joins smoothly onto the dominant blackbody component at @xmath138 kev ( e.g. , see fig . a comptonization model fit to data below @xmath139 kev will give something like @xmath140 kev as the best fit electron temperature in the model . at first sight , a low temperature like this is exactly what one wants since in the soft state , less energy is being dissipated in the corona and there are more soft photons to cool on . unfortunately , such a low temperature also predicts an exponential cutoff in the upscattered spectrum starting at @xmath141 kev . such a cutoff is _ not _ seen in the osse data for the soft spectra of galactic black hole candidates , which in some cases clearly extend to at least 500 kev and above ( see grove et al . 1997 , 1998 ) . recently gierliski et al . ( 1999 ) has taken osse data for cyg x-1 and tried to make as simultaneous fits as possible to lower energy asca and rxte data . figure 6 shows an example of joint rxte - osse data that indicates no strong break out to @xmath130 kev . in order to produce such a unbroken spectrum , the temperature of the comptonizing electrons must be comparably high . we illustrate this in figure 7 , where we show the comptonized spectrum from electrons with temperature 200 kev . at @xmath142 kev , this spectrum has the right slope to match the soft state spectrum , and it starts cutting off exactly when the statistics of most detectors become very poor , i.e. , it looks like an acceptable model . however , notice the large photon deficit at @xmath1 1 kev ( briefly mentioned above ) . the deficit occurs because the mean fractional energy change a blackbody photon undergoes in one scattering is @xmath143 for @xmath74 kev , this exceeds unity and implies that one can begin to see the shapes of the individual compton scattering orders , particularly the first one . ( approximating the blackbody soft photon distribution as a delta function at energy @xmath144 there are few upscattered photons between @xmath145 and @xmath146 since @xmath147 is so large hence we see a deficit . ) note the good agreement between the eqpair result and the monte carlo simulations shown there . ( when making comptonization calculations , especially using a kinetic code like eqpair , one has to be very careful not to create spurious features similar to this one by using an approximate compton redistribution function that does not spread scattered photons sufficiently in energy . ) such a feature is real and generic to high temperature comptonization models and is strongly ruled out by data like that of figure 6 at the many sigma level . ( remember , flux determinations in the kev range are now good down to the @xmath1 few percent level . ) the discrepancy with purely thermal models is probably even greater because gierliski et al . ( 1999 ) have averaged together several days of soft state osse data and find that any cutoff must be at energies @xmath148 kev ( ! ) . if the soft state is not the result of purely thermal comptonization , what are the alternatives ? first , one might interpret the lack of a @xmath1 kev deficit as implying the existence of `` excess '' emission . by superposing emission from two spatially distinct regions ( or else one runs into the problems discussed in the previous section if photons can sample both regions ) , one can in principle reproduce the observed spectrum using a low temperature and a high temperature region . one then needs to explain , however , where the extra component comes from and why it has the temperature and optical depth it does . another possibility that has recently been revived by titarchuk and collaborators in the context of galactic black hole candidates is that of bulk comptonization ( e.g. , see blandford & payne 1981 , colpi 1988 , shrader & titarchuk 1998 , titarchuk & zannias 1998 ; psaltis & lamb , in these proceedings ) . the rationale for invoking bulk comptonization during the soft state is that the soft photon density appears to increase dramatically during the soft state , strongly cooling the coronal electrons responsible for the hard state emission . if there is quasi - spherical accretion near the black hole , then cold infalling coronal electrons could acquire substantial velocities @xmath149 and comptonization using the large bulk inflow velocity of the electrons ( as opposed to their assumed smaller thermal motions ) could give rise to power law spectra like those observed in the soft state . while elegant , this interpretation may also have significant problems fitting data , particularly in the case of cyg x-1 . one of the key difficulties is again how to produce unbroken power law emission well beyond 511 kev ( @xmath7 ) . as a first order estimate of the location of the high energy break in the bulk comptonization spectrum , titarchuk , mastichiadis , & kylafis ( 1997 ) give @xmath150 where @xmath151 is the mass accretion rate measured in units of the eddington luminosity and the scattering electrons are assumed to be radially free - falling . however , to have a predicted x - ray spectral index in the observed range ( @xmath152 ) apparently requires a very high mass accretion rate , @xmath153 ( titarchuk & zannias 1998 ) . the first order estimate thus predicts a strong break in the comptonized spectrum at energy @xmath154 it is currently still not clear , though , how good this first order estimate is . as pointed out in titarchuk , mastichiadis , & kylafis ( 1997 ) , the second order terms in their equations tend to push their cutoff to higher energies . however , their treatment and essentially all other treatments until now have worked in the diffusion approximation and used the thomson cross - section with a down - scattering correction instead of the full klein - nishina cross - section . as in the case of standard thermal comptonization , when electron and photon energies exceed @xmath155 the results obtained with these approximations become suspect . in particular , in the klein - nishina limit , compton scatterings with the more energetic ( higher velocity ) electrons are reduced , and scattered photons tend to keep traveling along their original direction , e.g. , into the black hole . in addition , even moderately relativistic electron velocities ( @xmath156 ) will strongly collimate incoming radiation along the inflow direction . gravitational redshift effects when infall velocities are @xmath157 might further lower the break energy . i also note that the soft state temperature esin et al . find for the central adaf region is still @xmath158 kev , i.e. , it is not that low at all . standard thermal comptonization effects may still dominate over bulk comptonization ones , and at the very least , may cause a significant deviation from a power law spectrum . although a better calculation is required for a definitive answer , right now it seems difficult to produce a spectrum beyond 511 kev ( @xmath7 ) via bulk comptonization . how serious a problem is the low predicted break energy ? in most galactic black hole candidates , the typical data above @xmath130 kev are not good enough to say anything . in cyg x-1 , however , we have a clear indication from osse data that the spectrum continues unbroken to at least @xmath1 800 kev . although the cyg x-1 data is not quite as certain at energies higher than this due to possible source confusion , as noted above there is a strong indication that the spectrum continues to several mev . if correct , this would seem to strongly rule out the bulk comptonization hypothesis , for cyg x-1 at least . break energy aside , bulk comptonization models have one other serious problem fitting cyg x-1 . gierliski et al . ( 1999 ) finds that the asca / rxte - osse data above @xmath0 kev is _ not _ well - fit by a power law and requires excess emission in the @xmath159 kev range . together with the presence of an apparent iron line and edge , this strongly suggests the presence of reflection with a covering factor @xmath160 i.e. , half the hard x - ray flux hits a cool disk . such a covering factor is natural in a corona - over - disk geometry , but it is not obvious how to arrange this in the bulk comptonization scenario . as discussed in poutanen & coppi ( 1998 ) , an easy way to avoid all the problems mentioned is simply to allow the comptonizing plasma to be a hybrid one , with a non - thermal tail . this introduces an extra parameter , the ratio of thermal heating to non - thermal acceleration power , but allows disk transition scenarios like those of poutanen et al . and esin et al . to go through largely unchanged . as shown by the solid curves in figures 5 and 6 , the hybrid model ( eqpair ) including compton reflection and relativistic line smearing can fit the broad band data extremely well ( @xmath161 per d.o.f @xmath138 ) in _ all _ three states of cyg x-1 , including the `` intermediate '' transitional one . to see how far we could push the hybrid model , we also applied it to public rxte data on grs 1915 + 105 and also were able to obtain a good fit , e.g. , see figure 8 ( although that data only extends to @xmath162 kev ) . we are not aware of another type of model that can currently do this . this is slightly surprising given the quality of the asca / rxte data in the 1 - 20 kev range and may indicate that the crude assumptions made in the eqpair model ( e.g. , that the source is homogeneous , isotropic , and static ) are true to first order . a detailed discussion of the model fits to cyg x-1 and their physical implications ( e.g. , constraints on the importance of electron - positron pairs ) can be found in gierliski et al . ( 1999 ) . i conclude by showing results a simple phenomenological model which gives roughly the right fit parameters for the states ( see poutanen & coppi 1998 for more details ) . the source geometry we envision is along the lines of that shown in figure 9 . the total power supplied to the disk and the @xmath163 hot @xmath163 coronal @xmath163 region remains roughly constant during the state transition , i.e. , @xmath164 is constant . as in the poutanen et al . and esin et al . models , we assume a transition radius @xmath165 which marks the boundary between a cool outer disk and a hotter inner disk / corona . we also assume that the sum of soft luminosity from the disk , @xmath166 , and the thermal dissipation rate in the corona , @xmath167 , remains approximately constant during transitions . in addition to thermal dissipation in the corona , we will assume a central source of non - thermal electrons with acceleration luminosity , @xmath168 . ( note that if the source turns out to have a relatively low compactness , the non - thermal acceleration could also occur in a more extended region that does not change size . one will obtain quite similar spectra to those shown here . ) for simplicity , we take @xmath168 to be constant ( its value is not that well - constrained in the hard state ) . in the soft state , essentially all the power goes into accelerating non - thermal electrons . the density and temperature of the thermalized electrons / pairs responsible for the excess emission at @xmath1 a few kev are determined self - consistently . the results of the simulations are shown in figure 10 . although we have interpreted the state transition as a change in the location of the inner edge of the cool disk , note that this interpretation is in fact not unique . the arguments used in gierliski et al . 1997 apply _ only _ to models where the corona is static , i.e. , there is no bulk motion of coronal electrons . as discussed in beloborodov ( 1999 ) , this is rather unlikely given the strong radiation fields likely to be present and also given the highly dynamic , flaring nature of the emission . ( the possibility of winds arising from the surface of disks has also been discussed for some time , e.g. , see narayan & yi 1995 , blandford & begelman 1998 in the context of adaf - type solutions ; and woods et al . 1996 for a numerical calculation of x - ray heated winds and coronae from accretion disks . ) when the electrons acquire even relatively small outflow velocities ( @xmath169 ) , this is enough to make their upscattered radiation significantly anisotropic ( see beloborodov 1999 ) . even if a cold disk extends all the way in to the last marginally stable orbit and is directly under the corona ( e.g. , the slab disk - corona geometry again ) , this anisotropy means that fewer hard photons reach the disk , i.e. , the inferred covering fraction ( @xmath170 ) of the compton reflecting material will be low . if most of the accretion power near the black hole is dissipated in the corona above the disk rather than inside the disk , then this also means the disk underneath the corona will be very cold and will emit few soft photons , i.e. , the coronal plasma will be photon - starved . ( this is in contrast to the static corona of haardt & maraschi 1993 where half the hard x - ray flux hits the corona and is reprocessed , implying a soft photon luminosity comparable to the hard photon luminosity . ) in other words , if there is bulk motion in the corona , then there is no need to have a transition from a cool to a hot disk . rather , we might have a transition from a cool to a _ cooler _ disk . in this case , the transition radius in our model is to be interpreted as the boundary between the region where most dissipation occurs inside the disk and the region where ( for some reason ) most dissipation occurs outside the disk , in the corona . as long as the coronal outflow velocities are not too large , the results are quite similar to those of the central hot sphere - disk geometry of figure 9 . most attempts at modeling the emission from accreting black hole systems have typically assumed the underlying particle energy distribution is either a maxwellian ( `` thermal '' ) or a power law ( `` non - thermal '' ) . while such an assumption may be convenient analytically , it is no longer required given the advent of powerful computers , and more importantly , it is does not appear to be well - justified . there are several examples in nature , e.g. , the phenomenon of solar flares , where it is clear the underlying distribution is neither a maxwellian nor a power law , but rather a quasi - maxwellian at low energies with a high - energy approximately power - law tail . this is exactly the type of particle energy distribution that is often predicted by theoretical particle heating / acceleration models . ( the acceleration process typically kicks only a few particles in the high energy tail of the particle distribution to much higher energies . ) re - examining the process of electron thermalization under the physical conditions likely to be found near a black hole , we find that the likely thermalization time scales are likely to be quite long unless some ( unknown ) collective plasma process is more effective than two - body coulomb collisions at exchanging energy between electrons . in particular , because the radiation field is likely to be so intense near an accreting black hole , the coulomb relaxation time for even moderately relativistic electrons may be much longer than the relevant cooling times . depending on the exact plasma parameters , they may also be longer than the characteristic source variability time . thus , the old arguments made against thermal models still stand . these arguments were largely brushed aside and forgotten when it became clear many of the classical non - thermal sources like ngc 4151 ( an agn that was supposed to have strong mev emission ) in fact showed strong cutoffs at @xmath171 kev energies and had spectra that could be successfully fit using purely thermal comptonization models . now that we are seeing hints that the emission in galactic black holes candidates may indeed extend to much higher energies ( albeit at much lower levels than previously thought ) , `` hybrid '' models involving both thermal particle heating and non - thermal particle acceleration are beginning to creep back . one of the strongest cases for the existence of a non - thermal particle distribution near black holes is the soft spectral state of galactic black hole candidates . in this state , one sees very strong , quasi - blackbody emission at @xmath1 1 kev , with a steep @xmath1 power law tail extending to at least @xmath1 511 ( @xmath7 ) in several objects . particularly in the case of cyg x-1 , such emission is very hard to explain either via pure thermal comptonization or bulk comptonization in the accretion flow . this has not been completely appreciated . data from future missions with improved sensitivity in the @xmath87 kev - @xmath172 mev range ( e.g. , integral and astro - e ) should be conclusive . if we relax the assumption that the energy distribution of the comptonizing electrons is a strict maxwellian , then many problems go away . using a newly developed , self - consistent hybrid plasma code , we show that the spectrum in the soft state ( as well as in the other spectral states ) can easily be modeled . with the proviso that there is always some small amount of non - thermal acceleration going on ( compared to the total source luminosity ) , models that explain the cyg x-1 state transitions in terms of a moving transition radius between cold and hot disk phases appear to work fairly well . ( as cui points out in these proceedings , however , these models only attempt to explain _ time - averaged _ spectra . what such a spectrum and the deductions one makes from such a spectrum have to do with reality is not yet clear , particularly if this time - averaged emission is the superposition of many individual flare events , e.g. , as discussed here by poutanen & fabian as well as mineshige & negoro . if we follow theoretical prejudice and assume that particle energy distributions are indeed not completely thermal , then we must explain why so many black hole sources still manage to look so thermal . clearly , one part of the answer must be that the efficiency with which power is channeled into relativistic electrons ( lorentz factors @xmath142 ) is relatively low . why this is so , depends on the unknown details of the acceleration mechanism . another part of the answer , however , may have to do with comptonization and pair plasma physics . if a source photon gains most of its energy in a single scattering event before escaping , then clearly the emergent radiation spectrum depends critically on the underlying scattering electron energy spectrum . however , if multiple compton scattering is important , i.e. , a photon gains its energy in several small steps before it escapes , then the exact details of the energy distribution turn out not to matter . to first order , if the scattering electrons have the same mean energy , be they thermal or non - thermal , then they tend to produce the same mean energy change in a scattered photon , which results in a spectrum with the same shape . ( one can imagine doing a fokker - planck expansion of the relevant equations . ) the details of the distribution typically only matter at the high - energy and low - energy tails of the output spectrum . the types of sources where multiple compton is most important are those that are relatively photon - starved , i.e. , where the power supplied to electrons is much larger than the power initially supplied to the low energy target photons . in this case , as long as the bulk of the heated / accelerated electrons do not have lorentz factors @xmath173 it is largely irrelevant whether the electrons thermalize or not before they cool . ( note that if the source is very optically thick to gamma - ray pair production , and many generations of pair cascading are important , then the condition that most electrons have low lorentz factors is automatically guaranteed even if the accelerated electrons which initiate the cascading have high initial energies . ) to model cyg x-1 completely , for example , we may need a non - thermal power law source of energetic electrons . to match the data , however , the non - thermal acceleration must also produce a very steep power law in energy , i.e. , most of the power resides in the lowest energy electrons . if we add such a particle distribution to , say , a hot background thermal plasma distribution , and then make the source photon starved , we will see virtually no difference in the final spectrum , except perhaps at the highest energies @xmath174 kev . ( for connoisseurs of pair plasmas , though , note that differences in the high energy photon spectrum can mean big differences in the pair balance and the pair thermostat . ) it is probably no accident , then , that the photon - starved hard state of galactic black holes looks so thermal , while the photon - rich soft state does not . in sum , whether we realize it or not , hybrid plasmas may be all around us ! i thank juri poutanen , andrei beloborodov , and roland svensson for organizing an excellent conference , for many useful discussions and hospitality in past , and most of all for their patience and help in getting this contribution out . i would also like to thank marek gierlinski and andrzej zdziarski for helpful discussions and allowing me to use their monte carlo comptonization code . this work was supported in part by nasa grants nag5 - 6691 and nag5 - 7409 .

we describe an approach to numerical simulation of spiral waves dynamics of large spatial extent , using small computational grids .

1001.4454

[ sec : introduction ] spiral waves are a type of self - organization observed in a large variety of spatially extended , thermodynamically non - equilibrium systems of physical , chemical and biological nature @xcite , where wave propagation is supported by a source of energy stored in the medium . if the system can be considered spatially uniform and isotropic and its properties do not depend on time , the corresponding mathematical models possess corresponding symmetries . for many practical applications , a considerable interest is in non - stationary dynamics of spiral waves , which is usually defined separately either as drift , which is displacement of the average position of the core of the spiral with time due to external symmetry - breaking perturbations , or meandering , which is spontaneous symmetry breaking due to internal instability rather than external forces and which is manifested by complicated movement of the spiral with the average position of the core typically unmoved . the numerical simulation of drift and meander of spiral waves , particularly when models are complicated and high accuracy is required , can be challenging . there are some theoretical considerations which suggest some way of dealing with this challenge . so it has been observed that as far as drift is concerned , spiral waves behave like particle - like objects , which results from effective localization of the critical eigenfunctions of the adjoint linearized operator @xcite , so it should be sufficient to do the computations only around the core of the spiral to predict its drift . on the other hand , in the absence of external symmetry breaking perturbations , meandering of spirals can be understood by explicitly referring to the euclidean symmetry of the unperturbed problem @xcite . specifically , an idea of dynamics in the space of symmetry group orbits @xcite , when applied to a reaction - diffusion system of equations and the euclidean symmetry group , leads to a description which is formally equivalent to considering the solution in a moving frame of reference ( for ) such that the spiral wave maintains a certain position and orientation in this frame @xcite . we shall call it comoving for for short . the purpose of this article is to present a computational approach based on these considerations . we calculate the dynamics of the spiral wave in a comoving for ; as a result , the core of the spiral never approaches the boundaries of the computation box , which allows computations of drift and meandering of large spatial extent using small numerical grids . a simple software implementation of this approach , which is based on the popular spiral wave simulator ` ez - spiral ' @xcite , and which we called ezride , is provided on the authors website @xcite . our approach can be compared to the approach proposed by beyn and thummler @xcite and further developed by hermann and gottwald @xcite . their approach also exploits symmetry group orbits , but is different in some essential details . we shall discuss the similarities and differences when we will have introduced our method . the structure of the paper is as follows . in section [ sec : maths ] we lay out mathematical basics of the approach and briefly compare it with @xcite . in section [ sec : numerics ] we describe the numerical method itself . in section [ sec : examples ] we illustrate the work of the method by simple and quick examples . the potential for numerical accuracy is demonstrated in section [ sec : convergence ] . the subsequent three section are dedicated to examples of applications of the methods to problems where the conventional methods would be struggling : section [ sec : resonance ] for the degenerate case of meandering which results in `` spontaneous drift '' of spirals ; section [ sec : largecore ] for the dynamics near to , and beyond , the parametric boundary at which the core radius of the spiral becomes infinite ; and section [ sec : drift ] for drift caused by a symmetry breaking perturbation . we conclude with a brief discussion of the results in section [ sec : discussion ] . [ sec : maths ] following @xcite , we start from a perturbed reaction - diffusion system of equations in a plane , @xmath0 where @xmath1 is a column - vector of reagent concentrations varying in space and time , @xmath2 is a column - vector of reaction rates , @xmath3 is the matrix of diffusion coefficients , @xmath4 represents symmetry - breaking perturbations , @xmath5 , @xmath6 , and @xmath7 . if @xmath8 , then equation ( [ eq : rdeu ] ) is equivariant with respect to euclidean transformations of the spatial coordinates @xmath9 . the following technical discussion is necessary to place our method in the context of other works in the field . readers not interested in technical details may skip down to system ( [ eq : ezfnum ] ) . the idea of the symmetry group reduction is convenient to describe if we view ( [ eq : rdeu ] ) as an ordinary differential equation in a suitably chosen functional space @xmath10 , @xmath11 where @xmath12 represents the dynamic field @xmath13 , @xmath14 represents the unperturbed right - hand side @xmath15 , and @xmath16 represents the perturbation @xmath17 . let us suppose that equation ( [ eq : rdeu ] ) at @xmath8 is equivariant with respect to a representation @xmath18 of a lie group @xmath19 in @xmath10 . this means that for any @xmath20 and any @xmath21 , we have @xmath22 in our case , @xmath23 , the special euclidean transformations of the plane @xmath24 ( including translations and rotations ) , and @xmath18 is its representation in the space of functions @xmath25 defined on this plane , acting as @xmath26 we consider a subset @xmath27 such that @xmath19 acts freely on @xmath28 , for a @xmath29 , any nontrivial transformation @xmath20 changes @xmath30 , in other words , @xmath31 . in the terminology of @xcite , @xmath28 is the principal stratum of @xmath10 , corresponding to the trivial isotropy subgroups . in our case , this means that the graph of the function @xmath32 is devoid of any rotational or translational symmetry , which is of course true for functions describing single - armed spiral waves . it is straightforward that at @xmath33 , the set @xmath28 is an invariant set of ( [ eq : rdeu ] ) . moreover , we shall restrict our consideration to such perturbations @xmath34 that resulting solutions @xmath35 remain in @xmath28 for all @xmath36 . this means , that the perturbations are supposed to be so small they can not impose incidental symmetry on the otherwise unsymmetric spiral wave solutions . a group orbit of a given @xmath30 is defined as the set @xmath37 . that is , it is a set of all such functions @xmath25 that can be obtained from one another by applying an appropriate euclidean transformation to @xmath9 . a group orbit is a manifold in @xmath28 , of a dimensionality equal to @xmath38 less the dimensionality of the isotropy group . in our case , @xmath39 , the isotropy group is trivial and the orbits are smooth three - dimensional manifolds . from the definition of the set @xmath28 it follows that this set is foliated by group orbits . the principal assumption for the following analysis is that there exists an open subset @xmath40 , also invariant with respect to @xmath19 , in which the foliation has a global transversal section , we can select one representative from each orbit in @xmath41 , such that all such representatives form a smooth manifold @xmath42 , which is everywhere transversal to the group orbits . we call this manifold a _ representative manifold _ ( rm ) . that would mean that any orbit in @xmath41 crosses @xmath43 transversally and exactly once . hence @xmath44 the rm has co - dimensionality equal to the dimensionality of the group orbits , in our case @xmath45 . it is assumed to be smooth and we expect that it can locally be described by equations @xmath46 , @xmath47 , where functions @xmath48 , are functionals when interpreted in terms of the original reaction - diffusion equation ( [ eq : rdeu ] ) . a convenient pictorial interpretation for our case is in terms of spiral wave solutions and their tips . suppose the conditions @xmath49 , @xmath50 determine that the tip of the spiral wave is located at the origin , and condition @xmath51 fixes its orientation , so @xmath43 consists of such functions that look like spiral waves @xmath52 which have the tip exactly at the origin and in a standard orientation . then equation ( [ eq : manifold ] ) states that any spiral wave solution @xmath25 , considered at a fixed moment of time , can be transformed by a euclidean transformation , in a unique way , to a solution @xmath53 which has its tip at the origin and in the standard orientation . this is equivalent to saying that @xmath53 is the same as @xmath25 only considered in a different system of coordinates , with the origin at the tip of @xmath25 and oriented accordingly to the orientation of that tip . we shall say this is the system of coordinates _ attached to the tip_. an example of @xmath54 , as used in @xcite , is [ eq : mus ] @xmath55=v^{(l_1)}({\vec{0}})-{u _ * } , { \label{eq : mu1}}\\ \mu_2[{{\mathbf{v}}}({{\vec r}})]=v^{(l_2)}({\vec{0}})-{v _ * } , { \label{eq : mu2}}\\ \mu_3[{{\mathbf{v}}}({{\vec r}})]=\partial_xv^{(l_3)}({\vec{0 } } ) , { \label{eq : mu3}}\end{aligned}\ ] ] where @xmath56 are suitably chosen components , and @xmath57 . this means that the tip of @xmath25 is defined as the point of intersection of isolines of the components @xmath58 and @xmath59 of the field @xmath13 at appropriately chosen levels @xmath60 and @xmath61 respectively , and the orientation of the attached coordinate system is such that gradient of component @xmath62 ( which may or may not coincide with @xmath58 or @xmath59 ) is along the @xmath63-axis in that system . this choice of @xmath54 is of course not prescriptive , and later in this paper we shall consider some variations . regardless of the exact definition of the tip , choice of functionals @xmath54 , an essential assumption that we have to make is that our spiral waves have one tip only , otherwise there would be more than one way to transform them to the standard position or to chose the attached system of coordinates . hence the reason for a further constraint to the subset @xmath40 , which we now can define as consisting of such one - tip spiral wave solutions , or functions that look like it : without such constraint , the whole set @xmath28 includes solutions with no tips or more than one tip , for which the decomposition ( [ eq : manifold ] ) would not hold . as before , we assume that set @xmath41 is invariant with respect to the dynamic equation ( [ eq : rdeu ] ) for not too big @xmath64 , that is , if @xmath65 , then @xmath66 for all @xmath67 and @xmath68 . a further restriction is on the manifold @xmath43 . it is easy to see that equations like ( [ eq : mus ] ) may not be sufficient to define the manifold with the required property that any orbit crosses it only once . for instance , if @xmath53 satisfies ( [ eq : mus ] ) , then @xmath69 also satisfies it , so a rotation by @xmath70 around the origin transfers a point on @xmath41 to another point on @xmath41 . so to make the representation ( [ eq : manifold ] ) unique , rather than just requiring that the gradient of the @xmath62-component of @xmath53 is along the @xmath63 axis , one would need to specify in which direction it is , say add to the definition of @xmath43 by the equations @xmath71=0 $ ] a further inequality @xmath72>0 , \quad \textrm{where } \quad \mu_4[{{\mathbf{v}}}({{\vec r}})]=\partial_yv^{(l_3)}. { \label{eq : mu4}}\ ] ] this comment extends to the variations of ( [ eq : mus ] ) which we consider later . by performing the decomposition ( [ eq : manifold ] ) for every @xmath73 , we decompose motion in @xmath41 to motion along the rm and motion along group orbits which are transversal to the rm ( see illustration in fig . [ fig : theory - banach ] ) . , which has exactly one transversal intersection with evey group orbit @xmath74 within the relevant stratum of the phase space @xmath10 and is diffeomorphic to the orbit manifold . trajectory @xmath75 of an equivariant flow in @xmath10 is a relative periodic orbit , since it projects onto the trajectory @xmath76 on @xmath43 which is periodic . the flow on @xmath43 is devoid of symmetry @xmath19 . ] [ fig : theory - banach ] so for all @xmath73 , we have @xmath77 substituting ( [ eq : utv ] ) into ( [ eq : rdeu ] ) and applying @xmath78 to both sides , we get @xmath79 where @xmath80 we note that if @xmath33 , the right - hand side of ( [ eq : rdev ] ) is independent of @xmath81 . by the assumptions made , intersection of the group orbit @xmath82 with the manifold @xmath43 at the point @xmath52 is transversal . this means that the vectors @xmath83 and @xmath84 can be uniquely decomposed into the sums of the components along the group and along the manifold , @xmath85 [ eq : fhsplit ] hence equation ( [ eq : rdev ] ) splits into two components , along the rm and along the group orbit ( go ) : @xmath86 [ eq : decomposed ] note that equation ( [ eq : mani ] ) is the equation on the infinite - dimensional manifold @xmath43 , corresponds to a partial differential equation , whereas the left- and right - hand sides of the equation ( [ eq : group ] ) are in the tangent space to the finite - dimensional group orbits , and the dynamic variable @xmath81 is an element of the finite - dimensional manifold @xmath19 , so ( [ eq : group ] ) is in fact a system of ordinary differential equations of order @xmath38 . at this point we comment on what we see as a significant difference between our approach and that proposed by beyn and thummler @xcite ( bt ) . using our notation , in place of our `` pinning '' conditions @xmath46 , @xmath87 , they defined `` phase conditions '' of the form @xmath88 ( see equation ( 2.22 ) in @xcite ) , subsequently further generalized to @xmath89 ( , equation ( 2.33 ) ) . this means that their decomposition @xmath90 is not uniquely determined by the current state @xmath30 , but depends on history . such generalization may have its advantages and , apparently , works well for relative equilibria , steadily rotating spirals @xcite . however , the situation is different if the solution is a meandering spiral , is periodic with period @xmath91 in the orbit space ( as illustrated in fig . [ fig : theory - banach ] ) . this means that @xmath92 is equivalent to @xmath35 up to some euclidean transformation . in our approach , it is then guaranteed , that @xmath93 , as by ( [ eq : manifold ] ) , @xmath94 has a unique solution for @xmath81 at a given @xmath30 . however , in the bt approach , typically @xmath95 , since @xmath96 does not uniquely define @xmath81 , as @xmath97 is not fixed . so in our approach , study of meandering spirals reduces to study of periodic solutions for @xmath98 , but it does not do so in the bt approach . a practical approach to the problem of decomposing the vector fields as in ( [ eq : fhsplit ] ) is as follows . equations ( [ eq : decomposed ] ) together with the definition of the rm via functionals @xmath54 can be re - written in an equivalent form [ eq : man ] @xmath99 where @xmath100 is a vector belonging to the three - dimensional tangent space of the group orbit @xmath101 at @xmath52 . in this formulation , at any given moment of time , equations ( [ eq : along - man ] ) and ( [ eq : on - man ] ) together define the evolution of @xmath52 and the current value of the vector @xmath102 , whereas equation ( [ eq : across - man ] ) defines the evolution of @xmath81 . by definition , vector @xmath102 is a result of action of a linear combination of the generators of the lie group @xmath103 as linear operators on @xmath52 . to write the explicit expression for the general form @xmath102 for our case , let us introduce coordinates @xmath104 on @xmath23 , where @xmath105 is the translation vector , @xmath106 is the rotation angle and a group element acts as @xmath107 where @xmath108}$ ] , so @xmath109 is the matrix of rotation by angle @xmath106 . using this representation , differentiating the definition of @xmath110 given by ( [ eq : t ] ) , and substituting the result into ( [ eq : across - man ] ) , we get @xmath111 where @xmath112 and @xmath113 is the polar angle in the @xmath114 plane , so @xmath115 . with this result , the system ( [ eq : man ] ) in the original pde notation states @xmath116 [ eq : ezfnum ] where the dynamic variables are @xmath117 , @xmath118 , @xmath119 , @xmath120 and @xmath121 . in terms of the tip of the wave , equation ( [ eq : ezfnum - rda ] ) is the original reaction - diffusion equation ( [ eq : rdeu ] ) written in the comoving for , equations ( [ eq : ezfnum - pina ] ) , ( [ eq : ezfnum - pinb ] ) define the attachment ( pinning ) of the tip to this for , and equations ( [ eq : ezfnum - tip ] ) describe the movement of the for and , therefore , of the tip . equations ( [ eq : ezfnum - pina ] ) , ( [ eq : ezfnum - pinb ] ) imply that the position @xmath122 and orientation @xmath123 of the tip during calculations in the laboratory for are defined as @xmath124 and the comoving for is chosen so that in it , @xmath125 and @xmath126 at all times . unlike other equations of system ( [ eq : ezfnum ] ) , these are not prescriptive and is essentially an arbitrary choice , dictated by properties of particular systems . we shall refer to the pinning conditions ( [ eq : ezfnum - pina],[eq : ezfnum - pinb ] ) as `` choice 1 '' , as below we shall consider a variation of these , which we call `` choice 2 '' . when @xmath8 , the system ( [ eq : ezfnum ] ) decouples , as its upper part including ( [ eq : ezfnum - rda ] ) , ( [ eq : ezfnum - pina ] ) and ( [ eq : ezfnum - pinb ] ) becomes independent of the lower part ( [ eq : ezfnum - tip ] ) . this is the `` skew - product '' decomposition , the upper part describing the dynamics in the space of group orbits , so called `` quotient system '' , and the lower part the `` symmetry group extension '' , dynamics along the group , which depends on but does not affect the quotient dynamic . the connection between the quotient system and the group extension is via the dynamic variables @xmath127 ; in the following , we refer to these three quantities as `` quotient data '' for brevity . the skew - product representation has been useful for the analysis of various types of meander of spiral waves @xcite . note that the approach used in @xcite ( also see references therein ) is based on the assumption that the meandering pattern in question is considered in the vicinity of a bifurcation from the rigidly rotating spiral wave solution , so that the quotient dynamics can be reduced to the centre manifold , hence instead of equations ( [ eq : ezfnum - rda ] ) , ( [ eq : ezfnum - pina ] ) and ( [ eq : ezfnum - pinb ] ) , these studies considered normal forms on the corresponding centre manifolds . however , as noted in @xcite , the centre manfold theorem is not applicable for spiral waves , so this approach seems to be fundamentally flawed . this technical difficulty of course does not in any way affect the validity of system ( [ eq : ezfnum ] ) , which , as we have just demonstrated , is derived by elementary means without recourse to any bifurcations . in the rest of the paper , we consider system ( [ eq : ezfnum ] ) as a computational tool , rather than an instrument of theoretical analysis . the disadvantage of the original system ( [ eq : rdeu ] ) as a computational tool is that it requires a big computational grid to simulate dynamics of a spiral in an infinite medium , particularly when the tip of the spiral performs excursions to large distances . this is actually not necessary , as the dynamics of the spiral is mostly determined by the events in some finite vicinity of its tip @xcite . the system ( [ eq : ezfnum ] ) takes advantage of this property so that the pde calculations are done always in some fixed vicinity of the spiral wave , whereas the movement of the tip is described by the ode part . [ sec : numerics ] [ [ discretization . ] ] discretization . + + + + + + + + + + + + + + + we use time discretization with constant step @xmath128 and square spatial grid with step @xmath129 , covering spatial domain @xmath130 ^ 2 $ ] , so that @xmath131@xmath132@xmath133 and the grid coordinates of the origin are @xmath134 ( we only use odd values of @xmath135 ) . we designate the @xmath136-th time layer of the numerical solution as @xmath137 . we discretize the ode dynamic variables on the same time grid , @xmath138 etc . [ [ operator - splitting . ] ] operator splitting . + + + + + + + + + + + + + + + + + + + we rewrite equation ( [ eq : ezfnum - rda ] ) in the form @xmath139 + { \mathcal{h}}[{{\mathbf{v}}};{{\vec r}},\theta ] + { \mathcal{a}}[{{\mathbf{v}}};{{\vec c}},\omega]\ ] ] where differential operators @xmath140 , @xmath141 and @xmath142 are defined as @xmath143 = { { \mathbf{d}}}\nabla^2{{\mathbf{v}}}+ { { \mathbf{f}}}({{\mathbf{v } } } ) , { \label{eq : rdpart } } \\ & { \mathcal{h}}[{{\mathbf{v}}};{{\vec r}},\theta ] = { { \mathbf{h}}}({{\mathbf{v}}},e^{{\hat{\boldsymbol{\gamma}}}\theta}\cdot\nabla{{\mathbf{v } } } , { { \vec r}}+e^{{\hat{\boldsymbol{\gamma}}}\theta}{{\vec r}},t ) , { \label{eq : ppart } } \\ & { \mathcal{a}}[{{\mathbf{v}}};{{\vec c}},\omega ] = ( { { \vec c}}\cdot\nabla){{\mathbf{v}}}+ \omega{\frac{\partial{{{\mathbf{v}}}}}{\partial{\theta } } } \nonumber \\ & \phantom{{\mathcal{a}}({{\mathbf{v}}};{{\vec c}},\omega ) } = ( c_x-\omega y)\,{\frac{\partial{{{\mathbf{v}}}}}{\partial{x } } } + ( c_y+\omega x)\,{\frac{\partial{{{\mathbf{v}}}}}{\partial{y } } } . { \label{eq : apart}}\end{aligned}\ ] ] let @xmath144 , @xmath145 and @xmath146 be discretizations of @xmath140 , @xmath141 and @xmath142 . our computations proceed as follows : @xmath147 [ eq : step ] [ [ kinetics . ] ] kinetics . + + + + + + + + + as specific examples , we consider two models , the fitzhugh - nagumo model @xcite : @xmath148 } \mapsto { \left[\begin{array}{cccccc } \alpha^{-1}\left(u - u^3/3-v\right ) \\ \alpha(u+\beta-\gamma v ) \end{array}\right]}\ ] ] and barkley s model @xcite : @xmath148 } \mapsto { \left[\begin{array}{cccccc } c^{-1}u(1-u)\left(u-(v+b)/a\right ) \\ u - v \end{array}\right]}\ ] ] both with @xmath149}$ ] . [ [ reaction - diffusion - step . ] ] reaction - diffusion step . + + + + + + + + + + + + + + + + + + + + + + + + the computational scheme is designed as an extention to the standard approach to simulation of spiral waves . specifically , we chose barkley s ez - spiral @xcite as the starting point , and extended it to add the other computational steps . so the reaction - diffusion step ( [ eq : rdstep ] ) is as implemented in ez - spiral , with central 5-point difference approximation of the laplacian , without any features specific to the barkley model , such as implicit treatment of the kinetic terms , and with appropriate modifications when fitzhugh - nagumo model is used . [ [ perturbations . ] ] perturbations . + + + + + + + + + + + + + + we consider one particular type of nonzero perturbation , the electrophoresis , @xmath150 where @xmath151 is a diagonal matrix , @xmath152}$ ] , @xmath153 . for a reaction - diffusion system this perturbation can describe movement of the reagents in response to electric field with velocities @xmath154 and @xmath155 along the @xmath156-axis . for @xmath157 , this perturbation can also approximately describe the movement of an axially symmetric scroll ring . for a cylindrical system of coordinates @xmath158 : @xmath159 , @xmath160 , @xmath161 , the diffusion term has the form @xmath162 , which for @xmath163 and large @xmath164 is equivalent to an unperturbed diffusion term with a 2-dimensional laplacian in @xmath165 plane plus a small perturbation @xmath166 . if the filament of the scroll is located at large values of @xmath167 and as the dynamics of the scroll is mostly determined by the events near its filament , then @xmath168 can be approximately replaced with @xmath169 . perturbation ( [ eq : perturbation ] ) violates only rotational symmetry of the problem , preserving symmetry with respect to translations in space and time . hence @xmath170 explicitly depends only on @xmath106 . this limitation is not principal and translation symmetry breaking perturbations can be considered similarly , in which case @xmath170 would also explicitly depend on @xmath171 and/or @xmath36 . we discretize the first spatial derivatives in the perturbation term using upwind second - order accurate differences , and use explicit euler timestepping . in the absence of perturbations , @xmath8 , the perturbation step ( [ eq : pstep ] ) is of course omitted and @xmath172 . [ [ tip - definition - and - pinning - conditions . ] ] tip definition and pinning conditions . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + discretization of the pinning conditions ( [ eq : ezfnum - pina ] ) , ( [ eq : ezfnum - pinb ] ) , using @xmath173 , and the right - side first - order discretization of the @xmath156-derivative , gives [ eq : musteps ] @xmath174 where @xmath175 are grid coordinates of the origin . this works in principle , but gives rather inaccurate and noisy approximations for @xmath176 , which get worse for finer discretizations . this is typical for numerical differentiation . we overcome this by enhancing the spatial discretization step , by replacing the condition ( [ eq : mustep3 ] ) with @xmath177 where the grid point @xmath178 was chosen some way away from the centre point , @xmath179 . this means replacing the third pinning condition ( [ eq : ezfnum - pinb ] ) with @xmath180 where @xmath181 . empirically , we have found that the length of the displacement @xmath182 should be of the order of , but not exceeding , one full wavelength of the spiral . this revised orientation - pinning condition still does not define the position uniquely , as illustrated by fig . [ fig : ezfnum - unique1 ] . an extra inequality is required to distinguish between different solutions satisfying conditions ( [ eq : mustep1],[eq : mustep2],[eq : mustep3a ] ) . we use @xmath183 corresponding to @xmath184 specifically , we chose @xmath185 and @xmath186 . conditions ( [ eq : mustep4 ] ) and ( [ eq : ezfnum - pind ] ) then mean that the third pinning condition ( [ eq : mustep3a],[eq : ezfnum - pinc ] ) ensures that the front , rather than the back , of the excitation wave passes though the grid point @xmath178 . so equations ( [ eq : ezfnum - pina],[eq : ezfnum - pinc],[eq : ezfnum - pind ] ) , with their discretizaions ( [ eq : musteps],[eq : mustep4 ] ) are our `` choice 2 '' pinning conditions . the choice 1 and choice 2 pinning conditions define different rms and different quotient data @xmath187 , for the same solution @xmath188 . however , the two fors they define have a common origin and differ only by the orientation angle . so if @xmath127 are quotient data for choice 1 pinning conditions , and @xmath189 are quotient data for choice 2 pinning conditions , then we have @xmath190 where @xmath123 the tip orientation angle in the choice 2 comoving for , so @xmath191 is angle of one for against the other . [ fig : ezfnum - unique1 ] [ [ advection . ] ] advection . + + + + + + + + + + we use an upwind second - order accurate approximation of the spatial derivatives in @xmath146 . the steps ( [ eq : astep ] ) and ( [ eq : mustep ] ) are done in conjunction with each other . the discretization of @xmath192 at the tip pinning points , resulting from ( [ eq : astep ] ) , is used in the three equations ( [ eq : mustep ] ) to find the three unknowns @xmath193 , @xmath194 and @xmath195 , so that the pinning conditions ( [ eq : mustep ] ) are always satisfied exactly ( to the processor precision ) after every step . [ [ boundary - conditions . ] ] boundary conditions . + + + + + + + + + + + + + + + + + + + + since the boundaries in the comoving for do not represent any physical reality but are only a necessity of numerical approximation , the results can only be considered to be reliable if they do not depend on the boundary conditions . so we use both dirichlet and neumann bondary conditions and compare the results . for dirichlet conditions , we use boundary values of the resting state @xmath196 , such that @xmath197 . [ [ tip - trajectory - reconstruction . ] ] tip trajectory reconstruction . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the steps ( [ eq : thetastep ] ) and ( [ eq : rstep ] ) are simple first - order implementations of the corresponding odes . the resulting @xmath198 is used in calculations of the @xmath141 step when the perturbation is on . otherwise , @xmath198 and @xmath199 are calculated only for the record . [ [ some - details - of - software - implementation . ] ] some details of software implementation . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for stability purposes , we ensure that the following inequalites are observed during computations , @xmath200 and @xmath201 found in ( [ eq : astep ] ) and ( [ eq : mustep ] ) are beyond these limits then they are restricted to the intervals stated above . also , we eliminated the need to restrict the values of @xmath202 and @xmath201 to their stability limits by moving the spiral wave solution so that the tip of the spiral wave is in the center of the box , using the standard ez - spiral s ` mover ' function , which performs translation of the solution by an integer number of grid steps , suitably extrapolating the solution where necessary near the boundaries . for @xmath176 , we implemented the restriction that if @xmath203 exceeded its maximum stability value , then then we set @xmath204 . effectively this means that unless the orientation of the spiral wave is already very near the standard orientation satisfying equation ( [ eq : mustep3a ] ) and inequality ( [ eq : mustep4 ] ) , the code computes a solution of the problem @xmath205 [ eq : ezfnumt ] instead of ( [ eq : ezfnum ] ) . that is , it performs reduction by the subgroup of translations of the euclidean group . a typical run of the program in the interactive mode starts from obtaining a spiral wave solution in the standard `` ride - off '' mode , by solving initial - value problem ( [ eq : rdeu ] ) . when the spiral wave is initiated so there is one tip in the solution , the user switches the program to the `` ride - on '' mode , with calculations according to the above scheme . on the switch , the program first of all moves the tip of the spiral to the centre of the box via ez - spiral s ` mover ' function , by parallel translations of the solution , supplementing the missing pieces near boundaries by duplicating the existing boundary values . from then on , the spiral continues to rotate with its tip fixed at the centre of the box , thus solving the problem ( [ eq : ezfnumt ] ) . in this regime , only the first two pinning conditions are satisfied , and only @xmath202 and @xmath201 are calculated and used , where as @xmath176 is calculated but replaced with zero , until it falls within the stability limit and the fourth inequality - type pinning condition is satisfied . from that point , the program proceeds in the fully engaged mode , calculating the problem ( [ eq : ezfnum ] ) . [ sec : examples ] first we illustrate how our approach works using two examples . one example uses barkley model with rigidly rotating spiral waves , and the other is fitzhugh - nagumo model with meandering spiral waves . , @xmath206 , @xmath207 , @xmath208 , @xmath209 , @xmath210 , @xmath211 . the runs @xmath212 $ ] and @xmath213 $ ] are direct simulations . the run @xmath214 $ ] is a quotient system simulation , the pinning points are indicated by small white crosses . the third pinning condition is engaged at @xmath215 . ] [ fig : bkl - movie ] figure [ fig : bkl - movie ] illustrates the work of ezride in the case of a rigidly rotating spiral wave . the panels represent three consecutive runs , in different regimes : the `` direct numerical simulations '' ( dns ) of system ( [ eq : rdeu ] ) , then the `` skew - product '' calculation in the comoving for , and then again the dns in the laboratory for . the skew - product calculation in turn consists of two parts . the first part is described by ( [ eq : ezfnumt ] ) where only the two translation pinning conditions are engaged , so that the position of the tip of the spiral is fixed , but not its orientation , so the for is co - translating but not co - rotating . the second part is where all four pinning conditions are engaged , and the for is co - translating and co - rotating . it is seen from fig . [ fig : bkl - movie ] , that after a transient period , the solution in the fully comoving for becomes stationary . this corresponds to the definition of a rigidly rotating spiral wave as a relative equilibrium . , @xmath216 , @xmath217 , @xmath218 , @xmath219 , @xmath220 , @xmath221 . the runs @xmath222 $ ] and @xmath223 $ ] are direct simulations . the run @xmath224 $ ] is a quotient system simulation , the pinning points are indicated by small white crosses . the third pinning condition is engaged at @xmath225 . ] [ fig : fhn - movie ] figure [ fig : fhn - movie ] shows a similar set of runs for a different case , where the spiral wave is not stationary but is meandering . in this case , the solution in the comoving for is not stationary , but periodic in time . this corresponds to the definition of a meandering spiral wave as a relative periodic orbit . and ( b ) simulation shown in fig . [ fig : fhn - movie ] . the pieces labelled 1 are trajectories obtained in direct simulations in the laboratory for . the pieces labelled 2 are trajectories obtained via co - translating simulations , with first two pinning conditions engaged . the pieces labelled 3 correspond to co - moving ( co - translating and co - rotating ) simulations with all three pinning conditions engaged . the final pieces labelled 4 correspond to direct simulations in a non - moving for , which has been displaced with respect to the laboratory for during the quotient system simulations . ( c ) same as ( a ) , with @xmath226 , @xmath227 . ( d ) same as ( b ) , with @xmath226 , @xmath227 . ] [ fig : clips ] figure [ fig : clips](a , b ) show selected pieces of tip trajectories obtained as a result of the runs shown in fig . [ fig : bkl - movie ] and fig . [ fig : fhn - movie ] . the discretization steps there are deliberately chosen crude , to allow very fast running simulations , and also to illustrate the difference introduced by the change of method of computation . the tip trajectories obtained by reconstruction from the quotient data are qualitatively similar to the tip trajectories obtained in dns . however , the quantitative difference is also quite evident . in the case of rigid rotation , the reconstructed trajectory radius is noticeably bigger than that from dns , and the centres of the meandering patterns in different runs are offset against each other . as panels ( c , d ) in the same figure show , these discrepancies decrease when the discretization steps are refined . ) . in ( b d ) , the projections of the quotient data are shown , which for the dns curves are obtained by numerical differentiation of the tip trajectory , using ( [ eq : quotdiff ] ) . ] [ fig : mean ] figure [ fig : mean ] shows the tip and quotient system trajectories , obtained in laboratory and comoving for calculations , for a meandering spiral . this is drawn for the finer discretization steps , as in fig . [ fig : clips](d ) . for comparison , quotient data for both the laboratory and comoving for calculations were recalculated for the choice 1 pinning conditions using ( [ eq : gauge ] ) . there is good agreement between the two methods of calculations , within the expected accuracy . more detailed analysis of the numerical accuracy of our method is given in the next session . [ sec : convergence ] [ fig : convergence ] figure [ fig : convergence ] illustrates the convergence of the results of calculations of rigidly rotating spiral , using ezride with neumann and dirichlet boundary conditions , and dns using neumann boundary conditions . in these calculations , the box size is fixed at @xmath228 and the timestep is changed with the spacestep so that @xmath229 . for @xmath230 dependence , we also show the angular velocity measured in direct numerical simulations . we do not show @xmath231 found in dns , since obtaining it involves numerical differentiation which gives accuracy insufficient for the convergence study . our discretizations are second order accurate in @xmath129 and first order accurate in @xmath128 both in dns and in the riding mode , which corresponds to linear dependence of any results on @xmath232 for @xmath233 . we see in fig . [ fig : convergence ] that this is indeed the case . linear extrapolation of the @xmath230 gives the values of @xmath234 for laboratory and comoving calculations coinciding to within @xmath235 . one of the advantages of ezride is the fact that the simulations can be done in a smaller box compared to dns . so , the last test is convergence in box size . we have calculated the rigidly rotating spiral by ezride at fixed @xmath236 , @xmath237 and @xmath238 varying through @xmath239 $ ] and found that both @xmath240 and @xmath203 vary by less than @xmath235 . [ sec : resonance ] one of the cases where the dns would meet with difficulties , is the study of the the meandering of spiral waves for parameters near the `` 1:1 resonance '' between the euclidean and the hopf frequencies . this case is marginal between meandering patterns with inward petals and outward petals . near the resonance , the spatial extent of the meandering trajectory becomes large , and for the case of exact resonance , infinite , and the spiral appears to be spontaneously drifting @xcite . hence , following the dynamics of the spiral wave in the comoving for presents an advantage . we illustrate this using the fhn model . we fix the discretization parameters at @xmath241 , @xmath242 and @xmath208 . the choice of model parameter is influenced by winfree s `` flower garden '' @xcite , which gives a rough estimate for the location of the 1:1 resonance line in the @xmath243 plane at @xmath217 . using this information , we have selected two values @xmath244 and @xmath245 , and scanned values of @xmath246 across the resonance value , which we determined as @xmath247 for @xmath244 , and @xmath248 for @xmath245 at our discretization parameters . , @xmath217 and varying @xmath246 . ] [ fig : eps02-tip ] [ fig : eps02-quot ] the results are presented on figures [ fig : eps02-tip][fig : eps025-quot ] . the shape of trajectories is well known from the theory , and is outward petals for @xmath249 and inward petals for @xmath250 , degenerating into spontaneous straightforward drift at @xmath251 . the trajectory at @xmath251 in fig . [ fig : eps02-tip ] is shown twice : once for the whole duration as it was calculated , fig . [ fig : eps02-tip](c ) , and then a close - up of small part of it , fig . [ fig : eps02-tip](d ) . calculation of this particular trajectory using dns would require , by our estimate , about five weeks , as opposed to 2.5 hours used by ezride . the change of the quotient dynamics with changing @xmath246 is illustrated in fig . [ fig : eps02-quot ] . as opposed to the tip trajectories , there is no evident qualitative changes in the shape of the limit cycle across @xmath251 . note the very elongated shape of the limit cycles in all three projections . we do not know whether this has some theoretical explanation or is merely incidental . , for @xmath245 . ] [ fig : eps025-tip ] [ fig : eps025-quot ] the parametric line @xmath245 exhibits similar behaviour , as shown in fig . [ fig : eps025-tip ] and [ fig : eps025-quot ] . this is closer to the hopf bifurcation line in the quotient system , called @xmath252 line in @xcite . correspondingly , the size of the limit cycles in the quotient system is smaller and they become more oval - shaped note that the scale of @xmath201-axis is disproportionately stretched in fig . [ fig : eps025-quot ] , the hopf central manifold appears to be nearly orthogonal to that axis . again , there is no qualitative change in the quotient system dynamics when crossing the 1:1 resonance . [ sec : largecore ] another example where the spatial extent of the spiral wave dynamics is large is the vicinity of winfree s `` rotors boundary '' @xmath253 in the parametric space @xcite . in the vicinity of this boundary , the period of rotation and the radius of the core of the spiral wave grow infinitely . there are at least two different asymptotic theories , based on different choice of small parameters , which aim to describe the vicinity of @xmath253 . hakim and karma @xcite have developed a `` free - boundary '' asymptotic theory applicable to fitzhugh - nagumo type models in the limit @xmath254 or @xmath255 in terms of our chosen kinetics , where angular velocity @xmath176 typically decreases as @xmath256 where @xmath257 is a parameter of the model such that @xmath258 corresponds to the @xmath253 boundary . elkin @xcite obtained an alternative asymptotic based on assumptions which were not restricted to kinetics of any particular kind , but which were not directly validated . their prediction was @xmath259 further analysis has suggested that these two alternatives are not actually antagonistic and may be even observed in the same system in different parametric regions @xcite . reliably distinguishing between the two asymptotics is challenging for dns as it requires a rather close approach to the critical point @xmath258 , which is not known _ a priori _ , implying large tip trajectory radii and correspondingly significant computational resources . in here we present an example of studying this dependence using calculations in the comoving for , which is free from the above complication , as it can be performed within the box of fixed size for all @xmath257 . for this study , we use barkley s model with varying parameter @xmath257 chosen to be @xmath260 , varying from @xmath261 downwards with step @xmath262 until @xmath263 , with other parameters fixed at @xmath206 and @xmath207 . the discretization parameters are @xmath218 , @xmath241 , @xmath242 and @xmath264 . . the arrows indicate the direction of the vector @xmath265 . ] [ fig : lc - snaps ] and @xmath266 of the relative equilibria , for different @xmath267 as indicated . on panel ( a ) , the symbols correspond to the selected values of @xmath260 used in fig . [ fig : lc - snaps ] . ] [ fig : lc - curves ] selected stationary solution obtained in this way are illustrated in fig . [ fig : lc - snaps ] , and the graphs of @xmath268 and @xmath266 are shown in fig . [ fig : lc - curves ] . we compare the features of the observed solutions with those that are given by the two asymptotic theories @xcite , and observe that 1 . [ finger ] there is a critical value of the parameter @xmath269 , at which the behaviour of the solution changes qualitatively . at @xmath270 , we observe a nearly straight broken excitation wave . [ rotating ] for @xmath271 , the solutions are spiral waves , that is , broken excitation wavelets , which become less and less convex as @xmath272 , and have macroscopic angular velocity which however diminishes in the same limit . 3 . [ translating ] for @xmath273 the solutions are retracting nearly straight but slightly concave wavelets , with very small angular velocity . 4 . [ fingerrate ] for @xmath270 , the direction of movement of the tip seems approximately orthogonal to the overall orientation of the wave itself . [ retraction - rate ] for @xmath273 , the vertical component of vector @xmath265 depends on @xmath260 in a way which is consistent with the asymptotic @xmath274 , see fig . [ fig : lc - curves](e , f ) . since the overall orientation of the wavelets , as seen in fig . [ fig : lc - snaps](a c ) , is nearly vertical we can take @xmath201 as a crude estimate of the `` global tip growth rate '' as defined in @xcite . [ rotation - velocity ] for @xmath271 , the angular velocity of solutions depends on @xmath260 in a way which is consistent with the asymptotic @xmath275 , see fig . [ fig : lc - curves](b ) but not @xmath276 , see fig . [ fig : lc - curves](c ) . all these observations are in agreement with the theory in @xcite and can be used to empirically distinguish between the elkin asymptotics ( corresponding to the `` i / v '' parametric boundary in @xcite ) and hakim - karma asymptotics ( respectively , `` j / c '' boundary in @xcite ) . feature [ finger ] is inconclusive : existence of a critical solution , called `` critical finger '' by hakim and karma , is common for both j / c and i / v boundaries , but the shape of this solution is different . it is asymptotically linear for i / v boundary , and asymptotically logarithmic for j / c boundary . looking at fig . [ fig : lc - snaps](d ) and considering the effect of the boundary conditions , it is not clear which case is nearer to the observed reality . feature [ rotating ] is common for i / v and j / c boundaries . the phenomenological difference is that spirals close to i / v boundary can be `` growing '' or `` shrinking '' , while spirals close to j / c boundary can only be `` growing '' . the movement of the tip in fig . [ fig : lc - snaps](d f ) seems approximately orthogonal to the orientation of the wavelet near the tip , which is consistent with both cases . feature [ translating ] tips the balance in favour the i / v boundary since the broken wavelets are concave . according to @xcite , the translating waves near an i / v boundary should be concave , and those near an j / c boundary should be convex . feature [ fingerrate ] is common for i / v and j / c , as in both cases the critical fingers should have zero `` global growth rate '' . feature [ retraction - rate ] is common for i / v and j / c boundaries . feature [ rotation - velocity ] is , in our opinion , a convincing evidence in favour of an i / v boundary , since according to @xcite , near i / v boundary the dependence @xmath277 is linear , whereas near j / c boundary it is @xmath278 . an unequivocal interpretation of all theoretical predictions in the view of our present numerical results would require further investigation , as the asymptotics of @xcite operate with a `` crest line '' of an excitation wave . there is no obvious operational definition of this line which would be valid up to the tip , and some of the predictions concern the mutual orientation of this line and the tip velocity . however the predictions that can be unambiguously interpreted , seem to indicate that for the model considered here , we have the case of i / v boundary , elkin asymptotics , rather than j / c boundary corresponding to hakim - karma asymptotics . the last observation here is that of the small angular velocity @xmath176 calculated for the `` retracting waves '' at @xmath273 , seen on fig . [ fig : lc - snaps](a c ) . as we already noted , the smallness of these @xmath176 values is consistent with the theoretical prediction of translating but not rotating waves . however when these values are magnified , we observe that they demonstrate a peculiar power law @xmath279 where @xmath280 , see fig . [ fig : lc - curves](d ) . a theoretical explanation of this requires further study ; it is clear , however , that @xmath176 in this area is strongly affected by the boundaries , as the curves for @xmath218 and @xmath281 differ quite significantly . [ sec : drift ] finally , we illustrate calculation of the movement of spiral waves in a perturbed reaction - diffusion system . we consider fitzhugh - nagumo kinetics at the same parameters as in fig . [ fig : fhn - movie ] , and add to it the `` electrophoresis '' perturbation ( [ eq : perturbation ] ) in the right - hand side , with @xmath157 . [ fig : fhndr ] results of the simulations are presented in fig . [ fig : fhndr ] . the unpertubed spiral waves for these parameters are meandering , so with the perturbation present , we observe meandering with drift . the drift proceeds with a constant average velocity , which is consistent with the fact that the perturbation violates only the rotational but not the translational symmetry of the problem . the average drift is to the left , which corresponds to a collapsing scroll wave . so at these parameter values , the scroll waves have positive tension , inasmuch as this concept can be applied to meandering scrolls . in the calculations in the laboratory for , the time during which the drift can be observed is limited , as when the spiral reaches the left boundary , it terminates . in the comoving for , this drift can be observed indefinitely . comparing the traces in fig . [ fig : fhndr ] we see that although , as we know from figures [ fig : clips ] and [ fig : mean ] , the discretization is too crude to give quantitative agreement between laboratory and comoving calculations in detail , the drift velocities obtained in these two ways are very similar . we illustrate the relative advantages of the two methods of calculation by comparing the computation costs . the laboratory for simulation , for @xmath218 and @xmath282 $ ] has taken 325 sec ( the spiral has annihilated at the left at @xmath283 ) . the time taken by the comoving for simulation for the same boxsize @xmath238 and the same @xmath36 interval is 462 sec , is naturally somewhat longer due to the extra effort required for the advection term calculations . however , the comoving for calculation proceeded unabated where the laboratory for calculation failed due to annihilation with the border . to continue the laboratory for calculation to the same extent we would have to increase the box size @xmath238 with a corresponding increase in computation cost . moreover , virtually the same result , as far as drift velocity is concerned , can be obtained by comoving for calculation with @xmath208 , and it takes only 202 sec . of course the drift in the laboratory for with @xmath208 would terminate even earlier . [ sec : discussion ] we have described a numerical method of solving a reaction - diffusion system of equations describing a spiral wave , in a frame of reference which is moving with the tip of that wave . we have shown the method can provide accurate solutions , and that there are applications where the computational cost of our method can be considerably lower than that of the conventional approach , or the conventional approach is just inapplicable . as always , the computational advantages are particularly essential in case of parametric studies , for which the method is well suited . although the applications were chosen just to provide some meaningful examples of use of the method , the results obtained there can be of scientific value themselves . so , we have investigated the vicinity of the `` 1:1 resonance '' manifold in the parametric space , which corresponds to spontaneous drift of spirals , and which separates meandering patterns with outward petals and inward petals . henry @xcite has proposed a theory which implies that this manifold coincides or is an analytical continuation of the manifold where the filament tension of scroll waves vanishes . there are reports in literature confirming that change of sign of filament tension is associated with change from outward to inward petals in meandering patterns , but also examples where there are no such correlation , @xcite and references therein . our simulations indicate that as far as orbit manifold dynamics of the spiral is concerned , the 1:1 resonance is not characterized by any special features . hence any special features of this resonance ought to be due to the euclidean extension of the orbifold dynamics . since scroll filament tension can also be defined via properties of the spiral wave solutions within the comoving for , any genetic and generic relationship between the two manifolds seems unlikely ( but , of course , can not exclude the possibility of such relationship in some special cases ) . we have also investigated the vicinity of the `` @xmath253 '' manifold in the parametric space , which has provided a strong evidence towards one of the two theoretical possible asymptotics , namely elkin @xcite asymptotics as opposed to hakim - karma @xcite asymptotics . it should be noted here that while hakim - karma asymptotic theory was based on assumptions which have been well established , the elkin asymptotic theory was using assumptions , validity of which could not be asserted at that moment . here we have presented firm evidence that elkin asymptotis is not a mere theoretical possibility but is indeed observed in reality ( see also @xcite and a discussion below ) . a direct confirmation would be via calculation of the `` response functions '' , critical eigenfunctions of the adjoint linearized operator of the critical finger solution . this would require obtaining first a good quality critical finger solution , so the method described here can be a significant step towards this goal , too . finally , we have demonstrated that calculations in the comoving for can be efficiently used to study perturbation - caused drift of spirals , including meandering spirals . although the asymptotic theory of drift of meandering spirals is yet to be developed ( see , however , a preliminary draft of such theory in @xcite ) , we can expect , for instance , that scroll waves in the fitzhugh - nagumo model with the parameters as in fig . [ fig : fhn - movie ] , [ fig : fhndr ] will have `` positive tension '' , tend to collapse , rather than develop a scroll wave turbulence . the advantage of calculating drift in the comoving for , apart from computation cost , is absence of `` pinning '' effects of spatial discretization , both in terms of discrete space steps and discrete spatial directions , on the drift . our approach can be compared to the approach proposed by beyn and thummler ( bt ) @xcite . bt use a similar mathematical idea of decomposing the evolution of the nonlinear wave into the motion of the wave and evolution of its shape , which in the functional space appears as decomposition into motion along and across the euclidean symmetry group orbits . but there are also differences . there are technical details of implementations which are probably of lesser importance , such as choice of polar vs cartesian grid , central vs upwind discretization of spatial derivatives and explicit vs semi - implicit discretization in time . more significant differences are in the `` phase conditions '' they use , which play the same role as , but are qualitatively different in nature from , our `` pinning conditions '' . one aspect is that the phase conditions involve integral functionals . we show here that this is not necessary , and local conditions like ( [ eq : musteps ] ) are simpler . the other aspect is the one we discussed in section [ sec : maths ] : the bt phase conditions appear to be well suited for calculation of relative equilibria ( rigidly rotating spirals ) but not necessarily for relative periodic solutions ( meandering spirals ) . further , the phase conditions proposed by bt were not intended for use with symmetry breaking perturbations that produce drift of spirals . and indeed , bt comment in their paper that `` it seems quite a challenging task to freeze drifting spirals or recognize meandering spirals as periodic orbits . '' as we have demonstrated , our approach works both for meandering spirals and for drifting spirals . after completing this study we became aware of a work by hermann and gottwald ( hg ) @xcite who also investigated the large core limit , using a further development of the bt method . hg have paid a great deal of attention to refining the boundary conditions so as to minimize the effect of boundaries onto the quotient dynamics . this has allowed them , in particular , to verify the linear scaling law ( [ eq : ebhlaw ] ) for seven decades of variation of @xmath203 , compared to mere one decade as shown in fig . [ fig : lc - curves ] . notice that as shown in the same figure , our progress towards smaller values of @xmath203 is limited precisely by the influence of boundaries . hg also have explicitly addressed the issue of the numerical stability of the computations , which we treat in this study purely empirically . we believe that combining the advantageous features of the approach developed by bt and hg , and the one proposed here , is an interesting topic for future work , which may yield further results about spiral wave dynamics , that are not possible , or very difficult , to obtain by direct numerical simulations . this study has been supported in part by a ph.d studentship from the university of liverpool and by epsrc grant ep / d074789/1 . the preliminary stage for this study was done in collaboration with b.n . vasiev , who proposed the idea of solving for quotient data simultaneously with the advection substep . vnb is also grateful to a.m. pertsov and g.a . gottwald for inspiring discussions , and to g.a . gottwald also for informing us about the results of @xcite prior to publication .

we present mid - infrared spectroscopy of a sample of 16 optically faint infrared luminous galaxies obtained with the infrared spectrograph ( irs ) on the . these sources were jointly selected from and imaging surveys in the ndwfs botes field and were selected from their bright x - ray fluxes to host luminous agn . none of the spectra show significant emission from polycyclic aromatic hydrocarbons ( pahs ; 6.2 equivalent widths.2 ) , consistent with their infrared emission being dominated by agn . nine of the x - ray sources show 9.7 silicate absorption features . their redshifts are in the range , implying infrared luminosities of log(l)=12.5.6 . the average silicate absorption strength is not as strong as that of previously targeted optically faint infrared luminous galaxies with similar mid - infrared luminosities implying that the x - ray selection favors sources behind a smaller column of si - rich dust than non - x - ray selection . seven of the x - ray sources have featureless power - law mid - ir spectra . we argue that the featureless spectra likely result from the sources having weak or absent silicate and pah features rather than the sources lying at higher redshifts where these features are shifted out of the irs spectral window . we investigate whether there are any correlations between x - ray and infrared properties and find that sources with silicate absorption features tend to have fainter x - ray fluxes and harder x - ray spectra , indicating a weak relation between the amount of silicate absorption and column density of x - ray - absorbing gas .

0801.4578

mid - infrared spectroscopy is a useful tool in investigating the extent to which starbursts and agn contribute to the luminosity of luminous infrared galaxies ( lirgs ; @xcite ) . the infrared spectra of local lirgs have been studied extensively with isophot , isocam and sws on the @xmath9 @xmath10 @xmath11 ( iso ; e.g. , @xcite ; @xcite ; @xcite ; @xcite ) and more recently with the infrared spectrograph ( irs ; @xcite ) on the @xmath12 ( @xcite ; @xcite ; @xcite ; @xcite ) . @xmath1 irs has also extended these studies to fainter sources at higher redshifts . a particularly interesting population of high redshift optically faint infrared luminous sources ( @xcite ; @xcite ; @xcite ; @xcite ) has been discovered using the multiband imaging photometer ( mips ; @xcite ) . various observing programs with @xmath1 irs have found that mips sources at flux density levels of f@xmath13(24@xmath3 ) @xmath14 0.8 mjy with optical magnitudes @xmath15 @xmath16 24 are typically at z @xmath17 2@xmath73 ( @xcite ; @xcite @xcite ; @xcite ) . the infrared spectra exhibit a wide range of properties from deep 9.7@xmath3 silicate absorption features and/or strong polycyclic aromatic hydrocarbon ( pah ) emission features to featureless power - law continua . this obscured population is important in determining how much light originates from starbursts and agn at epochs in the universe when their luminosity density was at its maximum . given that their optical faintness makes optical follow - up difficult , understanding how to characterize these sources using their infrared spectra is important in understanding their energetics . one approach in understanding the contribution of starbursts and agn to the infrared emission of optically faint infrared sources is to investigate the infrared spectra of sources also selected in other wavebands . for example , at one extreme we can study sources selected on the basis of large far - infrared flux densities , which we might expect to contain cool dust and show signatures of starbursts . in @xcite , we presented the irs spectra of 11 70@xmath3-selected optically faint infrared luminous galaxies and found that seven had large 6.2@xmath3 pah equivalent widths , consistent with them being powerful starburst galaxies . four of the galaxies showed deep silicate absorption features and their properties suggested that agn are likely to be the dominant origin of their large mid - infrared luminosities . at the other extreme , we can consider sources selected on the basis of x - ray flux , which we might expect to show spectral signatures of agn . @xcite presented irs spectra of 9 x - ray - loud optically faint infrared luminous galaxies selected from the @xmath1 wide - area infrared extragalactic survey ( swire ; @xcite ) . eight sources have silicate absorption features and one source has a featureless power - law mid - ir spectrum . in this paper , we present @xmath1 irs spectra of 16 optically faint infrared galaxies in the botes field of the noao deep wide - field survey ( ndwfs ; @xcite ) which are selected to be x - ray loud . this significantly increases the number of x - ray loud optically faint infrared luminous galaxies thus far studied , and extends the sample to higher x - ray and 24@xmath3 luminosities . a cosmology of @xmath18 , @xmath19=0.3 , and @xmath20=0.7 is assumed throughout . our sample is selected from the botes field of the noao deep wide - field survey ( ndwfs ; @xcite ) . the combination of the large survey area ( 8.2 deg@xmath21 with overlapping mips and optical coverage ) and existing @xmath1 mips and irac , @xmath2 , and deep optical imaging data makes this an ideal data - set for multi - wavelength comparisons of sources . the mips data were obtained with an effective integration time at 24@xmath3 of @xmath1790s per sky pixel , reaching a 5 @xmath22 detection limit of @xmath17 0.3 mjy for unresolved sources . the resulting source lists contain @xmath2328,000 sources with f@xmath13(24@xmath3)@xmath240.3 mjy . the xbotes survey was undertaken using acis - i on the @xmath25 with 5 ks extosures over the entire botes field ( @xcite ; @xcite ; @xcite ) . it comprises 3293 point sources with 4 or more x - ray counts , corresponding to a limiting flux of f@xmath26=8.1@xmath27 . an 8.5 deg@xmath28 region of the botes field has been mapped in all four @xmath1 irac bands by the irac shallow survey @xcite , reaching 5@xmath22 limits of 6.4 , 8.8 , 51 , and 50 @xmath29 at 3.6 , 4.5 , 5.8 , and 8 @xmath3 , respectively . using the xbotes and mips 24@xmath3 catalogs , we selected a sample of 16 x - ray loud sources in the botes field for @xmath1 irs observations . we selected all 13 of the sources with @xmath244 x - ray counts ( f@xmath30 ) , f@xmath13(24@xmath3)@xmath24 2 mjy , and @xmath31 mag ( the `` main '' sample ) . we supplemented this sample with 3 extra sources ( bootesx13 , 14 , and 15 ) which were slightly fainter at 24 @xmath3 ( f@xmath13(24@xmath3)@xmath241 mjy ) but which were particularly powerful x - ray sources ( hereafter denoted as the `` x - ray bright '' sub - sample with 48 , 34 , and 19 x - ray counts corresponding to f@xmath32= 9.2 , 6.5 , and 3.7 @xmath33 respectively ) . because of their different selection criteria , we identify the x - ray bright sub - sample separately throughout the analysis and consider the conclusions with and without their inclusion . the basic optical and infrared properties of the sample are shown in table [ tab:70sample ] . 10/16 of the sources have fairly red optical to infrared colors and qualify as dust obscured galaxies ( dogs ) based in the criterion ( @xmath15-[24]@xmath2414 ) presented in @xcite . none of our sample have redshifts from optical spectroscopy . the x - ray properties are presented in table [ tab : xabs ] . at @xmath34 , the x - ray luminosities are far larger than can be attributed to a starburst and must be attributed to an agn . at @xmath35 ( the lowest measured source redshift ; see below ) , the minimum x - ray flux , f@xmath36 corresponds to an x - ray luminosity of @xmath37 ergs s@xmath38 . if this was all due to star - formation , it would imply an unfeasibly large star - formation rate of @xmath23 5000 m@xmath39 yr@xmath38 @xcite . llllllllllllll bootesx1 & sst24j143048.3 + 322532 & 24.00 & 23.11 & 22.33 & @xmath7 & 0.13 & 0.22 & 0.77 & 0.68 & 3.15 & @xmath412 & @xmath4120 & 14.70 + bootesx2 & sst24j142731.3 + 324434 & 25.54 & 23.04 & 21.51 & @xmath7 & 0.09 & 0.12 & 0.20 & 0.59 & 2.61 & @xmath415 & @xmath4100 & 14.42 + bootesx3 & sst24j142609.0 + 333303 & 23.33 & 22.80 & 21.84 & 18.29@xmath40 & 0.05 & 0.06 & 0.10 & 0.18 & 2.49 & @xmath420 & @xmath4120 & 14.13 + bootesx4 & sst24j143234.9 + 333637 & 24.53 & 22.96 & 21.53 & 18.65@xmath40 & 0.18 & 0.35 & 0.73 & 1.68 & 2.92 & 30@xmath416&110@xmath4122 & 14.47 + bootesx5 & sst24j142625.5 + 334051 & 25.14 & 23.53 & 22.04 & 18.00@xmath40 & 0.21 & 0.32 & 0.52 & 0.82 & 2.11 & @xmath7 & @xmath4120 & 14.69 + bootesx6 & sst24j143232.6 + 335903 & 23.09 & 22.79 & 22.10 & 17.85@xmath40 & 0.04 & 0.05 & 0.16 & 0.38 & 2.42 & 10@xmath412&80@xmath4116 & 14.09 + bootesx7 & sst24j143038.4 + 340929 & 24.34 & 22.86 & 21.67 & 18.53@xmath40 & 0.06 & 0.10 & 0.17 & 0.35 & 2.40 & 12@xmath412&@xmath4100 & 14.15 + bootesx8 & sst24j143301.5 + 342341 & 23.48 & 22.85 & 22.54 & 18.32@xmath40 & 0.10 & 0.16 & 0.35 & 0.62 & 2.17 & @xmath415 & @xmath4120 & 14.04 + bootesx9 & sst24j142849.0 + 343432 & 23.58 & 22.35 & 21.14 & 18.11@xmath42 & 0.08 & 0.10 & 0.13 & 0.33 & 2.36 & @xmath415 & @xmath4100 & 13.60 + bootesx10 & sst24j143834.9 + 343839 & 22.76 & 22.44 & 22.58 & @xmath7 & 0.06 & -1.0@xmath43&0.25&-1.0@xmath43 & 2.25 & @xmath420 & @xmath4150 & 13.66 + bootesx11 & sst24j143715.1 + 345323 & 24.15 & 23.73 & 23.06 & @xmath7 & 0.16 & 0.27 & 0.49 & 0.95 & 2.51 & @xmath415 & @xmath4120 & 15.07 + bootesx12 & sst24j142537.8 + 351735 & 24.21 & 22.12 & 20.69 & 18.30@xmath42 & 0.09 & 0.12 & 0.21 & 0.38 & 2.28 & 25@xmath415&@xmath4120 & 13.36 + houck13 & sst24j143644.2 + 350627 & 24.71 & 23.69 & 23.36 & 19.11@xmath44 & 0.04 & 0.10 & 0.32 & 0.69 & 2.34 & 9@xmath413@xmath45&43@xmath4112@xmath45 & 14.96 + bootesx13 & sst24j143723.4 + 334308 & 24.96 & 22.78 & 22.49 & @xmath7 & 0.03 & 0.03 & 0.04 & 0.11 & 1.44 & 20@xmath414&@xmath4120 & 13.52 + bootesx14 & sst24j142707.0 + 325213 & @xmath7 & 22.03 & @xmath7 & @xmath7 & 0.19 & 0.16 & 0.19 & 0.25 & 1.22 & @xmath412 & @xmath4100&12.59 + bootesx15 & sst24j143359.0 + 331301 & 23.95 & 22.40 & 21.06 & 17.15@xmath40 & 0.23 & 0.29 & 0.42 & 0.61 & 1.92 & @xmath415 & @xmath4120 & 13.45 + lllllll bootesx1 & cxoxb j143048.3 + 322531 & [email protected] & 1.00@xmath46 & [email protected] & 0.67 & 44.22 + bootesx2 & cxoxb j142731.4 + 324435 & [email protected] & 0.68@xmath47 & [email protected] & 0.74 & 43.63 + bootesx3 & cxoxb j142609.0 + 333304 & [email protected] & 0.20@xmath48 & [email protected] & 0.60 & 44.44 + bootesx4 & cxoxb j143234.9 + 333636 & [email protected] & 0.51@xmath49 & [email protected] & 0.59 & 43.75 + bootesx5 & cxoxb j142625.5 + 334052 & [email protected] & @xmath70.16@xmath50 & [email protected] & 1.07 & 44.23 + bootesx6 & cxoxb j143232.6 + 335902 & [email protected] & @xmath70.44@xmath51 & [email protected] & 1.26 & 44.25 + bootesx7 & cxoxb j143038.5 + 340928 & [email protected] & 0.72@xmath52 & [email protected] & 0.79 & 43.97 + bootesx8 & cxoxb j143301.5 + 342343 & [email protected] & 0.00@xmath53 & [email protected] & 1.07 & 44.95 + bootesx9 & cxoxb j142849.0 + 343432 & [email protected] & 0.67@xmath47 & [email protected] & 0.54 & @xmath7 + bootesx10 & cxoxb j143835.0 + 343839 & [email protected] & 0.11@xmath54 & [email protected] & 0.73 & 44.99 + bootesx11 & cxoxb j143715.1 + 345323 & [email protected] & @xmath70.20@xmath55 & [email protected] & 1.30 & @xmath7 + bootesx12 & cxoxb j142537.8 + 351735 & [email protected] & @xmath70.10@xmath56 & [email protected] & 1.10 & @xmath7 + houck13 & cxoxb j143644.2 + 350626 & [email protected] & 0.43@xmath57 & [email protected] & 1.10 & 44.55 + bootesx13 & cxoxb j143723.5 + 334307 & [email protected] & @xmath70.21@xmath58 & [email protected] & 1.58 & @xmath7 + bootesx14 & cxoxb j142707.0 + 325212 & [email protected] & @xmath70.47@xmath59 & [email protected] & 1.21 & @xmath7 + bootesx15 & cxoxb j143359.1 + 331300 & [email protected] & 0.47@xmath60 & [email protected] & 1.03 & @xmath7 + the spectroscopic observations were made with the irs short low module in order 1 only ( sl1 ) and with the long low module in orders 1 and 2 ( ll1 and ll2 ) , described in @xcite . these orders give low resolution spectral coverage from @xmath178@xmath61 m to @xmath1735@xmath61 m . the integration times for individual sources are given in table [ tab : irs ] . data were processed with version 13.0 of the ssc pipeline , and extraction of the source spectra was done with the smart analysis package @xcite . for details of the observational and data analysis techniques , we refer the reader to @xcite . the irs spectra of all 16 x - ray selected sources are shown in figure [ fig : xspec ] . although the irs spectrum of houck13 was presented in @xcite , it satisfies the selection criteria of the main sample and we include it again here for completeness . 9/13 of the main sample ( and 9/16 of the total `` main + x - ray bright '' sample ) exhibit identifiable 9.7@xmath3 silicate absorption features ( s@xmath62 - 0.4 is defined in table [ tab : irs ] ] ; hereafter described as `` absorption - dominated '' sources and denoted by `` abs '' in table [ tab : irs ] ) . a further 4/13 or the main sample ( and 7/16 of the total `` main + x - ray bright '' sample ) exhibit power - law spectra with no convincing features ( hereafter described as `` power - law '' sources and denoted by `` pow '' in table [ tab : irs ] ) ; either they have only weak silicate absorption features or they are at @xmath63 and the absorption feature does not fall within the observable irs bandwidth . bootesx6 may have a weak silicate emission feature if it is at @xmath64=0.9 . given that this is highly uncertain without other convincing features in the spectrum , we classify bootesx6 as a power - law source . none of the sources show convincing pah emission features , although , given their uncertain or unknown redshifts , there may be weak pah emission in some individual sources . we discuss each class in turn . lllllllllllll bootesx1 & 240&480 & 1.7 & 12.2 & 12.6 & 13.3 & @xmath7 & @xmath70.4 & @xmath40.05 & @xmath41380 & @xmath412.8 & abs + bootesx2 & 360&720 & 0.9 & 11.5 & 11.8 & 12.5 & 3.3 & @xmath71.0 & @xmath40.04 & @xmath4190 & @xmath412.0 & abs + bootesx3 & 480&960 & 2.0@xmath44 & 12.2 & 12.6 & 13.3 & @xmath7 & @xmath70.4 & @xmath40.20 & @xmath41170 & @xmath412.9 & abs + bootesx4 & 360&720 & 1.12 & 12.2 & 12.4 & 13.1 & 2.4 & @xmath71.4 & @xmath40.04 & @xmath4450 & @xmath412.4 & abs + bootesx5 & 480&960 & 1.4 & 12.0 & 12.3 & 13.0 & 2.8 & @xmath70.4 & @xmath40.09 & @xmath4590 & @xmath412.5 & abs + bootesx6 & 360&720 & 0.9 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & pow / em ? + bootesx7 & 480&960 & 1.1 & 11.5 & 11.9 & 12.6 & 6.3 & @xmath70.5 & @xmath40.05 & @xmath4110 & @xmath411.7 & abs + bootesx8 & 480&960 & 2.1 & 12.3 & 12.5 & 13.2 & @xmath7 & @xmath70.6 & @xmath40.05 & @xmath4560 & @xmath412.5 & abs + bootesx9 & 480&960 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & pow + bootesx10 & 480&960 & 2.6 & 12.6 & 12.9 & 13.6 & @xmath7 & @xmath65 & @xmath40.10 & @xmath41670 & @xmath412.9 & abs + bootesx11 & 480&960 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & pow + bootesx12 & 480&960 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & pow + houck13 & 480&720 & 1.95 & 12.3 & 12.5 & 13.2 & @xmath7 & @xmath72.9 & @xmath40.03 & @xmath4420 & @xmath412.3 & abs + bootesx13 & 480&1200 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & pow + bootesx14 & 480&1200 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & pow + bootesx15 & 480&1200 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & pow + + fig . 1 . continued . for the absorption - dominated sources , redshifts are determined primarily from the localized maximum in the irs continuum , or `` hump '' , which is blueward of the absorption feature . this feature is produced by absorption on either side of the feature , and with a possible contribution of 7.7@xmath61 m pah emission in composite sources . we determine redshifts by assuming this hump has a rest - frame wavelength of 7.9@xmath61 m ( see average spectra in @xcite and @xcite ) . comparison of irs redshifts with optical redshifts for similar absorbed sources ( e.g. , @xcite ; @xcite ) suggests that they can be uncertain to @xmath410.2 in @xmath64 . for the absorption - dominated sources , the overall nature of the infrared spectra can be best considered by assembling the average irs spectrum . this allows consistent features within individual spectra to be seen with improved signal - to - noise . this was achieved by wavelength correcting each individual unsmoothed spectrum to the rest - frame ( @xmath64=0 ) , interpolating them to a common wavelength scale of @xmath170.1@xmath3/pix , and taking a straight ( non - weighted ) average at each wavelength position . the averaged spectrum is shown in figure [ fig : xstack]a . the most dominant feature is the silicate absorption feature at 9.7@xmath3 . the average silicate absorption strength ( as defined in table [ tab : irs ] ) is @[email protected] ( with individual values ranging from @xmath70.4 to @xmath72.9 ) . for comparison , we also show the average spectrum of the absorption - dominated 70@xmath3-selected sources presented in @xcite ( figure [ fig : xstack]b ) . even though the x - ray - selected absorption - dominated sources have a similar average @xmath67[24 ] color to that of the 70@xmath3-selected absorption - dominated sources , they have a much shallower average silicate absorption ( @[email protected] compared to @[email protected] ) . the average silicate absorption of the x - ray sample is substantially less than the median value for local ulirgs ( @xcite ; @xcite ) and the optically obscured sample observed by @xcite and @xcite , and is slightly larger that the median value for seyfert 2 agn ( @xmath70.6 ; @xcite ) . the individual spectra show no evidence for pah emission ( 6.2@xmath3 equivalent width @xmath40.2@xmath3 ; see table [ tab : irs ] ) , although there is a hint of the 6.2 and 7.7@xmath3 pah features in the average spectrum . however , because of the redshift uncertainties for the individual spectra , it is difficult to be confident in any weak narrow emission line features that appear in the average . the possible feature at 7.7@xmath3 could also be due to the [ ne vi ] emission line at 7.6@xmath3 . for each individual source , we use the noise spectrum to estimate the maximum flux from the 6.2@xmath3 pah emission feature ( we estimate the flux of the nearest noise feature to 6.2@xmath3 ) . we then use the relation of @xcite to estimate the corresponding maximum sfr and infrared luminosity associated with star - formation . the values are shown in table [ tab : irs ] . in some cases , the spectra are not sensitive enough to put a useful limit on the sfr . bootesx2 and bootesx7 have the lowest limits of sfr@xmath4190 m@xmath68 yr@xmath38 and sfr@xmath4100 m@xmath68 yr@xmath38 respectively , corresponding to infrared luminosities associated with star - formation of log [email protected] and @xmath411.7 respectively . comparing these values to the total infrared luminosity estimated from the rest - frame 8@xmath3 flux density , suggests that less than 30% and 13% of the total infrared luminosities are from star formation processes . weak indications are present in the average for [ ne ii ] , [ ne v ] , and [ ne iii ] emission lines . the relative intensities of these lines are @xmath170.7:1:1 . the presence of [ ne v ] is a strong agn indicator ( @xcite ; @xcite ) . the relative intensities show that [ ne v ] is stronger than that of local seyferts such as ngc 4151 , markarian 3 , and cena @xcite . if there is a contribution of 12.7@xmath3 pah emission to the [ ne ii ] line in the stacked spectrum , this would make this effect even larger . this may indicate a higher ionization of the narrow - line region by the more powerful agn in these sources . average irs spectrum of x - ray sources with redshifts ( top ) . for comparison , the average irs spectra of absorption - dominated sources 70@xmath3 sources from @xcite is also shown ( bottom ) . the expected positions of typically strong emission lines and the silicate absorption line are shown . the average spectrum combines the individual un - smoothed rest - frame spectra and is not smoothed.,title="fig:",height=226 ] average irs spectrum of x - ray sources with redshifts ( top ) . for comparison , the average irs spectra of absorption - dominated sources 70@xmath3 sources from @xcite is also shown ( bottom ) . the expected positions of typically strong emission lines and the silicate absorption line are shown . the average spectrum combines the individual un - smoothed rest - frame spectra and is not smoothed.,title="fig:",height=226 ] we have measured the luminosities at 5.8@xmath3 to determine the continuum luminosity just shortward of the 6.2@xmath61 m pah feature and at a wavelength without strong absorption ( e.g. , @xcite ; @xcite ; @xcite : see table [ tab : irs ] for individual values ) . the absorption dominated sources have median luminosity @xmath70l@xmath13(6@xmath61 m ) = @xmath71 ( 1.5 @xmath72 l@xmath73 ) . these are very luminous sources , exceeding that of most local agn in the sample of @xcite : sy 2 in that sample have median @xmath70l@xmath13(6@xmath61m)@xmath74l@xmath73 , and ulirgs have @xmath70l@xmath13(6@xmath61m)@xmath17 @xmath75l@xmath73 . @xcite find that local ulirgs span the range @xmath70l@xmath13(6@xmath61 m ) = 1@xmath76 l@xmath68 . our sources lie at the upper end of this range . the most luminous x - ray source , at @xmath70l@xmath13(6@xmath61m)=4 @xmath77 10@xmath78 l@xmath68 , is nearly as luminous as the most luminous source yet published from irs spectra : source 9 in @xcite with @xmath70l@xmath13(6@xmath61m)=5.9 @xmath77 10@xmath78 l@xmath68 . we estimate the total infrared luminosity , l@xmath6=l(8 - 1000@xmath3 ) of each of the absorption dominated sources . the conversion factors vary significantly depending on the template adopted for the conversion @xcite . because we expect the infrared emission to be dominated by obscured agn activity , we use the conversion for the template mrk 231 : l@xmath6=5@xmath79l@xmath13(8@xmath61 m ) , where the rest - frame 8@xmath3 flux density is measured directly from the irs spectra . the estimates are about a factor of 7 lower than one would obtain using templates of starburst - dominated galaxies such as arp 220 , and this explains our smaller l@xmath6 compared to that of @xcite for houck13 which was estimated from f00183 - 7111 . the infrared luminosities confirm that the absorption - dominated sources are all ultra - luminous ( @xmath8010@xmath81 ) or hyper - luminous ( @xmath8010@xmath82 ) infrared galaxies . given the lack of features and/or optical follow - up of the power - law sources , we do not have redshifts for these sources . we therefore can not construct an average spectrum or determine luminosities . the spectra of the power - law sources must have weak or absent silicate features ( 6.2@xmath3 pah equivalent widths @xmath830.2@xmath3 ) or they lie beyond @xmath842.6 , causing their absorption features to fall longward of the irs wavelength limit . it is difficult to determine whether silicate emission exists in these sources given our lack of redshift information and that the silicate emission feature is generally weaker than absorption feature . however bootesx6 may have a weak silicate emission feature if it is at @xmath64=0.9 . all 3 of the `` x - ray bright '' subsample are power - law sources . all the x - ray selected sources have infrared spectra consistent with their mid - ir luminosities being dominated by emission originating from agn activity . this is expected given that their high x - ray luminosities imply that they host luminous agn which should dominate their infrared spectra . we find a larger fraction of power - law sources ( 4/13 of the main sample and 7/16 of the total ( main + `` x - ray bright '' ) sample ) than is found in the swire sample of @xcite ( they find only 1/9 to have a featureless power - law spectrum in their sample which has an x - ray limit which is @xmath85 fainter than that of our sample ) . this may in part be due to them having accurate optical redshifts , which make it easier to identify features . it could also be due to our selection of more luminous x - ray sources , which are less likely to be heavily obscured and exhibit silicate absorption . this hypothesis is supported by the fact that 6/7 of the power - law sources are also the brightest x - ray sources in the sample . to explain the lack of a silicate absorption feature in their irs spectra , the power - law sources must have weak or absent silicate features or they must lie beyond @xmath842.6 . we argue that the latter is unlikely given that they are among the brightest x - ray sources in our sample . if at @xmath862.6 , they must be extremely intrinsically luminous . assuming a canonical power - law spectrum with photon index , @xmath87=1.7 , the typical x - ray flux of f@xmath32=4 @xmath88 corresponds to a rest - frame x - ray luminosity of l@xmath32=2 @xmath89 at @xmath64=2.6 which would suggest that they were among the most powerful quasars known . if at @xmath63 , these sources would be significantly more x - ray luminous than any of the sources at @xmath90 . we would not expect a sharp difference in the luminosities of sources at higher redshifts . furthermore , only 0.7% of the @xmath171000 x - ray sources in the @xmath2 xbotes survey @xcite with spectroscopic redshifts from ages ( kochanek et al . in prep ) have luminosities this high . in the following discussion , we assume that these sources are at similar redshifts to the rest of the x - ray selected sources and that they must have weak or absent silicate features . in figure [ fig : irac ] , we show the irac color - color diagram for our irs sources . to increase the number of x - ray selected sources , we also plot the 9 x - ray selected irs sources from the lockman hole field of the swire survey presented by @xcite . the swire sample is selected to have f@xmath13(24@xmath3)@xmath240.9 mjy , f@xmath91 , and @xmath31 mag . the x - ray flux limit still corresponds to typical agn luminosities at these redshifts . to determine the location in color - color space of these sources in comparison to the general 24 @xmath3 population , we also plot the distribution of the @xmath2310,000 mips botes sources with f@xmath13(24@xmath3)@xmath240.5 mjy . the rough regions expected to be inhabited by different populations are shown . we also plot the empirical line of @xcite ( see also @xcite ) to separate agn - dominated sources from galactic stars and normal galaxies . 20/25 of the x - ray selected sources lie within the agn region of @xcite . @xcite show that large amounts of extinction ( a@xmath9250 ) toward the active nucleus and/or a large contribution from a starburst can cause the irac colors of an object to fall outside of the @xcite region . because our sources are largely within the agn region , they are unlikely to be heavily extincted or have a large starburst contribution to their mid - ir emission . there appears to be no correlation between the position in irac color - color space with the presence of the silicate absorption strength , again supporting the hypothesis that the silicate feature arises in the cooler gas which is traced by longer infrared wavelengths . the sources in the x - ray bright sub - sample appear to have slightly bluer colors which may demonstrate that they are less dust obscured . the irac color - color classification is expected to deteriorate at @xmath931 as the observed irac bands are pushed towards lower wavelengths in which the stellar light dominates over hot dust emission ( @xcite ; @xcite ) . however , it likely holds to higher redshifts for the most luminous sources because the near - infrared emission is still dominated by the hot dust , even in the integrated spectral energy distribution . in figure [ fig : opt_col ] , we show the 24-to-8 @xmath3 vs. 24-to-0.7 @xmath3 color - color diagram for our sources ( see @xcite for the location of different populations in this color - color space ) . the 24-to-8 @xmath3 and 24-to-0.7 @xmath3 colors are defined as f(24:8)=@xmath94[@xmath95/@xmath96 and f(24:r)=@xmath94[@xmath95/@xmath97 respectively . f(24:r ) is a good indicator of obscuration and/or redshift . f(24:8 ) has been used as a crude measure of the spectral slope and hence the dust temperature distribution and whether the source is agn- or starburst - dominated . @xcite find that agn - dominated sources tend to have f(24:8)@xmath40.3 and starburst - dominated sources tend to have f(24:8)@xmath240.3 . even though the x - ray selected sources must harbor a powerful agn , their 24-to-8 @xmath3 colors suggest that the sources are dominated by a combination of agn and starbursts . we suggest that this diagnostic breaks down for obscured sources due to the heavy absorption at near - ir wavelengths which will steepen the infrared slope regardless of the underlying power source . in addition , at @xmath981 - 2 , the silicate absorption feature is moving through the 24@xmath99 filter , complicating the interpretation further . the @xmath15-[24 ] colors of the x - ray selected sources are typically bluer than that of the most extreme irs sources presented in @xcite and @xcite which typically have @xmath100>$]15 ( f(24:r)@xmath241.83 ) . the faintest optical magnitude is relatively bright ( @xmath15=23.7 ) despite the fact that there is no optical selection criteria against including optically faint sources . we suggest that the x - ray - selection is biasing us against the most heavily absorbed ( and therefore optically faintest ) sources . indeed , the sources in the x - ray bright sub - sample have among the bluest @xmath15-[24 ] colors due at least partially to their relatively bright @xmath15-band magnitudes . figure [ fig : xabs ] shows the observed x - ray flux as a function of x - ray hardness ratio for the combined sample of botes and swire x - ray irs sources . there are no sources with both small x - ray fluxes and small x - ray hardness ratios . the grey - scale shows that this is not seen in the larger population of x - ray sources in the same redshift range , and only becomes apparent when we jointly select x - ray sources that are both infrared bright and optically faint . our selection criteria result in a sample of bolometrically luminous high redshift agn . in these cases , small observed x - ray fluxes will be associated with more obscured sources which will also have larger hardness ratios . for our combined sample , the absorption - dominated sources tend to have fainter x - ray fluxes and larger x - ray hardness ratios ( with a median value of hr=0.55 ) which are indicative of more absorbed x - ray spectra and larger hi column densities . conversely , the power - law sources have brighter x - ray fluxes and smaller hardness ratios ( with a median value of [email protected] ) indicative of less obscured x - ray spectra and smaller hi column densities . this suggests that the dusty absorbing regions responsible for the silicate absorption also contain sufficient gas to cause x - ray absorption or that both the gas and dust lie along similar lines of sight . deeper silicate absorption features may be associated with more embedded agn with higher column densities of gas . our results are consistent with those of @xcite who measure the silicate absorption depth for a sample of 85 nearby agn with x - ray data and find a correlation between the strength of the silicate feature and the hi column density . our lack of redshifts for the power - law sources means that we must compare the x - ray fluxes and hardness ratios rather than the x - ray luminosities and fitted hi column densities . any large difference in the redshift distribution of the two populations could affect our results ( for example higher redshift sources with identical x - ray luminosities will have softer x - ray spectra and smaller x - ray fluxes ) . however , in section [ sec : pow ] , we argue that the power - law sources lie at @xmath1012.6 and there is no reason to think that they lie at significantly different redshifts to the absorption - dominated sources given their similar selection criteria . although we find a trend in the x - ray and infrared properties of these sources , the correlations are only weak . the difficulty in establishing a clear relation between x - ray and infrared properties may be explained by the distribution of their gas and dust . @xcite show that the x - ray absorbing column density of seyfert 2 galaxies varies on yearly timescales for most sources . this suggests that most of the x - ray absorption must occur on sub - parsec scales . they obtain even tighter constraints in @xcite for a seyfert galaxy which exhibits spectral variation on timescales of days . this suggests that , at least for less powerful sources , the obscuring gas is on a similar scale to that of the broad - line region . significant amounts of dust are unlikely to lie at these distances because of the dust sublimation temperatures . the difference in the location of much of the gas and dust may explain the large scatter between parameters tracing the x - ray and mid - ir obscuration ( e.g. , @xcite ) . @xcite suggest that the obscuring material in dusty agn is likely to consist of both dense clumpy clouds of dust and a smoother inter - cloud medium . they argue that deep silicate absorption can only be produced in a source embedded in a smooth medium since clumpy clouds will provide direct views of portions of illuminated clouds which will fill in the resultant silicate absorption feature . since our x - ray selected sources have shallower silicate absorption features than typical ulirgs , this may suggest that they are dominated by clumpy obscuration . alternatively , they may simply have a smaller line of sight obscuration . @xcite obtained irs spectra for a sample of 7 x - ray selected agn which were specifically chosen to have large x - ray luminosities and large column densities of gas ( due to their large x - ray hardness ratios ) . given the correlation between hi column density and silicate absorption strength ( see discussion above and @xcite ) , we might expect them to exhibit deep silicate features . however , all of their sources exhibit featureless power - law spectra . the solution to this apparent problem may be that they selected a highly unusual population of sources which were both extremely x - ray luminous and x - ray obscured . figure [ fig : xabs ] shows that although x - ray obscured sources are more likely to have silicate absorption features , x - ray luminous sources are likely to have featureless power - law spectra . perhaps more luminous agn are more capable of breaking up the surrounding dust into discrete clouds which would fill in the silicate absorption feature in these cases despite the large column densities of gas . we compare the fraction of power - law dominated to absorption dominated sources in different samples . @xcite probe the lowest x - ray luminosities and find only 1/9 ( 11% ) are power - law dominated . our sample has higher x - ray luminosities and we find the fraction of power - law to absorption dominated sources to be 4/13 ( 31% ) in the main sample and 7/16 ( 43% ) in the total ( main + x - ray bright ) sample . all of the sources in the @xcite study are power - law dominated . their sources are particularly bright in the x - ray , with intrinsic x - ray luminosities in the range logl@xmath102=44.0 - 45.5 . this suggests that more x - ray luminous sources are more likely to have shallower silicate absorption ( which will appear as power - law sources ) . 0.5 - 7 kev band x - ray flux against x - ray hardness ratio for x - ray selected irs sources . the hardness ratios can range from 1 ( hardest x - ray spectrum ) to -1 ( softest x - ray spectrum ) . the botes x - ray selected sources are denoted by large blue symbols ( with the largest symbols denoting the x - ray bright sub - sample ) . the fainter x - ray sources from @xcite are also shown ( smaller red symbols ) . the sources are split into silicate absorption - dominated sources ( downwards triangles ) , and power - law dominated sources ( stars ) . poisson error bars are also shown . the grey scale shows the distribution of all x - ray sources in the xbotes survey with redshifts @xmath64=0.9 - 2.6.,height=188 ] @xcite , @xcite , and @xcite show that there is a strong relation between agn hard x - ray emission and mid - ir continuum in low redshift active galaxies . figure [ fig : xvs6 ] shows the agn continuum luminosity at 6@xmath3 as a function of 2 - 7 kev x - ray luminosity for the absorption - dominated sources ( the power - law sources could not be included because we do not have redshifts to derive the luminosities ) . although the absorption - dominated sources do not show a clear correlation between these two quantities , this lack is likely to be due to the large luminosity errors and relatively narrow luminosity range of the sample . when we compare our sources with lower luminosity sources at low redshift @xcite , we find that they form the higher luminosity extension of the relation . we have chosen not to correct the x - ray luminosity for absorption given the large uncertainties . 7/9 of our sources lie at l(2 - 7kev)/@xmath70l@xmath103(6@xmath3)@xmath40.1 , compared to only 8/47 of the @xcite sources ( before correcting for absorption ) . this suggests that our sources are generally more obscured than the low redshift , lower luminosity agn presented in @xcite . rest - frame 6@xmath3 luminosity against rest - frame absorption - uncorrected 2 - 7 kev x - ray luminosity for absorption - dominated botes sources ( blue triangles ) . the poisson error bars are shown for the x - ray luminosity . the low redshift iso sources from @xcite are plotted as small squares ( connected empty and filled squares denote x - ray luminosities uncorrected and corrected for x - ray absorption respectively ) . the solid line shows a direct correspondence between the x - ray and 6@xmath3 luminosities . the dotted lines show where l(2 - 7kev)=0.1 @xmath104(6@xmath105 and l(2 - 7kev)=0.01 @xmath104(6@xmath105.,height=188 ] we have presented mid - infrared spectra obtained using @xmath1 irs of a sample of 16 x - ray loud 24 @xmath3 sources . the spectra of these sources are characterized by either silicate absorption or a featureless power - law , with no evidence of pah emission . this suggests that the mid - infrared emission from these sources is dominated by warm dust emission surrounding the active nucleus with little contribution from star formation . the absorption - dominated sources lie in the range [email protected] and have very high x - ray and infrared luminosities ( log l@xmath6=12.5 - 13.6 l@xmath68 ) . the average absorption depth in our absorption - dominated x - ray sources is similar to that of local seyfert 2 agn and much smaller than that of typical ulirgs . this suggests a difference in the distribution of the dusty material between our sample and typical ulirgs . the weak silicate absorption feature in our sample could indicate a more clumpy dust distribution or a lower line of sight obscuration . there is evidence in the average spectrum for weak [ ne ii ] , [ ne v ] , and [ neiii ] lines . the [ ne v ] is relatively stronger than is seen in local seyferts , suggesting a higher ionization of the narrow - line region in these more powerful agn . given their higher x - ray fluxes , it is unlikely that the power - law sources lie at @xmath862.6 . instead , they must have only weak silicate absorption or emission features . comparison with other samples suggests that the fraction of power - law sources appear to increase with x - ray luminosity and decrease with x - ray obscuration . both the infrared and infrared to optical colors suggest that the sources are relatively unobscured compared with @xmath15-[24]@xmath2415 sources . our x - ray selection is likely to be preferentially biased to less obscured and/or more powerful agn . we find that sources with silicate absorption features tend to have fainter x - ray fluxes and larger hardness ratios . this suggests that more absorbed sources also have higher column densities of absorbing gas along the same line of sight . we thank our colleagues on the ndwfs , xbotes , mips , irs , ages , and irac teams . kb is supported by the giacconi fellowship at stsci . this research is supported by the national optical astronomy observatory which is operated by the association of universities for research in astronomy , inc . ( aura ) under a cooperative agreement with the national science foundation . support for this work by the irs gto team at cornell university was provided by nasa through contract number 1257184 issued by jpl / caltech . this work is based on observations made with the _ spitzer space telescope _ , which is operated by the jet propulsion laboratory , california institute of technology under a contract with nasa . spitzer / mips _ survey of the botes region was obtained using gto time provided by the _ spitzer _ infrared spectrograph team ( james houck , p.i . ) and by m. rieke . the irs was a collaborative venture between cornell university and ball aerospace corporation funded by nasa through the jet propulsion laboratory and the ames research center . we thank the anonymous referee for his / her useful comments . , l. , charmandaris , v. , bernard - salas , j. , spoon , h. w. w. , marshall , j. a. , higdon , s. j. u. , desai , v. , teplitz , h. i. , hao , l. , devost , d. , brandl , b. r. , wu , y. , sloan , g. c. , soifer , b. t. , houck , j. r. , & herter , t. l. 2007 , , 656 , 148 , k. , brown , m. j. i. , dey , a. , jannuzi , b. t. , kochanek , c. s. , kenter , a. t. , fabricant , d. , fazio , g. g. , forman , w. r. , green , p. j. , jones , c. j. , mcnamara , b. r. , murray , s. s. , najita , j. r. , rieke , m. , shields , j. c. , & vikhlinin , a. 2006 , , 641 , 140 , k. , dey , a. , weedman , d. , desai , v. , le floch , e. , jannuzi , b. t. , soifer , b. t. , brown , m. j. i. , eisenhardt , p. , gorjian , v. , papovich , c. , smith , h. a. , willner , s. p. , & cool , r. j. 2006 , , 644 , 143 , b. r. , bernard - salas , j. , spoon , h. w. w. , devost , d. , sloan , g. c. , guilles , s. , wu , y. , houck , j. r. , weedman , d. w. , armus , l. , appleton , p. n. , soifer , b. t. , charmandaris , v. , hao , l. , higdon , j. a. m. s. j. , & herter , t. l. 2006 , , 653 , 1129 , m. j. i. , brand , k. , dey , a. , jannuzi , b. t. , cool , r. , le floch , e. , kochanek , c. s. , armus , l. , bian , c. , higdon , j. , higdon , s. , papovich , c. , rieke , g. , rieke , m. , smith , j. d. , soifer , b. t. , & weedman , d. 2006 , , 638 , 88 , v. , armus , l. , spoon , h. w. w. , charmandaris , v. , bernard - salas , j. , brandl , b. r. , farrah , d. , soifer , b. t. , teplitz , h. i. , ogle , p. m. , devost , d. , higdon , s. j. u. , marshall , j. a. , & houck , j. r. 2007 , , 669 , 810 , a. , soifer , b. t. , desai , v. , brand , k. , lefloch , e. , brown , m. j. , jannuzi , b. t. , armus , l. , bussmann , s. , brodwin , m. , bian , c. , eisenhardt , p. , higdon , s. , weedman , d. , & willner , s. 2008 , arxiv e - prints , 0801 , arxiv:0801.1860 , p. r. , stern , d. , brodwin , m. , fazio , g. g. , rieke , g. h. , rieke , m. j. , werner , m. w. , wright , e. l. , allen , l. e. , arendt , r. g. , ashby , m. l. n. , barmby , p. , forrest , w. j. , hora , j. l. , huang , j .- s . , huchra , j. , pahre , m. a. , pipher , j. l. , reach , w. t. , smith , h. a. , stauffer , j. r. , wang , z. , willner , s. p. , brown , m. j. i. , dey , a. , jannuzi , b. t. , & tiede , g. p. 2004 , , 154 , 48 , r. j. , gonzalez , a. h. , mckenzie , e. , brodwin , m. , brown , m. j. i. , cardona , g. , dey , a. , dickinson , m. , eisenhardt , p. r. , jannuzi , b. t. , lin , y .- t . , mohr , j. j. , raines , s. n. , stanford , s. a. , & stern , d. 2006 , , 639 , 816 , r. c. , jones , c. , forman , w. r. , murray , s. s. , brodwin , m. , brown , m. j. i. , eisenhardt , p. r. , stern , d. , kochanek , c. s. , eisenstein , d. , cool , r. j. , jannuzi , b. t. , dey , a. , brand , k. , gorjian , v. , & caldwell , n. 2007 , arxiv e - prints , 708 , j. r. , roellig , t. l. , van cleve , j. , forrest , w. j. , herter , t. , lawrence , c. r. , matthews , k. , reitsema , h. j. , soifer , b. t. , watson , d. m. , weedman , d. , huisjen , m. , troeltzsch , j. , barry , d. j. , bernard - salas , j. , blacken , c. e. , brandl , b. r. , charmandaris , v. , devost , d. , gull , g. e. , hall , p. , henderson , c. p. , higdon , s. j. u. , pirger , b. e. , schoenwald , j. , sloan , g. c. , uchida , k. i. , appleton , p. n. , armus , l. , burgdorf , m. j. , fajardo - acosta , s. b. , grillmair , c. j. , ingalls , j. g. , morris , p. w. , & teplitz , h. i. 2004 , , 154 , 18 , j. r. , soifer , b. t. , weedman , d. , higdon , s. j. u. , higdon , j. l. , herter , t. , brown , m. j. i. , dey , a. , jannuzi , b. t. , le floch , e. , rieke , m. , armus , l. , charmandaris , v. , brandl , b. r. , & teplitz , h. i. 2005 , , 622 , l105 , a. , murray , s. s. , forman , w. r. , jones , c. , green , p. , kochanek , c. s. , vikhlinin , a. , fabricant , d. , fazio , g. , brand , k. , brown , m. j. i. , dey , a. , jannuzi , b. t. , najita , j. , mcnamara , b. , shields , j. , & rieke , m. 2005 , , 161 , 9 , m. , storrie - lombardi , l. j. , sajina , a. , appleton , p. n. , armus , l. , chapman , s. c. , choi , p. i. , fadda , d. , fang , f. , frayer , d. t. , heinrichsen , i. , helou , g. , i m , m. , marleau , f. r. , masci , f. , shupe , d. l. , soifer , b. t. , surace , j. , teplitz , h. i. , wilson , g. , & yan , l. 2004 , , 154 , 166 , c. , polletta , m. d. c. , surace , j. , shupe , d. , fang , f. , xu , c. k. , smith , h. e. , siana , b. , rowan - robinson , m. , babbedge , t. , oliver , s. , pozzi , f. , davoodi , p. , owen , f. , padgett , d. , frayer , d. , jarrett , t. , masci , f. , olinger , j. , conrow , t. , farrah , d. , morrison , g. , gautier , n. , franceschini , a. , berta , s. , perez - fournon , i. , hatziminaoglou , e. , afonso - luis , a. , dole , h. , stacey , g. , serjeant , s. , pierre , m. , griffin , m. , & puetter , r. 2004 , , 154 , 54 , s. s. , kenter , a. , forman , w. r. , jones , c. , green , p. j. , kochanek , c. s. , vikhlinin , a. , fabricant , d. , fazio , g. , brand , k. , brown , m. j. i. , dey , a. , jannuzi , b. t. , najita , j. , mcnamara , b. , shields , j. , & rieke , m. 2005 , , 161 , 1 , g. h. , young , e. t. , engelbracht , c. w. , kelly , d. m. , low , f. j. , haller , e. e. , beeman , j. w. , gordon , k. d. , stansberry , j. a. , misselt , k. a. , cadien , j. , morrison , j. e. , rivlis , g. , latter , w. b. , noriega - crespo , a. , padgett , d. l. , stapelfeldt , k. r. , hines , d. c. , egami , e. , muzerolle , j. , alonso - herrero , a. , blaylock , m. , dole , h. , hinz , j. l. , le floch , e. , papovich , c. , prez - gonzlez , p. g. , smith , p. s. , su , k. y. l. , bennett , l. , frayer , d. t. , henderson , d. , lu , n. , masci , f. , pesenson , m. , rebull , l. , rho , j. , keene , j. , stolovy , s. , wachter , s. , wheaton , w. , werner , m. w. , & richards , p. l. 2004 , , 154 , 25 , j. r. , rieke , g. h. , maiolino , r. , gilli , r. , papovich , c. , prez - gonzlez , p. g. , alonso - herrero , a. , le floch , e. , engelbracht , c. w. , gordon , k. , hines , d. c. , hinz , j. l. , morrison , j. e. , muzerolle , j. , rieke , m. j. , & su , k. y. l. 2004 , , 154 , 160 , d. , eisenhardt , p. , gorjian , v. , kochanek , c. s. , caldwell , n. , eisenstein , d. , brodwin , m. , brown , m. j. i. , cool , r. , dey , a. , green , p. , jannuzi , b. t. , murray , s. s. , pahre , m. a. , & willner , s. p. 2005 , , 631 , 163 , d. w. , soifer , b. t. , hao , l. , higdon , j. l. , higdon , s. j. u. , houck , j. r. , le floch , e. , brown , m. j. i. , dey , a. , jannuzi , b. t. , rieke , m. , desai , v. , bian , c. , thompson , d. , armus , l. , teplitz , h. , eisenhardt , p. , & willner , s. p. 2006 , , 651 , 101 , l. , helou , g. , fadda , d. , marleau , f. r. , lacy , m. , wilson , g. , soifer , b. t. , drozdovsky , i. , masci , f. , armus , l. , teplitz , h. i. , frayer , d. t. , surace , j. , storrie - lombardi , l. j. , appleton , p. n. , chapman , s. , choi , p. , fan , f. , heinrichsen , i. , i m , m. , schmitz , m. , shupe , d. l. , & squires , g. k. 2004 , , 154 , 60

we investigate the gev emission from gamma - ray bursts ( grbs ) , using the results from the energetic gamma ray experimental telescope ( egret ) , and in view of the _ gamma - ray large area space telescope ( glast)_. assuming that the conventional prompt and afterglow photons originate from synchrotron radiation , we compare an accompanying inverse - compton component with egret measurements and upper limits on gev fluence , taking klein - nishina feedback into account . we find that egret constraints are consistent with the theoretical framework of the synchrotron self - compton model for both prompt and afterglow phases , and discuss constraints on microphysical parameters in both phases . based on the inverse - compton model and using egret results , we predict that _ glast _ would detect grbs with gev photons at a rate yr from each of the prompt and afterglow phases . this rate applies to the high - energy tail of the prompt synchrotron emission and to the inverse - compton component of the afterglow . theory predicts that in a large fraction of the cases where synchrotron gev prompt emission would be detected by _ glast _ , inverse - compton photons should be detected as well at high energies ( gev ) . therefore _ glast _ will enable a more precise test of the high - energy emission mechanism . finally , we show that the contribution of grbs to the flux of the extragalactic gamma - ray background measured with egret is at least 0.01% and likely around 0.1% .

0807.0012

cosmological gamma - ray bursts ( grbs ) have released a tremendous amount of energy in the past and present universe . their emission covers very wide range of frequencies : a highly variable prompt phase radiates @xmath3100 kev gamma rays , while a subsequent afterglow radiates radio to x - ray photons . it is likely that the bulk of these photons are emitted by gyration of relativistic electrons in magnetic fields e.g . , synchrotron radiation . the relativistic electrons are accelerated in either internal dissipation ( for prompt emission ) or external shocks ( for afterglows ) . for reviews , see , @xcite . gev photons were detected as well from several grbs by the energetic gamma ray experimental telescope ( egret ) on board the _ compton gamma ray observatory ( cgro ) _ the data are still not sufficient for us to firmly infer emission mechanisms of these gev gamma rays , but the most promising mechanism is synchrotron self - compton ( ssc ) scattering ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? this is because the relevant emission parameters such as the energy fraction of the grb jets going to electrons ( @xmath4 ) and magnetic fields ( @xmath5 ) are relatively well measured from the afterglow spectra as well as light curves ; the typical values are @xmath6 and @xmath7 ( e.g. , * ? ? ? * ; * ? ? ? * ) . in the prompt emission , @xmath4 is similar or even higher , as evident from the high efficiency of this phase , while @xmath5 is not well constrained . thus , there should be a significant inverse - compton ( ic ) component accompanying the synchrotron radiation in both the afterglow and prompt emission . the luminosities of the synchrotron and ic are expected to be comparable as ic - to - synchrotron luminosity ratio is roughly given by @xmath8 , according to theory ( e.g. , * ? ? ? * ) . in this paper , we explore the gev gamma - ray emission of grbs in the context of ssc mechanism . besides the several grbs detected by egret , there are many others for which upper bounds on the fluence were obtained @xcite . these @xmath3100 grbs should also be compared with the predictions of ssc model , because the fluence upper limits in the egret energy band are comparable to the fluence of prompt emission collected by burst and transient source experiment ( batse ) instrument onboard _ cgro_. as the experimental bound is already strong , while theoretical models of ssc process predict a large fluence for the egret energy range , we derive meaningful constraints from egret data analysis on the physics of the high - energy emission mechanisms of grbs . this approach is different from ( and therefore complementary with ) that in previous studies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) , where the prediction of gamma - ray flux relies only on theoretical models and sub - gev observations . we instead use egret data in order to infer the gev emission and constrain the theoretical models . we use our results to predict the expected number of grbs that would be detected by the _ gamma - ray large area space telescope ( glast)_. the _ glast _ satellite is equipped with the large area telescope ( lat ) , which is an upgraded version of egret . since revealing the high - energy emission mechanisms of grbs are one of the important objectives of _ glast _ , our prediction should give a useful guideline . finally , we apply our results to estimate the contribution of grbs to the diffuse extragalactic gamma - ray background ( egb ) , which was also measured by egret ( * ? ? ? * ; * ? ? ? * see , however , @xcite for a subtle issue of galactic foreground subtraction ) . this paper is organized as follows . in [ sec : ic ] , we summarize the predictions of ssc model for the prompt ( [ sub : prompt ] ) and afterglow ( [ sub : afterglow ] ) phases . section [ sec : constraint on high - energy emission with egret ] is devoted for analysis of the grb fluence data by egret , from which distributions of fluence in the gev band are derived . we then use these distributions to argue prospects for grb detection with _ glast _ in [ sec : glast ] , and implications for egb from grb emissions in [ sec : egb ] . in [ sec : conclusions ] , we give a summary of the present paper . if the prompt and/or afterglow emission is due to synchrotron radiation from relativistic electrons ( with lorentz factor @xmath9 ) , then there must be an accompanying ic component from the same electrons scattering off the synchrotron photons . the spectral shape of the ic emission is almost the same as the synchrotron radiation ( shifted by @xmath10 ) , and is expected to fall around the gev range during both the prompt and afterglow phases . for @xmath11 , and assuming that there is no `` klein - nishina suppression '' and that the emitting electrons are fast cooling , the ic fluence is related to the synchrotron fluence simply through @xmath12 . thus , assuming that the microphysics do not vary much from burst to burst , it is natural to assume proportionality between the synchrotron mev fluence ( observed by batse ) and the gev synchrotron plus ic fluence ( observed by egret and in the future by _ glast _ ) : @xmath13 where @xmath14 and @xmath15 are coefficients for the proportionality due to synchrotron and ic processes . note that the synchrotron fluence in the gev range can be extrapolated relatively easily , if we assume that the spectrum extends up to such high energies . thus , we here focus on theoretical evaluation of the ic component . at first approximation , the coefficient @xmath15 is roughly @xmath8 from considerations above , and thus we define @xmath16 where for the prompt emission @xmath17 while for the afterglow @xmath18 is the afterglow fluence within the radio to x - ray energy bands . correction factors @xmath19 and @xmath20 represent the effect of klein - nishina suppression and detector energy window , respectively , which are given below . we define typical frequencies for both synchrotron ( @xmath21 ) and ic ( @xmath22 ) as the frequencies where most of the energies are radiated in case that the klein - nishina cross section does not play an important role ; i.e. , where @xmath23 for each component is peaked in this case . from relativistic kinematics , these two typical frequencies are related through @xmath24 where @xmath25 is a characteristic lorentz factor of the electrons that dominate the synchrotron power @xcite ; this is true in the fast cooling regime , which is the case in the most of our discussions @xcite . the klein - nishina effect is relevant if a photon energy in the electron rest frame exceeds the electron rest mass energy , and this condition is formulated as @xmath26 where @xmath27 is the bulk lorentz factor of the ejecta , which is on the order of 100 in the prompt phase of grbs and their early afterglows . upscattering synchrotron photons to energies above @xmath28 is highly suppressed , which results in ic cutoff at @xmath29 . besides producing a spectral cutoff , the klein - nishina effect also modifies the way electrons cool , which is relevant for the gev emission and is also included in @xmath30 . electrons with energies above klein - nishina threshold ( for a given seed - photon energy ) can lose their energies only through synchrotron radiation , while the lower - energy ones can cool through both processes . such an effect has been studied in the case where the seed photons for ic scattering are provided by an external sources ( e.g. , * ? ? ? * ; * ? ? ? * and references therein ) . however , in the case of ssc mechanism , since the seed photons are emitted from synchrotron process due to the same electron population , we should properly take into account feedback . giving full details on this is beyond the scope of the present paper , but some results are summarized briefly in appendix [ app : kn ] ( see also @xcite ) . here we only show the approximate analytic form of @xmath30 : @xmath31 where @xmath32 is the lorentz factor of electrons for which photons at @xmath33 are in the klein - nishina regime . the energy of an observed photon with frequency @xmath34 as measured in the rest frame of an electron with lorentz factor @xmath35 is @xmath36 where the @xmath37 factor converts the photon energy from the observer frame to the plasma rest frame and the @xmath35 factor converts it to the electron rest frame . since such a photon is in the klein - nishina regime of an electron with lorentz factor @xmath35 once its energy in the electron rest frame is larger than @xmath38 we obtain : @xmath39 this klein - nishina feedback effect modifies the spectrum shape of both synchrotron and ic emissions ( in addition to the klein - nishina cutoff for ic ) . we note that equation ( [ eq : eta kn ] ) provides a solution that agrees within a factor of @xmath32 with the one obtained by numerically solving equation ( [ eq : y ] ) . this precision is sufficiently good for our purpose , especially because it is well within the uncertainty ranges of other parameters . by @xmath20 , we take into account the fraction of the ic fluence that falls into the gev detector energy bands . egret window is between @xmath40 mev and @xmath41 gev while _ glast_-lat window is between @xmath42 mev and @xmath43 gev . we here assume that the frequency where most of the ic energy is released , @xmath44 $ ] , is always larger than lower limit of the frequency band , @xmath45 , as expected for both egret and _ glast _ , and thus consider the cases in which @xmath46 is within or above the detector frequency band . in the former case where @xmath47 , we have @xmath48 . on the other hand , if @xmath49 , then most of the energy comes from the upper frequency limit @xmath50 , and we have @xmath51 , where @xmath52 is the photon spectral index below peak frequency . thus we may approximate @xmath53 as @xmath54 } { \nu_{w , u}}\right)^{\alpha_1 - 2 } \nonumber\\ & = & \left\ { \begin{array}{ccc } \left(1+\frac{\gamma_{m}^2 \nu_{\rm syn}}{\nu_{w , u } } \right)^{\alpha_1 - 2 } & \mbox{for } & \gamma_{m } \le \gamma_{\rm kn},\\ \left(1+\frac{\gamma_b\gamma_{m}m_e c^2 } { h\nu_{w , u}}\right)^{\alpha_1 - 2 } & \mbox{for } & \gamma_{m } > \gamma_{\rm kn},\\ \end{array } \right . \label{eq : eta_w 1}\end{aligned}\ ] ] where @xmath55 as one can easily show . the discussion above assumes that the density of the synchrotron photon field is proportional to the instantaneous synchrotron emissivity . in the case of relativistically expanding radiation front , this assumption is valid when the duration over which the emissivity vary significantly , @xmath56 , is comparable to the time that passed since the expanding shell was ejected , @xmath57 . in this case the ratio between the synchrotron emissivity and the synchrotron photon field density is in a steady state . when @xmath58 the synchrotron photon field density may be significantly lower than in the steady state case @xcite , thereby suppressing the ic component . the exact suppression factor depends on the detailed spatial and temporal history of the emissivity . theoretically , in the afterglow phase we expect @xmath59 . also in the prompt emission phase , internal shock models generally predict @xmath60 ( * ? ? ? * and references therein ) . thus , in the internal - external shock model corrections to the ic component due to this effect are expected to be on the order of unity . therefore , in the present paper , we assume that such an effect can be neglected and that the synchrotron photon field is proportional to the instantaneous synchrotron emissivity . one should keep in mind , however , that @xmath61 is a viable possibility ( see , e.g. , * ? ? ? * ; * ? ? ? * for a more detailed study in such cases ) , especially in the highly variable prompt phase . in principle , detailed glast observations of an ic emission may be able to constrain @xmath62 during the prompt phase . in addition , towards the higher end of the egret or _ glast _ energy band , photons may start to be subject to absorption due to pair creation in the source or during propagation ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? although such a mechanism might be relevant for the ic yields ( especially in the prompt phase ) depending on some parameters that are not well constrained yet , we assume that it is not the case in the present paper . _ glast _ will hopefully provide information that enables better handle on this issue . erg @xmath63 and the standard deviation is @xmath64.,width=325 ] batse ( as well as _ swift _ satellite ) detected so far a large number of grbs in prompt phase with gamma rays in the energy band of 20 kev1 mev . the spectrum is well described by a smoothly broken power law with a typical lower - energy index of @xmath65 and higher - energy index of @xmath66 ; the spectral break typically occurs around @xmath67 kev , where the energy of the prompt emission @xmath68 peaks @xcite . as we show in figure [ fig : fluence ] , the distribution of the fluence integrated over the batse energy band follows log - normal function . the peak of this distribution is @xmath69 erg @xmath63 , and its standard deviation is @xmath70 . the average of the batse fluence is therefore @xmath71 erg @xmath63 . therefore , for the prompt emission phase , using @xmath72 , @xmath73 , and @xmath74 kev , we find : @xmath75 where @xmath76 . in addition , for @xmath20 , considering _ glast_-lat energy window ( 20 mev300 gev ) in equation ( [ eq : eta_w 1 ] ) , we obtain @xmath77^{-1 } & \mbox{for } & \gamma_{m } \le 170\gamma_{b,2},\\ \left(1+\frac{\gamma_{b,2 } \gamma_{m}}{5900}\right)^{-1 } & \mbox{for } & \gamma_{m } > 170 \gamma_{b,2}. \end{array}\right . \label{eq : eta_w 2}\ ] ] now assuming that all electrons are accelerated in the shocks , the typical value for the lorentz factor of the relativistic electrons are given as @xmath78 where @xmath79 is the relative lorentz factor of the colliding ejecta portions and @xmath80 . in the internal shock model for the prompt emission , @xmath81 is of order unity . if we adopt @xmath82 and @xmath83 , we obtain @xmath84 . furthermore , assuming @xmath85 , equation ( [ eq : eta_w 2 ] ) gives @xmath86 , and equation ( [ eq : eta kn ] ) with equation ( [ eq : break lorentz factor ] ) gives @xmath87 . by substituting these values and assuming @xmath7 in equation ( [ eq : eta_ic ] ) , we obtain @xmath88 , which implies that under the most straightforward assumptions a comparable fluence is expected in both _ glast_-lat and batse windows . in this case , the klein - nishina cutoff energy is in _ glast_-lat band as well as in egret band ( @xmath89 gev ) , and thus we also obtain another comparable value of @xmath90 in egret case . note that in the case of prompt emission , the synchrotron spectrum is not negligible in egret and _ glast_-lat energy bands . for canonical parameters ( @xmath74 kev , @xmath72 , and @xmath73 ) , the ratio @xmath91 at 100 mev is about @xmath92 , assuming that the synchrotron spectrum continues into the gev window without a break and ic is not much suppressed by the klein - nishina effect . therefore , the synchrotron component dominates around the lower - energy limit where most of the photons ( although not most of the fluence ) are observed . in the case of egret , since only a handful of photons were detected in all egret events , these are expected to be dominated by the synchrotron low - energy ( @xmath3100 mev ) photons . this indicates that the quantity we can constrain using the egret fluence upper limits is not @xmath15 but @xmath93 , the ratio of synchrotron fluence around 100 mev and that in the mev range . in addition , this picture is indeed consistent with the fact that the spectral indices of gev photons for several grbs measured with egret are @xmath943 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? note however , that the _ energy fluence _ in _ glast_-lat and egret bands can be dominated by a much harder ic component ( @xmath952 ) that peaks above @xmath31 gev and may carry up to @xmath310 times more energy than the one observed at 100 mev without being detected . this is because even when the @xmath310 gev fluence is ten times larger , the small photon number at such high - energies is still small enough to avoid detection . thus , egret observations , which are consistent with measurement of the synchrotron high energy tail , can only put an upper limit on @xmath15 . the afterglow is considered to be a synchrotron emission from electrons accelerated in the external shock , which is caused by the interaction between the relativistic ejecta and the interstellar medium . in this model , the synchrotron emission dominates the spectrum from radio to x - ray . the associated ic emission is expected to dominate the gev energy range ( i.e. , @xmath96 ) , since the electron lorentz factor is much larger than the case of prompt emission ( see eq . [ [ eq : gamma_e ] ] , where the relative and bulk lorentz factors are the same , @xmath97 ) , compensating the smaller @xmath21 ( eq . [ [ eq : frequency ] ] ) . during the first several minutes ( observer time ) , electrons might be cooling fast ( @xmath98 ) with @xmath99 kev , while @xmath10010@xmath101 . this implies that the fraction of the ic energy that falls in _ glast_-lat energy window is close to unity , i.e , @xmath1020.9 from equation ( [ eq : eta_w 1 ] ) ( for egret @xmath1030.5 ) and @xmath871 from equations ( [ eq : eta kn])([eq : gamma kn ] ) . since @xmath104 at early time is close to the upper limit of the energy window the effective photon index of the ic emission within the detector window during this time is @xmath1051.52 . at later times the electrons are at the slow - cooling regime and @xmath21 is the cooling frequency , while a typical @xmath106 is the lorentz factor of electrons that cooled significantly ( e.g. , * ? ? ? * ) . in this regime the ssc peak is very broad and its location is almost constant with time . for typical parameters , the klein - nishina effect do not play a major role while the peak of the ssc emission falls within _ glast_-lat and egret windows . therefore , at late time @xmath107 and the effective photon index within the energy windows of these detectors is @xmath1052 . one should , however , note that on long time scales the gev background becomes important , making it hard to detect the gev afterglow . therefore , the optimal time scale for gev afterglow search would be @xmath310010@xmath108 s @xcite . the afterglow gev fluence , @xmath109 in equation ( [ eq : egret fluence ] ) , is that integrated over a given time scale , while @xmath110 is collected over roughly @xmath111 , during which 90% of the mev photons are counted . the total energy radiated away by the radio to x - ray afterglow during every decade of time is roughly 0.010.1 of the energy emitted in the prompt phase . therefore we expect a bright gev afterglow which radiate about @xmath112@xmath113 every decade of time for hours and days after the bursts . in this paper when considering egret observations , we adopt 200 s after @xmath111 , when electrons are in the fast cooling regime , as the duration over which @xmath109 is integrated . @xcite analyzed grbs that were detected by batse and observed by egret . since the field of view of egret was much smaller than that of batse and the observation was limited by the life time of the spark chamber , egret covered only about 100 grbs out of @xmath33000 batse bursts . but this is still a reasonably large number to get statistically meaningful result . the analysis of the prompt burst in egret data was performed around the error circles of batse bursts for the first @xmath111 , and spectral index of @xmath114 is assumed within egret window ( the upper limits are higher by a factor of @xmath10510 for a spectral index of @xmath115 ) . the same analysis was performed for the afterglow phase , for 200 s after @xmath111 ( not including @xmath111 ) . @xcite measured the fluence of 6 and 12 grbs , in prompt and afterglow phases respectively . for all other grbs only fluence upper limits were obtained in the range @xmath116@xmath117 erg @xmath63 . @xmath118 space , obtained with the analysis of egret data assuming a spectral index of @xmath114 during @xmath111 ( _ top _ ) and 200 s after @xmath111 ( _ bottom _ ) . @xmath119 and @xmath118 are the central value and standard deviation , respectively , for the log - normal distribution of the fluence ratio @xmath120 ( eq . [ [ eq : eta distribution ] ] ) . the best fit points ( a ) are marked as crosses , and other representative points ( b and c ) are also indicated in both panels.,title="fig:",width=325 ] @xmath118 space , obtained with the analysis of egret data assuming a spectral index of @xmath114 during @xmath111 ( _ top _ ) and 200 s after @xmath111 ( _ bottom _ ) . @xmath119 and @xmath118 are the central value and standard deviation , respectively , for the log - normal distribution of the fluence ratio @xmath120 ( eq . [ [ eq : eta distribution ] ] ) . the best fit points ( a ) are marked as crosses , and other representative points ( b and c ) are also indicated in both panels.,title="fig:",width=325 ] here we interpret these results in the framework of the ssc model , which implies that the fluences in batse and egret bands are likely to be positively correlated through equation ( [ eq : egret fluence ] ) ( @xmath121 and @xmath122 ) . we further assume that the coefficient @xmath120 ( @xmath14 for prompt and @xmath15 for afterglow phases ) follows some probability distribution function @xmath123 which is independent of @xmath124 . we consider a log - normal distribution with the central value @xmath119 and standard deviation @xmath118 : @xmath125 d\log\eta . \label{eq : eta distribution}\ ] ] constraining @xmath119 and @xmath118 then leads to implications of grb parameters such as @xmath126 , @xmath5 , and @xmath127 , through their relations given in the previous section . , during ( a ) @xmath111 and ( b ) 200 s after @xmath111 ( assuming a spectral index of @xmath114 ) . the histogram represents cumulative fraction of grbs whose fluence limits are below a given value , which can be interpreted as the detector efficiency , while the solid curves are fitting function.,width=325 ] we used the observations to constrain @xmath119 and @xmath118 by carrying out a maximum likelihood analysis . fluence data , we used the results of fig . 2.3 of @xcite rather than tables 2.1 and 2.2 there . ] figure [ fig : mu_sig ] shows the contour plot of the most likely region on the @xmath119@xmath118 plane for @xmath111 ( _ top _ ) and 200 s after @xmath111 data ( _ bottom _ ) assuming a spectral index of @xmath114 ( if the spectral index is @xmath115 then @xmath119 increases by @xmath1051 ) . in that procedure , detection efficiency of egret as a function of fluence , @xmath128 , is obtained from the distribution of the egret upper limits ( for undetected grbs ) , which is shown in figure [ fig : efficiency ] ; i.e. , a cumulative fraction of bursts whose fluence limits are below a given fluence . in the case of detected grbs , on the other hand , the size of the error bars for the fluence is interpreted as measurement accuracy of egret . then , in order to test the consistency of the assumption that equation ( [ eq : eta distribution ] ) fits the data , we carried out a monte carlo simulation that draws @xmath129 realizations of egret observations assuming that the distribution of @xmath130 follows equation ( [ eq : eta distribution ] ) with the most likely values of @xmath119 and @xmath118 . by comparing the likelihood of these monte carlo realizations with that of the actual egret observations , we find that 70% of the realizations have a lower likelihood , suggesting that equation ( [ eq : eta distribution ] ) with its most likely values is indeed consistent with the observations . and ( b ) 200 s after @xmath111 . c correspond to the points on @xmath119@xmath118 plots in fig . [ fig : mu_sig ] . the batse fluence distribution is also plotted for comparison . the distribution for the prompt phase ( a ) is for the high - energy tail of the synchrotron radiation . the prompt ic fluence may be larger by up to about one order of magnitude ( see text).,width=325 ] given @xmath119 and @xmath118 , we can obtain the distribution of fluence in egret band by convolving batse fluence distribution ( @xmath131 ; fig . [ fig : fluence ] ) and @xmath132 : @xmath133 as representative models , we use three sets of @xmath134 for both the prompt and afterglow cases . these are labeled as a@xmath135 , b@xmath135 , and c@xmath135 ( a@xmath136 , b@xmath137 , and c@xmath137 ) , and shown in figure [ fig : mu_sig ] . in figure [ fig : dndf_egret ] , we show the resulting fluence distribution corresponding to each of these models . and @xmath25 from egret data for ( a ) prompt and ( b ) afterglow phases , obtained with canonical values for other parameters . the left and right regions of the solid curve in panel ( a ) is excluded and allowed regions respectively , and regions between two dashed ( dotted ) curves in panel ( b ) show allowed regions corresponding to @xmath138 ( @xmath139 ) . note , however , that these regions could easily change depending on values of other parameters.,title="fig:",width=325 ] and @xmath25 from egret data for ( a ) prompt and ( b ) afterglow phases , obtained with canonical values for other parameters . the left and right regions of the solid curve in panel ( a ) is excluded and allowed regions respectively , and regions between two dashed ( dotted ) curves in panel ( b ) show allowed regions corresponding to @xmath138 ( @xmath139 ) . note , however , that these regions could easily change depending on values of other parameters.,title="fig:",width=325 ] egret results imply that during the prompt emission phase , @xmath140 . as we discussed in [ sub : prompt ] , the low number of photons in the bursts detected by egret , as well as their spectrum , implies that the detections of prompt photons are most likely to have been dominated by the high - energy tail of the synchrotron emission ; i.e. , @xmath141 in figure [ fig : mu_sig](_top _ ) . in fact , simply extrapolating synchrotron tail of many batse bursts up to @xmath3100 mev regime , using inferred values for their @xmath21 and @xmath142 , gives a value of @xmath14 which is consistent with the one obtained here for the prompt phase . the harder ic prompt emission , however , can still have as much as 10 times larger fluence than that of the synchrotron emission in egret window , without being detected . therefore , this figure also sets an upper limit on the ratio of the ic and synchrotron components of @xmath143 , as larger @xmath15 gives enough photon fluence detectable by egret . as we showed in [ sub : prompt ] , theoretically we predict @xmath144 ( for egret ) with a canonical set of parameters . although this appears to imply that the current bound from egret already excludes the canonical model , we can not make such a strong statement given the current uncertainties of many relevant parameters . therefore , a more conservative statement would be that the current egret bound is barely consistent with the predictions of the ssc within the internal shock model . we may interpret the bound @xmath143 as constraints on @xmath145 and @xmath25 , which is shown in figure [ fig : const](a ) . as the klein - nishina suppression ( @xmath30 ) becomes significant for large @xmath25 , we have only modest limit on @xmath146 in such a regime . however , one should keep in mind that these are order of magnitude constraints , which may farther vary with other parameters , such as @xmath147 and @xmath27 . much better constraint plot is expected with the future _ glast _ data , where hopefully , @xmath15 will be measured for many individual bursts . during the afterglow the synchrotron emission is much softer than during the prompt phase , and therefore , the ic component is expected to dominate egret observations also near its lower energy - band limit . moreover , the fact that the number of bursts detected by egret during the afterglow is higher than the number detected during the prompt emission suggests that here egret is likely to have detected the actual ic component of the afterglow . the spectral index of the gev afterglow in egret window during the first 200 s is expected to be @xmath1482 , implying that the evaluation of @xmath119 in the bottom panel of figure [ fig : mu_sig ] , which assumes a spectral index of @xmath114 , might be larger by at most a small factor ( @xmath323 ) . thus , for the afterglow , @xmath149@xmath150 . we then compare this result with the theoretical expectation of @xmath15 in equation ( [ eq : eta_ic ] ) . but first we need to estimate the value of @xmath151 where @xmath18 is measured during the first 200 s following @xmath111 and @xmath110 is the prompt emission fluence . we use the _ swift _ grb table which provides x - ray afterglow fluences several tens to several hundreds of seconds after the bursts , as well as the prompt mev fluences . using only bursts where the x - ray observation starts after @xmath111 but no more than @xmath152 s after the burst we find a distribution of @xmath153 that ranges from @xmath117 to @xmath150 , with the central value of @xmath310@xmath154 . thus afterglow theory with canonical parameters predicts @xmath155 with a large spread , consistent with egret constraints . figure [ fig : const](b ) shows the interpretation of egret constraint on @xmath15 ( fig . [ fig : mu_sig ] ) as that for @xmath146 and @xmath25 , assuming canonical parameters and @xmath156 . although this allowed region may change with other model parameters , again one can not have too large value of @xmath25 because of the klein - nishina suppression factor @xmath30 . cccc 2.3 & 650 & @xmath157 & @xmath158 + 2.0 & 650 & @xmath159 & @xmath160 + 1.0 & 650 & @xmath161 & @xmath162 + [ table : glast ] we now move on to discussions on implications for _ glast _ using the obtained constraints on @xmath120 in the previous section . first we estimate the sensitivity of lat on board _ glast _ for prompt and afterglow gev emission , based on its published sensitivity to steady point sources , which is @xmath163 @xmath63 s@xmath1 above 100 mev at @xmath164 with a power - law index of @xmath165 . this sensitivity is obtained by a one - year all - sky survey during which any point source is observed for @xmath370 d ( the lat field of view is 2.4 sr ) . therefore during the background - limited regime ( when @xmath166 is large enough that many background photons are observed ) the flux limit scale with @xmath166 as @xmath163 @xmath63 s@xmath1 @xmath167 . during the photon - count - limited regime ( when @xmath166 is so small that less than one background photon is expected ) , in contrast , the detection limit is at a constant fluence . therefore the fluence sensitivity of the _ glast_-lat detector is @xmath168,\\ f_{\rm lim } ( t_0 ) \left(\frac{t}{t_0}\right)^{1/2 } & [ t > t_0 ] , \end{array } \right . \label{eq : glast limit}\ ] ] where @xmath169 s represents the time when the transition from photon - count - limited to background - limited regime occurs in the lat case . note that equation ( [ eq : glast limit ] ) is for the limiting _ fluence _ , the time - integrated flux , rather than the flux . this limit is more natural in the photon - count - limited regime and it is more relevant to egret constraints that we derived in the previous section . detailed derivation of this sensitivity is given in appendix [ app : glast ] . in table [ table : glast ] , we summarize the values of @xmath57 and @xmath170 for a few cases of power law index @xmath171 and integration time @xmath166 . the values of @xmath172 for @xmath173 in the table are determined by criteria of five - photon detection , while those for @xmath174 are by @xmath164 significance . the fluence we argue here is the one integrated over 30 mev30 gev , in order to compare with the egret fluence upper bounds . ccc a@xmath135 & 15 yr@xmath1 & @xmath175 ) + b@xmath135 & 20 yr@xmath1 & @xmath176 ) + c@xmath135 & 10 yr@xmath1 & @xmath177 ) + a@xmath178 & 20 yr@xmath1 & @xmath179 + b@xmath178 & 30 yr@xmath1 & @xmath180 + c@xmath178 & 15 yr@xmath1 & @xmath181 + [ table : result ] in the case of background - limited regime , it might be more appropriate to use higher energy threshold ( instead of 30 mev ) especially for hard source spectrum , because the background spectrum falls steeply with frequency ( @xmath182 ) . we may find optimal low - frequency threshold depending on spectral index of grb emissions ; it is higher for harder spectrum . thus , we should be able to improve the fluence sensitivity for background - limited regime , compared with the figures given in table [ table : glast ] . in addition , transition from photon - count to background limited regime would occur later than 650 s. for our purpose , however , as time scales we consider ( @xmath111 for prompt emission and 200 s after @xmath111 for afterglows ) are both during photon - count - limited regime , the consideration above does not apply and we can use full energy range ( 30 mev30 gev for egret ) to collect as many photons as possible . _ glast _ is also equipped with the glast burst monitor ( gbm ) instrument , dedicated for the detection of grbs . it detects photons of 8 kev to more than 25 mev and its field of view is @xmath38 sr . the expected rate of grbs that trigger gbm is @xmath3200 yr@xmath1 @xcite , which is almost as high as batse rate . each year , about 70 out of these @xmath3200 bursts should fall within the lat field of view . given the distribution of fluences ( fig . [ fig : dndf_egret ] ) and the lat sensitivity ( table [ table : glast ] ) , we can estimate the fraction of grbs that would be detected with lat . in table [ table : result ] , we show the expected lat detection rate for @xmath183 , which is @xmath320 yr@xmath1 for the best - fit models of the egret data for both the prompt and afterglow emissions . the prompt phase estimates are for detections of the synchrotron component in the @xmath3100 mev range . given the large effective area of the lat it is expected also to detect @xmath184 gev photons from the ic component and identify the spectral break associated with the transition from the synchrotron to ic component , thereby directly testing the ssc model . the estimates given in table [ table : result ] are fairly conservative . first , while we used five - photon criterion for the detection , even two - photon detection should be quite significant , because the expected background count is much smaller than one photon during @xmath111 and the following 200 s that we considered . second , _ swift _ can find dimmer bursts than gbm . although the discovery rate is not as high as that of gbm or batse , it would still be able to find tens of new grbs in the lat field of view . thus the true rate would likely be larger than the figures given in table [ table : result ] . from grbs of given fluence , for ( a ) prompt ( during @xmath111 ) and ( b ) afterglow emission ( 200 s after @xmath111 ) phases . in each panel , three models a c are shown . note that for the prompt phase , the fluence is that for synchrotron radiation , and that for ic component could be even larger ( see text).,width=325 ] all the grbs except for those detected by egret should contribute to the egb flux to a certain extent @xcite . this may be computed as @xmath185 , \label{eq : egb}\ ] ] where @xmath186 is egret fluence in 30 mev30 gev , @xmath187 is the normalized distribution of egret fluence ( eq . [ [ eq : fluence distribution ] ] and fig . [ fig : dndf_egret ] ) , and @xmath188 d@xmath1 is the occurrence rate of grbs from all sky . the factor @xmath189 takes into account the fact that very bright grbs can not contribute to the egb because they would be identified as point sources ( but see discussions below ) . figure [ fig : egb_dist ] shows differential egb intensity @xmath190 that represents contribution from grbs of a given fluence , for prompt and afterglow phases . in the third column of table [ table : result ] , we show the egb intensity due to prompt and afterglow phases of grbs which is @xmath310@xmath191 gev @xmath63 s@xmath1 sr@xmath1 . on the other hand , in the same energy range , egret measured the egb flux to be @xmath192 gev @xmath63 s@xmath1 sr@xmath1 @xcite . therefore , grbs that were detected by batse but were not detected as point sources by egret contribute to the egb at least @xmath30.01% . again , we note that the estimates for the prompt phase are those of synchrotron component . we thus need to take the predicted ic contribution into account , which is represented by a correction factor @xmath193 in table [ table : result ] . since this factor could be as large as @xmath310 according to the discussion in [ sec : constraint on high - energy emission with egret ] , egb flux due to prompt phase of grbs could also becomes @xmath310 times larger , which makes grb contribution as large as @xmath30.1% of the observations above @xmath3gev . in any case , the contributions from other astrophysical sources such as blazars are expected to be more significant than grbs ( e.g , * ? ? ? * and references therein ) . additional contribution to egb is expected from a large number of grbs that point away from us and therefore would not have been detected with batse . the emission from these bursts points towards us once the external shock decelerates @xcite . since the total gev energy emitted every decade of time during the afterglow is roughly constant , the contribution of these grbs to egb can be estimated by the gev emission of the bursts that were detected by batse . similar contribution is expected from bursts that points towards us but that are too faint to be detected by batse , if the grb luminosity function behaves as @xmath194 as suggested by the universal structured jet model ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * see however @xcite ) . therefore the contribution of bursts that were not detected by batse to egb can be estimated by the afterglow fluence of the detected bursts , assuming no contribution from bursts with only an upper limit . this is a reasonable estimate since the gev flux is dominated by the few brightest bursts in gev which are the most likely to be detected . taking the fluence of the detected gev bursts as the logarithmic mean of these upper and lower limits implies @xmath195 gev @xmath63 s@xmath1 sr@xmath1 , a grb contribution being @xmath30.1% of the egb . finally , we note that there is a big uncertainty in removing the galactic foreground contamination from the total diffuse flux @xcite . additionally , egret observations do not constrain tev emission that cascades down into the gev range for grbs at cosmological distances @xcite . thus , if the foreground subtraction was indeed underestimated or if grb tev emission is not negligible , then grb contribution might be much more significant than the estimates here . the _ glast _ satellite would enable us to test high - energy emission mechanisms of grbs . if this emission will be found to be consistent with ssc then its observations would constrain physical parameters such as @xmath196 ratio and the bulk lorentz factor of the jet , @xmath197 . the egret instrument on board _ cgro _ , while less sensitive than the _ glast_-lat detector , identified several batse grbs with gev photons . in addition , stringent upper limits for @xmath3100 grbs were put on fluences in the gev band by analyzing the egret data @xcite . in this paper , we further extended this egret result , comparing with the ssc emission model . following theoretical models of ssc , we assumed that there is a linear correlation between fluences in batse and egret energy bands , and that the proportionality coefficient @xmath120 follows a log - normal distribution . we found that the predictions from the ssc model using canonical parameter values is fully consistent with egret fluence measurements and upper limits for both the prompt and afterglow phases . during the course of showing this result , we properly took the klein - nishina feedback effect into account in the theoretical calculation . the best - fit value of the coefficient was @xmath198 for both the prompt and afterglow emissions , and it is already stringent enough to test the ssc scenario . the limits for the prompt emission phase are for the synchrotron radiation , and thus if we consider the ic component as well , the value of @xmath120 could be larger by up to one order of magnitude . the obtained @xmath120 distribution , together with the batse fluence distribution , gives the expected fluence distribution in the gev band , which is shown in figure [ fig : dndf_egret ] . as the _ glast_-lat detector covers egret energy band , we can predict the detectable number of grbs with _ glast _ from the distribution of @xmath199 , given the _ glast_-lat sensitivity . our conservative estimate using the five - photon criterion is that about @xmath320 grbs among those detected with gbm would be detected with _ glast_-lat each year . this number could be even larger if we use fewer - photon criteria . the fluence distribution can also be used to estimate the grb contribution to the egb intensity . we found that the contribution would be at least @xmath30.01% but is likely to be as large as @xmath30.1% . we are grateful to b. l. dingus and m. m. gonzlez snchez for very helpful comments and discussions . we thank the referee for useful comments . this work was supported by sherman fairchild foundation ( sa and en ) , nasa swift grant ( en ) , alfred p. sloan foundation , packard foundation , and a nasa atp grant ( rs ) . we shall find an analytic expression for @xmath30 due to the klein - nishina feedback . to simplify the argument such that we can treat it analytically , we make the following approximations : ( i ) an electron with a fixed lorentz factor @xmath9 radiates mono - energetic synchrotron photons ; ( ii ) the same electron upscatter a given synchrotron photon to another monochromatic energy , which is increased by a factor of @xmath10 ; ( iii ) @xmath68 of both synchrotron and ic photons peaks at @xmath21 ( a synchrotron frequency corresponding to @xmath25 ) and @xmath200 ; if there is no klein - nishina suppression ) , respectively ; ( iv ) the klein - nishina cutoff occurs quite sharply above its threshold ; ( v ) both cooling and self - absorption frequencies are much smaller than the frequency region of our interest ; and ( vi ) electrons cool so quickly that any dynamical effects can be neglected . with these approximations , expressions for the ratio of power of synchrotron and ic radiations from a given electron @xmath201 simplifies significantly . in particular , according to the assumption ( iii ) above , we have @xmath202 . this is given as @xmath203^{-(p-1)/2 } ( \nu^\prime)^{-1/2 } } { 1+y([\nu^\prime / \nu_{\rm syn}^\prime]^{1/2}\gamma_m ) } \theta\left(-\nu^\prime + \frac{m_ec^2}{h\gamma_m}\right ) , \label{eq : y}\ ] ] where @xmath204 is electron spectral index , @xmath205 is the step function , and primed quantities are evaluated in the rest frame of the ejecta ( e.g. , @xmath206 , where @xmath34 is the frequency in an observer frame ) . a detailed derivation as well as numerical approaches are given elsewhere ( nakar , ando , & sari , in preparation ) , but at least this equation can be understood qualitatively . for a given electron with lorentz factor @xmath25 , the synchrotron power does not depend on whether the klein - nishina suppression is effective or not . on the other hand , the ic power does , because it is proportional to the energy density of seed ( synchrotron ) photons integrated up to some cutoff frequency ; synchrotron photons above this frequency can not be ic scattered efficiently by the electron with @xmath25 because of the klein - nishina suppression . the integrand of equation ( [ eq : y ] ) represents the synchrotron spectrum . more specifically , assuming there is no klein - nishina suppression , the spectrum is simply given by @xmath207^{-(p-1)/2 } ( \nu^\prime)^{-1/2}$ ] ; the step function then represents the klein - nishina cutoff . the factor @xmath208 in the denominator of the integrand accounts for the suppression of the electron distribution function due to the enhanced ic cooling ; i.e. , @xmath209^{-1 } \propto [ 1 + y(\gamma_e)]^{-1}$ ] . these electrons are ones that emit synchrotron photons of a given frequency @xmath210 . recalling the relation @xmath211 , their lorentz factor is given by @xmath212 , which appears in the argument of @xmath213 in the integrand . finally , the other constants in equation ( [ eq : y ] ) are chosen so that we have a proper relation for the fast cooling , @xmath214 , if we turn off the klein - nishina cutoff and have constant @xmath213 . now we shall find analytic expressions of equation ( [ eq : y ] ) in asymptotic regions . we start from the case of @xmath215 , which is equivalent to @xmath216 . the integration then becomes @xmath217^{1/2 } \gamma_m ) } + \int_{\nu_{\rm syn}^\prime}^{\frac{m_ec^2}{h\gamma_m } } d\nu^\prime \frac{(\nu_{\rm syn}^\prime)^{p/2 - 1 } ( \nu^\prime)^{-p/2}}{1+y([\nu^\prime/\nu_{\rm syn}^\prime]^{1/2 } \gamma_m ) } \right].\ ] ] we assume that the function @xmath218 varies rather mildly in the integrand , so that in the argument of @xmath213 we may use @xmath219 . then the integral can be evaluated analytically , and gives @xmath220 = \epsilon_e / \epsilon_b$ ] . when @xmath221 , we have @xmath222 , which is the same result as in the case of no klein - nishina suppression . this makes sense because the condition @xmath223 indicates that the electrons with @xmath25 is below the klein - nishina threshold with seed photons at frequency @xmath224 that dominate the synchrotron power . on the other hand , when @xmath225 ( or @xmath226 ) , equation ( [ eq : y ] ) becomes @xmath227^{1/2}\gamma_m ) } \approx \frac{\epsilon_e}{\epsilon_b}\frac{p-2}{p-1 } \left(\frac{\gamma_m}{\gamma_{\rm kn}}\right)^{-1/2 } \frac{1}{1+y([\gamma_m\gamma_{\rm kn}]^{1/2 } ) } , \label{eq : y for large gamma_m}\ ] ] where in the second equality , we used @xmath228 for the argument of @xmath213 . when @xmath229 is large enough so that @xmath230^{1/2 } ) \ll 1 $ ] , then equation ( [ eq : y for large gamma_m ] ) immediately gives asymptotic solution for @xmath231 . when @xmath25 is in the intermediate regime , we can still get analytic expressions , which however are given elsewhere because they are somewhat complicated . here we simply show numerical solutions of equation ( [ eq : y ] ) as a function of @xmath232 for various values of @xmath146 . we show these results as well as a simple fitting form ( given by eq . [ [ eq : eta kn ] ] ) in figure [ fig : ym ] . thus , equation ( [ eq : eta kn ] ) provides fairly good fit to the results of numerical integration of equation ( [ eq : y ] ) . by an electron with lorentz factor @xmath25 as a function of @xmath233 . points represent numerical solutions of eq . ( [ eq : y ] ) for various values of @xmath234 , and solid line is an analytic fit ( eq . [ [ eq : eta kn]]).,width=325 ] for a steady point source with a spectral index of @xmath165 , the sensitivity of _ glast_-lat to its flux above 100 mev is @xmath163 @xmath63 s@xmath1 at @xmath164 significance for a one - year all - sky survey . considering the field of view of _ glast_-lat , 2.4 sr , this survey time corresponds to 70 d exposure time to the source , and therefore , the sensitivity to the number fluence integrated over this time scale is @xmath235 @xmath63 . in this section , we generalize this limit to an arbitrary spectral index , @xmath171 , and exposure time , @xmath166 . before starting the discussion , we define the differential number and energy fluences , and integrated number and energy fluences ( all quantities are time - integrated ) : @xmath236 @xmath237 where @xmath238 is a coefficient , and @xmath239 and @xmath240 are the energy band boundaries . the fluence sensitivity for a one year exposure is within the background - limited regime namely within one year many background photons are expected to be detected within the point - spread - function of the detector . in the case of _ glast_-lat , backgrounds are the egb or galactic foreground emissions . therefore , we start our discussion from this background - limited case . let us define this background rate of _ glast _ by @xmath241 , for which we assume @xmath242 spectrum and use the energy - dependent angular resolution and on - source effective area @xmath243 . the criterion of point - source detection is @xmath244 where @xmath118 represents significance of detection , and photon count from the source is obtained by @xmath245 therefore , using equation ( [ eq : dnde ] ) in equations ( [ eq : photon count ] ) and ( [ eq : detection criterion ] ) , we can obtain the sensitivity to the coefficient @xmath246 as follows : @xmath247^{-1 } , \label{eq : c_lim}\ ] ] and then using equation ( [ eq : n ] ) , this can be translated into the sensitivity to the number and energy fluences , @xmath248 and @xmath249 . we here note that @xmath246 depends on @xmath166 , @xmath250 , @xmath239 , and @xmath251 , while @xmath252 depends only on @xmath166 , @xmath253 , and @xmath240 . in this background - limited regime , the time dependence is @xmath254 from equation ( [ eq : detection criterion ] ) . we confirmed that , using the egb intensity measured by egret @xcite and energy - dependent angular resolution of lat , we could obtain the limit comparable to @xmath255 @xmath63 , in the case of @xmath94 , @xmath256 d , @xmath257 mev , @xmath258 , and @xmath259 . the results of this procedure for several values of interest of @xmath250 are summarized in equation ( [ eq : glast limit ] ) and table [ table : glast ] . here , we used egret energy range , i.e. , @xmath260 mev and @xmath261 gev , but we can instead adopt different values . if the time scale is short such that @xmath262 , then the argument above does not apply , but the sensitivity is simply obtained by the expected photon count from the source . in this photon - count - limited regime , we can evaluate the fluence sensitivity by requiring @xmath263 to be a few ; here we use @xmath264 . one can obtain the corresponding @xmath246 by solving this criterion using equation ( [ eq : photon count ] ) . this time , @xmath246 is independent of @xmath166 . then again using equation ( [ eq : n ] ) , one can get the fluence sensitivity in this regime as shown in table [ table : glast ] . ando , s. 2004 , , 354 , 414 ando , s. , komatsu , e. , narumoto , t. , & totani , t. 2007 , , 75 , 063519 asano , k. , & inoue , s. 2007 , , 671 , 645 band , d. , et al . 1993 , , 413 , 281 baring , m. g. , & harding , a. k. 1997 , , 491 , 663 casanova , s. , dingus , b. l. , & zhang , b. 2007 , , 656 , 306 chiang , j. , & dermer c. d. 1999 , , 512 , 699 derishev , e. v. , kocharovsky , v. v. , kocharovsky , v. v. , & mszros , p.2003 , gamma - ray burst and afterglow astronomy 2001 : a workshop celebrating the first year of the hete mission , 662 , 292 dermer , c. d. , chiang , j. , & mitman , k. e. 2000 , , 537 , 785 dermer , c. d. 2006 , preprint ( astro - ph/0605402 ) dermer , c. d. 2007 , american institute of physics conference series , 921 , 122 fan , y .- z . , piran , t. , narayan , r. , & wei , d .- 2008 , , 384 , 1483 gonzlez , m. m. , dingus , b. l. , kaneko , y. , preece , r. d. , dermer , c. d. , & briggs m. s. 2003 , , 424 , 749 gou , l .- j . , & mszros , p. 2007 , , 668 , 392 granot , j. , cohen - tanugi , j. , & silva , e. d. c. 2007 , preprint ( arxiv:0708.4228 [ astro - ph ] ) guetta , d. , & granot , j. 2003 , , 585 , 885 guetta , d. , piran , t. , & waxman e. 2005 , , 619 , 412 hurley , k. , et al . 1994 , , 372 , 652 ioka , k. , murase , k. , toma , k. , nagataki , s. , & nakamura , t. 2007 , , 670 , l77 kaneko , y. , preece , r. d. , briggs , m. s. , paciesas , w. s. , meegan , c. a. , & band , d. l. 2006 , , 166 , 298 keshet , u. , waxman , e. , & loeb , a. 2004 , j. cosmology astropart . , 04 , 006 , v. m. , postnov , k. a. , & prokhorov , m. e. 2001 , astronomy reports , 45 , 236 . lithwick , y. , & sari , r. 2001 , , 555 , 540 mcenery , j. , & ritz , s. 2006 , american institute of physics conference series , vol . 836 , gamma - ray bursts in the swift era , pp . 660663 medvedev , m. v. 2000 , , 540 , 704 , p. , rees , m. j. , & papathanassiou , h. 1994 , , 432 , 181 mszros , p. 2006 , rep . , 69 , 2259 moderski , r. , sikora , m. , coppi , p. s. , & aharonian , f. 2005a , , 363 , 954 moderski , r. , sikora , m. , coppi , p. s. , & aharonian , f. 2005b , , 364 , 1488 murase , k. , asano , k. , & nagataki , s. 2007 , , 671 , 1886 murase , k. , & ioka , k. 2008 , , 676 , 1123 nakar , e. 2007 , , 442 , 166 panaitescu , a. , & kumar , p. 2000 , , 543 , 66 , a. , & kumar , p. 2001 , , 560 , l49 panaitescu , a. 2008 , , 385 , 1628 peer , a. , & waxman , e. 2004 , , 613 , 448 peer , a. , & waxman , e. 2005 , , 633 , 1018 perna , r. , sari , r. , & frail , d. 2003 , , 594 , 379 piran , t. 1999 , , 314 , 575 piran , t. 2005 , rev . phys . , 76 , 1143 , r. d. , briggs , m. s. , mallozzi , r. s. , pendleton , g. n. , paciesas , w. s. , & band , d. l. 2000 , , 126 , 19 razzaque , s. , mszros , p. , & zhang , b. 2004 , , 613 , 1072 fireballs and surroundings , rhoads , j. e. 1997 , , 487 , l1 , e. , lazzati , d. , & rees , m. j. 2002 , , 332 , 945 . rybicki , g. b. , & lightman , a. p. 1979 , radiative processes in astrophysics . new york , wiley - interscience sreekumar , p. , et al . 1998 , , 494 , 523 strong , a. w. , moskalenko , i. v. , & reimer , o. 2004 , , 613 , 956 , e. 1997 , , 485 , l5 , d. m. , & lu , t. 1998 , , 505 , 252 , s. a. , harrison , f. a. , sari , r. , & frail , d. a. 2003 , , 597 , 459 zhang , b. , & mszros , p. 2001 , , 559 , 110 , b. , & mszros , p. 2002 , , 571 , 876 .

we calculate the density of states of the 2 + 1 dimensional btz black hole in the micro- and grand - canonical ensembles . our starting point is the relation between 2 + 1 dimensional quantum gravity and quantised chern - simons theory . in the micro - canonical ensemble , we find the bekenstein hawking entropy by relating a kac - moody algebra of global gauge charges to a virasoro algebra with a classical central charge via a twisted sugawara construction . this construction is valid at all values of the black hole radius . at infinity it gives the asymptotic isometries of the black hole , and at the horizon it gives an explicit form for a set of deformations of the horizon whose algebra is the same virasoro algebra . in the grand - canonical ensemble we define the partition function by using a surface term at infinity that is compatible with fixing the temperature and angular velocity of the black hole . we then compute the partition function directly in a boundary wess - zumino - witten theory , and find that we obtain the correct result only after we include a source term at the horizon that induces a non - trivial spin - structure on the wzw partition function . 8.5 in 6.5 in -.35 in ^(-)a # 1#1 mximo baados 0.2 cm _ departamento de fsica terica , facultad de ciencias , + universidad de zaragoza , zaragoza 50009 , spain _ + e - mail : [email protected] thorsten brotz 0.2 cm _ blackett laboratory , imperial college of science , technology and medicine , + prince consort road , london sw7 2bz , uk _ + e - mail : [email protected] miguel e. ortiz 0.2 cm _ blackett laboratory , imperial college of science , technology and medicine , + prince consort road , london sw7 2bz , uk _ + e - mail : [email protected] 0.5truecm 0.5truecm

hep-th9802076

the chern - simons formulation of 2 + 1 dimensional gravity @xcite has provided many interesting new insights into the problem of quantum gravity . ( for a review see @xcite . ) most notably , making use of the relationship between 2 + 1 dimensional chern - simons theory and the 1 + 1 dimensional wzw model @xcite , carlip has argued for a statistical mechanical interpretation of the entropy of a 2 + 1 dimensional black hole , backed up by a pair of tantalising calculations which yielded the correct value for the black hole entropy , both in lorentzian @xcite , and euclidean @xcite formalisms . furthermore , the derivation given in @xcite has been recently applied with success to de sitter space @xcite , and may provide a tool for understanding black hole entropy in string theory far from extremality @xcite . the most important assumption in carlip s analysis is that those degrees of freedom responsible for the black hole entropy are located at the horizon . this idea is certainly appealing and has been advocated by many authors . however , at a technical and conceptual level , it is difficult to see what states are counted in carlip s calculation , and what is being held fixed . in principle , it should be possible to count states in a micro - canonical ensemble , holding the mass and spin of the black hole fixed , or to infer the number of states in a grand - canonical ensemble , holding fixed the black hole temperature and angular velocity . a different approach to understanding the statistical mechanical origin of the 2 + 1 dimensional black hole entropy was recently proposed by strominger @xcite . in this approach , the basic ingredient is the discovery by brown and henneaux @xcite that the asymptotic symmetry group of 2 + 1 dimensional gravity with a negative cosmological constant @xcite is the conformal group with a ( classical ) central charge @xmath0 where @xmath1 is the cosmological constant . note that in the weak coupling limit @xmath2 , @xmath3 becomes very large . strominger has pointed out that if one counts states by regarding the theory as equivalent to @xmath3 free bosons , then at a fixed value of @xmath4 and @xmath5 , the degeneracy of states gives rise to exactly the bekenstein hawking entropy . since in 2 + 1 dimensions @xmath6 and @xmath7 where @xmath8 and @xmath9 are , respectively , the black hole mass and angular momentum @xcite , strominger s computation is clearly a micro - canonical calculation . in this approach one would like to find the underlying conformal field theory with a central charge equal to ( [ c ] ) , and its connection is to a counting of states at the black hole horizon . since it is clear that one can not obtain the correct black hole entropy in general by looking only at asymptotic isometries , it seems that strominger s calculation succeeds because the trivial nature of gravity in 2 + 1 dimensions results in an isomorphism between boundary theories at infinity and at the horizon . this paper has two goals . on the one hand , in sec . 2 we present a micro - canonical calculation of the black hole entropy . starting with a chern - simons theory , we find the algebra of global charges present at any constant radius boundary surface in the black hole spacetime . we prove that a subset of this infinite set of conserved charges satisfies the virasoro algebra with a central charge equal to ( [ c ] ) . this central charge , as in @xcite , arises classically @xcite . the advantage of considering the chern - simons formulation is that the underlying conformal field theory is , at least at the classical level , an @xmath10 wzw model whose relation to the virasoro generators is via a twisted sugawara construction @xcite . further , it is possible , to show that this virasoro algebra arises from a reduction of the wzw theory to a liouville theory @xcite . the counting of black hole microstates then follows just as in @xcite . as stressed before , since this counting needs @xmath8 and @xmath9 fixed , the relevant states live in a micro - canonical ensemble . the subset of global charges satisfying the virasoro algebra are shown to be precisely those charges that , at infinity , leave the leading order form of the metric invariant , agreeing with the asymptotic isometries found in ref . @xcite . however , the construction leading to a virasoro algebra of surface deformations is equally valid at any radius , and in particular at the horizon . thus , we are able to derive the bekenstein hawking entropy from the algebra of horizon deformations . we give the explicit form of the generators of the virasoro algebra and of the diffeomorphisms that they generate , valid at a boundary surface located at any radius . our second goal is to find the grand - canonical partition function for the 2 + 1 dimensional black hole . in sec . 3 , we compute the partition function @xmath11 for three dimensional euclidean gravity on a solid torus with a fixed value of the modular parameter @xmath12 of the torus . we show that this parameter is related to the black hole inverse temperature @xmath13 and angular velocity @xmath14 by @xmath15 and therefore @xmath11 is clearly the grand - canonical partition function . the computation of @xmath11 is not straightforward because the relevant group is @xmath16 which is not compact . we use here the trick of replacing @xmath17 @xcite and show that the partition function correctly accounts for the 2 + 1 dimensional black hole entropy , after continuing the spin of the @xmath18 representations back to complex values . in this case , we find the correct answer only if we include a source term at the black hole horizon which has a particularly interesting interpretation in terms of the wzw theory . it tracks the spin of each representation and is the analogue of a @xmath19 operator in fermionic theories , twisting the wzw theory in the time direction . in this section we canonically quantise the degrees of freedom associated with the gravitational field , by using the relation between 2 + 1 dimensional gravity and chern - simons theory . we then relate the kac - moody algebra of the boundary wzw theory that emerges from the chern - simons theory to a virasoro algebra that describes asymptotic isometries of the metric . this relation , achieved by a twisted sugawara construction , results in a theory whose degeneracy at fixed mass and spin of the black hole leads to the bekenstein hawking entropy @xcite . let us begin with some general remarks on chern - simons theory , always motivated by an application to the 2 + 1 dimensional black hole . consider the chern - simons action @xmath20={k\over 4\pi}\int \varepsilon^{\mu\nu\rho}\tr \left(a_\mu\partial_\nu a_\rho + { 2\over 3 } a_\mu a_\nu a_\rho\right)d^3x,\ ] ] where as we shall see below , we shall be interested in @xmath21 . up to a boundary term , ( [ a0 ] ) can be put into the canonical form @xmath22={k\over 4\pi } \int_{\sigma\times r } \varepsilon^{ij } g_{ab } \left ( a^a_i \dot a^b_j - a^a_t f^b_{ij}\right)dt d^2 x , \ ] ] ( here @xmath23 and @xmath24 ) . we take ( [ a1 ] ) , without additional boundary terms , as our starting point . the variation of this action leads to @xmath25={k\over 4\pi}\int \varepsilon^{\mu\nu\rho}\tr \left(\delta a_\mu f_{\nu\rho}\right)d^3x -{k\over 2\pi}\int_{{\partial } \sigma } \tr a_t \ , \delta a_\phi \ , , \ ] ] where we have assumed , in @xmath26 coordinates , that there is a single outer boundary at fixed @xmath27 . in order to make the variation of this action well defined , we choose to fix one or other of the conditions @xmath28 at the boundary ( in the @xmath29 chern - simons theory we will choose @xmath30 ) however , this is not enough to ensure the differentiability of ( [ mm14a ] ) since under ( [ chiral ] ) the boundary term reduces to @xmath31 . we thus need to imposse the equation @xmath32 or , in other words , we need to fix the value of @xmath33 at the boundary . under ( [ chiral ] ) and ( [ nn1 ] ) the action has well defined variations and the variational problem is then well posed . we now analise the meaning of these boundary conditions . the chirality conditions ( [ chiral ] ) should be regarded as specifying the class of spacetimes or field configurations that will be considered . in terms of lightlike coordinates @xmath34 these conditions read @xmath35 and it is well known that they are satisfied by the btz black hole solutions ( see , for example , @xcite ) . these conditions leave a large residual symmetry group , namely chiral ( anti chiral ) gauge transformations which are generated by kac - moody fields , and the corresponding boundary degrees of freedom are described by a chiral @xmath36 action . since the black hole is asymptotically anti - de sitter , it is natural to further reduce the boundary theory by imposing polyakov s reduction conditions on the @xmath37 currents leading to an effective liouville theory , as done in @xcite . we shall mention this possibility in sec . 2.8 , but in this paper we shall mainly consider the full kac - moody theory . let us finally point out that the chirality boundary conditions ( [ chiral ] ) are not known to be in one to one correspondence with the existence of a black hole , or an event horizon . the definition of black holes states " is a delicate problem which needs some information about the topology as well as local fields . our strategy here is to consider the chirality conditions , which include the black hole solutions , as a starting point and quantise the theory on that hilbert space . condition ( [ nn1 ] ) fixes the ensemble to be micro - canonical . indeed , in the @xmath29 theory , the values of @xmath38 are proportional to linear combinations of the black hole mass and angular momenta . an alternative procedure would be to add the term @xmath39 to the action ( [ mm14a ] ) and leave its value undetermined . this corresponds to a canonical ensemble and will be studied in detail in sec . there will in general also be a boundary variation at any inner boundary , but in the following discussion we shall only consider charges on an outer boundary . in @xcite it was shown that the algebra of gauge constraints leads to a set of global charges at the boundary whose poisson bracket algebra is a classical kac - moody algebra . these global charges are equivalent to the global charges obtained by a reduction of the chern - simons theory to a boundary wzw theory . from the point of view of the three dimensional theory in hamiltonian form , the global charges arise from considering the generators of gauge transformations @xmath40 where @xmath41 is a boundary term . because of the presence of the boundary , the functional derivative of @xmath42 is well - defined only if the boundary term @xmath43 has the variation @xmath44 the poisson bracket of two generators of the form ( [ a4 ] ) then becomes , @xmath45 where @xmath46 . one expects the boundary term in the right hand side of ( [ a6 ] ) to be equal to the charge @xmath47 , plus a possible central term @xcite . but to check this we first need to give boundary conditions in order to integrate ( [ a5 ] ) and extract the value of @xmath41 . we shall consider two different classes of boundary conditions . assuming that @xmath48 is fixed at the boundary the charge is @xmath49 one can then check that , indeed , the boundary term in ( [ a6 ] ) contains the charge associated to the commutator @xmath50 $ ] plus a central term . imposing the constraints , the algebra of global charges becomes @xmath51 ) + { k\over 2\pi}\int g_{ab } \eta^a \partial_k \lambda^b dx^k.\ ] ] since we are interested in the dirac bracket algebra , we should not only solve the constraint on the bulk but also fix a gauge . it is convenient to use the gauge chosen in @xcite , and in @xcite in a parallel treatment using the wzw formulation , @xmath52 where the boundary is taken to be at fixed @xmath27 . this gauge choice together with the constraint @xmath53 imply @xmath54 the gauge choice ( [ a8a ] ) is preserved only by gauge transformations whose parameters are of the form @xmath55 since @xmath56 still contains an arbitrary function of time @xmath57 , it seems that the gauge freedom has not been fixed completely yet . the extra requirement that fixes the time dependence of the gauge parameters comes from the boundary condition on @xmath58 . we have chosen our action in order to impose one or other of the conditions @xmath59 these conditions remove the gauge invariance since the group of transformations leaving ( [ aa8 ] ) invariant does not contain any arbitrary function of time . @xmath60 is constrained to depend only on @xmath61 or @xmath62 . setting , @xmath63 equation ( [ a8 ] ) leads to the classical kac - moody algebra @xmath64 note that the central term has the usual sign for @xmath21 . we could have obtained the same algebra by inserting ( [ a8a ] ) into the action ( [ a1 ] ) and computing the resulting poisson brackets ( see the next section for a detailed discussion of this reduction ) . we have seen above that global charges associated to gauge transformations that do not vanish at the boundary give rise to an infinite number of conserved charges satisfying the kac - moody algebra . we shall now investigate those charges associated to the group of diffeomorphisms at the boundary . since the boundary is a circle , one expects to find the virasoro algebra . furthermore , since in chern - simons theory diffeomorphisms and gauge transformations are related , one expects the virasoro and kac - moody generators to be related by the sugawara construction . actually , for those diffeomorphisms with a non - zero component normal to the boundary , one finds a twisted sugawara construction which induces a classical central charge in the virasoro algebra @xcite . this central charge was first found in @xcite in the adm formulation of 2 + 1 dimensional gravity and has recently been shown to play an important role in understanding the statistical mechanical origin of the 2 + 1 dimensional black hole entropy @xcite . recall that in chern - simons theories , diffeomorphisms with parameter @xmath65 are related to gauge transformations with parameter @xmath66 by the equations of motion . in a canonical realisation of gauge symmetries , @xmath67 is a lagrange multiplier , and so we must instead consider gauge transformations @xmath68 as we did in the last section , we fix the gauge by fixing @xmath69 . we also choose coordinates for the on - shell solution for which @xmath70 which implies @xmath71 this will be a good choice of the radial coordinate at infinity for the black hole . as a consequence of this choice , diffeomorphisms @xmath72 that preserve the gauge choice ( [ a8a ] ) and ( [ a8aa ] ) must be independent of @xmath27 . since the gauge choice only fixes the gauge freedom in the interior of the manifold , we are able to derive the algebra of global boundary diffeomorphisms in complete generality by looking only at @xmath73 , subject to the constraints imposed by the boundary condition on the gauge field . since from ( [ mm1b ] ) the gauge parameter is field - dependent , we replace ( [ a7 ] ) by @xmath74 this is a good choice of @xmath75 since @xmath76 is left unchanged by the action of the global charges . in other words @xmath77 is fixed at the boundary . the algebra of charges then becomes @xcite @xmath78 ) + { k\over 2\pi}g_{ab}\alpha^a\alpha^b\int \xi^r\partial_\phi\zeta^r \ : d\phi .\ ] ] if we restrict the diffeomorphisms to be of the specific form @xcite ( see below for a geometrical justification for this restriction ) , @xmath79 then the algebra of these restricted diffeomorphisms is the continuous form of the virasoro algebra with central charge @xmath80 ) - { k\over 2\pi } \alpha^2 \beta^2 \int \xi\partial^3_\phi \zeta \ : d\phi\ ] ] where @xmath81 denotes @xmath82 . defining @xmath83 or equivalently @xmath84 gives the usual poisson bracket version of the virasoro algebra @xmath85 so that the central charge is @xmath86 which is positive for @xmath21 in the virasoro algebra is due to the shift of the @xmath4 operator by @xmath87 in ( [ a19b ] ) . ] . hence , as expected , those diffeomorphisms that lead to global charges , after the restriction ( [ a16 ] ) , induce an infinite number of conserved charges satisfying the virasoro algebra with a classical central charge . ( [ a19b ] ) is an example of a twisted sugawara construction @xcite . note that from ( [ a19 ] ) we see that the boundary condition ( [ nn1 ] ) fixes the value of @xmath4 at the boundary . in 2 + 1 dimensional gravity with a negative cosmological constant , the einstein hilbert action is represented by the difference of two chern - simons actions @xmath88-i\left[\aa\right ] , \label{mm13}\ ] ] for a pair of @xmath89 gauge fields @xmath90 and @xmath91 , where , @xmath92 , @xmath93 so that @xmath94 = { \varepsilon^c}_{ab}j_c$ ] , and @xmath95 where @xmath96 and @xmath97 . ] @xmath98={k\over 4\pi}\int \varepsilon^{ij}\tr \left(^{(\pm)}a_i { ^{(\pm)}\dot{a}_j}-^{(\pm)}a_t \ , ^{(\pm)}f_{ij}\right)d^3x \ , . \label{mm14}\ ] ] the einstein hilbert action is recovered by defining @xmath99 from which it follows that @xmath100 ignoring boundary terms . this relates the level @xmath101 of the chern - simons theories to newton s constant , @xmath102 we see that for the black hole , @xmath21 , explaining why we have developed our arguments for negative @xmath101 . in the gauge theory representation of 2 + 1 dimensional gravity with a negative cosmological constant , it is possible to reproduce the full diffeomorphism transformation properties of @xmath103 and @xmath104 by a gauge transformation in both the covariant and canonical formalisms . in the covariant formalism , this gauge transformation must be chosen so that the gauge parameters for the connections @xmath105 are equal to @xmath106 where @xmath65 is the same in both cases . in the canonical formalism , the situation is a little more complicated , and perhaps not well - known , so we devote some space to it . if in this case we set @xmath107 , then there are only two arbitrary functions that parametrise diffeomorphisms , and it is easy to see that these two parameters only generate spatial diffeomorphisms . we are thus led to consider the case @xmath108 . in a canonical theory , diffeomorphisms and lorentz transformations are realised by the action of the constraints . in this case , the hamiltonian is equal to @xmath109\ , , \ ] ] via the two lagrange multipliers @xmath110 and @xmath111 , the hamiltonian induces a diffeomorphism defined by @xmath112 and a lorentz transformation defined by @xmath113 . explicitly , @xmath114 let us now compare these transformations laws with those obtained by a gauge transformations parametrised by @xmath115 . it is convenient to define @xmath116 the transformation equations for @xmath117 and @xmath118 read @xmath119 . \label{40}\end{aligned}\ ] ] comparing this with ( [ mm5 ] ) and ( [ mm5a ] ) we recognise these expressions as the canonical formulae for diffeomorphisms parametrised by @xmath120 along with a lorentz transformation with parameter @xmath121 . using @xmath122 ( @xmath123 and is used to raise and lower spatial indices ) , we can pick out @xmath124 and @xmath125 we can now see explicitly that if we had set @xmath126 , then the gauge transformations ( [ 40 ] ) would not generate timelike diffeomorphisms . the classical black hole solution @xcite in lorentzian signature can be conveniently written in proper radial coordinates as @xmath127 in these coordinates , the horizon is at @xmath128 . @xmath129 is an angular coordinate with period @xmath130 . note that the above metric represents only the exterior of the black hole . the inner regions can be obtained by replacing some hyperbolic functions by their trigonometric partners . the mass @xmath8 and angular momentum @xmath9 of the black hole are given in terms of @xmath131 as @xmath132 and the relation between the schwarzschild radial coordinate @xmath133 and the proper radial coordinate @xmath27 is @xmath134 by going to its euclidean section ( see eq . ( [ mm19eu ] ) below ) , the black hole ( [ mm19 ] ) can be seen to have a temperature @xmath135 and , using the first law of black hole mechanics , @xmath136 , we find an entropy @xmath137 the metric can be written in first order form as @xmath138 so that after computing the spin connection , the gauge connection representing the black hole is given by @xmath139 or in matrix form , @xmath140 where @xmath141 and @xmath142 . we can put this solution into the gauge ( [ a8a ] ) , @xmath143 so that @xmath144 we then see that for the black hole @xmath145 ( note that this value of @xmath81 can be changed by a rescaling of @xmath27 . ) we can see from ( [ mm25 ] ) that the gauge connection @xmath146 leads to a non - trivial holonomy around the closed loops of constant @xmath27 and @xmath147 , @xmath148 the first step in discussing the algebra of global charges for the 2 + 1 dimensional black hole is to choose appropriate boundary conditions for the gauge fields @xmath149 . from the on - shell gauge fields ( [ mm22 ] ) , it is natural to impose the conditions @xmath150 which lead to the conditions @xmath151 on the diffeomorphism parameters . this , along with the constraint @xmath152 then defines the complete set of diffeomorphisms that leave the boundary conditions invariant and preserve the gauge ( [ a8a ] ) and ( [ a8aa ] ) . we have seen above that if we impose on these diffeomorphisms the supplementary condition @xmath153 then the algebra of global charges associated with the remaining diffeomorphisms , @xmath154 leads to a pair of virasoro algebras , one sector coming from each of the gauge fields , with central charge @xmath155 . in terms of the gauge field ( [ mm22 ] ) representing the black hole , the global charges ( without condition ( [ mm25cc ] ) ) generate the transformations @xmath156 let us now look at the form of these transformations at infinity ( corresponding to placing the boundary at infinity ) . then , focusing on the leading order terms of order @xmath157 , we find that conditions ( [ mm25cc ] ) with @xmath158 are precisely what is required for them to vanish . it is easy to check that rescaling the coordinate @xmath159 ( this means changing the value @xmath81 ) introduces the ( more general ) condition @xmath160 which of course includes the above discussion of the black hole solution in the gauge ( [ mm22 ] ) . thus the sub - algebra of global charges defined by ( [ dif ] ) and the condition ( [ condition ] ) generates asymptotic isometries of the black hole metric . note that these automatically include the anti - de sitter group @xmath161 . let us make a direct comparison with the asymptotic isometries found in @xcite . using the results of the last subsection , we can translate the action of the transformations ( [ dif ] ) into the action of a temporal and spatial diffeomorphism . using the on - shell values of @xmath118 and @xmath162 , we see from ( [ mm10e ] ) and ( [ mm10f ] ) , and from the appropriate coordinate relations that @xmath163 so that we get the diffeomorphism @xmath164 accompanied by a lorentz transformation with parameter @xmath165 . comparing ( [ mm25ae ] ) with the asymptotic isometries found in ref . @xcite , @xmath166 we see that there is exact agreement to leading order . ( here @xmath167 , and note that @xmath168 . ) the disagreement to sub - leading order is because the diffeomorphisms ( [ mm25a ] ) do not preserve our gauge choice in the interior . it seems that up to a choice of gauge in the interior ( which is irrelevant since we are trying to isolate the boundary dynamics ) , ( [ mm25a ] ) and ( [ dif ] ) are equivalent . we know from the analysis of global charges that they lead to a virasoro algebra with central charge @xmath169 . inserting ( [ condition ] ) and the value of @xmath101 given by ( [ mm18 ] ) , we find that @xmath170 which agrees with the result obtained in @xcite for the algebra of asymptotic isometries . as a result , we see that the algebra of diffeomorphisms obtained by brown and henneaux is related to the kac - moody algebra of the boundary wzw theory at infinity by the twisted sugawara construction ( [ a19b ] ) . perhaps the most important point about the analysis of global charges is that it goes through for a boundary located on any surface of constant @xmath27 . thus the set of diffeomorphisms defined by ( [ dif ] ) and ( [ condition ] ) leads to a virasoro algebra of global charges with @xmath171 on any such boundary . however , the connection between the virasoro algebra and isometries of the three metric appears to be valid only at infinity . at finite @xmath27 , the virasoro algebra is a subset of all deformations of that surface , but without any obvious property to distinguish it . in particular , if we take the boundary to be at the horizon , we find that the global charges that generate the virasoro algebra generate a particular subset of deformations of the horizon with components both tangential and normal to the horizon , described by ( [ mm25ad ] ) and ( [ mm25ae ] ) . we can not , of course , rule out the possibility that these deformations may have some distinguishing properties that remain undiscovered . if at finite @xmath27 , the subset of global charges which give rise to the virasoro algebra are not special in any way , perhaps one should consider all generators @xmath172 on an equal footing , and regard the condition ( [ mm25cc ] ) as a technical step that leads to the virasoro algebra . the fact that the virasoro algebra is a subalgebra of the algebra of deformations then suggests that the number of states generated by the larger algebra should be greater than or equal to the bekenstein value . we shall discuss this issue briefly in the conclusions . we have so far in this section derived the algebra of global charges on any boundary of fixed @xmath27 , and shown that a subset of them leads to the virasoro algebra with a classical central term . we have also made explicit the relation between the asymptotic isometries of ref . @xcite and this subset of global charges when defined at infinity . we may now count states in the conformal field theory , by looking at representations of the virasoro algebra , as done in @xcite . we must look for representations with a specific value of @xmath4 and @xmath173 , since according to ( [ a19b ] ) , these two quantities are related to the mass and spin of the black hole as @xmath174 or , neglecting the @xmath175 terms , @xmath176 note that fixing ( [ nn1 ] ) is equivalent to fixing @xmath8 and @xmath9 , justifying our choice of action ( [ a1 ] ) for the micro - canonical ensemble . as pointed out in @xcite , since @xmath3 is large , one can use the degeneracy formula for @xmath3 free bosons to deduce a density of states equal to @xmath177 which agrees with the bekenstein hawking entropy of the 2 + 1 dimensional black hole . it is interesting that now this analysis is not necessarily related to diffeomorphisms at infinity . we can think of these states as living on any surface of constant @xmath27 , and in particular they could be defined at the horizon . as far as we are aware , this system then provides the first explicit realisation of a set of deformations of a black hole horizon that can be quantised to yield the correct bekenstein hawking entropy . we end this section with a brief remark about the relation between the virasoro algebra ( [ a19c ] ) and the reduction of wzw theory to liouville theory as discussed in @xcite in the context of 2 + 1 dimensional gravity , and by a number of other authors in a more general context ( see @xcite for an extensive list of references ) . as explained in @xcite , the first step is to join the two chiral wzw theories into a single non - chiral wzw theory . then the reduction takes place by imposing certain constraints on the currents of the wzw theory and interpreting a second set of constraints as gauge fixing conditions . referring back to the conditions ( [ mm25cc ] ) and ( [ m223 ] ) , we see that ( [ m223 ] ) are just the chirality conditions for each @xmath89 sector . as we shall see below , in the wzw theories , these conditions arise from the dynamics of the wzw currents . ( [ mm25cc ] ) are equivalent to holding fixed @xmath178 and are equivalent to the constraints usually imposed in the reduction from wzw to liouville theory @xcite . the reduction is completed by a set of gauge fixing conditions on the currents . a direct application of the results of @xcite uses the simplest gauge fixing condition , @xmath179 . looking at the set of diffeomorphisms ( [ a16 ] ) that lead to ( [ a19c ] ) , one can see from ( [ mm27aa ] ) that although @xmath179 on - shell , @xmath180 thus , dirac brackets will be required to compute the operator algebra for the liouville theory . we can invoke ref . @xcite and see that for any gauge fixing condition , the constraints ( [ mm25cc ] ) lead to a liouville theory with a central charge that for large @xmath101 is equal to @xcite @xmath181 ( since we must use the non - standard convention that has @xmath21 ) . this agrees with the result we have obtained for the central charge from ( [ a19ca ] ) . we conclude that the virasoro algebra ( [ a19c ] ) can be interpreted as coming from an underlying liouville theory , as predicted in refs . @xcite and @xcite . the goal of this section is to compute the grand - canonical partition function for three dimensional gravity @xmath182d[\omega ] \exp\left ( -{1\over \hbar}i_{eh}[e,\omega;\beta,\omega]\right)\ ] ] where @xmath13 and @xmath14 are , respectively , the inverse temperature and angular velocity of the black hole . these two ( intensive ) parameters define the grand - canonical ensemble . the thermodynamic quantities such as average energy and entropy can then be obtained from @xmath183 through the thermodynamic formulae @xmath184 since @xmath185 the problem now requires two steps : first , we need to impose boundary conditions in the action principle such that the action has well defined variations for @xmath13 and @xmath14 fixed , and second , we need to actually compute @xmath186 . the grand - canonical partition function will be defined as a sum over euclidean metrics . the einstein hilbert action for euclidean gravity with a negative cosmological constant may again be represented by the difference of two chern simons actions , but now for the group @xmath16 @xcite . we shall use this property to compute the partition function . defining , @xmath187 and = \epsilon_{abc } j^c$ ] and @xmath188 . ] @xmath189 , @xmath190 , then up to boundary terms , @xmath191- i\left[\bar a\right]\right),\ ] ] where @xmath192 $ ] is the chern - simons action written in a 2 + 1 decomposition , @xmath193 = \frac{k}{4\pi } \int_m \varepsilon^{ij}\tr\left(-a_i\dot{a}_j + a_0 f_{ij}\right)d^3x . \label{cs1}\ ] ] the coupling constant @xmath101 is given by @xmath194 just as in the lorentzian case . the euclidean black hole solution is obtained by defining @xmath195 and @xmath196 in ( [ mm19 ] ) , giving @xmath197 for the euclidean calculation it is helpful to change coordinates to @xmath198 where @xmath199 here , @xmath200 is the minkowskian angular velocity . the angular coordinate @xmath201 has the standard period @xmath202 , while the time coordinate @xmath203 , which is also periodic , has the range @xmath204 . the @xmath205 surfaces in the black hole manifold have thus the topology of a torus with the identifications @xmath206 with @xmath207 integers . the radial coordinate @xmath27 has the range @xmath208 , with @xmath128 representing the black hole horizon . thus the euclidean black hole manifold is represented by a solid torus . the line @xmath128 represents the horizon , and is a circle at the centre of the solid torus . we shall discuss below whether in the sum over metrics in the partition function , this line should be regarded as an inner boundary of the solid torus ( see fig . the gauge field representing the euclidean black hole is given by @xmath209 where @xmath12 is a complex dimensionless number given by , @xmath210 the corresponding formulae for @xmath211 are obtained simply by complex conjugation . it is now natural to define a complex coordinate @xmath212 by @xmath213 the identifications ( [ i d ] ) induce in the complex plane the identifications @xmath214 we thus find that the black hole has a natural complex structure with a modular parameter @xmath12 given by ( [ tau ] ) . the important point is that by the coordinate transformation ( [ max1 ] ) , we have introduced a second pair of parameters into the metric , which by virtue of the periodicity relations ( [ i d ] ) , should be thought of as the intensive parameters @xmath13 and @xmath14 . on - shell , they are related to @xmath131 by ( [ bo ] ) , conditions which emerge by either imposing the absence of conical singularities in the euclidean manifold or using the first law @xmath215 . off - shell , @xmath12 and @xmath131 can be taken to be independent . let us first consider the outer boundary of the solid torus which represents spatial infinity . the correct boundary term at infinity should be consistent with boundary conditions that fix @xmath13 and @xmath14 at infinity . in the complex coordinates @xmath216 , the on - shell gauge field ( [ abh ] ) and its complex conjugate have the property that @xmath217 in terms of the spacetime coordinates @xmath218 , these conditions read , @xmath219 as can be verified using @xmath220 and that im@xmath221 . ( [ b11 ] ) , or ( [ b1 ] ) , are just the euclidean version of the chirality conditions ( [ chiral ] ) used in sec . 2 , and we shall also use them here as part of our boundary conditions . the residual gauge group is generated by the kac - moody currents ( @xmath222 ) @xmath223 which are ( anti- ) holomorphic functions of ( @xmath224 ) @xmath212 . the main difference with the analysis of sec . 2 is that we shall not fix the value of @xmath225 because we now work in a grand - canonical ensemble . we thus need to add to the action ( [ cs1 ] ) a boundary term at infinity in order to make it well defined . the appropiate action for the grand - canonical ensamble defined by the boundary conditions ( [ b1 ] ) and a fixed @xmath12 ( note that this fixes @xmath13 and @xmath14 at infinity ) is @xmath226 = \frac{k}{4\pi } \int_m \epsilon^{ij } \mbox{tr}\ , ( -a_i \dot a_j + a_0 f_{ij } ) - \frac{k\tau}{4\pi } \int_{t^2_\infty } \mbox{tr } a^2_\varphi \ , , \label{cs - h}\ ] ] plus a boundary term at the horizon that we discuss in the next paragraph . [ note that conditions ( [ b1 ] ) were also used in @xcite , with a slight modification that makes the modular parameter time dependent , in an attempt to obtain a better understanding of carlip s original paper @xcite . ] using the expressions ( [ m225 ] ) for the mass and spin , it is easy to see that the boundary term ( the only term that survives in the semiclassical limit ) gives the correct weight factor @xmath227 for the grand - canonical ensemble . the boundary conditions at the horizon are more subtle . in the lorentzian theory it is quite natural to introduce a boundary term at the horizon , since only the ` outer ' part of the black hole may be viewed as physical . in the hamiltonian formulation this means that one has to fix the hypersurfaces at the bifurcation point which leads to an isolated , non - smooth boundary , often referred to as joint or edge . in @xcite it was shown that such a non - smooth boundary gives rise to additional boundary terms . in the case of black hole spacetimes this joint contribution at the bifurcation point is responsible for the appearance of a non - zero charge located at the horizon , equal to one quarter the black hole area , which then can be interpreted as the entropy of the black hole @xcite . in the euclidean spacetime the situation is different since the ` inner ' part of the black hole is already cut off from the manifold . nevertheless , it has been argued ( see , for example , @xcite ) that one also has to introduce a boundary at the origin of the euclidean spacetime by removing a point from the euclidean @xmath228-plane because the foliation using the vector field @xmath229 is not well defined at @xmath230 . since we are using a hamiltonian action , the @xmath231 surfaces are annuli with two boundaries and it is necessary to give some boundary conditions at the horizon in order to ensure that the hamiltonian is a well defined functional and its derivatives exist . we must now decide which boundary conditions to impose at the horizon . note firstly that the on - shell black hole field ( [ abh ] ) has a time component at @xmath232 that does not vanish , @xmath233 since @xmath203 is an angular coordinate , one may suspect that this indicates the presence of a non - trivial holonomy in the temporal direction . however , an explicit calculation using ( [ timecoordinate ] ) reveals that this is not the case and @xmath234 can be set equal to zero by a globally well - defined gauge transformation . the situation changes if one allows a conical singularity at the horizon as advocated in @xcite . in this case , the non - vanishing of @xmath235 at the horizon does imply the existence of a holonomy . to handle this situation classically , one can either remove the horizon from the manifold and hence change the topology or , alternatively , one can introduce a source term or wilson line along the horizon and work with a solid torus topology . the introduction of a wilson line at the core of the solid torus was also suggested in @xcite , but as far as we know its consequences were never explored . the above discussion suggests that we should look to fix @xmath235 at the horizon . this can be achieved using the canonical action with no boundary term . however , this does not then fix the condition ( [ timecoordinate ] ) ( as opposed to allowing conical singularities ) . this can be ensured only by introducing a wilson line term whose variation , when coupled to the bulk action , fixes ( [ timecoordinate ] ) . the wilson line term is @xmath236 which is localised along a non - contractible loop in the solid torus defined by @xmath237 . here @xmath12 is a parameter along the wilson line . @xmath238 is an element in the lie algebra of the group that specifies the vector charge of the source . we could also have included a dynamical source with its own kinetic term @xcite , but it is not clear that this is necessary . to conform with the usual choice of coordinates , we shall take the wilson line to be located along the curve @xmath128 , and to be parametrised by the variable @xmath201 . since we want the action principle to contain the black hole in its space of solutions , the strength of the source is fixed by looking at the classical field ( [ abh ] ) at the point @xmath128 . we take @xmath239 and the source term becomes @xmath240 where now @xmath241 and @xmath242 refer to the @xmath243 plane ( not @xmath244 ) . note that the choice of orientation of @xmath238 in the lie algebra is unimportant , since it only fixes the orientation of the solution in internal space . the source term ( [ m2 ] ) is the analogue of the geometrical horizon term proportional to the area that was mentioned above . indeed , on - shell , the value of this term ( plus the other copy ) is equal to one fourth of the area . in the off - shell language of carlip and teitelboim @xcite , the field equations will lead to @xmath245 and @xmath246 as expected , since no conical singularities are allowed . the total euclidean action for the black hole is therefore the sum of ( [ cs - h ] ) and ( [ m2 ] ) giving , @xmath247 = i[a,\tau ] - i[\bar a,\bar\tau ] , \ ] ] with @xmath248 & = & \frac{k}{4\pi } \int_m \epsilon^{ij } \mbox{tr}\ , \left ( -a_i \dot a_j + a_0 f_{ij } \right ) \cr & & - \frac{k\tau}{4\pi } \int_{t^2_{\rho=\infty } } \mbox{tr } a^2_\varphi + \frac{k}{2 } \int_{s_{\rho=0}^1 } a^3_\varphi , \end{aligned}\ ] ] ( where since @xmath249 , in euclidean space , the sign of the bulk term is different to that in ( [ a1 ] ) ) . it is worth stressing here , once again , that this action has well defined variations provide @xmath12 is fixed . note , finally , that the micro - canonical action ( [ a1 ] ) and the grand - canonical action ( [ y1 ] ) differ by exactly the boundary term at infinity equal to @xmath250 that one would expect on general grounds , and that has been advocated by brown and york @xcite . we shall now explore the semi - classical and quantum mechanical consequences of this action . the action ( [ cs - h ] ) with the source term ( [ m2 ] ) has the right semiclassical value . since the canonical action is zero on the classical black hole background ( [ abh ] ) one obtains , @xmath251 where @xmath252 is given by @xmath253 as expected , and comes entirely from the source term at the horizon . this partition function is grand - canonical because in the action principle only @xmath13 and @xmath14 ( or @xmath12 ) were fixed . this means that @xmath183 is a function of @xmath13 and @xmath14 . using ( [ mm19a ] ) and ( [ bo ] ) one can write @xmath183 as a function of @xmath13 and @xmath14 , @xmath254 . \label{gh}\ ] ] this is the semiclassical value of the grand - canonical partition function . in a chern - simons formulation of three dimensional euclidean quantum gravity , the partition function involves a sum over an @xmath16 gauge field . we must therefore deal with the fact that the group @xmath16 is not compact and the black hole manifold has a boundary . ( for manifolds without boundaries it has been proved in @xcite that @xmath183 can be understood as a complexified @xmath18 problem . ) as has been stressed in @xcite , one can hope to make progress by treating each of the complex connections @xmath255 and @xmath256 as real , so that the partition function becomes just the product of two , complex conjugate , @xmath18 partition functions . following @xcite we shall write @xmath257 and compute @xmath258 , hoping to make sense of its relation to the trace over states in @xmath259 by some form of analytic continuation . we thus consider the functional integral , @xmath260d[a_0 ] \exp\left(\frac{i}{\hbar } i[a_i , a_0;\tau]\right),\ ] ] where @xmath261 is given in ( [ y1 ] ) , and we integrate over all gauge fields satisfying the boundary conditions ( [ b1 ] ) . integrating over @xmath235 gives the constraint that @xmath262 , which implies that @xmath263 where @xmath264 is a map from @xmath8 to the group . we also have to fix a gauge and we can do this using the gauge fixing choice ( [ a8a ] ) which fixes @xmath265 since the first homotopy group of the solid torus is non - trivial , @xmath266 could be multi - valued . after inserting the gauge fixed and flat @xmath267 into the functional integral one obtains an expression that only depends on the boundary values of the map @xmath266 @xcite , @xmath268\right),\ ] ] where the chiral wzw action is given by , @xmath269 & = & -\frac{k}{4\pi } \int_{t^2 } \mbox{tr } ( { \partial } _ \varphi g^{-1 } \dot g ) - \frac{k}{12\pi } \int_m \mbox{tr}(g^{-1}dg)^3 - \nonumber\\ & & \frac{\tau k}{4\pi}\int \mbox{tr } ( g^{-1 } { \partial } _ \varphi g)^2 + \frac{k}{4\pi } \int \tr ( kg^{-1}{\partial } _ \varphi g ) . \label{cwzw}\end{aligned}\ ] ] here @xmath238 is related to the original @xmath238 of ( [ m1 ] ) by conjugation by @xmath270 and so may be taken equal to @xmath238 without loss of generality . the reduction of the three dimensional problem to a two dimensional conformal theory is a consequence of the absence of propagating degrees of freedom in the three dimensional field theory . this allowed us to solve the constraint . a second consequence of the absence of degrees of freedom is that the boundary term at the horizon is now linked to the boundary term at infinity . the conformal field theory lives on a torus with no reference at all to the radial coordinate . the chiral wzw action ( [ cwzw ] ) has two pieces . the kinetic term ( first line ) defines the commutation relations of the theory . as is well known @xcite , these commutation relations are given by the @xmath18 kac - moody algebra , @xmath271 = i \hbar \epsilon^{ab}_{\ \ c } t^c_{n+m } + n \hbar \frac{k}{2 } \delta^{ab } \delta_{n+m,0 } , \label{km}\ ] ] where the @xmath272 are the fourier components of the gauge field , @xmath273 note that it is more convenient to define the @xmath272 in this way rather than as in ( [ a10 ] ) in euclidean space . the second piece ( second line in ( [ cwzw ] ) ) is the hamiltonian . since the hamiltonian involves @xmath274 it has to be regularised by choosing a normal ordering . moreover , it is well known that the coefficient of @xmath4 in the non - abelian theory is not @xmath275 but rather @xmath276 . in the following we shall be interested in the limit of large @xmath101 and therefore this shift can be neglected . the partition function can then be calculated as @xmath277 where for large @xmath101 , @xmath4 is given by @xmath278 and @xmath279 labels the spin of the different @xmath18 representations ( it can be integer or half - integer ) . the symbol tr@xmath280 represents a trace over states belonging to the representation with spin @xmath279 . the spin structure term can be thought of as being equivalent to having a non - zero flux through the hole created by closing the time direction . the problem of computing @xmath281 for a given value of @xmath282 and @xmath283 is well known and explicit formulae are available . for @xmath18 , writing @xmath284 and taking @xmath285 , one has @xcite , @xmath286 the denominator in ( [ ch ] ) does not depend on @xmath279 and it is therefore a global factor in the partition function . this factor provides a quantum correction to the entropy that does not depend on newton s constant . the value of this contribution can be calculated in the limit of small @xmath12 ( large black holes ) as @xmath287 inserting @xmath284 and using ( [ s ] ) and ( [ z1 ] ) , the contribution to the entropy from this term is equal to @xmath288 this correction , which does not depend on @xmath289 , has already appeared in the literature @xcite . let us now consider the numerator in ( [ ch ] ) . in the limit of large @xmath101 , the sum over @xmath290 is suppressed because @xmath291 exponentially for @xmath292 . we thus keep only the term @xmath293 . this means that the bekenstein hawking entropy does not come from the higher kac - moody modes but from the sum over representations . this is quite different from the analysis in @xcite in which the entropy comes from the term with zero spin calculation . ] . setting @xmath293 and defining @xmath294 , we arrive at @xmath295 where @xmath296 represents the correction that does not depend on @xmath289 , while @xmath297 the contribution from the sum over @xmath290 whose logarithm vanishes at least as @xmath298 . note that in the quantum calculation the term @xmath299 , arising from the boundary term at the horizon , produces the factor @xmath300 which may be interpreted as a @xmath19 operator that alternates bosonic and fermionic representations . we shall see that this operator has an important role in producing the right contribution to the entropy . the sum ( [ s1 ] ) is not what we want because the black hole does not belong to the set of states that we are considering when we calculate the @xmath18 partition function . indeed , @xmath301 labels unitary @xmath18 representations for which @xmath302 however , the value of @xmath4 on the black hole background is @xmath303 , from which it follows that @xmath304 for large @xmath301 . thus , the states we are interested in for the black hole belong to an @xmath18 representation with complex spin . even in the non - rotating case , @xmath305 , @xmath306 is negative and thus can not be real . we shall not attempt to give any interpretation to such a representation here , but the reason behind it lies in the fact that euclidean gravity is a chern - simons theory for the group @xmath16 rather than two copies of @xmath18 . let us analytically continue @xmath301 to the complex plane and set @xmath307 . the sum reduces to @xmath308 note that the @xmath19 operator now has eigenvalues @xmath309 . remarkably , this term which produced the right entropy in the semiclassical calculation , also provides the right degeneracy in this quantum mechanical calculation . consider the total partition function @xmath310 . since @xmath311 is complex we define @xmath312 . using @xmath313 and @xmath314 we find @xmath315 with @xmath316 since , for a black hole , @xmath8 and @xmath9 are related to the inner and outer horizons by ( [ mm19a ] ) ( or in the euclidean version by replacing @xmath196 ) , we obtain @xmath317 the term @xmath318 outside the exponential in ( [ s4 ] ) combines with the sum to give the required measure over @xmath8 and @xmath9 . the entire partition function becomes @xmath319 and implies that the density of states is @xmath320 . using ( [ r1 ] ) , we obtain @xmath321 in complete agreement with the bekenstein hawking value . finally , the semiclassical grand - canonical partition function ( [ gh ] ) and thus the entropy @xmath252 can be obtained by a simple saddle point approximation ( [ s2 ] ) . noticing that the sum ( [ s2 ] ) has a saddle point at @xmath322 we find that @xmath323 where we have inserted @xmath314 . computing the complex modulus of @xmath183 , and taking into account the value of @xmath12 given in ( [ tau ] ) , we find that @xmath324 in complete agreement with ( [ gh ] ) . it is interesting to note that this calculation can be repeated in the case where @xmath325 , the semiclassical limit , in a purely abelian theory . details of this calculation are given in the appendix . we have performed two separate calculations of the entropy of the 2 + 1 dimensional black hole using the relation between 2 + 1 dimensional gravity and chern - simons theory . in sec . 2 , we have worked in the micro - canonical ensemble , and have calculated the density of states starting from the kac - moody algebra of global charges ( wzw theory ) . we have computed the correct density of states by relating the global charges to a particular virasoro algebra via a twisted sugawara construction , in a way first considered in @xcite . this virasoro algebra turns out to generate the same asymptotic isometries considered in refs . @xcite , if the analysis of global charges is performed at infinity . we have shown that it is also present on any other boundary surface at constant radius , including the black hole horizon . in sec . 3 , we have worked in the grand - canonical ensemble , which we have defined by adding an appropriate boundary term at infinity . we have found that in order to obtain the correct partition function we must also add a source term at the horizon . this source term gives the correct value of the partition function both semi - classically and in an exact quantum mechanical calculation . it is the analogue of the term equal to @xmath326 that is sometimes added to the canonical einstein hilbert action to yield the correct semi - classical partition function for black holes in arbitrary dimensions . in the micro - canonical calculation we saw that we obtain the correct density of states at a given value of mass and spin after we make a reduction from the wzw theory to a theory of boundary deformations that satisfies the virasoro algebra , or equivalently to a liouville theory . since this reduction involves additional constraints on the allowed global charges , it one expects that the density of states should be greater in the wzw theory ( although possibly equal to leading order ) . why , then , can one not calculate the density of states directly in terms of representations of the kac - moody algebra ? in the wzw theory , it seems clear that there are an insufficient number of states ( and for this reason , in the grand - canonical ensemble the correct partition function required the addition of a source term to give a larger apparent degeneracy ) . the answer presumably lies in the use of a twisted sugawara construction to connect the kac - moody and virasoro algebras . although the states we eventually count are unitary states with respect to the standard quantisation of the virasoro algebra , they most probably correspond to a non - unitary , twisted representation of the kac - moody algebra only if the vacuum has @xmath327 , which translates to a negative eigenvalue for @xmath328 defined in an untwisted sugawara construction from the underlying wzw theory . however , it is interesting to note from ( [ m224 ] ) that the condition @xmath327 corresponds to @xmath329 , @xmath330 ( anti - de sitter space ) and @xmath331 , while the black hole vacuum @xmath332 has @xmath333 but @xmath334 . this strongly suggests that our ensemble includes anti - de sitter space as its true vacuum . ] . thus in order to find the correct density of states , we should look at a different set of states to those usually constructed in representations of the kac - moody algebra . of course , this calculation is further complicated by the fact that representations of @xmath89 wzw theory are poorly understood . in contrast to the micro - canonical case , in the grand - canonical calculation we have managed to obtain the correct density of states directly in a standard ( not twisted ) wzw theory , using standard expressions for the partition function ( and an analytic continuation ) . however , this result came from a partition function with a `` spin - structure '' term , twisting the wzw theory in the time direction . it seems likely that the twists in the space and time directions that we have discussed are related by a modular transformation is the time coordinate and to verify that this can yield the same partition function . ] . in this context , it is also interesting to speculate on the correct interpretation of the source or wilson line term that gives rise to the non - trivial `` spin - structure '' . we saw in our semiclassical and quantum mechanical calculations that this term produces the black hole entropy not as a density of states , but rather as an operator eigenvalue . we conjecture that this source term can be understood in a different context as an effective action term . if we begin with the micro - canonical picture of a twisted wzw model with trivial spin - structure , then it should be possible to get back to an untwisted wzw theory by integrating out the additional degrees of freedom arising from the spatial twisting of the wzw theory . the effect of integrating out these states would be to introduce the spin - structure term . it would be extremely nice to see this connection explicitly . finally , we comment on what these various calculations tell us about the location of the degrees of freedom giving rise to the black hole entropy . while carlip @xcite has advocated that these degrees of freedom should be located on the black hole horizon , strominger @xcite has shown that the algebra of asymptotic isometries of the metric leads to the correct density of states . our discussion of the global charges and of the reduction to liouville theory in sec . 2 showed explicitly that the virasoro algebra responsible for the density of states can live at any value of the radius @xmath27 . a similar conclusion is suggested by the @xmath27 independence of the grand - canonical calculation . we were able to obtain an explicit form for a set of deformations of the horizon whose classical algebra is the virasoro algebra , with the same classical central charge as the set of asymptotic isometries , that gives the correct bekenstein hawking entropy . the discovery of an algebra of operators at the horizon , whose representations yield the correct density of states , makes an extension of these ideas to higher dimensions look more plausible . whereas it is unlikely that the algebra of asymptotic isometries of black holes in higher dimensions could lead to the correct density of states , it seems more likely that an algebra of deformations of the horizon could have the required properties . although these two algebras are identical in our case , this is because of the trivial dynamics of 2 + 1 dimensional gravity and would certainly not be true in general . in higher dimensional applications , one would then have to relate these charges at the horizon to the mass and spin of the black hole . this problem is solved in our case by the same trivial dynamics of the theory . what is not so clear from our analysis is the physical interpretation of the subset of deformations that lead to the virasoro algebra at any finite radius . however , as mentioned above in the context of wzw theory , the restriction to this subset of deformations will reduce the density of states , but it may well not change it to the leading semi - classical order . in that case , the complete algebra of deformations of the horizon would lead to the correct density of states . however , checking this probably requires getting a handle on the problem of state counting in the wzw theory . we are hopeful that a generalisation to arbitrary dimensions of the calculations that have been developed for the 2 + 1 dimensional black hole may come about through the algebra of deformations of the horizon . indeed , this algebra may have a very direct application for black holes in higher dimensions whose near - horizon behaviour is similar to the 2 + 1 dimensional black hole . in this case we would be able to talk of states localised at the horizon . we thank r. baeza , s. carlip , f. falceto , a.gomberoff , m. gaberdiel , m. henneaux , t. jacobson , c. martnez , f. mendez , i. sachs , s. stanciu , c. teitelboim , a. tseytlin , l. vergara and j. zanelli for helpful discussions . mb was supported by grants # 1970150 from fondecyt ( chile ) , cicyt ( spain ) # aen-97 - 1680 , and also thanks the spanish postdoctal program of ministerio de educacin y cultura . tb acknowledges financial support from the german academic exchange service ( daad ) . meo was supported by the pparc , uk . in the weak coupling limit the non abelian nature of the theory can be neglected . therefore it should also be possible to derive the semiclassical value of the partition function from the abelian theory , and we shall now show that this is indeed the case . if we define @xmath335 where @xmath336 are now the generators of the abelian lie algebra , the chiral wzw action ( [ cwzw ] ) reduces to @xmath337=\frac{k}{8\pi } \int \left ( -\partial_0 x^a \partial_\varphi x_a + \tau \left(\partial_\varphi x\right)^2 - 4\pi \partial_\varphi x^3 \right).\ ] ] from this action we find the field equations @xmath338 thus the general solution of these field equations contain only right moving modes which expresses the chirality of our theory . motivated by the field equation we may expand @xmath339 in normal modes as , @xmath340 note that the action ( [ c1 ] ) has a gauge symmetry , @xmath341 , which we use to set the function @xmath342 to zero . replacing the mode expansion into the chiral wzw action , one obtains @xmath343= \int dx^0 \left ( - \sum_{n\neq 0 } \frac { \dot{\alpha}_0^a \alpha_{a\:n}}{ikn } + \sum_{n\ge 1 } \frac{2 \dot{\alpha}_{-n}^a \alpha_{a\:n}}{ikn}+ \tau l_0 - 2\pi \alpha_0 ^ 3 \right),\ ] ] where @xmath4 is a virasoro generator , @xmath344 and where we have eliminated total derivative terms . the action ( [ c13 ] ) gives rise to the abelian kac - moody algebra @xmath345 = n \hbar \frac{k}{2 } \delta^{ab } \delta_{n+m,0 } , \ ] ] from which we can define a fock space in the standard manner . the partition function can be calculated as , @xmath346 \exp \left ( \frac{i}{\hbar } i_{cwzw}[\alpha,\tau ] \right ) = \int d\alpha_0 \tr \exp\left ( \frac{i\tau}{\hbar } l_0 -\frac{i}{\hbar } 2\pi \alpha_0 ^ 3 \right).\ ] ] from the field equation ( [ c2 ] ) we deduce again that semi - classically only the zero modes contribute to the partition function . hence we may split the partition function ( [ cz ] ) into a leading and a sub - leading part @xmath347 where we have introduced the number operator @xmath348 in the saddle point approximation the integral over @xmath349 ( @xmath350 and @xmath351 ) is easily evaluated @xmath352 note that the integral over @xmath349 is only well defined , if we assume @xmath353 . for the calculation of the prefactor we need to know the number of states at each level @xmath354 . this calculation is well known and can be found in @xcite . in the limit @xmath355 ( large black holes ) one finds @xmath356 the total partition function is thus given by @xmath357 thus the abelian calculation not only produces the bekenstein- hawking part of the partition function but also leads to the @xmath296 correction discussed above . s. hyun , _ u duality between three - dimensional and higher dimensional black holes _ , hep - th/9704005 ; k. sfetsos and k. skenderis , _ microscopic derivation of the bekenstein hawking entropy formula for non - extremal black holes _ , hep - th/9711138 . a. strominger , _ black hole entropy from near horizon microstates , _ hep - th/9712251 ; see also d. birmingham , i. sachs and s. sen , _ entropy of three - dimensional black holes in string theory , _ hep - th/9801019 .

we calculate theoretical isochrones in a consistent way for five filter pairs near the and band atmospheric windows ( , , , f110w f205w , and f110w f222 m ) using the padova stellar evolutionary models of girardi et al . we present magnitude transformations between various-band filters as a function of color . isochrones with extinction of up to 6 mag in the band are also presented . as found for the filter pairs composed of & band filters , we find that the reddened isochrones of different filter pairs behave as if they follow different extinction laws , and that the extinction curves of _ hubble space telescope _ nicmos filter pairs in the color - magnitude diagram are considerably nonlinear . because of these problems , extinction values estimated with nicmos filters can be in error by up to 1.3 mag . our calculation suggests that the extinction law implied by the observations of rieke et al for wavelengths between the and bands is better described by a power - law function with an exponent of 1.66 instead of 1.59 , which is commonly used with an assumption that the transmission functions of and filters are dirac delta functions .

astro-ph0601470

kim et al . ( 2005 , hereafter paper i ) have calculated theoretical isochrones with extinction for some @xmath4 and @xmath1 band filters using the padova stellar evolutionary models by girardi et al . ( 2002 ) . in paper i , we found that the reddened isochrones of different filter pairs in @xmath4 and @xmath1 bands behave as if they follow different extinction laws , and that care is needed when applying an extinction law obtained with one filter pair to other , similar filter pairs . for example , if the extinction law for the johnson - glass @xmath4 and @xmath1 filters obtained by rieke , rieke , & paul ( 1989 ) is directly applied to the photometry from the _ hubble space telescope _ ( _ hst _ ) nicmos filters ( f160w , f205w , and f222 m ) , estimated extinction values can be in error by up to 0.3 mag for true extinction at @xmath1 of 6 mag or less . to reduce this error , paper i introduced an `` effective extinction slope '' for each filter pair and isochrone model . it was also found that the extinction behavior of isochrones in the color - magnitude diagram ( cmd ) for filter pair f160w f222 m is highly nonlinear ( i.e. , the amount of extinction is not proportional to color excess ) because of a significant width difference in the two filters . these problems are certainly not limited to the isochrones for filter pairs in the @xmath4 and @xmath1 bands . this problem will apply to any situation in which one applies an extinction law deduced from one filter pair to other similar filter pairs . furthermore , the nonlinear behavior of the extinction vector in the cmd will be problematic for filter pairs with significant difference in width . in the present paper , we extend the calculations performed in paper i to the isochrones for filter pairs in the @xmath0 and @xmath1 bands . the filters considered here are the four ground - based filters @xmath0 , @xmath1 ( johnson et al . 1966 ) , @xmath2 ( wainscoat & cowie 1992 ) , and @xmath3 ( @xmath1-short ; developed by m. skrutskie ; see the appendix of persson et al . 1998 ) , and the three nicmos filters f110w , f205w , and f222 m ( transmission functions of these filters are shown in figure [ fig : filter ] ) . out of these seven filters , we consider five filter pairs : @xmath0@xmath1 , @xmath0@xmath2 , @xmath0@xmath3 , f110w f205w , and f110w f222 m . we adopt a vega - based photometric system ( vegamag system ) , which uses vega ( @xmath5 lyr ) as the calibrating star . for photometric zero points of nicmos filters , we adopt @xmath6 values from the nicmos data handbook ( ver . 5.0 ) : 1775 jy for f110w , 703.6 jy for f205w , and 610.4 jy for f222 m . for the spectra of synthetic stellar atmospheres , we adopt kurucz atlas9 no - overshoot models ( kurucz 1993 ) calculated by castelli et al . the metallicities of these models cover the values of [ m / h ] = @xmath7 to @xmath8 . a microturbulent velocity @xmath9 and a mixing length parameter @xmath10 are adopted in the present study . for the temporal evolution of effective temperature and luminosity as functions of stellar mass ( i.e. , stellar evolutionary tracks ) , we adopt the basic set " of the padova models ( girardi et al . we consider isochrones with a metallicity @xmath11 = 0.0001 , 0.001 , 0.019 , and 0.03 . the stellar spectral library and the evolutionary tracks we adopted assume a solar chemical ratios . for more details on the magnitude system , stellar spectral library , and evolutionary tracks that we adopt here , readers are referred to paper i. throughout this paper , we generically refer to the atmospheric wavebands centered near 1.25 , 1.65 , and 2.2 , as the @xmath0 , @xmath4 , and @xmath1 bands , whereas we refer to the johnson - glass filters ( johnson et al . 1966 ; glass 1974 ) as the @xmath0 , @xmath4 , and @xmath1 filters . we first prepare a table of magnitudes for all spectra in atlas9 models in the @xmath0 and @xmath1 band filters , covering a large range in @xmath12 , @xmath13 , and [ m / h ] , using equations ( 5 ) and ( 6 ) of paper i. we use this table as a set of interpolates for the @xmath12 , @xmath13 , and @xmath11 values predicted by the stellar evolution models for a given age in order to estimate synthetic isochrones . isochrones for @xmath14 , calculated in this way , are shown in figures [ fig : iso1]@xmath15[fig : iso4 ] for four different metallicities and four ages . the color differences between filters are more prominent for the highest metallicity isochrones . in most cases , isochrones for @xmath2 and @xmath3 are nearly indistinguishable , and those for f205w and f222 m are quite close to each other . in general , for red giants , intrinsic color differences between the atmospheric and nicmos filters are 0.20.4 mag . as an independent check of our procedure , in figure [ fig : padova ] we compare our @xmath16 versus @xmath1 isochrones to those calculated by girardi et al . the isochrones match nicely , except at the extremes . the discrepancy in the bright end is caused from the empirical m giant spectra that girardi et al . ( 2002 ) added to their spectral library , and that in the faint end is by the addition of late m dwarf spectra . the discrepancies are considerable only at the top and bottom @xmath17 1 mag of the isochrone , where only a small fraction of giants reside , or else stars are too faint for most observational situations . magnitude transformations between @xmath1-band filters can be obtained from our isochrones . we find that the magnitude difference can be well fitted by a third - order polynomial for @xmath18 mag , and by a separate second - order polynomial for @xmath19 mag . the largest residuals from the fit are 0.012 mag for the former and 0.008 mag for the latter . the coefficients of the best - fit functions are presented in tables [ table : trans1 ] and [ table : trans2 ] , along with the residuals and fitting ranges . one useful way of using these tables would be to compare the magnitudes of helium - burning clump giant stars , which are rather insensitive to metallicity or age and are often used as distance indicators , observed with different photometric systems ( the clump stars show a small variation with age , however ; see figer et al . 2004 ) . we present here isochrones with @xmath1-band extinctions of up to 6 mag , some of which are shown in figures [ fig : red1]@xmath15[fig : red5 ] . for the extinction between the @xmath0 and @xmath1 bands , we adopt a power law , @xmath20 where we choose @xmath21 m , and @xmath22 is the extinction at @xmath23 . when assuming that the transmission functions of the @xmath0 and @xmath1 filters are dirac delta functions centered at 1.24 and 2.21 @xmath24 m , respectively , the extinction law by rieke et al . ( 1989 ) gives @xmath25 . however , as discussed below in this section , the apparent extinction behavior of isochrones in the cmd can differ from the actual extinction law , as a result of a nonzero width and asymmetry of the filter transmission functions . we find that @xmath26 makes the isochrone for the @xmath27 , age = @xmath28 yr model behave in the cmd as if it followed an extinction law with @xmath25 . we choose this particular isochrone for calibrating the extinction law , with the assumption that the stars used in rieke et al . ( 1989 ) to derive their extinction law , which are the stars in the central parsec of our galaxy , can be represented by the same metallicity and age . for the sake of comparison , isochrones in figures [ fig : red1]@xmath15[fig : red5 ] have been dereddened by the amount @xmath29 , where the central wavelength of the filter @xmath30 is defined by equation ( 8) of paper i , and given in table [ table : lambdac ] . since we have dereddened the isochrones with the known amount of extinction at @xmath30 , all the dereddened isochrones with different extinction values in figures [ fig : red1]@xmath15[fig : red5 ] should be coincident if the filter transmission functions were dirac delta functions centered at @xmath30 . as in paper i , the dereddened isochrones misalign significantly , and this implies that the amount of extinction inferred from a cmd is sensitively dependent on the shape of the filter transmission function . when estimating the amount of extinction from an observed cmd , one converts an observed color excess to an extinction value , following an assumed extinction law , which usually has the form of a power law . when one has photometric data from a pair of two filters , @xmath31 and @xmath32 , the amount of extinction can be estimated by @xmath33 where @xmath34 , @xmath35 and @xmath36 , @xmath37 are the magnitudes and the central wavelengths of the two filters , respectively , and subscript 0 denotes the intrinsic value . for estimating extinction from our isochrones , we first use @xmath25 . figure [ fig : adiff1 ] shows the difference between the inferred extinction values , using equation ( [ a_est ] ) and colors from our reddened isochrones , and the actual extinction values . here the extinction of each isochrone has been calculated using the mean color ( for @xmath38 ) and magnitude ( for @xmath39 ) of the reddened isochrone data points having intrinsic @xmath1-band magnitudes between @xmath156 and 0 mag . as the figure shows , the differences between estimated and actual extinction values are much larger for the nicmos filter pairs . the largest relative difference is @xmath40% , and the largest absolute difference is 1.25 mag . note that the extinction estimates for the @xmath27 and age = @xmath28 yr model inferred from @xmath4 and @xmath1 are very close to the actual extinction values , justifying our choice of @xmath26 for equation ( [ extinction ] ) . the error bar in the figure represents the standard deviation of @xmath41 values . some of the f110w isochrones show quite large deviations , as pointed out in appendix a of lee et al . ( 2001 ) . to reduce the problems seen in figure [ fig : adiff1 ] , paper i introduced an `` effective extinction slope '' @xmath42 for each filter pair and isochrone model , which is defined such that it better describes the extinction behavior in the cmd : @xmath43 where @xmath44 is the slope of the straight line that fits the distribution of reddened magnitudes versus reddened colors , as in figure [ fig : extlaw ] . this figure shows reddened @xmath1-band magnitudes and colors for the @xmath27 and age = @xmath28 yr isochrone ( the figure only shows an isochrone data point whose intrinsic @xmath1 magnitude is 0 , as an example ) . we calculate @xmath44 for data points of each isochrone whose intrinsic @xmath1 magnitudes are between @xmath45 and 0 mag , and take an average for each isochrone model . table [ table : alpha_eff ] shows the averages and standard deviations of @xmath42 values for each isochrone model . for atmospheric filters , the standard deviations of @xmath42 in an isochrone is generally much smaller than the differences of average @xmath42 values between different isochrones , while those for nicmos filters are relatively larger . the average @xmath42 values range from 1.403 to 1.610 , which are 15% to 0.02% smaller than the original @xmath5 value we adopted for extinction , 1.66 . as seen in figure [ fig : adiff3 ] , extinction values estimated by equation ( [ a_est ] ) with @xmath42 are closer to the actual values for atmospheric filters , but still deviate significantly from the actual values for nicmos filters , because of the nonlinear extinction seen in figure [ fig : extlaw ] . as pointed out in paper i , the nonlinear extinction behavior of nicmos filters is due to a significant difference in relative widths of the two filters : the width to central wavelength ratio @xmath46 is @xmath47 for f110w , while those for f205w and f222 m are @xmath48 and @xmath49 , respectively . figure [ fig : nonlin ] shows the effect of the filter width by comparing the extinction behavior of six imaginary filter pairs . filter pair @xmath50 represents @xmath0 and @xmath1 , whose @xmath46 values are both @xmath51 , and its extinction behavior in the cmd is nearly linear . on the other hand , filter pairs @xmath44 and @xmath52 , which represent filter pairs f110w f205w and f110w f222 m , show considerable nonlinearity . when the @xmath46 of the short - wavelength filter is reduced by @xmath53% , however , the extinction behaves much more linearly ( @xmath54 and @xmath55 ) . this shows that the nonlinear extinction in filter pairs f110w f205w and f110w f222 m is due to a relatively larger @xmath46 value of the f110w filter . when both filters have the same large @xmath46 values ( @xmath47 ) , the extinction becomes almost linear again ( @xmath56 ) . the introduction of effective extinction slopes does not alleviate the nonlinear extinction problem of nicmos filter pairs . so in table [ table : nonlin ] we provide the coefficients of best - fit third - order polynomials of the extinction curves for nicmos filter pairs shown in figure [ fig : adiff1 ] so that one can accurately estimate the extinction value for nicmos filter pairs as well . note that we assumed @xmath57 for all isochrone models when estimating @xmath58 in figure [ fig : adiff1 ] . as in paper i , we find that for the filters whose extinction behavior is relatively linear , the transformation of extinction values from filter @xmath32 to filter @xmath59 can be obtained by @xmath60 if @xmath39 is estimated with @xmath42 , and the original @xmath5 value of 1.66 is used in the above equation . we have calculated in a consistent way five near - infrared theoretical isochrones for filter pairs composed of @xmath0 and @xmath1 filters : @xmath0@xmath1 , @xmath0@xmath2 , @xmath0@xmath3 , f110w f205w , and f110w we presented isochrones for a @xmath11 of 0.00010.03 and an age of @xmath61@xmath62 yr . even in the same vega magnitude system , near - infrared colors of the same isochrone can be different by up to @xmath63 mag at the bright end of the isochrone for different filter pairs . the difference in intrinsic colors for a red giant for atmospheric filters and the _ hst _ nicmos filters is generally 0.20.4 mag . we have provided magnitude transformations between @xmath1-band filters as a function of color from @xmath0 and @xmath1 band filters . we also presented isochrones with @xmath64 of up to 6 mag . we found that care is needed when comparing extinction values that are estimated using different filter pairs , in particular when comparing those of atmospheric and nicmos filter pairs : extinction values inferred using nicmos filters can be in error by up to 1.3 mag . to alleviate this problem , we introduced an `` effective extinction slope '' for each filter pair and isochrone model , which describes the extinction - dependent behavior of isochrones in the observed cmd . we also provided a procedure to accurately estimate the extinction value for nicmos filter pairs , whose extinction curves in the cmd are highly nonlinear . we thank jae - woo lee for a helpful discussion . s. s. k. was supported by the astrophysical research center for the structure and evolution of the cosmos ( arcsec ) of the korea science and engineering foundation through the science research center ( src ) program . m. g. l. was in part supported by the abrl ( r14 - 2002 - 058 - 01000 - 0 ) and the bk21 program . castelli , f. , gratton , r. g. , & kurucz , r. l. 1997 , a&a , 318 , 841 figer , d. f. , rich , r. m. , kim , s. s. , morris , m. , & serabyn , e. 2004 , apj , 601 , 319 girardi , l. , bertelli , g. , bressan , a. , chiosi , c. , groenewegen , m. a. t. , marigo , p. , salsasnich , b. , & weiss , a. 2002 , a&a , 391 , 195 glass , i. s. 1974 , mon . notes astr . afr . , 33 , 53 johnson , h. l. , iriarte , b. , mitchell , r. i. , & wisniewskj , w. z. 1966 , comm . lunar plan . lab . , 4 , 99 kim , s. s. , figer , d. f. , lee , m. g. , & oh , s. 2005 , pasp , 117 , 445 ( paper i ) kurucz , r. l. 1993 , in the stellar populations of galaxies , eds . b. barbuy , & a. renzini , ( dordrecht , kluwer ) , iau symp . , 149 , 225 lee , j .- w . , carney , b. w. , fullton , l. k. , & stetson , p. b. 2001 , aj , 122 , 3136 persson , s. e. , murphy , d. c. , krzeminski , w. , roth , m. , & rieke , m. j. 1998 , aj , 116 , 2475 rieke , g. h. , rieke , m. j. , & paul , a. e. 1989 , apj , 336 , 752 wainscoat , r. j. , & cowie , l. l. 1992 , aj , 103 , 332 cclrrrrcc @xmath65 & @xmath2 @xmath15@xmath1 & 0.0001 & @xmath66 & @xmath67 & @xmath68 & @xmath69 & @xmath70 & @xmath71 + @xmath65 & @xmath2 @xmath15@xmath1 & 0.001 & @xmath66 & @xmath72 & @xmath73 & @xmath74 & @xmath75 & @xmath76 + @xmath65 & @xmath2 @xmath15@xmath1 & 0.019 & @xmath77 & @xmath78 & @xmath79 & @xmath80 & @xmath81 & @xmath82 + @xmath65 & @xmath2 @xmath15@xmath1 & 0.03 & @xmath83 & @xmath84 & @xmath85 & @xmath86 & @xmath75 & @xmath87 + @xmath65 & @xmath3@xmath15@xmath1 & 0.0001 & @xmath88 & @xmath89 & @xmath90 & @xmath91 & @xmath92 & @xmath71 + @xmath65 & @xmath3@xmath15@xmath1 & 0.001 & @xmath88 & @xmath93 & @xmath94 & @xmath95 & @xmath96 & @xmath76 + @xmath65 & @xmath3@xmath15@xmath1 & 0.019 & @xmath77 & @xmath97 & @xmath98 & @xmath99 & @xmath100 & @xmath82 + @xmath65 & @xmath3@xmath15@xmath1 & 0.03 & @xmath83 & @xmath101 & @xmath102 & @xmath103 & @xmath96 & @xmath87 + @xmath65 & f205w@xmath15@xmath1 & 0.0001 & @xmath104 & @xmath105 & @xmath106 & @xmath107 & @xmath96 & @xmath71 + @xmath65 & f205w@xmath15@xmath1 & 0.001 & @xmath104 & @xmath108 & @xmath109 & @xmath67 & @xmath110 & @xmath76 + @xmath65 & f205w@xmath15@xmath1 & 0.019 & @xmath111 & @xmath112 & @xmath79 & @xmath80 & @xmath75 & @xmath82 + @xmath65 & f205w@xmath15@xmath1 & 0.03 & @xmath111 & @xmath103 & @xmath79 & @xmath113 & @xmath70 & @xmath87 + @xmath65 & f222m@xmath15@xmath1 & 0.0001 & @xmath114 & @xmath115 & @xmath66 & @xmath116 & @xmath110 & @xmath71 + @xmath65 & f222m@xmath15@xmath1 & 0.001 & @xmath104 & @xmath83 & @xmath117 & @xmath118 & @xmath96 & @xmath76 + @xmath65 & f222m@xmath15@xmath1 & 0.019 & @xmath119 & @xmath77 & @xmath120 & @xmath112 & @xmath121 & @xmath82 + @xmath65 & f222m@xmath15@xmath1 & 0.03 & @xmath119 & @xmath122 & @xmath123 & @xmath124 & @xmath75 & @xmath87 + @xmath125 & @xmath1 @xmath15@xmath2 & 0.0001 & @xmath83 & @xmath111 & @xmath126 & @xmath127 & @xmath70 & @xmath128 + @xmath125 & @xmath1 @xmath15@xmath2 & 0.001 & @xmath83 & @xmath129 & @xmath130 & @xmath101 & @xmath96 & @xmath131 + @xmath125 & @xmath1 @xmath15@xmath2 & 0.019 & @xmath66 & @xmath132 & @xmath133 & @xmath134 & @xmath81 & @xmath135 + @xmath125 & @xmath1 @xmath15@xmath2 & 0.03 & @xmath66 & @xmath136 & @xmath137 & @xmath138 & @xmath96 & @xmath139 + @xmath125 & @xmath3@xmath15@xmath2 & 0.0001 & @xmath83 & @xmath140 & @xmath141 & @xmath142 & @xmath143 & @xmath128 + @xmath125 & @xmath3@xmath15@xmath2 & 0.001 & @xmath83 & @xmath144 & @xmath145 & @xmath140 & @xmath92 & @xmath131 + @xmath125 & @xmath3@xmath15@xmath2 & 0.019 & @xmath146 & @xmath116 & @xmath147 & @xmath148 & @xmath143 & @xmath135 + @xmath125 & @xmath3@xmath15@xmath2 & 0.03 & @xmath83 & @xmath116 & @xmath149 & @xmath150 & @xmath92 & @xmath139 + @xmath125 & f205w@xmath15@xmath2 & 0.0001 & @xmath111 & @xmath151 & @xmath152 & @xmath153 & @xmath143 & @xmath128 + @xmath125 & f205w@xmath15@xmath2 & 0.001 & @xmath111 & @xmath101 & @xmath154 & @xmath155 & @xmath70 & @xmath131 + @xmath125 & f205w@xmath15@xmath2 & 0.019 & @xmath114 & @xmath101 & @xmath115 & @xmath83 & @xmath110 & @xmath135 + @xmath125 & f205w@xmath15@xmath2 & 0.03 & @xmath104 & @xmath156 & @xmath89 & @xmath157 & @xmath110 & @xmath139 + @xmath125 & f222m@xmath15@xmath2 & 0.0001 & @xmath104 & @xmath158 & @xmath159 & @xmath160 & @xmath110 & @xmath128 + @xmath125 & f222m@xmath15@xmath2 & 0.001 & @xmath111 & @xmath158 & @xmath72 & @xmath161 & @xmath143 & @xmath131 + @xmath125 & f222m@xmath15@xmath2 & 0.019 & @xmath104 & @xmath162 & @xmath78 & @xmath163 & @xmath70 & @xmath135 + @xmath125 & f222m@xmath15@xmath2 & 0.03 & @xmath111 & @xmath162 & @xmath164 & @xmath90 & @xmath143 & @xmath139 + @xmath165 & @xmath1 @xmath15@xmath3 & 0.0001 & @xmath146 & @xmath161 & @xmath166 & @xmath167 & @xmath92 & @xmath168 + @xmath165 & @xmath1 @xmath15@xmath3 & 0.001 & @xmath146 & @xmath163 & @xmath169 & @xmath147 & @xmath96 & @xmath170 + @xmath165 & @xmath1 @xmath15@xmath3 & 0.019 & @xmath115 & @xmath140 & @xmath171 & @xmath172 & @xmath100 & @xmath173 + @xmath165 & @xmath1 @xmath15@xmath3 & 0.03 & @xmath66 & @xmath174 & @xmath175 & @xmath142 & @xmath96 & @xmath176 + @xmath165 & @xmath2 @xmath15@xmath3 & 0.0001 & @xmath66 & @xmath97 & @xmath95 & @xmath177 & @xmath143 & @xmath168 + @xmath165 & @xmath2 @xmath15@xmath3 & 0.001 & @xmath66 & @xmath178 & @xmath132 & @xmath97 & @xmath92 & @xmath170 + @xmath165 & @xmath2 @xmath15@xmath3 & 0.019 & @xmath88 & @xmath156 & @xmath136 & @xmath179 & @xmath143 & @xmath173 + @xmath165 & @xmath2 @xmath15@xmath3 & 0.03 & @xmath66 & @xmath178 & @xmath73 & @xmath180 & @xmath92 & @xmath176 + @xmath165 & f205w@xmath15@xmath3 & 0.0001 & @xmath111 & @xmath181 & @xmath182 & @xmath183 & @xmath110 & @xmath168 + @xmath165 & f205w@xmath15@xmath3 & 0.001 & @xmath104 & @xmath84 & @xmath184 & @xmath185 & @xmath70 & @xmath170 + @xmath165 & f205w@xmath15@xmath3 & 0.019 & @xmath114 & @xmath84 & @xmath186 & @xmath179 & @xmath96 & @xmath173 + @xmath165 & f205w@xmath15@xmath3 & 0.03 & @xmath114 & @xmath187 & @xmath188 & @xmath189 & @xmath110 & @xmath176 + @xmath165 & f222m@xmath15@xmath3 & 0.0001 & @xmath104 & @xmath190 & @xmath89 & @xmath191 & @xmath110 & @xmath168 + @xmath165 & f222m@xmath15@xmath3 & 0.001 & @xmath104 & @xmath190 & @xmath192 & @xmath91 & @xmath70 & @xmath170 + @xmath165 & f222m@xmath15@xmath3 & 0.019 & @xmath104 & @xmath193 & @xmath194 & @xmath169 & @xmath110 & @xmath173 + @xmath165 & f222m@xmath15@xmath3 & 0.03 & @xmath104 & @xmath193 & @xmath117 & @xmath169 & @xmath70 & @xmath176 + f110w@xmath15f205w & @xmath1 @xmath15f205w & 0.0001 & @xmath195 & @xmath196 & @xmath197 & @xmath134 & @xmath75 & @xmath198 + f110w@xmath15f205w & @xmath1 @xmath15f205w & 0.001 & @xmath195 & @xmath199 & @xmath177 & @xmath117 & @xmath110 & @xmath200 + f110w@xmath15f205w & @xmath1 @xmath15f205w & 0.019 & @xmath201 & @xmath172 & @xmath202 & @xmath129 & @xmath96 & @xmath203 + f110w@xmath15f205w & @xmath1 @xmath15f205w & 0.03 & @xmath201 & @xmath172 & @xmath202 & @xmath204 & @xmath70 & @xmath205 + f110w@xmath15f205w & @xmath2 @xmath15f205w & 0.0001 & @xmath201 & @xmath144 & @xmath206 & @xmath199 & @xmath143 & @xmath198 + f110w@xmath15f205w & @xmath2 @xmath15f205w & 0.001 & @xmath201 & @xmath144 & @xmath207 & @xmath104 & @xmath143 & @xmath200 + f110w@xmath15f205w & @xmath2 @xmath15f205w & 0.019 & @xmath164 & @xmath140 & @xmath88 & @xmath88 & @xmath110 & @xmath203 + f110w@xmath15f205w & @xmath2 @xmath15f205w & 0.03 & @xmath195 & @xmath90 & @xmath117 & @xmath208 & @xmath110 & @xmath205 + f110w@xmath15f205w & @xmath3@xmath15f205w & 0.0001 & @xmath201 & @xmath114 & @xmath209 & @xmath210 & @xmath110 & @xmath198 + f110w@xmath15f205w & @xmath3@xmath15f205w & 0.001 & @xmath195 & @xmath158 & @xmath211 & @xmath74 & @xmath70 & @xmath200 + f110w@xmath15f205w & @xmath3@xmath15f205w & 0.019 & @xmath164 & @xmath158 & @xmath212 & @xmath213 & @xmath96 & @xmath203 + f110w@xmath15f205w & @xmath3@xmath15f205w & 0.03 & @xmath164 & @xmath188 & @xmath214 & @xmath192 & @xmath110 & @xmath205 + f110w@xmath15f205w & f222m@xmath15f205w & 0.0001 & @xmath66 & @xmath215 & @xmath197 & @xmath216 & @xmath75 & @xmath198 + f110w@xmath15f205w & f222m@xmath15f205w & 0.001 & @xmath88 & @xmath136 & @xmath217 & @xmath129 & @xmath70 & @xmath200 + f110w@xmath15f205w & f222m@xmath15f205w & 0.019 & @xmath83 & @xmath199 & @xmath178 & @xmath218 & @xmath96 & @xmath203 + f110w@xmath15f205w & f222m@xmath15f205w & 0.03 & @xmath83 & @xmath132 & @xmath219 & @xmath115 & @xmath70 & @xmath205 + f110w@xmath15f222 m & @xmath1 @xmath15f222 m & 0.0001 & @xmath164 & @xmath83 & @xmath83 & @xmath169 & @xmath110 & @xmath220 + f110w@xmath15f222 m & @xmath1 @xmath15f222 m & 0.001 & @xmath195 & @xmath66 & @xmath221 & @xmath180 & @xmath110 & @xmath222 + f110w@xmath15f222 m & @xmath1 @xmath15f222 m & 0.019 & @xmath67 & @xmath94 & @xmath223 & @xmath191 & @xmath81 & @xmath224 + f110w@xmath15f222 m & @xmath1 @xmath15f222 m & 0.03 & @xmath67 & @xmath225 & @xmath226 & @xmath111 & @xmath96 & @xmath227 + f110w@xmath15f222 m & @xmath2 @xmath15f222 m & 0.0001 & @xmath195 & @xmath180 & @xmath129 & @xmath72 & @xmath110 & @xmath220 + f110w@xmath15f222 m & @xmath2 @xmath15f222 m & 0.001 & @xmath201 & @xmath179 & @xmath90 & @xmath208 & @xmath143 & @xmath222 + f110w@xmath15f222 m & @xmath2 @xmath15f222 m & 0.019 & @xmath195 & @xmath101 & @xmath140 & @xmath219 & @xmath70 & @xmath224 + f110w@xmath15f222 m & @xmath2 @xmath15f222 m & 0.03 & @xmath201 & @xmath101 & @xmath190 & @xmath221 & @xmath143 & @xmath227 + f110w@xmath15f222 m & @xmath3@xmath15f222 m & 0.0001 & @xmath195 & @xmath228 & @xmath218 & @xmath229 & @xmath110 & @xmath220 + f110w@xmath15f222 m & @xmath3@xmath15f222 m & 0.001 & @xmath195 & @xmath122 & @xmath219 & @xmath94 & @xmath70 & @xmath222 + f110w@xmath15f222 m & @xmath3@xmath15f222 m & 0.019 & @xmath195 & @xmath229 & @xmath230 & @xmath157 & @xmath110 & @xmath224 + f110w@xmath15f222 m & @xmath3@xmath15f222 m & 0.03 & @xmath195 & @xmath228 & @xmath122 & @xmath167 & @xmath70 & @xmath227 + f110w@xmath15f222 m & f205w@xmath15f222 m & 0.0001 & @xmath83 & @xmath145 & @xmath138 & @xmath231 & @xmath75 & @xmath220 + f110w@xmath15f222 m & f205w@xmath15f222 m & 0.001 & @xmath146 & @xmath149 & @xmath232 & @xmath195 & @xmath70 & @xmath222 + f110w@xmath15f222 m & f205w@xmath15f222 m & 0.019 & @xmath66 & @xmath181 & @xmath140 & @xmath219 & @xmath96 & @xmath224 + f110w@xmath15f222 m & f205w@xmath15f222 m & 0.03 & @xmath66 & @xmath78 & @xmath157 & @xmath77 & @xmath70 & @xmath227 + cclrrrcc @xmath65 & @xmath2 @xmath15@xmath1 & 0.0001 & @xmath161 & @xmath130 & @xmath94 & @xmath233 & @xmath234 + @xmath65 & @xmath2 @xmath15@xmath1 & 0.001 & @xmath235 & @xmath236 & @xmath237 & @xmath143 & @xmath238 + @xmath65 & @xmath2 @xmath15@xmath1 & 0.019 & @xmath239 & @xmath240 & @xmath241 & @xmath70 & @xmath242 + @xmath65 & @xmath2 @xmath15@xmath1 & 0.03 & @xmath243 & @xmath244 & @xmath245 & @xmath143 & @xmath246 + @xmath65 & @xmath3@xmath15@xmath1 & 0.0001 & @xmath117 & @xmath84 & @xmath116 & @xmath233 & @xmath234 + @xmath65 & @xmath3@xmath15@xmath1 & 0.001 & @xmath178 & @xmath247 & @xmath211 & @xmath92 & @xmath238 + @xmath65 & @xmath3@xmath15@xmath1 & 0.019 & @xmath211 & @xmath248 & @xmath249 & @xmath143 & @xmath242 + @xmath65 & @xmath3@xmath15@xmath1 & 0.03 & @xmath250 & @xmath251 & @xmath252 & @xmath92 & @xmath246 + @xmath65 & f205w@xmath15@xmath1 & 0.0001 & @xmath253 & @xmath254 & @xmath146 & @xmath233 & @xmath234 + @xmath65 & f205w@xmath15@xmath1 & 0.001 & @xmath212 & @xmath255 & @xmath256 & @xmath70 & @xmath238 + @xmath65 & f205w@xmath15@xmath1 & 0.019 & @xmath237 & @xmath257 & @xmath258 & @xmath75 & @xmath242 + @xmath65 & f205w@xmath15@xmath1 & 0.03 & @xmath259 & @xmath260 & @xmath261 & @xmath70 & @xmath246 + @xmath65 & f222m@xmath15@xmath1 & 0.0001 & @xmath262 & @xmath174 & @xmath208 & @xmath233 & @xmath234 + @xmath65 & f222m@xmath15@xmath1 & 0.001 & @xmath188 & @xmath193 & @xmath192 & @xmath233 & @xmath238 + @xmath65 & f222m@xmath15@xmath1 & 0.019 & @xmath104 & @xmath188 & @xmath77 & @xmath233 & @xmath242 + @xmath65 & f222m@xmath15@xmath1 & 0.03 & @xmath162 & @xmath263 & @xmath115 & @xmath233 & @xmath246 + @xmath125 & @xmath1 @xmath15@xmath2 & 0.0001 & @xmath89 & @xmath264 & @xmath221 & @xmath233 & @xmath265 + @xmath125 & @xmath1 @xmath15@xmath2 & 0.001 & @xmath132 & @xmath266 & @xmath267 & @xmath143 & @xmath268 + @xmath125 & @xmath1 @xmath15@xmath2 & 0.019 & @xmath269 & @xmath270 & @xmath271 & @xmath110 & @xmath272 + @xmath125 & @xmath1 @xmath15@xmath2 & 0.03 & @xmath273 & @xmath274 & @xmath275 & @xmath143 & @xmath276 + @xmath125 & @xmath3@xmath15@xmath2 & 0.0001 & @xmath77 & @xmath218 & @xmath140 & @xmath233 & @xmath265 + @xmath125 & @xmath3@xmath15@xmath2 & 0.001 & @xmath144 & @xmath80 & @xmath160 & @xmath143 & @xmath268 + @xmath125 & @xmath3@xmath15@xmath2 & 0.019 & @xmath277 & @xmath278 & @xmath106 & @xmath92 & @xmath272 + @xmath125 & @xmath3@xmath15@xmath2 & 0.03 & @xmath160 & @xmath279 & @xmath280 & @xmath233 & @xmath276 + @xmath125 & f205w@xmath15@xmath2 & 0.0001 & @xmath281 & @xmath78 & @xmath221 & @xmath233 & @xmath265 + @xmath125 & f205w@xmath15@xmath2 & 0.001 & @xmath218 & @xmath282 & @xmath283 & @xmath92 & @xmath268 + @xmath125 & f205w@xmath15@xmath2 & 0.019 & @xmath149 & @xmath284 & @xmath285 & @xmath143 & @xmath272 + @xmath125 & f205w@xmath15@xmath2 & 0.03 & @xmath113 & @xmath286 & @xmath287 & @xmath92 & @xmath276 + @xmath125 & f222m@xmath15@xmath2 & 0.0001 & @xmath90 & @xmath288 & @xmath230 & @xmath233 & @xmath265 + @xmath125 & f222m@xmath15@xmath2 & 0.001 & @xmath289 & @xmath290 & @xmath291 & @xmath70 & @xmath268 + @xmath125 & f222m@xmath15@xmath2 & 0.019 & @xmath292 & @xmath293 & @xmath294 & @xmath70 & @xmath272 + @xmath125 & f222m@xmath15@xmath2 & 0.03 & @xmath295 & @xmath296 & @xmath297 & @xmath143 & @xmath276 + @xmath165 & @xmath1 @xmath15@xmath3 & 0.0001 & @xmath228 & @xmath199 & @xmath298 & @xmath233 & @xmath299 + @xmath165 & @xmath1 @xmath15@xmath3 & 0.001 & @xmath144 & @xmath112 & @xmath300 & @xmath92 & @xmath301 + @xmath165 & @xmath1 @xmath15@xmath3 & 0.019 & @xmath277 & @xmath302 & @xmath106 & @xmath143 & @xmath303 + @xmath165 & @xmath1 @xmath15@xmath3 & 0.03 & @xmath304 & @xmath305 & @xmath306 & @xmath92 & @xmath307 + @xmath165 & @xmath2 @xmath15@xmath3 & 0.0001 & @xmath115 & @xmath91 & @xmath189 & @xmath233 & @xmath299 + @xmath165 & @xmath2 @xmath15@xmath3 & 0.001 & @xmath97 & @xmath308 & @xmath185 & @xmath143 & @xmath301 + @xmath165 & @xmath2 @xmath15@xmath3 & 0.019 & @xmath141 & @xmath309 & @xmath310 & @xmath92 & @xmath303 + @xmath165 & @xmath2 @xmath15@xmath3 & 0.03 & @xmath231 & @xmath311 & @xmath312 & @xmath233 & @xmath307 + @xmath165 & f205w@xmath15@xmath3 & 0.0001 & @xmath196 & @xmath84 & @xmath156 & @xmath233 & @xmath299 + @xmath165 & f205w@xmath15@xmath3 & 0.001 & @xmath228 & @xmath313 & @xmath314 & @xmath143 & @xmath301 + @xmath165 & f205w@xmath15@xmath3 & 0.019 & @xmath197 & @xmath315 & @xmath316 & @xmath110 & @xmath303 + @xmath165 & f205w@xmath15@xmath3 & 0.03 & @xmath249 & @xmath317 & @xmath318 & @xmath143 & @xmath307 + @xmath165 & f222m@xmath15@xmath3 & 0.0001 & @xmath163 & @xmath319 & @xmath101 & @xmath233 & @xmath299 + @xmath165 & f222m@xmath15@xmath3 & 0.001 & @xmath232 & @xmath320 & @xmath68 & @xmath92 & @xmath301 + @xmath165 & f222m@xmath15@xmath3 & 0.019 & @xmath321 & @xmath322 & @xmath323 & @xmath143 & @xmath303 + @xmath165 & f222m@xmath15@xmath3 & 0.03 & @xmath324 & @xmath325 & @xmath326 & @xmath92 & @xmath307 + f110w@xmath15f205w & @xmath1 @xmath15f205w & 0.0001 & @xmath141 & @xmath327 & @xmath88 & @xmath233 & @xmath328 + f110w@xmath15f205w & @xmath1 @xmath15f205w & 0.001 & @xmath264 & @xmath329 & @xmath330 & @xmath110 & @xmath331 + f110w@xmath15f205w & @xmath1 @xmath15f205w & 0.019 & @xmath332 & @xmath333 & @xmath334 & @xmath335 & @xmath336 + f110w@xmath15f205w & @xmath1 @xmath15f205w & 0.03 & @xmath337 & @xmath338 & @xmath339 & @xmath110 & @xmath340 + f110w@xmath15f205w & @xmath2 @xmath15f205w & 0.0001 & @xmath149 & @xmath104 & @xmath94 & @xmath233 & @xmath328 + f110w@xmath15f205w & @xmath2 @xmath15f205w & 0.001 & @xmath115 & @xmath341 & @xmath288 & @xmath92 & @xmath331 + f110w@xmath15f205w & @xmath2 @xmath15f205w & 0.019 & @xmath74 & @xmath107 & @xmath342 & @xmath143 & @xmath336 + f110w@xmath15f205w & @xmath2 @xmath15f205w & 0.03 & @xmath319 & @xmath343 & @xmath344 & @xmath92 & @xmath340 + f110w@xmath15f205w & @xmath3@xmath15f205w & 0.0001 & @xmath84 & @xmath129 & @xmath163 & @xmath233 & @xmath328 + f110w@xmath15f205w & @xmath3@xmath15f205w & 0.001 & @xmath262 & @xmath345 & @xmath182 & @xmath70 & @xmath331 + f110w@xmath15f205w & @xmath3@xmath15f205w & 0.019 & @xmath109 & @xmath245 & @xmath248 & @xmath110 & @xmath336 + f110w@xmath15f205w & @xmath3@xmath15f205w & 0.03 & @xmath346 & @xmath347 & @xmath295 & @xmath143 & @xmath340 + f110w@xmath15f205w & f222m@xmath15f205w & 0.0001 & @xmath195 & @xmath348 & @xmath77 & @xmath233 & @xmath328 + f110w@xmath15f205w & f222m@xmath15f205w & 0.001 & @xmath216 & @xmath349 & @xmath350 & @xmath110 & @xmath331 + f110w@xmath15f205w & f222m@xmath15f205w & 0.019 & @xmath351 & @xmath352 & @xmath353 & @xmath335 & @xmath336 + f110w@xmath15f205w & f222m@xmath15f205w & 0.03 & @xmath354 & @xmath355 & @xmath356 & @xmath110 & @xmath340 + f110w@xmath15f222 m & @xmath1 @xmath15f222 m & 0.0001 & @xmath151 & @xmath178 & @xmath115 & @xmath233 & @xmath357 + f110w@xmath15f222 m & @xmath1 @xmath15f222 m & 0.001 & @xmath358 & @xmath228 & @xmath91 & @xmath233 & @xmath359 + f110w@xmath15f222 m & @xmath1 @xmath15f222 m & 0.019 & @xmath151 & @xmath201 & @xmath167 & @xmath233 & @xmath360 + f110w@xmath15f222 m & @xmath1 @xmath15f222 m & 0.03 & @xmath164 & @xmath361 & @xmath115 & @xmath233 & @xmath362 + f110w@xmath15f222 m & @xmath2 @xmath15f222 m & 0.0001 & @xmath229 & @xmath211 & @xmath94 & @xmath233 & @xmath357 + f110w@xmath15f222 m & @xmath2 @xmath15f222 m & 0.001 & @xmath363 & @xmath98 & @xmath364 & @xmath70 & @xmath359 + f110w@xmath15f222 m & @xmath2 @xmath15f222 m & 0.019 & @xmath365 & @xmath366 & @xmath367 & @xmath70 & @xmath360 + f110w@xmath15f222 m & @xmath2 @xmath15f222 m & 0.03 & @xmath368 & @xmath369 & @xmath312 & @xmath143 & @xmath362 + f110w@xmath15f222 m & @xmath3@xmath15f222 m & 0.0001 & @xmath166 & @xmath177 & @xmath190 & @xmath233 & @xmath357 + f110w@xmath15f222 m & @xmath3@xmath15f222 m & 0.001 & @xmath206 & @xmath199 & @xmath84 & @xmath92 & @xmath359 + f110w@xmath15f222 m & @xmath3@xmath15f222 m & 0.019 & @xmath209 & @xmath98 & @xmath126 & @xmath143 & @xmath360 + f110w@xmath15f222 m & @xmath3@xmath15f222 m & 0.03 & @xmath370 & @xmath267 & @xmath254 & @xmath92 & @xmath362 + f110w@xmath15f222 m & f205w@xmath15f222 m & 0.0001 & @xmath119 & @xmath86 & @xmath66 & @xmath233 & @xmath357 + f110w@xmath15f222 m & f205w@xmath15f222 m & 0.001 & @xmath371 & @xmath372 & @xmath373 & @xmath110 & @xmath359 + f110w@xmath15f222 m & f205w@xmath15f222 m & 0.019 & @xmath374 & @xmath375 & @xmath376 & @xmath96 & @xmath360 + f110w@xmath15f222 m & f205w@xmath15f222 m & 0.03 & @xmath377 & @xmath378 & @xmath379 & @xmath70 & @xmath362 + lcccccc 0.0001 & @xmath61 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.0001 & @xmath381 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.0001 & @xmath28 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.0001 & @xmath62 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.001 & @xmath382 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.001 & @xmath381 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.001 & @xmath28 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.001 & @xmath62 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.019 & @xmath61 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.019 & @xmath381 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.019 & @xmath28 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.019 & @xmath62 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.03 & @xmath382 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.03 & @xmath381 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.03 & @xmath28 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 0.03 & @xmath62 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + lccrrrrrcrrrrr 0.0001 & @xmath61 & & 7.07e-04 & 2.07e-01 & @xmath151.08e-01 & 7.77e-03 & 0.011 & & @xmath159.83e-05 & 2.67e-01 & @xmath151.18e-01 & 7.01e-03 & 0.013 0.0001 & @xmath381 & & 9.78e-04 & 1.75e-01 & @xmath151.05e-01 & 7.70e-03 & 0.077 & & 1.18e-04 & 2.37e-01 & @xmath151.16e-01 & 7.07e-03 & 0.080 0.0001 & @xmath28 & & 1.61e-03 & 1.14e-01 & @xmath151.00e-01 & 7.88e-03 & 0.082 & & 6.21e-04 & 1.84e-01 & @xmath151.14e-01 & 7.41e-03 & 0.083 0.0001 & @xmath62 & & 1.82e-03 & 8.93e-02 & @xmath159.82e-02 & 7.95e-03 & 0.049 & & 7.90e-04 & 1.62e-01 & @xmath151.13e-01 & 7.53e-03 & 0.050 0.001 & @xmath382 & & 9.79e-04 & 1.75e-01 & @xmath151.05e-01 & 7.69e-03 & 0.082 & & 1.01e-04 & 2.38e-01 & @xmath151.16e-01 & 7.05e-03 & 0.084 0.001 & @xmath381 & & 1.16e-03 & 1.58e-01 & @xmath151.04e-01 & 7.77e-03 & 0.115 & & 2.30e-04 & 2.21e-01 & @xmath151.16e-01 & 7.17e-03 & 0.118 0.001 & @xmath28 & & 1.75e-03 & 9.25e-02 & @xmath159.86e-02 & 7.93e-03 & 0.049 & & 6.72e-04 & 1.66e-01 & @xmath151.13e-01 & 7.53e-03 & 0.047 0.001 & @xmath62 & & 1.86e-03 & 7.52e-02 & @xmath159.74e-02 & 7.98e-03 & 0.057 & & 8.48e-04 & 1.51e-01 & @xmath151.12e-01 & 7.57e-03 & 0.054 0.019 & @xmath61 & & 7.77e-04 & 2.03e-01 & @xmath151.08e-01 & 7.74e-03 & 0.010 & & @xmath157.16e-05 & 2.64e-01 & @xmath151.18e-01 & 7.01e-03 & 0.011 0.019 & @xmath381 & & 1.33e-03 & 1.36e-01 & @xmath151.02e-01 & 7.82e-03 & 0.172 & & 3.57e-04 & 2.05e-01 & @xmath151.15e-01 & 7.23e-03 & 0.171 0.019 & @xmath28 & & 1.93e-03 & 6.95e-02 & @xmath159.71e-02 & 7.97e-03 & 0.068 & & 8.10e-04 & 1.46e-01 & @xmath151.12e-01 & 7.55e-03 & 0.066 0.019 & @xmath62 & & 2.07e-03 & 4.13e-02 & @xmath159.42e-02 & 7.93e-03 & 0.064 & & 8.75e-04 & 1.22e-01 & @xmath151.09e-01 & 7.48e-03 & 0.062 0.03 & @xmath382 & & 1.03e-03 & 1.72e-01 & @xmath151.05e-01 & 7.73e-03 & 0.120 & & 1.43e-04 & 2.35e-01 & @xmath151.16e-01 & 7.06e-03 & 0.122 0.03 & @xmath381 & & 1.30e-03 & 1.36e-01 & @xmath151.02e-01 & 7.83e-03 & 0.182 & & 3.61e-04 & 2.06e-01 & @xmath151.15e-01 & 7.25e-03 & 0.180 0.03 & @xmath28 & & 1.92e-03 & 6.25e-02 & @xmath159.61e-02 & 7.93e-03 & 0.071 & & 8.03e-04 & 1.40e-01 & @xmath151.11e-01 & 7.51e-03 & 0.068 0.03 & @xmath62 & & 2.04e-03 & 3.35e-02 & @xmath159.28e-02 & 7.84e-03 & 0.063 & & 8.45e-04 & 1.15e-01 & @xmath151.08e-01 & 7.44e-03 & 0.063

in 1930s paul erds conjectured that for any positive integer in any infinite sequence there exists a subsequence , for some positive integers and , such that . the conjecture has been referred to as one of the major open problems in combinatorial number theory and discrepancy theory . for the particular case of a human proof of the conjecture exists ; for a bespoke computer program had generated sequences of length of discrepancy , but the status of the conjecture remained open even for such a small bound . we show that by encoding the problem into boolean satisfiability and applying the state of the art sat solvers , one can obtain a discrepancy sequence of length and a _ proof _ of the erds discrepancy conjecture for , claiming that no discrepancy 2 sequence of length , or more , exists . in the similar way , we obtain a precise bound of on the maximal lengths of both multiplicative and completely multiplicative sequences of discrepancy .

1405.3097

discrepancy theory is a branch of mathematics dealing with irregularities of distributions of points in some space in combinatorial , measure - theoretic and geometric settings @xcite . the paradigmatic combinatorial discrepancy theory problem can be described in terms of a hypergraph @xmath15 , that is , a set @xmath16 and a family of its subsets @xmath17 . consider a colouring @xmath18 of the elements of @xmath16 in _ blue _ @xmath19 and _ red _ ( @xmath20 ) colours . then one may ask whether there exists a colouring of the elements of @xmath21 such that colours are distributed uniformly in every element of @xmath22 or a discrepancy of colours is always inevitable . formally , the discrepancy ( deviation from a uniform distribution ) of a hypergraph @xmath23 is defined as @xmath24 . discrepancy theory has found applications in computational complexity @xcite , complexity of communication @xcite and differential privacy @xcite . one of the oldest problems of discrepancy theory is the discrepancy of hypergraphs over sets of natural numbers with the subsets ( hyperedges ) forming arithmetical progressions over these sets @xcite . theorem @xcite , one of the main results in the area , states that for the hypergraph formed by the arithmetic progressions in @xmath25 , that is @xmath26 , where @xmath27 and elements of @xmath28 being of the form @xmath29 for arbitrary @xmath30 , the discrepancy grows at least as @xmath31 . surprisingly , for the more restricted case of _ homogeneous _ arithmetic progressions of the form @xmath32 , the question of the discrepancy bounds is open for more than eighty years . in 1930s paul erds conjectured @xcite that discrepancy is unbounded . independently the same conjecture has been raised by @xcite . proving or disproving this conjecture became one of the major open problems in combinatorial number theory and discrepancy theory . it has been referred to as the _ erds discrepancy problem _ ( edp ) @xcite . the expected value of the discrepancy of random @xmath1 sequences of length @xmath33 grows as @xmath34 and the explicit constructions of a sequence with slowly growing discrepancy at the rate of @xmath35 have been demonstrated @xcite . by considering cases , one can see that any @xmath36 sequence containing @xmath37 or more elements has discrepancy at least @xmath10 ; that is , erds s conjecture holds for the particular case @xmath7 ( also implied by a stronger result of mathias ) . until recently the status of the conjecture remained unknown for all other values of @xmath0 . although widely believed not to be the case , there was still a possibility that an infinite sequence of discrepancy 2 existed . the conjecture whether the discrepancy of an arbitrary @xmath36 sequence is unbounded is equivalent to the question whether the discrepancy of a completely multiplicative @xmath36 sequence is unbounded , where a sequence is completely multiplicative if @xmath38 for any @xmath39 @xcite . for completely multiplicative sequences the choices of how the sequence can be constructed are severely limited as the entire sequence is defined by the values of @xmath40 for prime @xmath41 . the longest completely multiplicative sequence of discrepancy @xmath10 has length @xmath42 @xcite . for discrepancy @xmath14 the bound was not known . the edp has attracted renewed interest in 2009 - 2010 as it became a topic of the fifth polymath project a widely publicised endeavour in collective math initiated by t. gowers . as part of this activity an attempt has been made to attack the problem using computers ( see discussion in @xcite ) . a purposely written computer program had successfully found @xmath36 sequences of length @xmath9 and discrepancy @xmath10 ; however , no further progress has been made leading to a claim `` given how long a finite sequence can be , it seems unlikely that we could answer this question just by a clever search of all possibilities on a computer '' @xcite . the status of the erds discrepancy conjecture for @xmath8 has been settled by the authors of this article @xcite , @xcite by reduction to sat . the method is based on establishing the correspondence between @xmath36 sequences that violate a given discrepancy bound and words accepted by of a finite automaton . traces of this automaton are represented then by a propositional formula and state of the art sat solvers are used to prove that the longest @xmath36 sequence of discrepancy @xmath10 contains @xmath11 elements . a @xmath43 long @xmath36 sequence of discrepancy @xmath14 was also constructed . this article is a revised and extended version of @xcite . we use a different smaller sat encoding of the erds discrepancy problem , which is based on the sequential counter encoding of the _ at most _ cardinality constraints . the impact of the new encoding is twofold . firstly , it allows us to significantly reduce the size of the machine - generated proof of the fact that any sequence longer than @xmath11 has discrepancy at least @xmath14 . secondly , by combining the new encoding with additional restrictions that the sequence is multiplicative , or completely multiplicative , we improve significantly the lower bound on the length of sequences of discrepancy @xmath14 . we prove the surprising result that @xmath13 , the length of the longest completely multiplicative sequence of discrepancy @xmath14 , is also the maximal length of a multiplicative sequence of discrepancy @xmath14 , which is not the case for @xmath7 and @xmath8 . the article also contains detailed argumentation , examples and complete proofs . the article is organised as follows . in section [ sec : preliminaries ] we introduce the main terms and definition . in section [ sec : encoding ] we describe the new sat encoding of the erds discrepancy problem . results and conclusions are discussed in sections [ sec : experiments ] and [ sec : conclusion ] respectively . to improve readability a number of technical proofs have been deferred to an appendix . we divide this section into three parts : main definitions for the erds discrepancy problem , some background and definitions for sat solving , and sequential counter - based sat encoding of cardinality constraints . since number @xmath44 is used both as an element of @xmath36 sequences and as the logical value _ true _ , to avoid confusion , in what follows we write @xmath44 to refer to the logical value _ true _ and @xmath45 to refer to elements of @xmath36 sequences . we also use the following naming convention : we write @xmath46 for @xmath36 sequences , @xmath47 for sequences of propositions , and @xmath48 for @xmath49 boolean sequences . a @xmath36 sequence of length @xmath33 is a function @xmath51 . an infinite @xmath36 sequence is a function @xmath52 , where @xmath53 is the set of positive natural numbers . we write @xmath54 to denote a finite @xmath36 sequence of length @xmath33 , and @xmath2 to denote an infinite sequence . we refer to the @xmath41-th element of a sequence @xmath55 , that is the value of @xmath56 , as @xmath40 . a ( finite or infinite ) @xmath36 sequence @xmath55 is _ completely multiplicative _ @xcite if @xmath57 the sequence is _ multiplicative _ if ( [ eq : multdef1 ] ) is only required for coprime @xmath58 and @xmath33 . it is easy to see that a sequence @xmath55 is completely multiplicative if , and only if , @xmath59 and for the canonical representation @xmath60 , where @xmath61 are primes and @xmath62 , we have @xmath63 . this observation leads to a more computationally friendly definition of completely multiplicative sequences : @xmath55 is completely multiplicative if , and only if , @xmath64 the erds discrepancy problem can be naturally described in terms of @xmath36 sequences ( and this is how erds himself introduced it ) . erds s conjecture states that for any @xmath65 in any infinite @xmath1 sequence @xmath2 there exists a subsequence @xmath66 , for some positive integers @xmath4 and @xmath5 , such that @xmath67 . the general definition of discrepancy given above can be specialised in this case as follows . the discrepancy of a finite @xmath1 sequence @xmath68 of length @xmath33 can be defined as @xmath69 . for an infinite sequence @xmath2 its discrepancy is the supremum of discrepancies of all its initial finite fragments . [ ex : edp1 ] it is easy to see why any @xmath36 sequence containing @xmath37 elements has discrepancy at least @xmath10 . for the proof of contradiction , suppose that the discrepancy of some @xmath36 sequence @xmath70 is @xmath44 . assume that @xmath71 is @xmath45 . we write @xmath72 to track progress in this example , that is , we put specific values @xmath45 or @xmath20 into positions @xmath73 , to indicate decisions on @xmath40 which have been taken so far , and mark positions @xmath40 , for which no decision has been made by an underscore . notice that @xmath74 must be @xmath20 for otherwise @xmath75 . so , we progress to @xmath76 then the @xmath77th element of the sequence must be @xmath45 for otherwise for @xmath78 the sum @xmath79 . so we progress to @xmath80 then the @xmath14rd element of the sequence must be @xmath20 for otherwise @xmath81 and so we come to @xmath82 repeating the reasoning above for @xmath83 and @xmath84 followed by @xmath85 , for @xmath86 and @xmath87 followed by @xmath88 , for @xmath85 and @xmath89 followed by @xmath90 and finally for @xmath84 and @xmath91 followed by @xmath92 we progress to @xmath93 but then for @xmath94 we have @xmath95 . so we derive a contradiction . it can be checked in a similar way that the other possibility of @xmath71 being @xmath20 also leads to a contradiction . the first eleven elements of the sequence ( [ eq : example1 ] ) form a discrepancy @xmath44 sequence . it is multiplicative but not completely multiplicative as @xmath90 is @xmath20 . reasoning similar to the one above shows that there exists a unique longest completely multiplicative @xmath36 sequence of discrepancy @xmath44 which has nine elements : @xmath96 we assume standard definitions for propositional logic ( see , for example , @xcite ) . propositional formulae are defined over boolean constants _ true _ and _ false _ , denoted by @xmath44 and @xmath97 , respectively , and the set of boolean variables ( or propositions ) @xmath98 as follows : boolean constants @xmath97 and @xmath44 , as well as the elements of @xmath98 , are formulae ; if @xmath99 and @xmath100 are formulae then so are @xmath101 ( conjunction ) , @xmath102 ( disjunction ) , @xmath103 ( implication ) , @xmath104 ( equivalence ) and @xmath105 ( negation ) . we typically use letters @xmath106 , @xmath107 and @xmath108 to denote propositions and capital greek letters @xmath99 and @xmath100 to denote propositional formulae . whenever necessary , subscripts and superscripts are used . we use @xmath109 to denote the set of all propositions occurring in the formula @xmath99 . every propositional formula can be reduced to conjunctive normal form . propositions and negations of propositions are called _ literals_. when the negation is applied to a literal , double negations are implicitly removed , that is , if @xmath110 is @xmath111 then @xmath112 is @xmath106 . a disjunction of literals is called a _ clause_. a clause containing exactly one literal is called a unit clause . a conjunction of clauses is called a propositional formula in _ conjunctive normal form _ , a cnf formula for short . a clause can be represented by the set of its literals and the empty clause correspond to @xmath97 ( _ false _ ) . a cnf formula can be represented by the set its clauses . these representations are used interchangeably . we typically use meaningful terms typeset in sans serif font , for example @xmath113 or @xmath114 , to highlight the fact that a propositional formula is a cnf formula of interest . for a propositional formula @xmath99 , we write @xmath115 to indicate that @xmath116 . propositions @xmath47 are designated as ` input ' propositions in this case , and the intended meaning is that formula @xmath99 encodes some property of @xmath47 . then the expression @xmath117 denotes the result of simultaneous replacement of every occurrence of @xmath118 in @xmath99 with @xmath119 , for @xmath120 . the semantics of propositional formulae is given by interpretations ( also termed assignments ) . an interpretation @xmath121 is a mapping @xmath122 extended to literals , clauses , cnf formulae and propositional formulae in general in the usual way . for an assignment @xmath121 and a formula @xmath99 we say that @xmath121 satisfies @xmath99 ( or @xmath121 is a model of @xmath99 ) if @xmath123 . a formula @xmath99 is satisfiable if there exists an assignment that satisfies it , and unsatisfiable otherwise . despite the non - tractability of the satisfiability problem , the tremendous progress in recent years made it possible to solve many interesting hard problems by first expressing them as a propositional formula and then using a sat solver for obtaining a solution @xcite . in addition to returning a satisfying assignment , if the input formula is satisfiable , some sat solvers are also capable to return a proof ( or certificate ) of unsatisfiability . reverse unit propagation ( rup ) proofs constitute a compact representation of the resolution refutation of the given formula @xcite in the following sense . _ unit propagation _ is a cnf formula transformation technique , which simplifies the formula by fixing the values of propositions occurring to its unit clauses to satisfy these clauses . that is , if the unit clause @xmath124 occurs in the cnf formula then all occurrences of @xmath106 are replaced by @xmath44 and if the unit clause @xmath125 occurs in the cnf formula , all occurrences of @xmath106 are replaced by @xmath97 . then the cnf formula is simplified in the obvious way . a clause @xmath126 is a rup inference from the input cnf formula @xmath100 if adding the unit clauses @xmath127 to @xmath100 makes the whole formula refutable by unit propagation . a rup unsatisfiability certificate is the sequence of clauses @xmath128 such that for every @xmath129 the clause @xmath130 is a rup inference from @xmath131 and @xmath132 is the empty clause . every unsatisfiable cnf formula has a rup unsatisfiability certificate @xcite . delete reverse unit propagation ( drup ) proofs extend rup proofs by including extra information about the proof search process , namely clauses that have been discarded by the solver . eliminating this extra information from a drup proof converts it to a valid rup proof . drup proofs are somewhat longer but they are significantly faster to verify than a rup proof @xcite . cardinality constraints @xcite are expressions that impose restrictions on interpretations by specifying numerical bounds on the number of propositions , from a fixed set of propositions , that can be assigned value @xmath44 . the _ at most @xmath133 _ constraint over the set of propositions @xmath134 , written as @xmath135 , holds for an interpretation @xmath121 if , and only if , at most @xmath133 propositions among @xmath47 are true under @xmath121 . a sat encoding for cardinality constraints of the form @xmath136 based on a sequential counter circuit has been suggested by sinz . in this encoding , auxiliary propositions @xmath137 are introduced to represent a unary counter storing the partial sums of prefixes of @xmath47 so that whenever @xmath138 , for some @xmath139 , we have @xmath140 . we slightly simplify the presentation of @xcite as follows . let formula @xmath141 be the conjunction of @xmath142 recall that we write @xmath141 to highlight the fact that @xmath47 are designated ` input ' propositions ; the set of all propositions of @xmath141 is @xmath143 . notice that instead of including formulae ( [ eq : sinz_card_2 ] ) and ( [ eq : sinz_card_3 ] ) explicitly in the encoding , one can directly modify ( [ eq : sinz_card_1 ] ) by replacing all occurrences of @xmath137 , for @xmath144 , with @xmath97 ( the truth value _ false _ ) and all occurrences of @xmath137 , for @xmath145 and all @xmath146 , with @xmath44 ( the truth value _ true _ ) . then , for example , for @xmath147 formula ( [ eq : sinz_card_1 ] ) simplifies to @xmath148 . we write ( [ eq : sinz_card_2 ] ) and ( [ eq : sinz_card_3 ] ) explicitly for the exposition purposes . the proof of the following statement can be extracted from @xcite . it is based on the observation that the sum of the first @xmath149 elements of the @xmath49 sequence @xmath47 exceeds @xmath4 if , and only if , either the sum of the first @xmath150 elements already exceeds @xmath4 , or the sum of the first @xmath150 elements is @xmath4 and the @xmath149-th element of the sequence is @xmath44 . we give the formal proof in an appendix for completeness of the presentation . [ prop : sinz ] let @xmath141 be as defined above . then 1 . for any assignment @xmath151 such that @xmath121 satisfies @xmath141 , any @xmath152 and @xmath153 we have @xmath154 2 . for any @xmath49-sequence @xmath155 there exists an assignment @xmath156 such that @xmath157 , for @xmath120 ; @xmath121 satisfies @xmath141 ; and for any @xmath158 and @xmath139 if @xmath159 then @xmath160 , for @xmath161 . it follows from proposition [ prop : sinz ] that the formula @xmath162 enforces the cardinality constraint @xmath135 . formula ( [ eq : sinz_card_1 ] ) can be equivalently rewritten into clausal form : @xmath163 where @xmath153 , @xmath152 . notice that the original encoding in @xcite only contains clauses ( [ eq : equiv_clause_3 ] ) and ( [ eq : equiv_clause_4 ] ) due to a polarity - based optimisation based on tseitin s renaming techniques . one can see that by unit propagation of clauses ( [ eq : sinz_card_2 ] ) , ( [ eq : sinz_card_3 ] ) and @xmath164 into ( [ eq : equiv_clause_3 ] ) and ( [ eq : equiv_clause_4 ] ) we obtain exactly the set of clauses used in @xcite , which consists of @xmath165 clauses and requires @xmath165 auxiliary propositions . this optimisation leads to a relaxation in item ( _ i _ ) of proposition [ prop : sinz ] : suppose that an assignment @xmath121 , which satisfies ( [ eq : sinz_card_2 ] ) , ( [ eq : sinz_card_3 ] ) , ( [ eq : equiv_clause_3 ] ) and ( [ eq : equiv_clause_4 ] ) , is such that @xmath166 . then the value of @xmath167 under @xmath121 can still be true , as long as @xmath168 , while not violating the cardinality constraint @xmath169 . for our purposes we require an unoptimised version of the encoding with a tighter restriction on the values of @xmath137 stated in proposition [ prop : sinz ] . we say that a @xmath36 sequence @xmath54 is _ @xmath0-bounded _ , for some @xmath170 , if @xmath171 , for all @xmath172 . we extend the notion of @xmath0-boundedness to boolean @xmath49 sequences using the relation between @xmath36 sequences @xmath173 and boolean @xmath174-sequences @xmath47 defined as follows : @xmath175 ( in other words , @xmath45 is encoded by the boolean value _ true _ , and @xmath20 is encoded by the boolean value _ false _ ) . then a @xmath49 sequence @xmath48 is @xmath0-bounded if for every @xmath176 the disbalance between the number of @xmath44s in @xmath177 and the number of @xmath97s in @xmath177 is at most @xmath0 . we build our sat encoding of the erds discrepancy problem on the sequential counter - based encoding of cardinality constraints described in section [ sec : sinz_card ] . we illustrate our approach by the following consideration . by proposition [ prop : sinz ] , an arbitrary assignment of @xmath49 values to propositions @xmath178 can be uniquely extended to a model of @xmath141 . we denote this model as @xmath121 . suppose that the value of some @xmath137 under @xmath121 is false . then , by proposition [ prop : sinz ] , the sequence @xmath179 contains at most @xmath180 occurrences of @xmath44 and so it contains at least @xmath181 occurrences of @xmath97 . therefore , the disbalance between the number of occurrences of @xmath97 and the number of occurrences of @xmath44 is at least @xmath182 . if @xmath183 then the sequence @xmath179 is not @xmath0-bounded thus , in our encoding of @xmath0-bounded sequences we need to exclude such a possibility . this can be achieved by conjoining formula @xmath141 with @xmath184 similarly , if the value of @xmath137 is true , for some @xmath185 , then the number of @xmath44s exceeds the number of @xmath97s by more than @xmath0 , so these possibilities should also be excluded by @xmath186 to summarise , let propositional formula @xmath187 be the conjunction of @xmath141 , with ( [ eq : equiv_clause_8 ] ) and ( [ eq : equiv_clause_6 ] ) . notice that , as in the case of @xmath141 , we write ( [ eq : equiv_clause_8 ] ) and ( [ eq : equiv_clause_6 ] ) explicitly for the ease of explanation . the proof of the following theorem can be found in appendix [ sec : proofs ] . [ th : cbounded ] for any assignment @xmath188 the following holds : there exists an extension of @xmath121 to @xmath189 that is a model of @xmath187 if , and only if , the sequence @xmath190 is @xmath0-bounded . as ( [ eq : sinz_card_1 ] ) is logically equivalent to the set of clauses ( [ eq : equiv_clause_1])([eq : equiv_clause_4 ] ) , formula @xmath187 is logically equivalent to the set of clauses @xmath22 consisting of ( [ eq : equiv_clause_1])([eq : equiv_clause_4 ] ) , ( [ eq : sinz_card_2 ] ) , ( [ eq : sinz_card_3 ] ) , ( [ eq : equiv_clause_8 ] ) and ( [ eq : equiv_clause_6 ] ) . let @xmath191 be the result of applying unit propagation to @xmath22 in an exhaustive manner . one can see that the set @xmath191 contains less than @xmath192 auxiliary propositions and less than @xmath193 clauses . we construct @xmath194 , a clausal representation of the statement that the sequence @xmath195 is @xmath10-bounded . we also demonstrate how clauses ( [ eq : sinz_card_2 ] ) , ( [ eq : sinz_card_3 ] ) , ( [ eq : equiv_clause_8 ] ) and ( [ eq : equiv_clause_6 ] ) are unit propagated into clauses ( [ eq : equiv_clause_1])([eq : equiv_clause_4 ] ) . notice that for @xmath196 every instance of clause ( [ eq : equiv_clause_1 ] ) contains literal @xmath197 , the only literal of the unit clause ( [ eq : sinz_card_3 ] ) . thus , every instance of clause ( [ eq : equiv_clause_1 ] ) for @xmath196 is redundant . for @xmath198 and @xmath199 , clause ( [ eq : equiv_clause_1 ] ) contains @xmath200 , the only literal of the unit clause ( [ eq : sinz_card_2 ] ) , so for @xmath198 and @xmath199 , the instance of clause ( [ eq : equiv_clause_1 ] ) is also redundant . for @xmath198 and @xmath201 , an instance of the unit clause ( [ eq : sinz_card_2 ] ) , namely @xmath202 , unit propagates into ( [ eq : equiv_clause_1 ] ) resulting in a 2-cnf clause @xmath203 . for @xmath198 and @xmath204 , ( [ eq : equiv_clause_1 ] ) instantiates to @xmath205 . finally , for @xmath198 and for @xmath206 , clause ( [ eq : equiv_clause_1 ] ) contains @xmath207 , the only literal of the unit clause ( [ eq : equiv_clause_8 ] ) . thus , for @xmath198 and @xmath206 the instances of clause ( [ eq : equiv_clause_1 ] ) are redundant . by a further consideration of cases , one can see that the set of all non - redundant simplified instance of clause ( [ eq : equiv_clause_1 ] ) , for @xmath208 and @xmath209 , consists of @xmath210 ( \lnot { \ensuremath{{s}}}^2_3 \lor { \ensuremath{{s}}}^2_2 \lor { \ensuremath{{s}}}^1_2 ) & ( \lnot { \ensuremath{{s}}}^3_5 \lor { \ensuremath{{s}}}^3_4 \lor { \ensuremath{{s}}}^2_4 ) . \end{array}\ ] ] we group here the clauses in such a way that all clauses in one column correspond to the same value of the parameter @xmath4 . similarly , instances of clause ( [ eq : equiv_clause_2 ] ) , for @xmath208 and @xmath211 are simplified with the help of unit clause ( [ eq : sinz_card_2 ] ) , ( [ eq : sinz_card_3 ] ) , ( [ eq : equiv_clause_8 ] ) and ( [ eq : equiv_clause_6 ] ) to @xmath212 ( \lnot { \ensuremath{{s}}}^1_2 \lor { \ensuremath{{s}}}^1_1 \lor p_2 ) & ( \lnot { \ensuremath{{s}}}^2_3 \lor { \ensuremath{{s}}}^2_2 \lor p_3 ) & ( \lnot { \ensuremath{{s}}}^3_5 \lor { \ensuremath{{s}}}^3_4 \lor p_5 ) . \\[2pt ] ( { \ensuremath{{s}}}^1_2 \lor p_3 ) & ( \lnot { \ensuremath{{s}}}^2_4 \lor { \ensuremath{{s}}}^2_3 \lor p_4 ) & \\[2pt ] \mbox { } & ( { \ensuremath{{s}}}^2_3 \lor p_5 ) & \\[2pt ] \end{array}\ ] ] the set of all non - redundant simplified instances of clause ( [ eq : equiv_clause_3 ] ) consists of @xmath213 & ( \lnot { \ensuremath{{s}}}^2_3 \lor { \ensuremath{{s}}}^2_4 ) & \end{array}\ ] ] and set of all non - redundant simplified instances of clause ( [ eq : equiv_clause_4 ] ) consists of @xmath214 ( \lnot p_2 \lor { \ensuremath{{s}}}^1_2 ) & ( \lnot { \ensuremath{{s}}}^1_2 \lor \lnot p_3 \lor { \ensuremath{{s}}}^2_3 ) & ( \lnot { \ensuremath{{s}}}^2_3 \lor \lnot p_4 \lor { \ensuremath{{s}}}^3_4)&\\[2pt ] \mbox { } & ( \lnot p_4 \lor { \ensuremath{{s}}}^2_4 ) & ( \lnot { \ensuremath{{s}}}^2_4 \lor \lnot p_5 \lor { \ensuremath{{s}}}^3_5 ) & \\[2pt ] \end{array } \ ] ] thus , the set @xmath194 consists of 26 clauses grouped in ( [ eq : ex1.1])([eq : ex1.4 ] ) above . any @xmath36-sequence containing less than or equal to @xmath0 elements is always @xmath0-bounded . it should be clear then that the discrepancy of a @xmath36 sequence @xmath215 is bounded by @xmath0 if , and only if , for every @xmath216 the subsequence @xmath217 is @xmath0-bounded . then we define @xmath218 we assume here that for the different values of @xmath5 sets @xmath219 share the same input propositions @xmath47 but use different auxiliary propositions @xmath137 . then the following theorem is a direct consequence of theorem [ th : cbounded ] [ th : discrepancy ] for any assignment @xmath188 the following holds : there exists an extension of @xmath121 to @xmath220 that is a model of @xmath221 if , and only if , @xmath190 encodes a @xmath36 sequence @xmath54 of length @xmath33 and discrepancy at most @xmath0 . we conclude this section with a description of two optimisations , which we present in the form of propositions . both reduce significantly the size of the unsatisfiability certificate and have some noticeable effect on the running time . the first optimisation allows one to remove the ` do nt care ' propositions , which do not affect the satisfiability of the problem . the second optimisation breaks the symmetry in the problem . [ prop : oddeven ] suppose that a sequence @xmath48 is @xmath0-bounded and either @xmath33 is odd and @xmath0 is even or @xmath33 is even and @xmath0 is odd . then for an arbitrary value @xmath222 the sequence @xmath223 is @xmath0-bounded . it suffices to notice that @xmath224 is odd if , and only if , @xmath149 is odd . thus , under the conditions of the proposition , @xmath225 , and the sequence can be extended arbitrarily . [ prop : symmetry ] there exists a @xmath36 sequence @xmath54 of length @xmath33 and discrepancy at most @xmath0 if , and only if , there exists a @xmath36 sequence @xmath226 of length @xmath33 and discrepancy at most @xmath0 , in which @xmath227 , for some arbitrary but fixed value of @xmath110 . the problem is symmetric , that is , the discrepancy of @xmath54 is bounded by @xmath0 if , and only if , the discrepancy of @xmath228 is bounded by @xmath0 . multiplicativity and complete multiplicativity of @xmath36 sequences can be encoded in sat in a rather straightforward way . assuming that a boolean sequence @xmath229 encodes a @xmath36 sequence @xmath173 so that the logical value @xmath44 encodes the numerical value @xmath45 and the logical value @xmath97 encodes the numerical value @xmath20 , a sat encoding of the fact that @xmath230 is captured by the following clauses , which enumerate all four combinations of values of @xmath231 and @xmath232 : @xmath233 then multiplicativity of @xmath46 is captured by instances of ( [ eq : mult2 ] ) for all coprime pairs @xmath41 and @xmath149 ; and , by ( [ eq : multdef2 ] ) , complete multiplicativity of the sequence @xmath54 is captured by instances of ( [ eq : mult2 ] ) for @xmath149 and @xmath4 such that every product @xmath234 is generated only once . for complete multiplicativity further optimisation is possible due to the fact that in any such sequence @xmath235 for any @xmath236 . it can be seen that the cnf formula @xmath237 defined below expresses the complete multiplicativity condition on @xmath41 , for every @xmath238 . @xmath239 \{(p_i)\ } & \textrm { if $ i = j^2 $ , for some $ j\geq 1 $ } \\[5pt ] \mathsf{prod}_{j , k}\quad & \textrm { if none of the cases above applies}\\ & \textrm { and $ j , k$ are some non - trivial divisors of $ i$. } \end{array } \right.\ ] ] then we define two sets of clauses @xmath240 and @xmath241 the following statement is a direct consequence of theorem [ th : discrepancy ] . [ th : mult_discrepancy ] for any assignment @xmath188 the following holds : there exists an extension of @xmath121 to @xmath242 ( or an extension of @xmath121 to @xmath243 ) , which is a model of @xmath244 ( or @xmath245 , respectively ) if , and only if , @xmath190 encodes a multiplicative ( or completely multiplicative , respectively ) @xmath36 sequence @xmath54 of length @xmath33 and discrepancy at most @xmath0 . finally we notice that the completely multiplicative case can be optimised based on the following observation . [ prop : cmult_opt ] the discrepancy of a completely multiplicative @xmath36 sequence @xmath54 is bounded by @xmath0 , for some @xmath65 , if , and only if , @xmath54 is @xmath0-bounded . the necessary condition is trivial by definition of discrepancy . for the sufficient condition we show that for any @xmath0-bounded sequence @xmath173 and any @xmath246 the subsequence @xmath247 is @xmath0-bounded . let @xmath248 . then @xmath249 results ------- in our experiments we use ` treengeling ` , a parallel cube - and - conquer flavour of the ` lingeling`sat solver @xcite version aqw , the winner of the _ sat - unsat _ category of the sat13 competition @xcite , and the ` glucose`solver @xcite version 3.0 , the winner of the _ certified unsat _ category of the sat13 competition @xcite . all experiments were conducted on pcs equipped with an intel core i5 - 2500k cpu running at 3.30ghz and 16 gb of ram . in our first series of experiments we investigate the discrepancy of unrestricted @xmath36 sequences . we encode the existence of a @xmath36 discrepancy @xmath0 sequence of length @xmath33 into sat as described in section [ sec : encoding ] . we deploy both optimisations described in proposition [ prop : oddeven ] and proposition [ prop : symmetry ] . we choose as @xmath110 , for which we fix @xmath250 to be @xmath45 , colossally abundant numbers @xcite , which have many divisors and thus contribute to many homogeneous sequences . specifically for @xmath8 , the choice of @xmath251 is more beneficial for satisfiable instances ; however , @xmath252 results in a better reduction of the size of the unsatisfiability proof described below . for consistency of presentation , we use @xmath253 in all our experiments for @xmath8 . for @xmath8 we establish that the maximal length of a @xmath1 sequence of discrepancy @xmath10 is @xmath11 . the cnf formula @xmath254 contains @xmath255 propositions and @xmath256 clauses . it takes the ` treengeling`system about @xmath257 seconds to find an example of such a sequence on our hardware configuration . one of the sequences of length @xmath11 of discrepancy @xmath10 can be found in appendix [ sec : sequence ] . when applied to the cnf formula @xmath258 , which contains @xmath259 propositions and @xmath260 clauses , ` treengeling`reports unsatisfiability . in order to corroborate this statement , we also use ` glucose ` . it takes the solver about @xmath261 seconds to generate a drup certificate of unsatisfiability . the correctness of the generated unsatisfiability certificate has been independently verified with the ` drup - trim`tool @xcite . the size of the certificate is about @xmath262 gb , and the time needed to verify the certificate is comparable with the time needed to generate it . the rup unsatisfiability certificate , that is the drup certificate with all information on the deleted clauses stripped , is @xmath263 mb ; it takes ` drup - trim`about five and a half hours to verify it . combined with theorem [ th : discrepancy ] , these two experiments yield a computer proof of the following statement . [ th : edp2 ] the length of a maximal @xmath36 sequence of discrepancy @xmath10 is @xmath11 . thus we prove that the erds discrepancy conjecture holds true for @xmath8 . when applied to @xmath264 for increasing values of @xmath33 our method could only produce sequences of discrepancy @xmath14 of length in the region of @xmath265 , even though solvers were allowed to run for weeks . since both multiplicativity and complete multiplicativity restrictions reduce severely the search space , in hope for better performance , we perform the second series of experiments to investigate the discrepancy bound for multiplicative and completely multiplicative sequences . notice that the optimisation described in proposition [ prop : symmetry ] is not applicable in this case as the fact that @xmath54 is multiplicative does not imply that @xmath266 is . we saw in example [ ex : edp1 ] that multiplicative sequences of discrepancy @xmath44 are longer than completely multiplicative sequences . the longest completely multiplicative sequence of discrepancy @xmath10 is known to contain @xmath42 elements @xcite ; tests with @xmath267 show that the longest multiplicative sequence of discrepancy @xmath10 has @xmath268 elements . thus it would nt be unreasonable to assume that the longest multiplicative discrepancy @xmath14 sequence is longer than the longest completely multiplicative one , but is probably harder to find . it turns out that this expectation is wrong on both accounts . we establish that the length of a maximal @xmath1 completely multiplicative discrepancy @xmath14 sequence coincides with the length of a maximal @xmath36 multiplicative discrepancy @xmath14 sequence and is equal to @xmath13 . it takes ` treengeling`about one hour and fifty minutes to find a satisfying assignment to @xmath269 , which contains @xmath270 propositions and @xmath271 clauses , and about one hour and thirty five minutes to find a satisfying assignment to @xmath272 , which also contains @xmath270 propositions but @xmath273 clauses . it takes the ` glucose`solver just under eight hours to generate an approximately 1.95 gb drup proof of unsatisfiability for @xmath274 , which contains @xmath270 propositions and @xmath275 clauses , and about nine and a half hours to generate an approximately 3.78 gb drup proof of unsatisfiability for @xmath276 , which again contains the same number of propositions but @xmath277 clauses . the optimisation of proposition [ prop : cmult_opt ] leads reduction in the problem size for the completely multiplicative case ( 446753 propositions and 1738125 clauses for length 127645 and 446759 propositions and 1738149 clauses for length 127646 ) and a significant reduction in the ` treengeling`running time ( about 20 and 30 minutes , respectively ) ; however , it does not reduce the size of the drup certificate for the unsatisfiable case , which is about 2.22 gb . this lack of reduction in size is due to the fact that unit propagation steps are not recorded as part of a drup certificate . so we get a computer - aided proof of another sharp bound on the sizes of maximal sequences of bounded discrepancy . [ th : edp3mult ] the length of a maximal multiplicative @xmath36 sequence of discrepancy @xmath14 equals to the length of a maximal completely multiplicative @xmath36 sequence of discrepancy @xmath14 and is @xmath13 . unrestricted sequences of discrepancy @xmath14 can still be longer than @xmath278 : by requiring that only first @xmath279 elements of a sequence are completely multiplicative , we generate a 130,000 long edp3 sequence in about one hour and fifty minutes thus establishing a slightly better lower bound on the length of @xmath36 sequences of discrepancy @xmath14 than the one from theorem [ th : edp3mult ] . the solvers struggle to expand it much further . notice that the optimisation of proposition [ prop : cmult_opt ] is not applicable here . we summarise known facts about discrepancy of unrestricted , multiplicative and completely multiplicative sequences in table [ tab : results ] . we highlight in boldface cases where the lengths of maximal sequences of different kinds are equal . we have demonstrated that sat - based methods can be used to tackle the longstanding mathematical questions related to discrepancy of @xmath1 sequences . not only were we able to identify the exact boundary between satisfiability and unsatisfiability of the encoding of the edp for @xmath8 , thus identifying the longest sequences of discrepancy @xmath10 , but also we have established the surprising fact that the lengths of the longest multiplicative and completely multiplicative sequences of discrepancy @xmath14 coincide . the latter result helps to establish a novel lower bound on the length of the longest discrepancy @xmath14 sequence . there is , however , a noticeable asymmetry in our findings . the fact that a sequence of length 1160 has discrepancy @xmath10 can be relatively easily checked manually . it is harder but not impossible to verify the correctness of the discrepancy bound for @xmath280-long sequences . on the other hand , even though improvements to our method shortened the wikipedia - size 13 gb proof reported in @xcite more than tenfold passing thus the psychological barrier of @xmath44 gb , it is still probably one of the longest proofs of a non - trivial mathematical result , and it is equally improbable that a mathematician would verify by hand ten billions or half a billion of automatically generated proof lines . it should be noted that this gigantic proof is a formal proof in a well - specified proof system and , as such , it has been verified by a third - party tool , and it can potentially be translated into a proof assistant such as coq @xcite . the reduction of proof size will be useful for any future analysis of the proof in an attempt to identify patterns and lemmas and produce a compact proof more amenable for human comprehension . [ [ acknowledgements ] ] acknowledgements + + + + + + + + + + + + + + + + the authors would like to thank armin biere , marijn heule , pascal fontaine , laurent simon and laurent thry and donald knuth for helpful discussions , comments and ideas following the publication of the preliminary version of this paper @xcite . leonidas alaoglu and paul erds . on highly composite and similar numbers . , 56(3):448469 , 1944 . noga alon . transmitting in the _ n_-dimensional cube . , 37/38:911 , 1992 . tom m. apostol . . undergraduate texts in mathematics . springer , new york , ny , usa , 1976 . michal armand , germain faure , benjamin grgoire , chantal keller , laurent thry , and benjamin werner . a modular integration of sat / smt solvers to coq through proof witnesses . in jean - pierre jouannaud and zhong shao , editors , _ certified programs and proofs - first international conference , cpp 2011 , kenting , taiwan , december 7 - 9 , 2011 . proceedings _ , volume 7086 of _ lecture notes in computer science _ , pages 135150 . springer , 2011 . gilles audemard and laurent simon . glucose 2.3 in the sat 2013 competition . in _ proceedings of sat competition 2013 _ , pages 4243 , helsinki , 2013 . university of helsinki . adrian balint , anton belov , marijn j. h. heule , and matti jrvisalo , editors . . university of helsinki , 2013 . jzsef beck and william w. l. chen . . cambridge university press , cambridge , 1987 . jsef beck and vera t. ss . discrepancy theory . in ronald . l. graham , martin grtschel , and lsl lovsz , editors , _ handbook of combinatorics _ , volume 2 , pages 14051446 . elsivier , amsterdam , 1995 . armin biere , marijn heule , hans van maaren , and toby walsh , editors . , volume 185 of _ fronteers in artificial intelligence and applications_. ios press , amsterdam , 2009 . armin biere . ingeling , plingeling and treengeling entering the sat competition 2013 . in _ proceedings of sat competition 2013 _ , pages 5152 , helsinki , 2013 . university of helsinki . peter borwein , stephen k. k. choi , and michael coons . completely multiplicative functions taking values in @xmath281 . , 362(12):62796291 , 2010 . bernard chazelle . . cambridge university press , new york , 2000 . paul erds . some unsolved problems . , 4(3):291300 , 1957 . evguenii i. goldberg and yakov novikov . verification of proofs of unsatisfiability for cnf formulas . in _ proceedings of design , automation and test in europe conference and exposition ( date 2003 ) , 3 - 7 march 2003 , munich , germany _ , pages 1088610891 , 2003 . timothy gowers . is massively collaborative mathematics possible ? , 2009 . retrieved april , 10 2014 from http://gowers.wordpress.com/2009/01/27/is-massively-collaborative-mathematics-possible/ timothy gowers . erds and arithmetic progressoins . in _ erds centennial conference _ , 2013 . retrieved april , 10 2014 from http://www.renyi.hu/conferences/erdos100/program.html marijn heule , warren a. hunt jr . , and nathan wetzler . trimming while checking clausal proofs . in _ proceedings of formal methods in computer - aided design , fmcad 2013 _ , pages 181188 . ieee , 2013 . boris konev and alexei lisitsa . a sat attack on the erds discrepancy conjecture . , abs/1402.2184 , 2014 . boris konev and alexei lisitsa . a sat attack on the erds discrepancy conjecture . in _ proceedings of the 17th international conference on theory and applications of satisfiability testing , sat 2014 _ , 2014 . to appear . a. r. d. mathias . on a conjecture of erds and udakov . combinatorics , geometry and probability , 1993 . ji matouek and joel spencer . discrepancy in arithmetic progressions . , 9(1):195204 , 1996 . ji matouek . , volume 18 of _ algorithms and combinatorics_. springer , 1999 . s. muthukrishnan and aleksandar nikolov . optimal private halfspace counting via discrepancy . in _ proceedings of the 44th symposium on theory of computing _ , stoc 12 , pages 12851292 , new york , ny , usa , 2012 . aleksandar nikolov and kunal talwar . on the hereditary discrepancy of homogeneous arithmetic progressions . , abs/1309.6034v1 , 2013 . erds discrepancy problem : polymath wiki , 2010 . retrieved april , 10 2014 from http://michaelnielsen.org/polymath1/index.php?title=the_erds_discrepancy_problem d.h.j . polymath . human proof that completely multiplicative sequences have discrepancy greater than 2 , 2011 . retrieved april , 10 2014 from http://michaelnielsen.org/polymath1/index.php?title=human_proof_that_completely_multiplicative_sequences_have_discrepancy_at_least_2 wolfgang rautenberg . . springer , new york , 3 edition , 2010 . klaus f. roth . remark concerning integer sequence . , 9:257260 , 1964 . olivier roussel and vasco m. manquinho . pseudo - boolean and cardinality constraints . in armin biere , marijn heule , hans van maaren , and toby walsh , editors , _ handbook of satisfiability _ , volume 185 of _ fronteers in artificial intelligence and applications _ , pages 695733 . ios press , amsterdam , 2009 . carsten sinz . towards an optimal cnf encoding of boolean cardinality constraints . in _ principles and practice of constraint programming - cp 2005 , 11th international conference , cp 2005 , sitges , spain , october 1 - 5 , 2005 , proceedings _ , volume 3709 of _ lecture notes in computer science _ , pages 827831 . springer , 2005 . grigory s. tseitin . on the complexity of derivation in propositional calculus . in anatol o. slisenko , editor , _ studies in constructive mathematics and mathematical logic , part 2 _ , pages 115125 , new york , 1970 . consultants bureau . nikolai udakov . theory of the characters of number semigroups . , 20:1115 , 1956 . in this section we give proofs of the technical results used in the main text . we re - state propositions and theorems here for the reader s convenience . proposition [ prop : sinz ] . _ let @xmath141 be as defined above . then _ 1 . for any assignment @xmath151 such that @xmath121 satisfies @xmath282 , any @xmath152 and @xmath153 we have @xmath154 2 . for any @xmath49-sequence @xmath155 there exists an assignment @xmath156 such that @xmath157 , for @xmath120 ; @xmath121 satisfies @xmath141 ; and for any @xmath158 and @xmath139 if @xmath159 then @xmath160 , for @xmath161 . 1 . the proof proceeds by induction on the lexicographical partial order @xmath283 on pairs of non - negative integers : @xmath284 iff @xmath285 . fix some @xmath286 . + consider cases : * suppose that @xmath287 . then formula ( [ eq : sinz_card_1 ] ) , one of the conjuncts of @xmath288 , instantiates to @xmath289 . therefore , for an assignment @xmath121 such that @xmath290 satisfies @xmath291 we have @xmath292 . furthermore , for the satisfying assignment we have @xmath293 and @xmath294 . it follows then that @xmath295 , which is equivalent to the statement of the proposition for the case @xmath147 . * suppose that @xmath199 and @xmath296 . for a satisfying assignment @xmath121 we have @xmath297 ( as ( [ eq : sinz_card_2 ] ) is a conjunct of @xmath288 ) . on the other hand for @xmath296 we have @xmath298 . thus the statement of the proposition holds true in this case . * suppose that @xmath299 , @xmath300 . for a satisfying assignment @xmath121 we have @xmath301 if and only if @xmath302 or @xmath303 ( by satisfaction of ( [ eq : sinz_card_1 ] ) ) . by induction hypothesis the later is equivalent to @xmath304 or @xmath305 and @xmath306 , which in turn is equivalent to @xmath307 . first notice that any assignment @xmath308 can be extended in a unique way to the assignment @xmath156 . indeed , satisfaction of ( [ eq : sinz_card_2 ] ) and ( [ eq : sinz_card_3 ] ) defines uniquely the values of satisfying assignment @xmath121 on @xmath309 for the cases @xmath310 and @xmath311 , respectively . further , using satisfaction condition for ( [ eq : sinz_card_1 ] ) the values of @xmath121 on the remaining variables @xmath309 with @xmath312 are defined uniquely by induction on @xmath283 . the remaining condition , that is for any @xmath158 and @xmath139 if @xmath159 then @xmath160 , for @xmath161 , now follows from item ( _ i _ ) above . theorem [ th : cbounded ] . _ for any assignment @xmath188 the following holds : there exists an extension of @xmath121 to @xmath189 that is a model of @xmath187 if , and only if , the sequence @xmath190 is @xmath0-bounded . _ @xmath313 assume that for an assignment @xmath314 the sequence @xmath315 is @xmath0-bounded . by item ( _ ii _ ) of proposition [ prop : sinz ] , @xmath121 can be extended to an assignment @xmath316 such that @xmath317 , for all @xmath318 ; @xmath319 satisfies @xmath320 ; and , furthermore , @xmath321 if , and only if , @xmath322 . we show that @xmath319 is , in fact , a model of @xmath323 . it suffices to demonstrate that clauses ( [ eq : equiv_clause_8 ] ) and ( [ eq : equiv_clause_6 ] ) are true under @xmath319 . 1 . assume to the contrary that @xmath324 , for some @xmath325 and @xmath326 . then by proposition [ prop : sinz ] we have @xmath327 as the number of occurrences of @xmath44 in @xmath328 is @xmath329 ( and , hence , the number of occurrences of @xmath97 in @xmath328 is @xmath330 ) , the difference between the number of occurrences of @xmath97 and the number of occurrences of @xmath44 is at least @xmath331 . + as we supposed that @xmath332 , the difference between the number of occurrences of @xmath97 and the number of occurrences of @xmath44 exceeds @xmath333 ( by ( [ eq : proof2 ] ) ) and therefore the sequence @xmath334 and hence @xmath315 is not @xmath0-bounded contradicting our assumption . 2 . assume to the contrary that @xmath321 for some @xmath325 and @xmath335 . then @xmath336 ( by proposition [ prop : sinz ] ) now we estimate the difference between the number of occurrences of @xmath44 in @xmath337 and the number of occurrences of @xmath97 in @xmath337 . we have + @xmath338 + it follows that the sequence @xmath315 is not @xmath0-bounded contradicting our assumption . @xmath339 consider an assignment @xmath340 and assume that its extension to @xmath341 is a model of @xmath323 . now we are to show that @xmath342 is @xmath0-bounded . assume to the contrary that it is not @xmath0-bounded . as @xmath319 is an extension of @xmath121 we have that @xmath343 is also not @xmath0-bounded . then we have @xmath344 for some @xmath149 . consider two cases : 1 . the number of occurrences of @xmath44 in @xmath345 , for @xmath346 , exceeds the number of occurrences of @xmath97 by more than @xmath0 . let @xmath347 . then we have @xmath348 that is @xmath349 and @xmath350 . now we have @xmath351 , by the satisfaction condition for ( [ eq : equiv_clause_8 ] ) , and @xmath321 , by @xmath347 . a contradiction . the number of occurrences of @xmath97 in @xmath345 , for @xmath346 , exceeds the number of occurrences of @xmath44 by more than @xmath0 . let @xmath352 . then we have @xmath353 that is , @xmath354 and @xmath355 . then , by the satisfaction condition for ( [ eq : equiv_clause_6 ] ) , we have @xmath321 and , @xmath352 , we have @xmath324 . a contradiction . we give a graphical representation of one of the sequences of length @xmath11 obtained from the satisfying assignment computed with the ` treengeling`solver . here @xmath356 stands for @xmath45 and @xmath357 for @xmath20 , respectively . - + + - + - - + + - + + - + - - + - - + + - + - - + - - + + + - + - - + + - + + - + - + + - - + + - + - - - + - + + - + + - - + - - + + + + - - + - - + + - + - - + + - + + - - - + - + + - + + - + - + + - - + + - + - + - - - + + - + - - + + + - + + - + - - + + - + - - + - - - + - + + - + - - + + - + + + - + - - + - - + + - + + - + - - + + - + - - + + + - + + - + - - - - + + + - + - - + - - + + + - - - + + - + + - + + - - + - - + + + - - + - + - + - - + - + + + - + + - + - - + + - - + + - + - - + + - + + - + - - + - - + + - - + + + - + - - + + + - + - - - + + - + - - + + - - + - + - - + - + + + + - + - - + + - + + - + - - + + - + - - + - - + + - + - - + + + - - + - + + - + - + - - + - + - + + - + - - + + - + - + - + - - + + - + - + - + + - + - + - + + - - - + - + - - + + + + + - - + - - - + + - + - + + - + - - + + - + - - + - - + + + - + - - + + + + - - + - - - + - + + + + - - + - - + + + - + + - + - - + + - + - - + - - + + - + - - + + - + + - + + - - + + - + - - + - - + + - + + - + - - - - + + + - + - - + + + - - + + + - - - + - + + - + - - + - + + - - - + - + + + - + + - + - - + - - + + - - + + + + - + - - + - - + - - + + + + + - - + - - + + + - - - + + - + + - + - - + + - - + - + + - - + - - + + - + + - + - - + - - + - + + + - + + - + - + - + - - + + - - + - + + - + + - + - - + - - + - - + + - + + + - + - - + + + - - - + + - + - - + + - + + - - - + + + - + - - + + - + + - - - - + + + - - + - + + - + - - + - - + + + - + - - + + - + + - + - + + - - + + - - + + - - - - + + + + - + + - - + + - - - - + + - + + + - - + + - - - + + + - - + - - + - + - + + - + + - + + - + - + - - - - + + + - - + + + - + - - + + - + + - + - - + - - + - - + + - + - - + + - + + + - + - - + + - - + - + - - + - + - + - + + + + - - - + - + + - + + - - + - - + - + - + - + + - + - + + + - - + - + - + - + - - + - + + + - - + - + + + - - - + + - + - - + - - + + + - + + - - + + - - - + + - + - + + - - + + - + - - - + - + + + - + - - + - + + - - + + - + - - + + - + - - + - + + + + - + - - + + - - + - + - + + + - - + - + - - + + - + + - + + - - + - - + - + + - - - + - + + - + - + + - - + + - + - - + + + + - + - - - - + + - - + - + + - + - + + - - + + - + - + - + + - + - + + - - + + - + - - - + - + + - + - - + + + - + - - - + - + - + + - - + + - + - - + + - + + - + + - + - - + + - - + - - + + + + - - - + + - - - + - + - + + - + - + + + + - - + - + + - - + - + - - + - + - + + - - - + + + - + + +

the multiband nature of iron pnictides gives rise to a rich temperature - doping phase diagram of competing orders and a plethora of collective phenomena . at low dopings , the tetragonal - to - orthorhombic structural transition is closely followed by a spin density wave transition both being in close proximity to the superconducting phase . a key question is the nature of high- superconductivity and its relation to orbital ordering and magnetism . here we study the nafecoas superconductor using polarization - resolved raman spectroscopy . the raman susceptibility displays critical enhancement of non - symmetric charge fluctuations across the entire phase diagram which are precursors to a-wave pomeranchuk instability at temperature . the charge fluctuations are interpreted in terms of quadrupole inter - orbital excitations in which the electron and hole fermi surfaces breathe in - phase . below , the critical fluctuations acquire coherence and undergo a metamorphosis into a coherent ingap mode of extraordinary strength .

1410.6456

an important aim in the study of iron - based superconductors is to elucidate the nature of the superconducting state and its relation to adjacent phases @xcite . most feas compounds share a common phase diagram in which the underdoped region is marked by a tetragonal - to - orthorhombic structural transition at @xmath5 followed by a magnetic ordering transition at @xmath6 of collinear spin stripes which either precedes or coincides with @xmath5 @xcite . on introducing dopant atoms , superconductivity emerges with a transition temperature @xmath0 of tens of degrees @xcite . the driving force behind the structural transition is widely debated with main proposals of either spin @xcite or ferro - orbital @xcite nematic ordering . in the spin - nematic scenario the structural transition at @xmath5 is driven by magnetic fluctuations which breaks fourfold rotational ( @xmath7 ) lattice symmetry @xcite . the latter induces a sharp increase of the spin correlation length for one spin stripe orientation and a decrease of the other . in the orbital - nematic scenario , @xmath7 symmetry is broken by ferro - orbital ordering in which strong inter - orbital interactions lead to inequivalent occupation of the @xmath8 and @xmath9 fe orbitals . an enhancement of spin susceptibility is observed in ins or nmr pnictide data upon approaching the sdw transition @xcite . however , at higher dopings away from the sdw phase , this enhancement is rapidly suppressed @xcite . hence , while the close proximity to magnetic order naturally favors spin fluctuations as a candidate in providing the glue for cooper pairs @xcite , suppressed spin fluctuations appear to be insufficient in explaining the whole temperature - doping ( @xmath10@xmath11x ) phase diagram @xcite . nmr measurements of the relaxation rate @xmath12 in fese , which has no sdw transition , revealed that spin fluctuations only emerge below @xmath5 and the nematic order was argued to be driven by orbital degrees of freedom @xcite . in elastic strain measurements of co- and k - doped bafe@xmath13as@xmath13 and fese , the shear modulus @xmath14 softens on cooling and 1/@xmath14 follows a curie - weiss - like behavior which is interrupted at @xmath15 ) . elastoresistivity measurements display a similar behavior @xcite . the weiss - temperature , which we define as @xmath4 is observed to increase towards zero doping @xcite . @xmath16 was related to the contribution of the lattice to the electronic nematic fluctuations in refs . . @xmath16 correlates with the downshift of the mean field transition temperature @xmath4 @xcite . these two temperatures being noticeably different and a non - vanishing 1/@xmath14 at @xmath5 leaves the origin of the transition @xmath4 as an open question . the 1/@xmath14 temperature dependence was attributed to the electric quadrupole fluctuations due to the 3@xmath3 inter - orbital fluctuations in refs . . this conjecture is supported by detection of fe - quadrupole orbital fluctuations by electron diffraction measurements @xcite . at present it is unsettled whether the @xmath4-line @xcite is associated with the structural instability or is a separate instability of a different nature that breaks @xmath7 symmetry . the nematic theories and analysis of experimental data are generally based on the assumption that @xmath7 symmetry is broken at @xmath5 while translational symmetry is broken at @xmath6 @xcite . most pnictides have @xmath5 and @xmath6 near - conjoint in the low - doping regime @xcite including the heavily studied 122-family , i.e. co- or k - doped bafe@xmath13as@xmath13 . however , nafe@xmath1co@xmath2as , which is a 111-system , has @xmath5 and @xmath6 separated by more than 10 k and presents a better suited material in which the nature of the structural and sdw transitions can be studied separately @xcite . temperature - dependent x - ray powder diffraction studies of nafe@xmath1co@xmath2as find that at @xmath5 the high - temperature tetragonal p4/nmm structure transforms into the orthorhombic cmma structure with the orthorhombic distortion @xmath17=a@xmath18-b@xmath18 emerging smoothly upon cooling @xcite . here , a@xmath18 and b@xmath18 are the lattice parameters of the orthorhombic unit cell , a@xmath18=@xmath19a@xmath20+@xmath21 and b@xmath18=@xmath19a@xmath20-@xmath21 ( fig . [ fig : feas](h ) ) , and a@xmath20 of the tetragonal unit cell ( fig . [ fig : feas](b ) ) . neutron diffraction and muon spin rotation data confirms that the lattice distortion starts above @xmath22 @xcite , and an arpes study reports brillouin zone ( bz ) folding and doubling of the unit cell at @xmath23 @xcite . the orthorhombic op is established at @xmath5 and development of sdw long - range order is established at @xmath6 @xcite . the smooth continuous op implies the occurrence of a single structural instability which sets in at @xmath23 . the structural transition appears to be subtle with the volume of the lattice changing only marginally @xcite while both transitions display anomalies in resistivity measurements @xcite . specific heat studies reveal anomalies at @xmath5 and @xmath6 which are characteristic of second - order phase transitions @xcite . the spin - nematic scenario predicts a jump in the magnetic correlation length at @xmath23 and formation of a pseudogap which has been refuted by ins measurements where the @xmath22 and @xmath23 transitions appear to be decoupled @xcite . this may be in contrast to that observed in the 122 systems ( ba , ca)fe@xmath13as@xmath13 @xcite implying the spin - nematic scenario is more applicable to 122 systems than to nafe@xmath1co@xmath2as . an alternative picture to the spin - nematic model is where critical ferroquadrupoles trigger the orthorhombic structure transition which involves a ferro - orbital density wave at @xmath23 @xcite . the proximity of the structural and superconducting phase transitions is universal which makes it necessary to investigate both instabilities in one setting . so far , no clear consensus has been reached on the symmetry of the superconducting op . theories building on spin fluctuations favor unconventional @xmath24 pairing in which the superconducting op changes sign between electron- and hole - like fss @xcite . yet , other theories embrace orbital fluctuations building on superconductivity with @xmath25-pairing in which there is no sign change @xcite . recently , orbital antiphase @xmath24 has been proposed in which the pairing function of the fe @xmath26 orbital has opposite sign to the @xmath8 and @xmath9 orbitals @xcite , as well as orbital triplet pairing @xcite . the type of doping leading to superconductivity can either have a nodeless ( @xmath27-wave ) or a nodal ( @xmath3-wave ) op ( fig . [ fig : feas](i ) ) . electron- or hole - doping bafe@xmath13as@xmath13 with respectively co or k leads to a nodeless op , except at high hole - dopings where a switch to a nodal op occurs . in contrast , isovalent substitution with p yields a nodal op @xcite . a long - standing issue that remains unresolved in many classes of unconventional superconductors , including cuprates @xcite , heavy fermions @xcite , and iron pnictides is whether a quantum critical point ( qcp ) lies beneath the superconducting dome @xcite . the quantum criticality related to the anti - ferromagnetic qcp was extensively studied within the spin fermion model@xcite . in this model the critical fluctuations related to the qcp were shown to affect the properties far into the normal state . hence , the existence and detection of a qcp may offer an understanding of the origin of unconventional superconductivity and its coexistence with either magnetic or exotic phases . it has recently been demonstrated in theoretical studies that cooper pairing is enhanced in the vicinity of a nematic qcp @xcite . experimentally , elastic anomalies of the @xmath14 shear modulus observed near a qcp in ba(fe@xmath1co@xmath2)@xmath13 as@xmath13 suggests the involvement of ferro - quadrupole fluctuations @xcite . a second - order quantum phase transition lying beneath the superconducting dome has been reported in bafe@xmath13(as@xmath1p@xmath2)@xmath13 by measurements of the london penetration depth @xcite . however , a study using nmr , x - rays and neutrons finds no signatures of a qcp @xcite raising questions as to its origin . identification of the charge multi - polar collective excitations and their symmetry associated with the nematic qcp is essential for understanding superconductivity and competing phases to which raman spectroscopy is the most suitable probe @xcite . we use polarization - resolved electronic raman spectroscopy to study the charge dynamics of the multiband nafe@xmath1co@xmath2as superconductors characterized by partially filled 3@xmath3-orbitals . we demonstrate that charge transfers between the degenerate @xmath8 and @xmath9 orbitals lead to collective intra - orbital quadrupole charge fluctuations in the normal and superconducting state . we find that the entire tetragonal phase is governed by the emergence of strong overdamped orbital quadrupole fluctuations which upon cooling display critical enhancement . these critical fluctuations foretell an approaching subleading second order phase transition with broken @xmath7 symmetry and an orbitally - ordered state . in the low doping region , the formation of this phase is intervened by the structural transition and becomes subleading . below @xmath0 , the fluctuations acquire coherence and undergo a metamorphosis into a sharp ingap mode of extraordinary strength . in section ii we introduce the raman experiments including sample preparations and the raman probe . in section iii we give an overview of the nafe@xmath1co@xmath2as raman data and establish the @xmath10@xmath11x phase diagram of the static raman susceptibility . in section iv we compare the static raman susceptibility to a two - component fit of the nmr relaxation rate . in section v we present and analyse the raman data in more details and discuss it in terms of critical quadrupole fluctuations and the pomeranchuk instability . in section vi we discuss a possible density wave state below the structural transition . in section vii we present raman data in the superconducting state which entails discussions of ingap collective modes and their connection to critical quadrupole fluctuations in the normal state . in section viii we present the bardasis - schrieffer mode and its interplay with the ingap exciton mode in the particle - hole channel . in section ix we discuss a quantum critical point inside the superconducting dome in terms of the pomeranchuk instability and the interplay of the bardasis - schrieffer mode with the strong ingap collective mode . in section x we present the main conclusion of the @xmath3-wave pomeranchuk quadrupole fluctuations and their relation to the ingap collective mode of extraordinary strength . in the appendices we present appendix a : analysis of raman spectra ; appendix b : coupling of pomeranchuk fluctuations to the raman probe ; appendix c : relaxational mode fitting procedure ; appendix d : first - principle band structure calculations ; and appendix e : symmetry modes in momentum space . nafe@xmath1co@xmath2as single crystals were grown by the self - flux method as described in ref . . the volume fractions of bulk superconductivity for compounds with a doping range between 0.015 and 0.06 , measured with a quantum design squid magnetometer , were larger than 80@xmath28 . @xmath5 , @xmath6 and @xmath0 versus doping were reported in ref . and are shown in the @xmath10@xmath11x phase diagram , fig . [ fig : phasediagram](a ) . the superconducting gap values 2@xmath29 determined by arpes in refs . are indicated by vertical dashed lines in fig . [ fig : scmodes ] . figure [ fig : sm_blue](a ) shows 2@xmath29 determined by raman . the samples were vetted for the highest quality surfaces and were handled in a protective argon atmosphere in a glovebox , where they were packed into sealed glass containers with a protective argon atmosphere . upon preparing to do the raman measurements the sample was unpacked inside a nitrogen - filled protective glovebag sealed to the entrance of the cryostat . the crystal was then cleaved and positioned in the continuous flow optical cryostat . all raman scattering measurements were performed in a quasi - back scattering geometry along the crystal @xmath30-axis and excited with a kr@xmath31 laser line . we used a laser excitation energy of @xmath32=2.6 ev , except for investigations of the ingap collective modes shown in figs . [ fig : scmodes_red],[fig : collectivemodes](a ) where @xmath32=1.93 ev was also used . the incident laser power was less than 12 mw focused to a 50@xmath33100 @xmath34m@xmath35 spot on the @xmath36-surface . in the superconducting state the power was reduced to less than 2 mw . for @xmath32=2.6 ev , being close to resonant condition , the lowest temperature was @xmath375 k. the lower excitation energy of @xmath32=1.93 ev being pre - resonant and at @xmath373 k allowed us to observe both the @xmath38 as well as the @xmath39 excitons at finite frequencies . the spectra of the collected scattered light were measured by a triple - stage raman spectrometer designed for high - straylight rejection and throughput equipped with a liquid nitrogen - cooled charge - coupled detector . the raman spectra were corrected for the spectral response of the spectrometer and detector in obtaining the raman scattering intensity , @xmath40 . here , @xmath41 is a small luminescence background and @xmath42 and @xmath43 the polarization vectors for the incident and scattered photons for a given scattering geometry with respect to the unit cell ( fig . [ fig : feas](b ) ) . the recorded raman intensity was background subtracted with a near - linear line and a constant determined for each polarization geometry ( see appendix a ) . in obtaining the static raman susceptibility @xmath44 in the @xmath45 symmetry channel shown in fig . [ fig : phasediagram ] we performed a k - k transformation of the @xmath46 data shown in figs . [ fig : ramandata2](d - f),[fig : ramandata3](a - e ) . for a given doping x , the @xmath47 spectra for each temperature was first divided by @xmath48 to obtain @xmath49 . the lower frequency cutoff is @xmath3720 @xmath50 and @xmath49 was therefore extended to zero frequency with a phenomenological even function which fits well to the data . the static raman susceptibility was then calculated from the kramers - kronig relation , @xmath51 at zero frequency . the integration was performed up to the highest measured frequency @xmath37750 @xmath50 at which point @xmath49 was near zero . the raman response function is sensitive to charge density fluctuations driven by the incident and scattered photon fields . for a given scattering geometry with polarization vectors @xmath42 and @xmath43 for the incident and scattered photons , the raman susceptibility is given by , @xmath52\right > \ , dt . \label{response}\ ] ] the symmetrized raman tensor @xmath53 for the different scattering geometries can be classified by the irreducible representations for the crystallographic point group @xcite . the symmetry channels accessible by raman scattering transform @xmath54 , @xmath55 , @xmath56 , @xmath45 and @xmath57 irreproducible representation of @xmath58 point group ( above @xmath5 ) and as @xmath59 , @xmath56 , @xmath45 and @xmath60 for @xmath61 ( below @xmath5 ) . below @xmath5 , @xmath54 and @xmath45 becomes @xmath62 . using circularly polarized light we confirmed that the contribution from the @xmath55 symmetry channel can be neglected . the scattering geometry is referenced to the @xmath63@xmath11@xmath64 coordinate system of the crystallographic ( as - as ) unit cell depicted in fig . [ fig : feas](b ) . the incident and scattered photon fields cross polarized along the @xmath65- and @xmath66-directions of the two - fe unit cell yields @xmath67 susceptibility . for nafe@xmath1co@xmath2as with @xmath58 point group symmetry in the tetragonal phase , @xmath67 probes excitations in @xmath45 symmetry . the cross polarized photon fields rotated by 45@xmath68 yields @xmath69 or @xmath56 susceptibility . @xmath70 can be obtained in two steps : first , by aligning both photon fields along one axis , and then by obtaining the @xmath71 susceptibility . @xmath70 is given by @xmath72@xmath11@xmath69 . the space group in the tetragonal and orthorhombic phase is respectively @xmath73 with point group @xmath58 and @xmath74 with point group @xmath61 @xcite . entering the orthorhombic phase from the tetragonal phase is associated with broken symmetry operators which includes @xmath7 rotations and mirror planes of the tetragonal phase with @xmath58 point group symmetry . the point group symmetry for the orthorhombic phase is @xmath61 in which both @xmath45 and @xmath54 symmetry of the tetragonal phase conforms to @xmath59 symmetry . the temperature or doping dependent electronic raman susceptibility @xmath75 reveals the dynamics of collective excitations and provides an unambiguous identification of their symmetry @xcite . the symmetrized raman tensor for the different scattering geometries can be classified by the irreducible representations for the crystallographic point group @xcite . the symmetry channels accessible by raman scattering are @xmath54 , @xmath55 , @xmath56 , @xmath45 and @xmath57 for pnictides with a tetragonal 2-fe unit cell , i.e. for nafe@xmath1co@xmath2as ( above @xmath5 ) . in fig . [ fig : ramandata ] we show raman susceptibility @xmath76 at representative temperatures and dopings for the @xmath54 , @xmath45 and @xmath56 symmetry channels to point out important features in relation to the tetragonal , orthorhombic , sdw and superconducting phases of the @xmath10@xmath11x phase diagram ( fig . [ fig : phasediagram](a ) ) which will be discussed in depth below . most of these features are reflected in the @xmath77 response , while @xmath78 contains important characteristics of superconducting nature , and @xmath79 mainly features a @xmath56 phonon . the detailed temperature and doping dependence is shown in figs . [ fig : ramandata2]-[fig : scmodes_red ] . using kramers - kronig ( k - k ) transformation , we calculate the real part of @xmath80 at @xmath48=0 , the static raman susceptibility @xmath81 , for @xmath45 symmetry . figure [ fig : phasediagram](a ) shows @xmath44 in a @xmath82x phase diagram where @xmath23 , @xmath22 , and @xmath83 obtained by transport measurements @xcite are superimposed on top . the enhancement of @xmath44 with cooling , observed for all x , starts from high temperatures and culminates in a maximum at the structural transition @xmath23 or at a smaller maximum before the @xmath84line for higher dopings . @xmath44 is suppressed below the structural transition @xmath23 @xcite . figure [ fig : phasediagram](b ) shows @xmath81 with a universal fit to @xmath85 where the temperature axis for each doping x is shifted by @xmath4 . the inset shows the inverse of @xmath81 versus temperature @xmath10@xmath11@xmath4 with a fit to a universal straight line . the two sharp modes in @xmath78 at @xmath37164 and @xmath37195 @xmath50 , and in @xmath79 at @xmath37211 @xmath50 observed in the spectra for all dopings and temperatures are phonon excitations ( figs . [ fig : ramandata](g , h , i ) ) , as they are expected for the 111-family crystallographic structure @xcite . the frequencies of these phonons increase slightly with cooling , typical of anharmonic behavior , and do not display any anomalies in self - energy upon crossing phase transition lines . for low dopings , the @xmath78 susceptibility displays an overall enhancement of the spectra upon traversing the high - temperature tetragonal phase to the orthorhombic and sdw phases which maximizes at lower temperatures . for [email protected] , the most important changes occur in the low - frequency region below @xmath37200 @xmath50 when crossing from the normal into the superconducting state . here @xmath78 displays markedly different dynamics above and below @xmath83 , with featureless spectra above @xmath83 and below , one or more superconducting features in the range of @xmath3770 @xmath50 . the @xmath45 symmetry channel , @xmath47 contains several characteristics : ( i ) a broad peak extending to about 400 @xmath50 , indicated by green shading , which is dominating in the entire tetragonal phase above the @xmath23 and @xmath83 lines ; ( ii ) a low - frequency suppression and coherence peak in the orthorhombic phase , indicated by light blue shading ; ( iii ) a sharp resonance of extraordinary strength in the superconducting phase at @xmath37@xmath87 @xmath50 ( 7.1 mev ) , indicated by blue shading ; ( iv ) a broad continuum which diminishes with doping ( fig . [ fig : rmfit](a ) ) . figure [ fig : sm_nqr ] displays the temperature dependence of the nmr relaxation rate @xmath88 for as compared to the static raman susceptibility @xmath89 . the nmr data for dopings x=0 , 0.025 and 0.06 are from refs . . the as nucleus has spin 3/2 and can relax into both an electronic spin or a charge quadrupole excitation@xcite . the latter is described by nuclear quadrupole resonance ( nqr ) . @xmath88 is decomposed into two contributions , @xmath90 where @xmath91 and @xmath92 @xcite . in this model , the former is the curie - weiss law due to intraband relaxation and the latter is due to interband - like excitations in which the gap @xmath93=240 @xmath50 is used . @xmath89 scales to @xmath94 for all three dopings x=0 , x=0.025 and x=0.06 and we attribute @xmath95 to correspond to the pomeranchuk transition temperature at @xmath4 . the used value for @xmath93 corresponds to the minor mode at 240 @xmath50 which is present in @xmath96 for all dopings and temperatures above @xmath23 . the self - consistency of the presented analysis of @xmath88 and its correspondence to @xmath89 suggests that @xmath88 for nafe@xmath1co@xmath97as originates from quadrupole excitations and not spin relaxation . these orbital singlet excitations have @xmath29l=2 and can be detected by raman spectroscopy and nqr but not ins experiments . the same scaling analysis , including elastic probes , applied to the 122-family ( sr , eu)fe@xmath13as@xmath13 in ref . implies this role of quadrupoles may be a general feature of pnictide materials . in the tetragonal phase above @xmath23 , @xmath46 reveals the emergence of broad quasielastic scattering ( qes ) peaked at @xmath98 ( figs . [ fig : ramandata2],[fig : ramandata3],[fig : rmfit ] ) . the intensity of this feature is weak at high temperatures . upon cooling , it softens and gains in intensity , where it reaches a maximum at the @xmath99line and near @xmath83 . below @xmath23 the static susceptibility drops rapidly without any observable inflection point at @xmath22 , which is congruent with a smoothly developing orthorhombic op demonstrated in x - ray and neutron diffraction studies @xcite . we apply a universal fit to @xmath46 with a simultaneous fit of our data as a function of frequency , temperature and doping ( see appendix c , fig . [ fig : rmfit2 ] ) . above the @xmath23 and @xmath83 lines , @xmath46 can be decomposed into three components ( fig . [ fig : rmfit](a ) ) which includes a broad qes peak which can be described as a relaxational mode ( rm ) , @xmath100^{-1}$ ] , a continuum and a minor peak at @xmath37240 @xmath50 . description of the rm is based on a phenomenological model , eqs . [ rpa]-[raman 1 ] ( see appendix c ) . both the intensity of the continuum and of the @xmath37240 @xmath50 mode diminishes rapidly with doping , and vanishes near [email protected] ( figs . [ fig : rmfit](c , d ) ) . importantly , the orbital content of the larger @xmath101 fs is mainly composed of @xmath26 orbitals , while the @xmath102 and @xmath103 fss primarily have @xmath8 and @xmath9 orbital character ( fig . [ fig : feas](c ) ) @xcite . at the m point , the inner ( outer ) part of the @xmath17/@xmath104 fs has @xmath8 and @xmath9 ( @xmath26 ) orbital character . the continuum and the @xmath37240 @xmath50 mode likely involve the @xmath103 band as its fs reduces with doping ( see figs . [ fig : feas](b , d ) ) with the former due to intraband excitations and the latter due to an interband - like excitation with a 240 @xmath50 gap consistent with quadrupole excitations as verified by scaling of @xmath44 to nqr data ( see section iv ) . this finding is consistent with first - principle calculations taking into account spin - orbit coupling ( fig . [ fig : feas](e ) ) . figure [ fig : rmfit](c ) displays the intensity dependence of the rm with doping which is seen to persist for all dopings . figure [ fig : rmfit](e ) shows the doping dependence of @xmath4 which is observed to decrease close - to linear for increasing dopings becoming negative near x=0.022 . this behavior is consistent with that obtained from the analysis of the static raman susceptibility @xmath81 shown in the @xmath10@xmath11@xmath105 phase diagram ( fig . [ fig : phasediagram](a ) ) . the intensity of the rm decreases with doping ( fig . [ fig : rmfit](c ) ) . the frequency decreases linearly upon cooling below @xmath37100 k for all dopings with the extension crossing the temperature axis at @xmath4 ( insets to figs . [ fig : ramandata2](d - f),[fig : ramandata3 ] ) . the decrease of @xmath4 with doping can be described by a function @xmath106 crossing zero at x = x@[email protected] and becoming negative for [email protected] ( fig . [ fig : rmfit](e ) ) . in the @xmath10@xmath11x phase diagram , [ fig : phasediagram](a ) , the @xmath108line is parallel to the @xmath22 and @xmath99lines @xcite , approximately 10 and 20 k below , respectively , for [email protected] . the critical behavior of the susceptibility @xmath80 manifests in : ( 1 ) the enhancement of the static raman susceptibility @xmath44 which scales to the universal response function @xmath110^{-1}$ ] upon cooling for all doping concentrations x with a linear temperature dependence of @xmath4 ; and ( 2 ) the gain in intensity and near - linear slowdown of the characteristic fluctuation frequency @xmath111 . the inverse of @xmath81 , shown in the inset to fig . [ fig : phasediagram](b ) , exhibits the same linear behavior with temperature as @xmath112 of the rm until @xmath23 or @xmath83 , below which @xmath81 rapidly falls off . next we reflect on the emergent critical enhancement of @xmath80 as a result of strong electronic interactions . potential reasons for the critical behavior include : ( 1 ) electronic coupling to lattice degrees of freedom ; ( 2 ) magnetic fluctuations @xcite which may invoke the ising spin - nematic scenario ; ( 3 ) charge fluctuations leading to charge order . here we consider the latter in terms of quadrupole pomeranchuk fluctuations as the most likely candidate . the partially filled fe - orbitals with 3@xmath113 configuration give rise to interorbital quadrupole charge fluctuations @xcite . the critical charge fluctuations in real space are manifested in electron - hole excitations between degenerate @xmath8 and @xmath9 orbitals on fe - sites which lead to charge transfers as illustrated in fig . [ fig : feas](f ) . this induces a dynamic quadrupole moment of @xmath45 symmetry with nodes along the @xmath63@xmath11@xmath64 directions ( fig . [ fig : feas](b ) ) . these are orbital singlet excitations ( @xmath29l=2 ) . in momentum space , the fluctuations lead to dynamic distortions of the fss around the @xmath114- and m - point ( fig . [ fig : feas](f ) ) resulting in fluctuating quadrupole moments with nodes along @xmath114x and @xmath114y ( fig . [ fig : sm_qp](a ) ) . in - phase synchronization of the two fss leads to @xmath115 quadrupole deformations which are favored over @xmath116 for a dominant repulsive interaction between the @xmath114- and m - points ( see appendix e for further details ) . these critical quadrupole fluctuations drive the system towards a pomeranchuk - like instability extended to multibands @xcite . in a fermi liquid , the pomeranchuk instability directly leads to a nematic transition via spontaneous quadrupole deformation of the fss which freeze with static distortions and in real space the 2-fe unit cell becomes monoclinic ( fig . [ fig : feas](g ) ) . the critical behavior of the @xmath45 raman response foretells the approaching second order pomeranchuk phase transition at @xmath4 which breaks rotational invariance while translational symmetry is preserved . it occurs when the attraction in the @xmath3-wave channel exceeds a critical threshold @xcite . the pomeranchuk instability in iron pnictides is special in that it breaks the discrete @xmath7 symmetry via orbital ordering , i.e. a quadrupole lattice in an ordered orbital pattern ( see fig . [ fig : feas](g ) ) @xcite but without instigating a density wave ( dw ) instability . similar to fermi liquids it requires an attraction in the @xmath3-wave ( @xmath45 ) channel , i.e. an interaction term of the form @xmath117 with @xmath118 favoring an occupation difference @xmath119 of the @xmath8 and @xmath9 orbitals . the low - energy anomalies in the @xmath45 raman data reflects the critical fluctuations associated with the pomeranchuk instability . the extraordinary large temperature and frequency range of these fluctuations is consistent with the presence of a qcp defined by a vanishing weiss - temperature @xmath120 . the scaling of @xmath44 in a two - component fit to nqr data provides compelling evidence of quadrupole - relaxation ( see section iv ) @xcite . the range of the critical fluctuations in the @xmath121-symmetry channel extend over a much wider temperature range ( figs . [ fig : phasediagram](a , b ) ) than the sdw fluctuations limited to a narrow temperature range above @xmath22 @xcite . in conclusion , we find fe - orbital quadrupole fluctuations display critical behavior foretelling an approaching new ground state below the @xmath95-line pomeranchuk instability . these results appear to be consistent with other pnictide materials including the 122 and fese families suggesting this conclusion may be more universal . in addition , the observation of a similar weiss - temperature @xmath4-like line in the @xmath10@xmath11x phase diagram of both na-111 and ba-122 systems by raman @xcite suggests the @xmath95-line to be a universal feature of pnictides . a @xmath95-line is likewise seen in elastic strain measurements @xcite . in nmr studies of fese , no curie - weiss behavior was observed in the relaxation rate @xmath12 above @xmath5 @xcite . in fese , nmr is only sensitive to spins since the @xmath122se nucleus has spin 1/2 and does not couple to quadrupoles . however , in the 122-family compounds mentioned above which do display curie - weiss behavior , the @xmath122as nucleus has spin 3/2 and does relax into quadrupolar excitations . this observation imply that the curie - weiss behavior originates from quadrupoles . the fact that curie - weiss behavior is indeed observed in fese when using a @xmath14 probe @xcite further underlines this conclusion . in raman studies of co - doped ba-122 @xcite ( as well as afe@xmath13as@xmath13 , a = eu , er @xcite ) , fluctuations were detected over a range from @xmath5 ( @xmath37138 k for x=0 ) to room temperature . in the orthorhombic phase @xmath46 is characterized by a low - frequency suppression of spectral weight and a peak at 2@xmath123 which develops upon cooling observed for x=0 and x=0.015 crystals ( fig . [ fig : ramandata2](a , b , d , e ) ) . the peak is at about @xmath37300 @xmath50 at low temperatures for x=0 . this is near the @xmath3733 mev gap value reported by stm studies @xcite . the evolution of the raman response as a function of frequency and temperature in the low - doping regime is captured in the color contour plots shown in figs . [ fig : ramandata2](a , b ) . whether 2@xmath123 starts to develop at @xmath23 or at @xmath22 on cooling is obscured by the quasielastic peak ( qep ) which rapidly decreases below @xmath23 at which point the quadrupole fluctuations freeze due to the broken @xmath7 symmetry . the energy of 2@xmath123 decreases with doping , and for [email protected] , this low - frequency suppression and peak are absent . the 2@xmath123 feature could potentially originate from an sdw gap similar to what has been reported for 122-systems @xcite . here , signatures of the gap develops below @xmath22 which in the 122-family are near - conjoint with @xmath23 . in nafe@xmath1co@xmath2as , @xmath23 and @xmath22 are separated by more than 10 k. in case the 2@xmath123 suppression develops below @xmath23 , @xmath7 and translational symmetry are broken together and the instability at @xmath23 is kugel - khomski - type or due to quantum mechanical interactions between orbital and spin degrees of freedom as described by kugel and khomski @xcite . if the density wave order develops below @xmath22 translational symmetry is first broken at @xmath22 . in the superconducting state , for [email protected] , the low frequency peak and suppression are absent and @xmath46 contains features in both @xmath54 and @xmath45 symmetry . below @xmath83 , a resonance emerges in @xmath45 symmetry which sharpens , gains in strength and hardens to @[email protected] mev upon cooling for x@xmath86x@xmath107 ( figs . [ fig : ramandata2](f),[fig : ramandata3],[fig : scmodes ] ) . @xmath39 is the strongest at the lowest temperatures and near x@xmath107 at x=0.0175 , and then decreases in strength for increasing doping still prevailing for x=0.05 and vanishes for x=0.08 . the doping dependence of the superconducting features at 5 k in both @xmath78 ( top row ) and @xmath47 ( bottom row ) in comparison to normal state spectra at 23 k is summarized in fig . [ fig : scmodes ] . in the top panel , @xmath124@xmath3768 @xmath50 ( 8.5 mev ) , 2@xmath125 and 2@xmath126 are present from x=0.0175 to x=0.05 , and are nearly independent of doping ( fig . [ fig : sm_blue](a ) ) . 2@xmath125 and 2@xmath126 are consistent with arpes @xcite and are assigned as pair - breaking excitations across the corresponding superconducting gaps ( fig . [ fig : scmodes ] ) . the width of @xmath39 is less than 1 mev for [email protected] ( fig . [ fig : sm_blue](b ) ) whereafter it broadens and its intensity diminishes gradually until it vanishes before x=0.08 . @xmath124 and @xmath39 qualify as true ingap excitations as their energy lies below the minimal quasiparticle gap , @xmath128 ( fig . [ fig : scmodes ] ) . in contrast to the superconducting gap feature at the 2@xmath29 threshold which is characterized by a square - root divergence @xcite , the ingap collective modes appear as sharp delta - function - like resonances . next we interpret the spectrum of collective modes as they may present pertinent information of the superconducting state @xcite . early studies focused on the bardasis - schrieffer mode in bcs single band @xmath27-wave superconductors where attraction in a non-@xmath27-wave particle - particle ( @xmath129-@xmath129 ) channel would result in the bardasis - schrieffer mode forming below the 2@xmath29 gap edge.@xcite in multiband superconductors with weak interband interactions , which applies to mgb@xmath13 , the leggett mode results from coherent cooper pair interband tunneling @xcite . in multiband superconductors with strong interband interactions , which applies to pnictides including nafe@xmath1co@xmath2as , the leggett mode is pushed above the 2@xmath29 gap edge where it becomes overdamped and is therefore undetectable . recently , chubukov _ et al . _ predicted a new ingap exciton in pnictides to appear in @xmath54 symmetry below 2@xmath29 consistent with a condensate with @xmath24 symmetry @xcite . rather than cooper pairs , this mode is composed of particle - hole ( @xmath129-@xmath130 ) pairs forming a bound exciton in @xmath54 symmetry . a raman study of collective modes in multiband superconductors predicted a new @xmath129-@xmath130 mode in @xmath45 symmetry below 2@xmath29 and also discusses the bardasis - schrieffer mode @xcite . we assign @xmath124 to the @xmath129-@xmath130 charge exciton predicted by chubukov _ this @xmath129-@xmath130 mode in @xmath54 symmetry is represented by in - phase breathing between the electron and hole fss ( fig . [ fig : collectivemodes](b ) and appendix e , fig . [ fig : sm_qp](b ) ) . this breathing mode entails periodic charge transfer between the particle and hole pockets . the out - of - phase breathing of the particle and hole pockets turns the repulsion into an effective attraction @xcite . the sign flip of the effective interaction is similar to the effective attraction in the cooper channel for opposite sign of the op for the particle and hole pockets . hence , both @xmath124 and @xmath24 depend on the inter - pocket interaction winning over intra - pocket repulsion @xcite . @xmath124 signifies attraction in the @xmath27-wave channel in much the same way as @xmath39 does in the @xmath3-wave channel . if strong enough , such attraction may lead to the pomeranchuk instability in the @xmath54 channel . the significant intensity of @xmath39 suggests that it couples to light directly , implying that it is the @xmath3-wave counterpart of the @xmath124 exciton @xcite . because @xmath80 is controlled by a large coupling constant @xmath131 , the @xmath39 resonance , which is facilitated by a positive feedback of the superconductivity , emerges from the normal state rm upon cooling through @xmath0 while retaining its identity as a bound state of @xmath115 @xmath129-@xmath130 oscillations . hence , at higher dopings , where the structural transition is suppressed , these @xmath3-wave pomeranchuk fluctuations grow strong and below @xmath0 where low - lying excitations are removed , the rm gains coherence and @xmath39 appears as a sharp resonance . the immediate consequence of attraction in the @xmath45 channel is the ingap resonant modes below the quasi - particle continuum in the superconducting state . hence the attraction in the @xmath121 channel leads to a sharp resonance below the @xmath129-@xmath130 continuum @xcite . we note that the attraction causing the resonance is operational in the @xmath129-@xmath130 channel , while it is well known that the @xmath129-@xmath130 and cooper channel do not have a separate existence and are combined into a single ingap mode @xcite . it was shown that if the superconducting op changes sign on different sheets of the fs the two channels disentangle @xcite . this explains the presence of two rather than one peaks in the underdoped regime . for dopings [email protected] and temperatures @xmath373 k , a new weak mode appears at @xmath38=@xmath3725 @xmath50 ( 3.1 mev ) which becomes stronger for decreasing doping while @xmath39 weakens considerably ( figs . [ fig : scmodes_red],[fig : collectivemodes](a ) ) . @xmath38 , which we attribute to a bardasis - schrieffer mode , exists only in a narrow doping window to the right for both the @xmath23 and @xmath22 lines @xcite . the maximum @xmath44 in fig . [ fig : phasediagram](a ) tracks the known part of the @xmath23-line and at higher dopings is then observed to slightly curve in towards lower dopings for decreasing temperatures but below @xmath4 in a region we will name sc2 ( see figs . [ fig : phasediagram](a),[fig : scmodes_red],[fig : collectivemodes](a ) ) . in contrast to the detrimental effect of the dw state with the dw gap depleting the density of states , superconductivity below @xmath4 is not obstructed by the pomeranchuk instability . thus , superconductivity in the orthorhombic phase appears in the narrow doping window below @xmath4 ( region sc2 ) . the bardasis - schrieffer mode is excited indirectly by photons as the transformation of a @xmath129-@xmath130 into a cooper pair requires assistance of the condensate @xcite . for @xmath131@xmath1320 , pairing in the @xmath3-wave channel provides the conditions for the bardasis - schrieffer mode to exist . figure [ fig : collectivemodes](a ) shows an energy diagram of the superconducting state including the superconducting gap 2@xmath29 and the ingap collective modes @xmath124 , @xmath39 and @xmath38 shown together with their spectroscopic signatures in the raman data . beneath the superconducting dome but above @xmath4 , which we will name sc1 , the susceptibility diverges upon approaching @xmath4 and the @xmath39 exciton acquires extraordinary strength . however , in region sc2 , below @xmath4 the pomeranchuk fluctuations are gapped and the @xmath39 exciton susceptibility is rapidly suppressed . upon decreasing the distance to @xmath4 , i.e. at lower dopings away from x@xmath107 , the bardasis - schrieffer mode gets sharper by borrowing spectral weight from the @xmath39 exciton , ( figs . [ fig : scmodes_red],[fig : collectivemodes](a ) ) . the interaction between @xmath39 and @xmath38 versus doping is similar to that discussed for fese in ref . . the existence of the two superconducting regions sc1 and sc2 which feature the doping - dependent @xmath39 and @xmath38 exciton modes ( fig . [ fig : scmodes_red ] ) , separated by @xmath133 , defines a qcp at x@xmath107 lying beneath the superconducting dome ( fig . [ fig : phasediagram](a ) ) . the location of the boundary between sc1 and sc2 is affected by the competition between the nematic and superconducting orders for carriers @xcite . below the pomeranchuk instability at @xmath4 , the pomeranchuk fluctuations vanish and sc2 is characterized by a rhombohedral primitive unit cell , broken @xmath7 symmetry and a quadrupole lattice ordered in an orbital pattern ( fig . [ fig : feas](g ) ) . in sc1 , the critical fluctuations become quantum in nature , and upon decreasing the nonthermal control parameter x from the overdoped regime , @xmath39 gain in strength upon approaching @xmath134 . when crossing into sc2 , the intensity of the @xmath39 resonance collapses and @xmath38 appears indicative of a qcp occurring at x@xmath107 . hence , with doping as a control parameter , we probe spectral weight transfer from the strong @xmath129-@xmath130 @xmath45 exciton to the emerging bardasis - schrieffer mode and find signatures of a qcp lying beneath the superconducting dome @xcite . the qcp is associated with non - fermi - liquid behavior and occurring at the pomeranchuk instability becoming quantum at @xmath135@xmath1360 suggests it is driven by quadrupole pomeranchuk fluctuations . the same scenario may prevail in bafe@xmath13(as@xmath1p@xmath2)@xmath13 where a qcp is clearly present below the superconducting dome @xcite , but where a study using nmr , x - rays and neutrons finds no signatures of a qcp @xcite . thus suggesting the criticality or the qcp arises from the quadrupole pomeranchuk qcp . we find that the criticality or the qcp does not arise from either the structural or sdw transitions in support of the quadrupole pomeranchuk qcp presented in the main text . this conclusion is supported by recent theoretical studies of superconductivity driven by nematic fluctuations at or near a nematic qcp which find that : pairing in the @xmath27-wave channel is boosted by @xmath3-wave symmetry fluctuations @xcite ; near a qcp and pomeranchuk transitions , superconductivity is strongly enhanced @xcite . this study concludes that superconductivity is determined by a delicate interplay between the two competing effects , the pairing tendencies of op fluctuations and strong non - fermi - liquid effects due to electronic fluctuations ; considering a microscopic model , the nematic and sdw transitions merge below a temperature @xmath137@xmath138@xmath0 and continues to zero temperature as a first - order single nematic - sdw transition line @xcite . this study finds superconductivity to have a strong effect on this quantum phase transition allowing strong fluctuations to exist near it ; these transition lines backbend due to superconductivity and there may be a shift of the qcp beneath the superconducting dome @xcite . the existence of a qcp has been linked to the occurrence of superconductivity across several classes of unconventional superconductors with a superconducting dome surrounding it in the @xmath10@xmath11x phase diagram and with optimal @xmath0 near the qcp . it is by now widely believed that critical quantum fluctuations are important for the superconductivity @xcite . these fluctuations enhance interactions and result in an enhancement of electronic correlations upon approaching the qcp @xcite . we have studied many - body effects leading to unconventional superconductivity and to competing phases of charge , orbital and spin ordering of the na-111 family of pnictides containing partially filled 3@xmath3-orbitals . using polarization - resolved raman spectroscopy we find that the inter - orbital attractive interaction , which can be tuned by isovalent co substitution for fe , makes the system receptive to the pomeranchuk - like instability with @xmath3-wave symmetry and that strong critical fluctuations towards this instability dominate the entire tetragonal phase . in the superconducting phase , these fluctuations acquire coherence and undergo a metamorphosis into ingap collective modes of extraordinary strength . our finding is an example of non - fermi - liquid behavior , unconventional superconductivity and electronic ordering emerging from strong multi - polar interactions among 3@xmath3 electrons , which should be a more generic phenomenon relevant to other compounds containing partially filled @xmath3 or @xmath139-orbitals . we thank a. v. chubukov , p. coleman , r. m. fernandes , y. gallais , k. haule , t. t. ong , i. paul , a. sacuto , j. schmalian and w .- l . zhang for discussions . we extend our special thanks to a. v. chubukov for careful reading of the manuscript and for giving important comments and suggestions . the crystal growth efforts at utk and rice were supported by the us doe , bes , through contract de - fg02 - 05er46202 . thanks the university of iowa for support and acknowledges support from the israel science foundation ( grant no . 1287/15 ) . v.k.t . acknowledges support from nsf through award dmr-1104884 . research at rutgers was supported by the us department of energy , office of basic energy sciences through award de - sc0005463 . symmetry channel . ( a ) raman scattering intensity ( black solid line ) shown together with a luminescence background @xmath41 ( red dashed line ) to be subtracted . ( b ) raman susceptibility after background subtraction and conversion to @xmath46 . ( see text ) . ] the raman spectra were corrected for the spectral response of the spectrometer and detector in obtaining the raman scattering intensity , @xmath40 . here , @xmath41 is a small luminescence background and @xmath42 and @xmath43 the polarization vectors for the incident and scattered photons for a given scattering geometry with respect to the unit cell ( fig . [ fig : feas](b ) ) . the recorded raman intensity was background subtracted with a near - linear line and a constant determined for each polarization geometry as illustrated in fig . [ fig : bg ] . the goal of this section is to show microscopically that the photons in the @xmath45 configuration are coupled to the local orbital fluctuations shown in fig . [ fig : feas](f ) . as the orbital character of raman driven excitations play a central role in our analysis we derive this coupling explicitly . apart from the transition between @xmath8 and @xmath9 orbitals the @xmath26 orbital excitations are accessible in the @xmath45 configuration . the raman response , eq . [ response ] , is determined by the raman operators @xmath140 discussed in details below . we point out that the standard effective mass approximation @xcite is applicable only to pockets derived from a single non - degenerate band with well defined orbital content . this is the case for the @xmath101-pocket derived predominantly from @xmath26 orbitals , see figs . [ fig : feas](c , d , e ) . within the effective mass approximation , however to the extent that this pocket is approximately circular the @xmath45 coupling to @xmath101 pocket is relatively weak , and we focus on the other two electron and hole pockets at the @xmath141 and at the @xmath114 points , respectively . consider the raman coupling to the @xmath142 hole pockets first . it is convenient to use the symmetry constrained @xmath143 luttinger hamiltonian @xcite , @xmath144 here the parameters @xmath145 , @xmath146 , @xmath147 and @xmath148 are determined by a fit to the five - band tight - binding model or by first principle calculations which are tabulated in ref . for selected iron superconductors . we set @xmath149 which corresponds to circular hole fss . at the @xmath114-point , @xmath150 , the two bloch states are degenerate . these states are characterized by well defined orbital content , and we denote the creation operators of these states by @xmath151 . in terms of these operators the hamiltonian , takes the form , @xmath152 . the representation is referred to as orbital to be contrasted with the band representation obtained by diagonalization of . the raman coupling to the @xmath142 pockets is a matrix in orbital space @xcite , @xmath153 in the single - band approximation the orbital indices are redundant , and the more familiar effective mass approximation results . substitution of eq . in eq . gives for the raman vertex in @xmath45 geometry @xmath154 therefore , the physical meaning of the @xmath45 raman probe is the quadrupole excitations causing orbital population imbalance as illustrated in fig . [ fig : feas](f ) . to understand the implications of the @xmath45 raman probe in the band representation one diagonalizes the hamiltonian which yields @xmath102 and @xmath103 bands with bloch states created by the operators @xmath155 and @xmath156 , respectively . in the band representation the raman vertex takes the form @xmath157 where @xmath158 is the angle formed by the vector @xmath159 and the @xmath160-direction in the bz . the first intra - band contribution in describes the out - of - phase breathing of the @xmath102 and @xmath103 bands with the amplitude changing as @xmath161 , as shown in fig . [ fig : feas](f ) . the nodes of the intraband @xmath45 excitation are along @xmath162 and @xmath163 as expected . we conclude , that the raman response in @xmath45 symmetry couples directly to the pomeranchuk fluctuations of the fs . the second , inter - band part of the coupling plays a role in temperature and frequency dependence of the @xmath45 response . the electron pockets coupling to photons can be analyzed along the same lines as is done above for holes using the same effective hamiltonian approach . instead of eq . we have for the electron pockets @xcite , @xmath164 where the upper and lower signs refer to the two electron pockets , @xmath165 , @xmath166 , @xmath167 , @xmath168 , @xmath169 are parameters to be fixed by matching to the band structure calculations . the matrix acts for the @xmath170 signs acts in the space of bloch states that have @xmath171 , @xmath71 ( @xmath172 , @xmath71 ) orbital content . again , the electron equivalent of eq . tells us that the @xmath45 coupling excites the @xmath173 phase shifted breathing of the two electron pockets . we obtain for an intra - band contribution @xmath174 eq . shows that photons in the @xmath45 configuration cause the two electron pockets to breathe with a phase difference of @xmath173 . the shape of the raman response @xmath46 with the relaxational mode ( rm ) and the emergent critical behavior above @xmath175 can be described by an expression for interacting susceptibilities given by , @xmath176 here , @xmath177 is the coupling of light to the quadrupole charge density fluctuations in @xmath121 symmetry ( mainly to the @xmath103 band ) , @xmath178 is the non - interacting susceptibility and @xmath131 is the coupling constant . in the @xmath45 symmetry channel . @xmath46 with the three component fit for x=0 in a range of temperatures above the structural transition showing intensity versus frequency and temperature . ] raman spectroscopy as a dynamic probe is well - suited to account for the relaxation processes that determine @xmath179''$ ] . to this end we assume that @xmath179''$ ] is controlled by a single energy scale , @xmath180 . in addition we assume it to saturate at large frequencies . since at low frequencies @xmath179 '' \propto \omega $ ] we model it as @xmath179 '' = c \mbox{arctan}(\omega/\gamma_t)$ ] , where @xmath181 is a constant . correspondingly , causality yields @xmath182 where @xmath183 is a high - frequency cutoff . the scale @xmath180 contains contributions from a few elastic and inelastic processes listed below . while for zero momentum the intraband processes are forbidden @xcite this is not so in the present case with finite optical penetration depth . in addition , the interband transitions at high temperatures make a contribution to @xmath179''$ ] that scales as @xmath184 . recently , the scattering mechanism that involves both the disorder and long wavelength critical quadrupole fluctuations was shown to give rise to a nearly @xmath10-linear scattering rate @xcite . this contribution is expected to grow with doping . we also note aslamazov - larkin corrections due to fluctuations at momentum @xmath185=(@xmath186,@xmath186 ) @xcite that are quite different from the quadrupolar fluctuations which also contribute to the scattering rate , @xmath187@xmath188 since the velocities of electrons and holes are antiparallel . unlike the above - mentioned contribution , the aslamazov - larkin contribution weakens with doping as the deviation from perfect nesting suppresses the coupling of fluctuations at @xmath185 to the zero momentum quadrupolar fluctuations @xcite . we also note that the elastic scattering off the disorder yields a constant @xmath10-independent contribution to the scattering rate @xcite which grows with doping . all of the above scattering processes contribute to @xmath179''$ ] . our results of the qep scaling are nevertheless universal because the exact temperature dependence of @xmath180 at bare level without the effects of quadrupole attraction is not essential . essential is that as the relaxation rate in the renormalized raman susceptibility eq . is reduced compared to the bare value @xmath180 therefore , the reduction in the observed width of @xmath189''$ ] reflects the tendency to order at the pomeranchuk instability . for @xmath190 , substitution of eq . [ choice_1 ] in eq . [ rpa ] yields the following relaxational expression for the rm of the @xmath191 susceptibility , @xmath192 where @xmath193 $ ] and @xmath194=@xmath195 . [ rpa ] ensures the critical behavior with temperature above @xmath23 and @xmath83 of the static susceptibility where @xmath196 . here @xmath4 is defined by @xmath131 in terms of @xmath197 . our basic assumption of attraction in the @xmath115 @xmath129-@xmath130 channel follows from the critical enhancement of @xmath80 and implies @xmath131@xmath1320 . for higher dopings , the electron and hole fss uncouple and our assumption of attraction in the @xmath115 @xmath129-@xmath130 channel eventually breaks down . hence , the criticality persists but weakens with doping . we use expression [ raman 1 ] with a simultaneous fit of the rm as a function of frequency , temperature and doping dependence . assuming a weak temperature dependence of the scattering rate we use the expansion @xmath198 . here @xmath48 and @xmath10 are fitting variables and @xmath199 , @xmath95 and @xmath102 are fitting parameters . figure [ fig : rmfit2](f ) shows a universal fit to @xmath200 in a range of temperatures above @xmath5 versus frequency and temperature . above the @xmath23 and @xmath83 lines , @xmath46 can be decomposed into three components ( fig . [ fig : rmfit](a ) ) . both the intensity of the continuum and of the @xmath37240 @xmath50 mode diminishes rapidly with doping , and vanishes near [email protected] ( figs . [ fig : rmfit](c , d ) ) . importantly , the orbital content of the larger @xmath101 fs is mainly composed of @xmath26 orbitals , while the @xmath102 and @xmath103 fss primarily have @xmath8 and @xmath9 orbital character ( fig . [ fig : feas](c ) ) @xcite . at the m point , the inner ( outer ) part of the @xmath17/@xmath104 fs has @xmath8 and @xmath9 ( @xmath26 ) orbital character . the continuum and the @xmath37240 @xmath50 mode likely involve the @xmath103 band as its fs reduces with doping ( see figs . [ fig : feas](b , d ) ) with the former due to intraband excitations and the latter due to an interband - like excitation with a 240 @xmath50 gap consistent with quadrupole excitations as verified by scaling of @xmath44 to nqr data ( see section iv ) . this finding is consistent with first - principle calculations taking into account spin - orbit coupling ( fig . [ fig : feas](e ) ) . figure [ fig : rmfit](c ) displays the intensity dependence of the rm with doping which is seen to persist for all dopings . figure [ fig : rmfit](e ) shows the doping dependence of @xmath4 which is observed to decrease close - to linear for increasing dopings becoming negative near x=0.022 . this behavior is consistent with that obtained from the analysis of the static raman susceptibility @xmath81 shown in the @xmath10@xmath11@xmath105 phase diagram ( fig . [ fig : phasediagram](a ) ) . the first - principles calculations use a combination of density functional theory and dynamical mean field theory ( dft+dmft ) @xcite as in ref . . it is based on the full - potential linear augmented plane wave method implemented in wien2k @xcite for carrying out first - principle calculations . the electronic charge is computed self - consistently in the dft+dmft density matrix . the continuous time quantum monte carlo method @xcite was used to solve the quantum impurity problem using the coulomb repulsion in its fully rotational form . we used the experimentally determined lattice structure for nafeas with the lattice constants @xmath201 , @xmath202 , and atomic positions @xmath203 for fe , @xmath204 for as , @xmath205 for na @xcite . the calculations were done in the paramagnetic state with coulomb interactions @xmath206=5.0 ev and @xmath207=0.8 ev at a temperature of @xmath10=116 k. in @xmath121 symmetry , modulations of the fss around the @xmath114- and m - point with nodes along @xmath114x and @xmath114y lead to a neutral quadrupole charge density mode with either @xmath116@xmath11 or @xmath115@xmath11symmetry of the bz , fig . [ fig : sm_qp](a ) . the critical charge fluctuations above the structural transition @xmath5 originate from local electron - hole excitations with charge transfer between the @xmath8 and @xmath9 orbitals on the fe - sites which introduces a quadrupole moment in @xmath45 or @xmath121 symmetry . these local @xmath8 and @xmath9 charge transfer processes are the primary excitations sustaining the quadrupole pattern . secondly , the phasing between the @xmath114- and m - point is dictated by interband interactions across the fermi level at the @xmath114 and m - point as illustrated in fig . [ fig : sm_qp](a ) . the fss elongate and squeeze along @xmath114 m and @xmath114m into quadrants ( @xmath116 ) or half quadrants ( @xmath115 ) defined by the nodes . the quadrupole mode is neutral where the charge at each of the @xmath114- and m - point are conserved as well as the overall charge of all participating fss . hence , the quadrupole mode results from deformations of the fss in which charge redistribution by intra and interband transitions causes a quadrupole pattern of changing positive and negative half - quadrant regions of more or less charge . out - of - phase charge modulation yields @xmath115@xmath11symmetry where the fss at the @xmath114- and m - point elongate and squeeze in - phase , while in - phase charge modulation corresponds to @xmath116-symmetry . simultaneous @xmath208 and @xmath209 two - electron - hole exchange interactions will promote @xmath115@xmath11symmetry . the in - phase breathing mode predicted by chubukov _ @xcite and later verified by klein _ et al . _ @xcite in @xmath54 symmetry is illustrated in fig . [ fig : sm_qp](b ) . this in - phase breathing mode is a particle - hole exciton which forms in @xmath54 symmetry and is consistent with an @xmath24 condensate . it is represented by in - phase breathing of the electron and hole fss and entails charge transfer between the two pockets .

the cloud radio access network ( c - ran ) concept , in which densely deployed access points ( aps ) are empowered by cloud computing to cooperatively support mobile users ( mus ) , to improve mobile data rates , has been recently proposed . however , the high density of active aps results in severe interference and also inefficient energy consumption . moreover , the growing popularity of highly interactive applications with stringent uplink ( ul ) requirements , e.g. network gaming and real - time broadcasting by wireless users , means that the ul transmission is becoming more crucial and requires special attention . therefore in this paper , we propose a joint downlink ( dl ) and ul mu - ap association and beamforming design to coordinate interference in the c - ran for energy minimization , a problem which is shown to be np hard . due to the new consideration of ul transmission , it is shown that the two state - of - the - art approaches for finding computationally efficient solutions of joint mu - ap association and beamforming considering only the dl , i.e. , group - sparse optimization and relaxed - integer programming , can not be modified in a straightforward way to solve our problem . leveraging on the celebrated ul - dl duality result , we show that by establishing a virtual dl transmission for the original ul transmission , the joint dl and ul optimization problem can be converted to an equivalent dl problem in c - ran with two inter - related subproblems for the original and virtual dl transmissions , respectively . based on this transformation , two efficient algorithms for joint dl and ul mu - ap association and beamforming design are proposed , whose performances are evaluated and compared with other benchmarking schemes through extensive simulations . cloud radio access network , green communication , uplink - downlink duality , group - sparse optimization , relaxed - integer programming , beamforming . [ section ] [ section ] [ section ] [ section ] [ section ] [ section ] [ section ]

1402.4238

to meet the fast growing mobile data volume driven by applications such as smartphones and tablets , the traditional wireless network architecture based on a single layer of macro - cells has shifted to one composed of smaller cells such as pico / femto cells with more densely deployed access points ( aps ) . therefore , cloud radio access network ( c - ran ) @xcite has recently been proposed and drawn a great deal of attention . in a c - ran , the distributed aps , also termed remote radio heads ( rrhs ) , are connected to the baseband unit ( bbu ) pool through high bandwidth backhaul links , e.g. optical transport network @xcite , to enable centralized processing , collaborative transmission , and real - time cloud computing . as a result , significant rate improvement can be achieved due to reduced pathloss along with joint scheduling and signal processing . however , with densely deployed aps , several new challenges arise in c - ran . first , close proximity of many active aps results in increased interference , and hence the transmit power of aps and/or mobile users ( mus ) needs to be increased to meet any given quality of service ( qos ) . second , the amount of energy consumed by a large number of active aps @xcite as well as by the transport network to support high - capacity connections with the bbu pool @xcite will also become considerable . such facts motivate us to optimize the energy consumption in c - ran , which is the primary concern of this paper . in particular , both downlink ( dl ) and uplink ( ul ) transmissions are considered jointly . the studied c - ran model consists of densely deployed aps jointly serving a set of distributed mus , where comp based joint transmit / receive processing ( beamforming ) over all active aps is employed for dl / ul transmissions . under this setup , we study a joint dl and ul mu - ap association and beamforming design problem to minimize the total energy consumption in the network subject to mus given dl and ul qos requirements . the energy saving is achieved by optimally assigning mus to be served by the minimal subset of active aps , finding the power levels to transmit at all mus and aps , and finding the beamforming vectors to use at the multi - antenna aps . this problem has not been investigated to date , and the closest prior studies are @xcite . however , the prior studies have all considered mu association and/or active ap selection problems for various objectives from the dl perspective . in particular , the problems studied in @xcite can be treated as the dl - only version of our considered joint dl and ul problem , in which the transmit beamforming vectors and the active set of aps are jointly optimized to minimize the power consumption at all the aps . note that the mu association and/or active ap selection based on dl only may result in inefficient transmit power of mus or even their infeasible transmit power in the ul considering various possible asymmetries between the dl and ul in terms of channel , traffic and hardware limitation . furthermore , with users increasingly using applications with high - bandwidth ul requirements , ul transmission is becoming more important . for example , the upload speed required for full high definition ( hd ) @xmath0p skype video calling is about @xmath1 mbps @xcite . therefore , we need to account for both dl and ul transmissions while designing the mus association and active ap selection scheme . the ul - only mu association problem has also been considered extensively in the literature @xcite ; however , their solutions are not applicable in the context of this work due to their assumption of one - to - one mu - ap association . it is worth noting that the joint mus association and active ap selection is mathematically analogous to the problem of antenna selection in large multiple - input multiple - output ( mimo ) systems @xcite , which aims to reduce the number of radio transmission chains and hence the energy consumption and signal processing overhead . the connection between these two problems can be recognized by treating the c - ran as a distributed large mimo system . in terms of other related work , there have been many attempts to optimize the energy consumption in cellular networks , but only over a single dimension each time , e.g. power control @xcite , ap `` on / off '' control @xcite , and coordinated multi - point ( comp ) transmission @xcite . to avoid an infeasible power allocation , it was suggested in @xcite to gradually remove the mus that can not be supported due to their limited transmit power budgets . in addition to achieving energy saving from mus perspective , @xcite proposed to switch off the aps that are under light load to save energy by exploiting the fact that the traffic load in cellular networks fluctuates substantially over both space and time due to user mobility and traffic burstiness . cooperation among different cells or aps could be another possible way to mitigate the interference and achieve energy - efficient communication . for example , if a certain cluster of aps can jointly support all the mus , the intercell interference can be further reduced especially for the cell - edge mus @xcite . a judicious combination of these techniques should provide the best solution , and this is the direction of our work . unfortunately , the considered joint dl and ul mu - ap association and beamforming design problem in this paper involves integer programming and is np hard as shown for a similar problem in ( * ? ? ? * theorem 1 ) . to tackle this difficulty , two different approaches , i.e. , group - sparse optimization ( gso ) and relaxed - integer programming ( rip ) , have been adopted in @xcite and @xcite , respectively , to solve a similar dl - only problem , where two polynomial - time algorithms were proposed and shown to achieve good performance through simulations . in particular , the gso approach is motivated by the fact that in the c - ran with densely deployed aps , only a small fraction of the total number of aps needs to be active for meeting all mus qos . however , due to the new consideration of ul transmission in this paper , we will show that the algorithms proposed in @xcite can not be applied directly to solve our problem , and therefore the methods derived in this paper are important advances in this field . the contributions of this paper are summarized as follows : 1 . to optimize the energy consumption tradeoffs between the active aps and mus , we jointly study the dl and ul mu - ap association and beamforming design by solving a weighted sum - power minimization problem . to our best knowledge , this paper is the first attempt to unify the dl and ul mu - ap association and beamforming design into one general framework . 2 . due to a critical scaling issue in the ul receive beamforming design , the gso based algorithm @xcite and the rip based algorithm @xcite can not be applied to solve our joint dl and ul design directly . to address this issue , we establish a virtual dl transmission for the original ul transmission in c - ran by first ignoring the individual ( per - ap and per - mu ) power constraints based on the celebrated ul - dl duality result @xcite . consequently , the considered joint dl and ul problem without individual power constraints can be transformed into an equivalent dl problem with two inter - related subproblems corresponding to the original and virtual dl transmissions , respectively . with the equivalent dl - only formulation , we extend the gso based and rip based algorithms to solve the relaxed joint dl and ul optimization problem . 3 . considering the fact that the optimal solution to the ul sum - power minimization is component - wise minimum , there is no tradeoff among different mus in terms of power minimization in the ul . consequently , we are not able to establish the duality result for the per - mu power constraints in the ul , which are thus difficult to incorporate into our developed algorithm . to resolve this issue , we propose a price based iterative method to further optimize the set of active aps while satisfying the per - mu power constraints . finally , we verify the effectiveness of our proposed algorithms by extensive simulations from three perspectives : ensuring feasibility for both dl and ul transmissions ; achieving optimal network energy saving with mu - ap association ; and flexibly adjusting various power consumption tradeoffs between active aps and mus . it is worth pointing out that as the baseband processing is migrated to a central unit , i.e. , bbu pool , the data exchanged between the aps and the bbu pool includes oversampled real - time digital signals with very high bit rates ( in the order of gbps ) . as a result , the capacity requirement for the backhaul / fronthaul links becomes far more stringent in the c - ran . given finite backhaul capacity , the optimal strategy for backhaul compression and quantization has been studied recently in e.g. @xcite . in this paper , however , we focus on addressing the energy consumption ( including both transmission and non - transmission related portions ) issue in the c - ran , which is also one of the major concerns for future cellular networks , by assuming that the backhaul transport network is provisioned with sufficiently large capacity . note that the optical network has been widely accepted as a good option to implement the high - bandwidth backhaul transport network @xcite . the rest of this paper is organized as follows . section [ sec : system model ] introduces the c - ran model , and the power consumption models for the aps and mus . section [ sec : problem formulation ] presents our problem formulation , introduces the two existing approaches , namely gso and rip , and explain the new challenges in solving the joint dl and ul optimization . section [ sec : joint ul dl ] presents our proposed algorithms based on the virtual dl representation of the ul transmission . section [ sec : numerical ] shows numerical results . finally , section [ sec : conclusion ] concludes the paper . _ notations _ : boldface letters refer to vectors ( lower case ) or matrices ( upper case ) . for an arbitrary - size matrix @xmath2 , @xmath3 , @xmath4 , and @xmath5 denote the complex conjugate , conjugate transpose and transpose of @xmath2 , respectively . the distribution of a circularly symmetric complex gaussian ( cscg ) random vector with mean vector @xmath6 and covariance matrix @xmath7 is denoted by @xmath8 ; and @xmath9 stands for `` distributed as '' . @xmath10 denotes the space of @xmath11 complex matrices . @xmath12 denotes the euclidean norm of a complex vector @xmath6 , and @xmath13 denotes the magnitude of a complex number @xmath14 . we consider a densely deployed c - ran @xcite consisting of @xmath15 access points ( aps ) , denoted by the set @xmath16 . the set of distributed aps jointly support @xmath17 randomly located mobile users ( mus ) , denoted by the set @xmath18 , for both downlink ( dl ) and uplink ( ul ) communications . in this paper , for the purpose of exposition , we consider linear precoding and decoding in the dl and ul , respectively , which is jointly designed at the bbu pool assuming the perfect channel knowledge for all mus . the results in this paper can be readily extended to the case of more complex successive precoding / decoding , e.g. dirty - paper coding ( dpc ) @xcite and multiuser detection with successive interference cancelation ( sic ) @xcite , with fixed coding orders among the users . we also assume that each ap @xmath19 , @xmath20 , is equipped with @xmath21 antennas , and all mus are each equipped with one antenna . it is further assumed that there exist ideal low - latency backhaul transport links with sufficiently large capacity ( e.g. optical fiber ) connecting the set of aps to the bbu pool , which performs all the baseband signal processing and transmission scheduling for all aps . the centralized architecture results in efficient coordination of the transmission / reception among all the aps , which can also be opportunistically utilized depending on the traffic demand . we consider a quasi - static fading environment , and denote the channel vector in the dl from ap @xmath19 to mu @xmath22 and that in the ul from mu @xmath22 to ap @xmath19 as @xmath23 and @xmath24 , respectively . let the vector consisting of the channels from all the aps to mu @xmath22 and that consisting of the channels from mu @xmath22 to all the aps be @xmath25 $ ] and @xmath26^{t}$ ] , respectively . there are two main techniques for separating dl and ul transmissions on the same physical transmission medium , i.e. , time - division duplex ( tdd ) and frequency - division duplex ( fdd ) . if tdd is assumed , channel reciprocity is generally assumed to hold between dl and ul transmissions , which means that the channel vector @xmath27 in the ul is merely the transpose of that @xmath28 in the dl , i.e. , @xmath29 . however , if fdd is assumed , @xmath30 s and @xmath27 s are different in general . in dl transmission , the transmitted signal from all aps can be generally expressed as @xmath31 where @xmath32 is the beamforming vector for all aps to cooperatively send one single stream of data signal @xmath33 to mu @xmath22 , which is assumed to be a complex random variable with zero mean and unit variance . note that @xmath34 . then , the transmitted signal from ap @xmath19 can be expressed as @xmath35 where @xmath36 is the @xmath19th block component of @xmath37 , corresponding to the transmit beamforming vector at ap @xmath19 for mu @xmath22 . note that @xmath38^{t}$ ] and @xmath39^{t}$ ] , @xmath40 . from ( [ eq : dl transmit signal from ap n ] ) , the transmit power of ap @xmath19 in dl is obtained as @xmath41 we assume that there exists a maximum transmit power constraint for each ap @xmath19 , i.e. , @xmath42 the received signal at the @xmath22th mu is then expressed as @xmath43 where @xmath44 is the receiver noise at mu @xmath22 , which is assumed to be a circularly symmetric complex gaussian ( cscg ) random variable with zero mean and variance @xmath45 , denoted by @xmath46 . treating the interference as noise , the signal - to - interference - plus - noise ratio ( sinr ) in dl for mu @xmath22 is given by @xmath47 in ul transmission , the transmitted signal from mu @xmath22 is given by @xmath48 where @xmath49 denotes the transmit power of mu @xmath22 , and @xmath50 is the information bearing signal which is assumed to be a complex random variable with zero mean and unit variance . with @xmath51 denoting the transmit power limit for mt @xmath22 , it follows that @xmath52 the received signal at all aps is then expressed as @xmath53 where @xmath54 denotes the receiver noise vector at all aps consisting of independent cscg random variables each distributed as @xmath55 . let @xmath56 denote the receiver beamforming vector used to decode @xmath57 from mu @xmath22 . then the sinr in ul for mu @xmath22 after applying @xmath58 is given by @xmath59 let @xmath60 denote the @xmath19th block component in @xmath58 , corresponding to the receive beamforming vector at ap @xmath19 for mu @xmath22 . we thus have @xmath61^{t}$ ] , @xmath40 . the total energy consumption in the c - ran comprises of the energy consumed by all aps and all mus . from ( [ eq : dl transmit signal from all aps ] ) and ( [ eq : ul transmit signal from mu i ] ) , the total transmit power of all aps in dl and that of all mus in ul can be expressed as @xmath62 and @xmath63 , respectively . besides the static power consumption at each ap @xmath19 due to e.g. real - time a / d and d / a processing , denoted as @xmath64 , in c - ran with centralized processing , the extensive use of high - capacity backhaul links to connect all aps with the bbu pool makes the power consumption of the transport network no more negligible @xcite . for example , consider the passive optical network ( pon ) to implement the backhaul transport network @xcite . the pon assigns an optical line terminal ( olt ) to connect to a set of associated optical network units ( onus ) , which coordinate the set of transport links connecting all the aps to the bbu pool , each through a single fiber . for simplicity , the resulting power consumption in the pon can be modeled as @xcite @xmath65 where @xmath66 and @xmath67 are both constant and denote the power consumed by the olt and the transport link associated with ap @xmath19 , respectively . moreover , we consider that for energy saving , some aps and their associated transport links can be switched into sleep mode @xcite ( compared with active mode ) with negligible power consumption ; thus , the total static power consumption of ap @xmath19 , denoted by @xmath68 , @xmath20 , can be saved if ap @xmath19 and its associated transport link are switched into sleep mode for both transmission in dl and ul . for convenience , we express the total static power consumption of all active aps as @xmath69 where @xmath70 , @xmath20 , is an indicator function for ap @xmath19 , which is defined as @xmath71 note that in practical pon systems , the olt in general can not be switched into sleep mode as it plays the role of distributor , arbitrator , and aggregator of the transport network , which has a typical fixed power consumption of @xmath72w @xcite . we thus ignore @xmath66 since it is only a constant . from ( [ eq : on off condition ] ) , mu @xmath22 is associated with an active ap @xmath19 if its corresponding transmit and/or receive beamforming vector at ap @xmath19 is nonzero , i.e. , @xmath73 and/or @xmath74 . under this setup , it is worth pointing out that each mu @xmath22 is allowed to connect with two different sets of aps for dl and ul transmissions , respectively , e.g. @xmath73 but @xmath75 for some @xmath20 , which is promising to be implemented in next generation cellular networks @xcite . furthermore , from ( [ eq : on off condition ] ) , ap @xmath19 could be switched into sleep mode only if it does not serve any mu . we aim to minimize the total energy consumption in the c - ran , including that due to transmit power of all mus ( but ignoring any static power consumption of mu terminals ) as well as that due to transmit power and static power of all active aps . therefore , we consider the following weighted sum - power as our design metric : @xmath76 where @xmath77 is a weight to trade off between the total energy consumptions between all the active aps and all mus . to minimize the weighted power consumption in ( [ eq : weighted sum power ] ) , we jointly optimize the dl and ul mu - ap association and transmit / receive beamforming by considering the following problem . @xmath78 where @xmath79 and @xmath80 are the given sinr requirements of mu @xmath22 for the dl and ul transmissions , respectively . in the rest of this paper , the constraints in ( [ eq : uldl dl power constraint ] ) and ( [ eq : uldl ul power constraint ] ) are termed per - ap and per - mu power constraints , respectively . problem ( p1 ) can be shown to be non - convex due to the implicit integer programming involved due to indicator function @xmath70 s in the objective . prior to solving problem ( p1 ) , we first need to check its feasibility . since the dl and ul transmissions are coupled only by the objective function in ( [ eq : uldl objective function ] ) , the feasibility of problem ( p1 ) can be checked by considering two separate feasibility problems : one for the dl and the other for the ul , which have both been well studied in the literature @xcite and thus the details are omitted here for brevity . for the rest of this paper , we assume that problem ( p1 ) is always feasible if all aps are active . as mentioned in section [ sec : introduction ] , the problem of joint mu - ap association and transmit beamforming subject to mus qos and per - ap power constraints for power minimization in the dl - only transmission has been recently studied in @xcite using the approaches of gso and rip , respectively , where two different polynomial - time algorithms were proposed and shown to both achieve good performance by simulations . in contrast , problem ( p1 ) in this paper considers both dl and ul transmissions to address possible asymmetries between the dl and ul in terms of channel realization , traffic load and hardware limitation . furthermore , considering that mus are usually powered by finite - capacity batteries as compared to aps that are in general powered by the electricity grid , we study the power consumption tradeoffs between aps and mus by minimizing the weighted sum - power @xmath81 in ( p1 ) . therefore , the problems considered in @xcite can be treated as special cases of ( p1 ) . in the following , we show that due to the new consideration of ul transmission , the algorithms proposed in @xcite based on gso and rip for solving the dl optimization can not be applied directly to solve ( p1 ) , which thus motivates us to find a new method to resolve this issue in section [ sec : joint ul dl ] . given the fact that the static power , i.e. , @xmath82 , is in practice significantly larger than the transmit power at each ap @xmath19 , to minimize the total network energy consumption @xcite , it is conceivable that for the optimal solution of ( p1 ) only a subset of @xmath15 aps should be active . as a result , a `` group - sparse '' property can be inferred from the following concatenated beamforming vector : @xmath83,\cdots,[\hat{\mathbf{w}}^{\text{dl}}_n,\hat{\mathbf{v}}^{\text{ul}}_n]\right]\end{aligned}\ ] ] in which @xmath84 $ ] and @xmath85 $ ] , @xmath86 , i.e. , the beamforming vectors are grouped according to their associated aps . if ap @xmath19 is in the sleep mode , its corresponding block @xmath87 $ ] in ( [ eq : group sparsity ] ) needs to be zero . consequently , the fact that a small subset of deployed aps is selected to be active implies that the concatenated beamforming vector in ( [ eq : group sparsity ] ) should contain only a very few non - zero block components . one well - known approach to enforce desired group sparsity in the obtained solutions for optimization problems is by adding to the objective function an appropriate penalty term . the widely used group sparsity enforcing penalty function , which was first introduced in the context of the group least - absolute selection and shrinkage operator ( lasso ) problem @xcite , is the mixed @xmath88 norm . in our case , such a penalty is expressed as @xmath89\right\|.\end{aligned}\ ] ] the @xmath88 norm in ( [ eq : uldl penalty ] ) , similar to @xmath90 norm , offers the closest convex approximation to the @xmath91 norm over the vector consisting of @xmath92 norms @xmath93\right\|\right\}^{n}_{n=1}$ ] , implying that each @xmath94\right\|$ ] is desired to be set to zero to obtain group sparsity . more generally , the mixed @xmath95 norm has also been shown to be able to recover group sparsity with @xmath96 @xcite , among which the @xmath97 norm , defined as @xmath98 has been widely used @xcite . compared with @xmath88 norm , @xmath97 norm has the potential to obtain more sparse solution but may lead to undesired solution with components of equal magnitude . in this paper , we focus on the @xmath88 norm in ( [ eq : uldl penalty ] ) for our study . we will compare the performance of @xmath88 and @xmath97 norms by simulations in section [ sec : numerical ] . according to @xcite , at first glance it seems that using the @xmath88 norm , problem ( p1 ) can be approximately solved by replacing the objective function with @xmath99 where @xmath100 can be treated as a convex relaxation of the indicator functions in ( [ eq : weighted sum power ] ) , and @xmath101 indicates the relative importance of the penalty term associated with ap @xmath19 . however , problem ( p1 ) with ( [ eq : scaling problem objective ] ) as the objective function is still non - convex due to the constraints in ( [ eq : uldl dl sinr constraint ] ) and ( [ eq : uldl ul sinr constraint ] ) . furthermore , since the ul receive beamforming vector @xmath58 s can be scaled down to be arbitrarily small without affecting the ul sinr defined in ( [ eq : ul sinr ] ) , minimizing ( [ eq : scaling problem objective ] ) directly will result in all @xmath58 s going to zero . to be more specific , let @xmath102 and @xmath103 denote the optimal solution of problem ( p1 ) with ( [ eq : scaling problem objective ] ) as the objective function . then , it follows that @xmath104 and @xmath105 , preserves the `` group - sparse '' property where the non - zero block components correspond to the active aps . two issues thus arise : first , the ul does not contribute to the selection of active aps ; second , the set of selected active aps based on the dl only can not guarantee the qos requirements for the ul . as a result , the @xmath88 norm penalty term in ( [ eq : scaling problem objective ] ) or more generally the @xmath95 norm penalty does not work for the joint dl and ul ap selection in our problem , and hence the algorithm proposed in @xcite , which involves only the dl transmit beamforming vector @xmath37 s , can not be modified in a straightforward way to solve our problem . next , we reformulate problem ( p1 ) by introducing a set of binary variable @xmath106 s indicating the `` active / sleep '' state of each ap as follows . @xmath107 where @xmath108 is a constant with arbitrary value . note that the active - sleep constraints in ( [ eq : bigm onoff ] ) are inspired by the well - known big-@xmath109 method @xcite : if @xmath110 , the constraint ( [ eq : bigm onoff ] ) ensures that @xmath111 ; if @xmath112 , the constraint has no effect on @xmath113 and @xmath114 , as @xmath115 represents an upper bound on the term @xmath116 . notice that @xmath117 can be chosen arbitrarily due to the scaling invariant property of ul receive beamforming vector @xmath58 s . with the active - sleep constraints in ( [ eq : bigm onoff ] ) , the equivalence between problems ( p1 ) and ( p2 ) can be easily verified . in @xcite , a similar problem to ( p2 ) was studied corresponding to the case with only dl transmission . for problem ( p2 ) without @xmath58 and @xmath118 and their corresponding constraints , the problem can be transformed to a convex second - order cone programming ( socp ) by relaxing the binary variable @xmath106 as @xmath119 , \forall n \in \mathcal{n}$ ] . under this convex relaxation , a bnc algorithm , which is a combination of the branch - and - bound ( bnb ) and the cutting plane ( cp ) methods @xcite , was proposed in @xcite to solve the dl problem optimally . however , the computational complexity of bnb is prohibitive for large networks in practice , which grows exponentially with the number of aps . to obtain polynomial - time algorithm with near - optimal performance , in @xcite , the authors further proposed an incentive measure based heuristic algorithm to determine the set of active aps . the incentive measure reflects the importance of each ap to the whole network and is defined as the ratio of the total power received at all mus to the total power expended for each ap . however , with both dl and ul transmissions , it is observed that problem ( p2 ) can no longer be transformed to a convex form by relaxing @xmath106 s as continuous variables due to the constraints in ( [ eq : uldl ul sinr constraint ] ) . furthermore , because of the scaling invariant property of ul receive beamforming vectors , solving the relaxed problem of ( p2 ) will result in all @xmath58 s going to zero , similar to the case of gso based solution . particularly , the value of the relaxed indicator @xmath120 , will not be related to @xmath58 s , which in fact contributes to the penalty incurred in the objective due to the static power of ap @xmath19 , i.e. , @xmath121 . finally , it is nontrivial to find an incentive measure that reflects the importance of each ap to both dl and ul transmissions . in this section , we provide two efficient algorithms to approximately solve problem ( p1 ) based on the gso and rip approaches , respectively . first , we consider the approach of gso and present a new method to address the joint dl and ul optimization . to obtain an efficient solution for problem ( p1 ) , we first assume that all the aps and mus have infinite power budget , i.e. , @xmath122 , @xmath123 and @xmath124 . the resulting problem is termed ( p1 - 1 ) . an equivalent reformulation of problem ( p1 - 1 ) is then provided to overcome the receive beamforming scaling issue mentioned in section [ sec : problem formulation ] . then , we discuss the challenges of dealing with finite per - ap and per - mu power constraints and provide efficient methods to handle them . first , we consider the following transmit sum - power minimization problem in the ul : @xmath125 from @xcite , it follows that problem ( [ eq : ul power minimization ] ) can be solved in a virtual dl channel as @xmath126 where @xmath127 is the virtual dl transmit beamforming vector over @xmath15 aps for mu @xmath22 . denote @xmath128 , @xmath129 and @xmath130 , @xmath131 as the optimal solutions to problems ( [ eq : ul power minimization ] ) and ( [ eq : virtual dl power minimizatin ] ) , respectively . then from @xcite it follows that @xmath128 and @xmath130 can be set to be identical , @xmath132 , and furthermore @xmath133 . by establishing a virtual dl transmission for the ul transmission based on the above ul - dl duality , we have the following lemma . [ lemma:1 ] problem ( p1 - 1 ) is equivalent to the following problem . @xmath134 for any given feasible solution to problem ( p3 ) , we can always find a corresponding feasible solution to problem ( p1 - 1 ) achieving the same objective value as that of problem ( p3 ) , and vice versa , similar as ( * ? ? ? * proposition 1 ) ; thus , problems ( p1 - 1 ) and ( p3 ) achieve the same optimal value with the same set of optimal dl / ul beamforming vectors . lemma [ lemma:1 ] is thus proved . since problem ( p3 ) is a dl - only problem that has the same `` group - sparse '' property as ( p1 - 1 ) , it can be approximately solved by replacing the objective function with @xmath135 comparing ( [ eq : reformulate problem objective ] ) and ( [ eq : scaling problem objective ] ) , we have successfully solved the scaling issue of ul receive beamforming vector , @xmath58 s , by replacing them with the equivalent dl transmit beamforming vector , @xmath136 s , since from ( [ eq : virtual dl power minimizatin ] ) it follows that the virtual dl sinr of each mu @xmath22 is no more scaling invariant to @xmath136 s . furthermore , since any arbitrary phase rotation of the beamforming vectors does not affect both ( [ eq : reformulate problem objective ] ) and the sinr constrains in ( [ eq : dl sinr r ] ) and ( [ eq : vdl sinr r ] ) , ( p3 ) with ( [ eq : reformulate problem objective ] ) as the objective function can be reformulated as a convex socp @xcite , which is given by @xmath137 where @xmath138 $ ] , @xmath139 $ ] , and @xmath140 s are auxiliary variables with @xmath141 and @xmath142 indicating that ap @xmath19 is in active and sleep mode , respectively . notice that without @xmath88 norm penalty or @xmath143 , @xmath123 , problem ( p4 ) can be decomposed into two separate minimum - power beamforming design problems : one for the original dl transmission , and the other for the virtual dl transmission . conventionally , the ul transmit sum - power minimization problem , as in ( [ eq : ul power minimization ] ) , has a convenient analytical structure and thus is computationally easier to handle , as compared to the dl minimum - power beamforming design problem , as in ( [ eq : virtual dl power minimizatin ] ) . consequently , most existing studies in the literature have transformed the dl problem to its virtual ul formulation for convenience . the motivation of exploiting the reverse direction in this work , however , is to overcome the scaling issue of ul receive beamforming in gso , so that we can solve the ap selection problem jointly for both dl and ul transmissions . next , we present the complete algorithm for problem ( p1 - 1 ) based on gso , in which three steps need to be performed sequentially . 1 . _ identify the subset of active aps denoted as @xmath144_. this can be done by iteratively solving problem ( p4 ) with different @xmath145 s . notice that how to set the parameter @xmath145 s in ( p4 ) plays a key role in the resulting aps selection . to optimally set the values of @xmath145 s , we adopt an iterative method similar as in @xcite , shown as follows . in the @xmath146th iteration , @xmath147 , @xmath148 s are obtained by solving problem ( p4 ) with @xmath149 . the @xmath150 s are derived from the solution @xmath151 s of the @xmath152th iteration as @xmath153 where @xmath154 is a small positive number to ensure stability . notice that the initial values of @xmath155 s are chosen as @xmath156 where @xmath157 and @xmath158 are the beamforming vector solution of problem ( p4 ) with @xmath159 . the above update is repeated until @xmath160 , @xmath123 , where @xmath117 is a small positive constant that controls the algorithm accuracy . + let @xmath161 $ ] denote the sparse solution after the convergence of the above iterative algorithm . then the nonzero entries in @xmath162 correspond to the aps that need to be active , i.e. , @xmath163 . _ obtain the optimal transmit / receive beamforming vectors @xmath164 and @xmath165 , @xmath131 , given the selected active aps_. this can be done by solving ( p4 ) with @xmath143 , @xmath166 and @xmath167 . 3 . _ obtain the optimal transmit power values of mus @xmath168 , @xmath131_. this can be done by solving problem ( [ eq : ul power minimization ] ) with @xmath169 , @xmath170 , which is a simple linear programming ( lp ) problem . the iterative update given in ( [ eq : iterative update ] ) is designed to make small entries in @xmath171 converge to zero . furthermore , as the updating evolves , the penalty associated with ap @xmath19 in the objective function , i.e. , @xmath172 , will converge to two possible values : @xmath173 in other words , only the active aps will incur penalties being the exact same values as their static power consumption , which has the same effect as the indicator function in problem ( p1 - 1 ) or ( p1 ) . convergence of this algorithm can be shown by identifying the iterative update as a majorization - minimization ( mm ) algorithm @xcite for a concave minimization problem , i.e. , using @xmath174 function , which is concave , to approximate the indicator function given in ( [ eq : on off condition ] ) . the details are thus omitted due to space limitations . it is first observed that the per - ap power constraints in ( [ eq : uldl dl power constraint ] ) , i.e. , @xmath175 are convex . therefore , adding per - ap power constraints to problem ( p1 - 1 ) does not need to alter the above algorithm . thus , we focus on the per - mu power constraints in the ul transmission in this subsection . again , we consider the following transmit sum - power minimization problem in the ul with per - mu power constraints : @xmath176 although it has been shown in @xcite that the sum - power minimization problem in the dl with per - ap power constraints can be transformed into an equivalent min - max optimization problem in the ul , we are not able to find an equivalent dl problem for problem ( [ eq : ul power minimization with per mu ] ) as in section [ sec : without ] which is able to handle the per - mu power constraints . the fundamental reason is that the power allocation obtained by solving problem ( [ eq : ul power minimization ] ) is already component - wise minimum , which can be shown by the uniqueness of the fixed - point solution for a set of minimum sinr requirements in the ul given randomly generated channels @xcite . the component - wise minimum power allocation indicates that it is not possible to further reduce one particular mu s power consumption by increasing others , i.e. , there is no tradeoff among different mus in terms of power minimization . consequently , solving problem ( [ eq : ul power minimization with per mu ] ) requires only one additional step compared with solving problem ( [ eq : ul power minimization ] ) , i.e. , checking whether the optimal power solution to problem ( [ eq : ul power minimization ] ) satisfies the per - mu power constraints . if this is the case , the solution is also optimal for problem ( [ eq : ul power minimization with per mu ] ) ; otherwise , problem ( [ eq : ul power minimization with per mu ] ) is infeasible . next , we present our complete algorithm for problem ( p1 ) with the per - ap and per - mu power constraints . compared to the algorithm proposed for problem ( p1 - 1 ) in section [ sec : without ] without the per - ap and per - mu power constraints , the new algorithm differs in the first step , i.e. , to identify the subset of active aps . the main idea is that a set of candidate active aps is first obtained by ignoring the per - mu power constraints but with a new sum - power constraint in the ul ( or equivalently its virtual dl ) , i.e. , we iteratively solve the following problem similarly as in the first step of solving problem ( p1 - 1 ) in section [ sec : without ] . @xmath177 the sum - power constraint in ( [ eq : sum user p4 ] ) is added to impose a mild control on the transmit powers of all mus in the ul . after obtaining the candidate set , the feasibility of the ul transmission is then verified . if the candidate set can support the ul transmission with the given per - mu power constraints , then the optimal solution of ( p1 ) is obtained ; otherwise , one or more aps need to be active for the ul transmission . to be more specific , denote the set of candidate active aps obtained by iteratively solving problem ( p5 ) as @xmath178 . problem ( [ eq : ul power minimization ] ) is then solved with @xmath178 , for which the feasibility is guaranteed due to the virtual dl sinr constraints in ( [ eq : vdl - socp sinr constraint ] ) . we denote the obtained power allocation as @xmath179 , @xmath180 . * if @xmath178 can support the ul transmission without violating any mu s power constraint , i.e. , @xmath181 the candidate set can be finalized as the set of active aps and the algorithm proceeds to find the optimal transmit / receive beamforming vectors similarly as that in section [ sec : without ] . * if @xmath178 can not support the ul transmission with the given mu s power constraints , we propose the following price based iterative method to determine the additional active aps . specifically , in each iteration , for those aps that are not in the candidate set each will be assigned a price @xmath182 , which is defined as @xmath183 where @xmath184 . the price @xmath185 is set to be the normalized ( by its corresponding static power consumption ) weighted - sum power gains of the channels from ap @xmath186 to all the mus that have their power constraints being violated . the weights are chosen as the ratios of mus required additional powers to their individual power limits . according to the definition of @xmath185 in ( [ eq : price ] ) , the ap having smaller static power consumption and better channels to mus whose power constraints are more severely violated will be associated with a larger price . the candidate set is then updated by including the ap that corresponds the largest @xmath185 as @xmath187 with updated @xmath178 , the feasibility of the ul transmission needs to be re - checked by obtaining a new set of power allocation , which will be used to compute the new @xmath185 s in next iteration if further updating is required . the above process is repeated until all the mus power constraints are satisfied . its convergence is guaranteed since problem ( p1 ) has been assumed to be feasible if all aps are active . combining with the algorithm in section [ sec : without ] , our complete algorithm for problem ( p1 ) based on gso is summarized in table [ table1 ] . for the algorithm given in table [ table1 ] , there are two problems that need to be iteratively solved , i.e. , problems ( [ eq : ul power minimization ] ) and ( p5 ) . since problem ( [ eq : ul power minimization ] ) can be efficiently solved by the fixed - point algorithm @xcite , the computation time is dominated by solving the socp problem ( p5 ) . if the primal - dual interior point algorithm @xcite is used by the numerical solver for solving ( p5 ) , the computational complexity is of order @xmath188 . furthermore , since the convergence of the iterative update in steps 4)-5 ) , governed by the mm algorithm , is very fast ( approximately @xmath189-@xmath190 iterations ) as observed in the simulations , the overall complexity of the algorithm in table [ table1 ] is approximately @xmath191 . ' '' '' 1 . set @xmath192 , initialize the set of candidate active aps as @xmath193 . 2 . obtain @xmath157 s and @xmath158 s by solving problem ( p5 ) with @xmath159 . 3 . set @xmath194 . * repeat : * * @xmath195 . * set @xmath196 , @xmath123 . * obtain @xmath197 $ ] by solving problem ( p5 ) with @xmath198 . 5 . * until * @xmath199 , @xmath123 or @xmath200 . 6 . set @xmath178 as @xmath201 . * repeat : * * obtain @xmath179 , @xmath131 , by solving problem ( [ eq : ul power minimization ] ) with @xmath178 . * set @xmath202 . * set @xmath203 . * set @xmath204 * until * @xmath205 . 9 . obtain @xmath164 and @xmath206 , @xmath131 , by solving ( p5 ) with @xmath143 , @xmath123 and @xmath207 . 10 . set @xmath208 , and compute the @xmath168 , @xmath131 , by solving problem ( [ eq : ul power minimization ] ) . ' '' '' [ table1 ] in this subsection , an alternative algorithm for problem ( p1 ) is developed based on rip by applying the same idea of establishing a virtual dl transmission for the original ul . similar to the case with gso , the per - mu power constraints are first replaced with a sum - power constraint in the ul . the resulting problem is further reformulated as a convex socp by relaxing the binary variables @xmath209 , which is given as follows . @xmath210 note that instead of implementing the active - sleep constraints jointly for the actual dl and virtual dl as ( [ eq : bigm onoff ] ) in problem ( p2 ) , i.e. , @xmath211 , we divide them into two sets of coupled active - sleep constraints as in ( [ eq : bigm1 ] ) and ( [ eq : bigm2 ] ) via @xmath106 s . for the non - relaxed problem of ( p6 ) with binary @xmath106 s , i.e. @xmath212 , it can be shown that these two formulations are equivalent . however , for the case of the relaxed problem ( p6 ) with continuous valued @xmath106 s , the separated active - sleep constraints are designed to avoid the situation that the difference between @xmath213 and @xmath214 is too large such that the optimal value of @xmath106 is dominated by either dl or ul transmission . to implement the big-@xmath109 method with active - sleep constraints @xcite , an appropriate upper bound for the term @xmath215 needs to be found . according to the ul - dl duality , the minimum sum - power achieved is the same for the ul and its virtual dl transmissions . therefore , @xmath214 can be chosen as the upper bound of @xmath216 . finally , it is evident that the optimal value of problem ( p6 ) serves as a lower bound of its non - relaxed problem with binary @xmath106 s . in order to further tighten this lower bound , one way is to reduce the feasible set of design variables . constraint in ( [ eq : red ] ) is introduced specifically to achieve this end , which can be shown to be redundant for the non - relaxed problem of ( p6 ) . we adopt the same idea of incentive measure based ap selection as in @xcite to design a polynomial - time algorithm for problem ( p2 ) . however , it remains to find an incentive measure that reflects the importance of each ap to both dl and ul transmissions based on problem ( p6 ) . it is interesting to observe that after transforming the ul related terms to their virtual dl counterparts , the optimal relaxed binary variable solution of problem ( p6 ) becomes a good choice to serve this purpose . let @xmath217 , @xmath218 and @xmath219 denote the optimal solution to problem ( p6 ) . intuitively , the ap that has larger static power consumption and worse channels to all mus is more desired to be switched into sleep mode from the perspective of energy saving . in ( p6 ) , for ap @xmath19 having larger @xmath82 , @xmath217 is desired to be smaller in order to achieve the minimum value of the objective function . furthermore , it is practically valid that for dl power minimization problem , the optimal transmit power of aps that have worse channels to mus is in general smaller . as a result , solving problem ( p6 ) yields smaller @xmath220 and @xmath221 and thus smaller @xmath217 for ap @xmath19 that has worse channels to mus for both the dl and ul transmissions . to summarize , the ap that corresponds to smaller @xmath217 is more desired to be switched into sleep mode . an iterative process is then designed to determine the set of active aps based on @xmath217 s that are taken as incentive measures . the process starts with assuming all aps are active . in each iteration , problem ( p6 ) is solved with a candidate set of active aps , and the ap corresponding to the smallest @xmath217 will be removed from the candidate set . this process is repeated until one of the following conditions occurs : * the weighted sum - power can not be further reduced ; * problem ( p6 ) becomes infeasible ; * problem ( [ eq : ul power minimization with per mu ] ) becomes infeasible . note that the feasibility checking for problem ( [ eq : ul power minimization with per mu ] ) is the same as that in algorithm i , which ensures the per - mu power constraints . an overall algorithm for problem ( p1 ) based on rip is summarized in table [ table2 ] . for the algorithm given in table [ table2 ] , the computation time is dominated by solving the socp problem ( p6 ) . if the primal - dual interior point algorithm @xcite is used by the numerical solver for solving ( p6 ) , the computational complexity is of order @xmath188 . furthermore , since the worst case complexity for the iteration in steps 2)-3 ) is @xmath222 , the overall complexity of the algorithm in table [ table2 ] is @xmath223 . ' '' '' 1 . set @xmath192 , @xmath224 a sufficiently large value , and initialize the set of candidate active aps as @xmath193 . * repeat : * * @xmath195 . * solve problem ( p6 ) with @xmath110 , @xmath225 . * set @xmath226 . * set @xmath227 as the optimal value of problem ( p6 ) with @xmath228 and @xmath229 . until * @xmath230 or problem ( p6 ) is infeasible or problem ( [ eq : ul power minimization with per mu ] ) is infeasible . 4 . obtain @xmath164 and @xmath206 , @xmath131 , by solving problem ( p6 ) with @xmath228 and @xmath229 . 5 . set @xmath208 , and compute the @xmath168 , @xmath131 , by solving problem ( [ eq : ul power minimization ] ) . ' '' '' [ table2 ] in this section , we present numerical results to verify our proposed algorithms from three perspectives : ensuring feasibility for both dl and ul transmissions ; achieving network power saving with optimal mu - ap association ; and adjusting minimum power consumption tradeoffs between active aps and mus . we consider two possible c - ran configurations : 1 . homogeneous setup : all aps are assumed to have the same power consumption model with @xmath231w and @xmath232w , @xmath233 , if not specified otherwise . heterogeneous setup : two types of aps are assumed , namely , high - power ap ( hap ) and low - power ap ( lap ) , where the static power consumption for hap and lap are set as @xmath234w and @xmath235w , respectively , and the transmit power budgets for hap and lap are set as @xmath1w and @xmath236w , respectively . we assume that each ap @xmath19 , @xmath20 , is equipped with @xmath237 antennas . for the single - antenna mu , we set the transmit power limit as @xmath238w , @xmath170 . for simplicity , we assume that the sinr requirements of all mus are the same in the ul or dl . all the aps ( except haps under the heterogeneous setup ) and mus are assumed to be uniformly and independently distributed in a square area with the size of @xmath239km@xmath240@xmath239 km . for all the simulations under heterogeneous setup , it is assumed that there are @xmath235 haps with fixed location at @xmath241 $ ] and @xmath242 $ ] , respectively . we assume a simplified channel model consisting of the distance - dependent attenuation with pathloss exponent @xmath243 and a multiplicative random factor ( exponentially distributed with unit mean ) accounting for short - term rayleigh fading . we also set @xmath244 if not specified otherwise , i.e. , we consider the sum - power consumption of all active aps and mus . finally , we set the receiver noise power for all the aps and mus as @xmath245dbm . first , we demonstrate the importance of active ap selection by jointly considering both dl and ul transmission in terms of the sinr feasibility in c - ran . since feasibility is our focus here instead of power consumption , it is assumed that the selected active aps will support all mus for both the dl and ul transmissions . the simulation results compare our proposed algorithms ( i.e. , algorithms i and ii ) with the following three ap selection schemes : * * ap initiated reference signal strength ( apirss ) based selection * : in this scheme , aps first broadcast orthogonal reference signals . then , for each mu , the ap corresponding to the largest received reference signal strength will be included in the set of active aps . note that this scheme has been implemented in practical cellular systems @xcite . * * mu initiated reference signal strength ( muirss ) based selection * : in this scheme , mus first broadcast orthogonal reference signals . then , for each mu , the ap corresponding to the largest received reference signal strength will be included in the set of active aps . note that since all mus are assumed to transmit reference signals with equal power and pathloss in general dominates short - term fading , the ap that is closest to each mu will receive strongest reference signal in general . also note that in the previous apirss based scheme , if all aps are assumed to transmit with equal reference signal power ( e.g. , for the homogenous setup ) , the selected active aps will be very likely to be the same as those by the muirss based scheme . * * proposed algorithm without considering ul ( paw / oul ) * : in this algorithm , the set of active aps are chosen from the conventional dl perspective by modifying our proposed algorithms . specifically , algorithm i is used here and similar results can be obtained with algorithm ii . note that algorithm i without considering ul transmission is similar to that proposed in @xcite . with the obtained set of active aps , the feasibility check of problem ( p1 ) can be decoupled into two independent feasibility problems : one for the dl and the other for the ul , while the network feasibility is achieved only when both the ul and dl sinr feasibility of all mus are guaranteed . in fig . [ fig : feasibilitydemo ] , we illustrate the set of active aps generated by different schemes under the heterogeneous setup , and also compare them with that by the optimal exhaustive search . it is assumed that there are @xmath235 haps and @xmath246 laps jointly supporting @xmath246 mus . the sinr targets for both dl and ul transmissions of all mus are set as @xmath246db . first , it is observed that algorithm i and algorithm ii obtain the same set of active aps as shown in fig . [ fig : feasibilitydemohetnetl12 ] , which is also identical to that found by exhaustive search . second , it is observed that the @xmath235 haps are both chosen to be active in fig . [ fig : feasibilitydemohetnethighsignal ] for the apirss based scheme . this is due to the significant difference between hap and lap in terms of transmit power , which makes most mus receive the strongest dl reference signal from the hap . the above phenomenon is commonly found in heterogenous network ( hetnet ) @xcite with different types of bss ( e.g. macro / micro / pico bss ) . third , from fig . [ fig : feasibilitydemohetnetclose ] , the active aps by the muirss based scheme are simply those closer to the mus , which is as expected . finally , in fig . [ fig : feasibilitydemohetnetdl ] , only two laps are chosen to support all mus with the paw / oul algorithm . this is because the algorithm does not consider ul transmission , and as a result fig . [ fig : feasibilitydemohetnetdl ] only shows the most energy - efficient ap selection for dl transmission . to compare the feasibility performance , we run the above algorithms with different dl and ul sinr targets it is assumed that @xmath247 and @xmath248 . the results are summarized in table [ table4 ] and table [ table3 ] , where the number of infeasible cases for each scheme is shown over @xmath249 randomly generated network and channel realizations , for homogeneous setup and heterogeneous setup , respectively . note that in these examples , algorithm i and algorithm ii have identical feasibility performance , since the system is infeasible only when the dl / ul sinr requirements can not be supported for given channels and power budgets even with all aps being active . from both table [ table4 ] and table [ table3 ] , it is first observed that the three comparison schemes , i.e. , apirss based scheme , muirss based scheme and paw / oul , all incur much larger number of infeasible cases as compared to our proposed algorithms . it is also observed that among the three comparison schemes , paw / oul has the best performance ( or the minimum number of infeasible cases ) when the dl transmission is dominant ( i.e. , @xmath250 ) ; however , it performs the worst in the opposite situation ( i.e. , @xmath251 ) . this observation indicates that dl oriented scheme could result in infeasible transmit power of mus in the ul for the cases with stringent ul requirements . from the last two rows of table [ table3 ] , it is observed that the apirss based scheme performs worse than the muirss based scheme when the ul sinr target is high . this is because that under heterogeneous setup , as shown in fig . [ fig : feasibilitydemohetnethighsignal ] , mus are attached to the haps under apirss based scheme although the haps may be actually more distant away from mus compared with the distributed laps . this imbalanced association causes much higher transmit powers of mus or even their infeasible transmit power in the ul . there has been effort in the literature to address this traffic imbalance problem in hetnet . for example in @xcite , the reference signal from picocell bs is multiplied by a factor with magnitude being larger than one , which makes it appear more appealing for mu association than the heavily - loaded macrocell bs . .feasibility performance comparison under homogeneous setup [ cols="^,^,^,^,^,^,^ " , ] [ table3 ] next , we compare the performance of the proposed algorithms in terms of sum - power minimization in c - ran with the following benchmark schemes : * * exhaustive search ( es ) * : in this scheme , the optimal set of active aps are found by exhaustive search , which serves as the performance upper bound ( or lower bound on the sum - power consumption ) for other considered schemes . with any set of active aps , the minimum - power dl and ul beamforming problems can be separately solved . since the complexity of es grows exponentially with @xmath15 , it can only be implemented for c - ran with small number of aps . * * joint processing ( jp ) among all aps*@xcite : in this scheme , all the aps are assumed to be active and only the total transmit power consumption is minimized by solving two separate ( dl and ul ) minimum - power beamforming design problems . * * algorithm i with @xmath97 norm penalty * : this algorithm is the same as that given in table [ table1 ] except that the sparsity enforcing penalty is replaced with @xmath97 norm as given in ( [ eq : infinity ] ) . in our simulations , we consider the homogeneous c - ran setup with @xmath247 and plot the performance by averaging over @xmath252 randomly generated network and channel realizations . the sinr requirements are set as @xmath253db for all mus @xmath254 . [ fig : performancemu ] and fig . [ fig : performancecomparepcdis ] show the sum - power consumption achieved by different algorithms versus the number of mus @xmath17 and ap static power consumption @xmath82 ( assumed to be identical for all aps ) , respectively . from both figures , it is observed that the proposed algorithms have similar performance as the optimal es and achieve significant power saving compared with jp . it is also observed that the penalty term based on either @xmath88 or @xmath97 norm has small impact on the performance of algorithm i. finally , algorithm i always outperforms algorithm ii although the performance gap is not significant . = 0.7w , @xmath123.,title="fig:",width=453 ] = 0.7 .,title="fig:",width=453 ] finally , we compare the sum - power consumption tradeoffs between active aps and all mus for the proposed algorithms as well as the optimal es , by varying the weight parameter @xmath255 in our formulated problems . we consider a homogenous c - ran setup with @xmath247 and @xmath248 , where @xmath253db for all mus @xmath254 . since it has been shown in the pervious subsection that algorithm i with @xmath88 norm and @xmath97 norm achieves similar performance , we choose @xmath88 norm in this simulation . furthermore , since js assumes that all the aps are active , which decouples dl and ul transmissions and thus has no sum - power consumption tradeoffs between aps and mus , it is also not included . from fig . [ fig : tradeoff ] , it is first observed that for all considered algorithms , as @xmath255 increases , the sum - power consumption of active aps increases and that of all mus decreases , which is as expected . it is also observed that algorithm i achieves trade - off performance closer to es and outperforms algorithm ii , which is in accordance with the results in figs . [ fig : performancemu ] and [ fig : performancecomparepcdis ] . in this paper , we consider c - ran with densely deployed aps cooperatively serving distributed mus for both the dl and ul transmissions . we study the problem of joint dl and ul mu - ap association and beamforming design to optimize the energy consumption tradeoffs between the active aps and mus . leveraging on the celebrated ul - dl duality result , we show that by establishing a virtual dl transmission for the original ul transmission , the joint dl and ul problem can be converted to an equivalent dl problem in c - ran with two inter - related subproblems for the original and virtual dl transmissions , respectively . based on this transformation , two efficient algorithms for joint dl and ul mu - ap association and beamforming design are proposed based on gso and rip techniques , respectively . by extensive simulations , it is shown that our proposed algorithms improve the network reliability / feasibility , energy efficiency , as well as power consumption tradeoffs between aps and mus , as compared to other existing methods in the literature . s. tombaz , p. monti , k. wang , a. vastberg , m. forzati , and j. zander , `` impact of backhauling power consumption on the deployment of heterogeneous mobile networks , '' in _ proc . ieee globecom _ , pp . 1 - 5 , dec . 2011 . m. hong , r. sun , h. baligh , and z. luo , `` joint base station clustering and beamformer design for partial coordinated transmission in heterogeneous networks , '' _ ieee j. sel . areas commun . 226 - 240 , feb . 2013 . y. cheng , m. pesavento , and a. philipp , `` joint network optimization and downlink beamforming for comp transmissions using mixed integer conic programming , '' _ ieee trans . signal process . 16 , pp . 3972 - 3987 , aug . 2013 . m. rasti , a. r. sharafat , s. member , and j. zander , `` pareto and energy - efficient distributed power control with feasibility check in wireless networks , '' _ ieee trans . inform . theory _ 245 - 255 , jan . 2011 . k. son , h. kim , y. yi , and b. krishnamachari , `` base station operation and user association mechanisms for energy - delay tradeoffs in green cellular networks , '' _ ieee j. sel . areas commun . 29 , no . 8 , pp . 1525 - 1536 , sep . d. gesbert , s. hanly , h. huang , s. s. shitz , o. simeone , and w. yu , `` multi - cell mimo cooperative networks : a new look at interference '' , _ ieee j. sel . areas commun . 9 , pp . 1380 - 1408 , dec . 2010 . s. h. park , o. simeone , o. sahin , and s. shamai , `` joint precoding and multivariate backhaul compression for the downlink of cloud radio access networks , '' _ ieee trans . signal process . 22 , pp . 5646 - 5658 , nov . 2013 . l. liu , r. zhang , and k. c. chua , `` achieving global optimality for weighted sum - rate maximization in the k - user gaussian interference channel with multiple antennas , '' _ ieee trans . wireless commun . 5 , pp . 1933 - 1945 . may 2012 . g. auer , v. giannini , c. desset , i. godor , p. skillermark , m. olsson , m. imran , d. sabella , m. gonzalez , o. blume , and a. fehske , `` how much energy is needed to run a wireless network ? '' _ ieee trans . wireless commun . _ , vol . 18 , no . 40 - 49 , oct . 2011 . m. yuan and y. lin , `` model selection and estimation in regression with grouped variables , '' _ journal of the royal statistical society : series b ( statistical methodology ) _ , vol . 49 - 67 , feb . 2006 . q. ye , y. chen , m. al - shalash , c. caramanis , and j. g. andrews , `` user association for load balancing in heterogeneous cellular network , '' _ ieee trans . wireless commun . 12 , no . 6 , 2706 - 2716 , jun . 2013 . j. sangiamwong , y. saito , n. miki , t. abe , s. nagata , and y. okumura , `` investigation on cell selection methods associated with inter - cell interference coordination in heterogeneous networks for lte - advanced downlink , '' in _ 11th european wireless conference sustainable wireless technologies _ , pp . 1 - 6

we study quantum quenches in the two - dimensional kitaev toric code model and compute exactly the time - dependent entanglement entropy of the non - equilibrium wave - function evolving from a paramagnetic initial state with the toric code hamiltonian . we find that the area law survives at all times . adding disorder to the toric code couplings makes the entanglement entropy per unit boundary length saturate to disorder - independent values at long times and in the thermodynamic limit . there are order - one corrections to the area law from the corners in the subsystem boundary but the topological entropy remains zero at all times . we argue that breaking the integrability with a small magnetic field could change the area law to a volume scaling as expected of thermalized states but is not sufficient for forming topological entanglement due to the presence of an excess energy and a finite density of defects .

1007.2014

cold atomic systems have provided an ideal playground for the experimental studies of the unitary evolution of thermally isolated quantum systems . @xcite apart from the experimental interest , studies of quantum evolution have shed light on some fundamental questions such as thermalization . @xcite a widely - studied scenario is the unitary evolution of a quantum state when the hamiltonian is suddenly changed a `` quantum quench '' , @xcite which has been the subject of many recent studies ( see refs . [ ] for a few examples ) . a challenging question , which has been addressed theoretically in one dimensional systems , is the evolution of the entanglement entropy following the quench . @xcite more generally , such studies are concerned with the fate of an out - of - equilibrium quantum state with a hamiltonian that changes with some time - dependent protocol . the case where the coupling constants are changed across a quantum phase transition is of special interest . in recent experiments with spinor condensates , for example , the formation of ferromagnetic domains was observed in a quench to a hamiltonian with a symmetry - breaking ferromagnetic ground state . @xcite even though the interactions after the quench favor a certain type of order , it is not at all obvious that the order should emerge out of equilibrium . since symmetry - breaking order has a local order parameter , the non - equilibrium state can support the formation of ordered domains . but what about quenches across a topological phase transition ? would any of the topological characteristics of the equilibrium phase show up in the non - equilibrium wave - function ? another interesting aspect of the non - equilibrium dynamics is the issue of thermalization and whether a large system can act as an effective heat bath for its subsystems . if thermalization occurs , due to the presence of some excess energy , the non - equilibrium state at long times should be in some sense _ close _ to a thermal state at a finite temperature , which implies volume scaling for the entanglement entropy . the qualitative discussions above motivate a detailed analysis of a quantum quench across a topological phase transition into the topological phase . we study the specific case of the kitaev toric code model @xcite with spin - polarized initial states . a related quantum quench problem with topologically ordered initial states was recently studied numerically in ref . we focus on the time evolution of the entanglement entropy because it is intimately related to topological order through the topological entanglement entropy . @xcite entanglement entropy is also related to thermalization since , as mentioned above , volume scaling of the entropy as opposed to the area law is generically expected of thermal states . we compute exactly the bipartite entanglement entropy as a function of time . this is an example where exact calculation can be done in 2 + 1d ( thus far , analytical results for the time - evolution of entanglement had focused on 1 + 1d systems ) . we show that the entanglement entropy respects the area law at all times and if one adds disorder to the coupling constants , the recurrences or revivals are suppressed at long times and the entanglement entropy per unit `` area '' ( boundary length ) @xmath0 equals @xmath1 in the thermodynamic limit when the initial polarization is in the @xmath2 or @xmath3 direction . we find the dependence of this saturation value on the initial state for an arbitrary direction of the initial magnetization . a new ingredient that comes in when considering these initial states is order - one corrections to the entanglement entropy from the convex and concave corners in the subsystem boundary . we find that the general form of the generated entanglement entropy is given by @xmath4 where @xmath0 is the perimeter of the subsystem and @xmath5 ( @xmath6 ) the number of convex ( concave ) corners in its boundary . the functions of time @xmath7 , @xmath8 and @xmath9 are independent of the subsystem geometry . from the form of the entanglement entropy above , we find that the topological entropy vanishes at all times . we study the effects of breaking the integrability in a perturbative approach . we identify the mechanism by which an integrability - breaking magnetic field can change the area law to the volume law as expected of thermalized states . the physical picture is that the area law survives at low orders in an expansion in the integrability - breaking field . going to higher orders however increases the thickness of a ring - like region around the subsystem boundary that contributes to the entanglement entropy . eventually at high enough orders , the ring will collapse onto itself making volume scaling possible . in the perturbative treatment , we explicitly find that the topological entanglement entropy remains zero at low orders . we expect that the excess energy and the finite density of defects should prevent the formation of any topological order in a sudden quench even if one breaks the integrability . the paper is organized as follows . in section [ sec : quenchesxz ] we discuss quantum quenches for spin - polarized initial states in the @xmath2 or @xmath3 direction . we calculate the entanglement entropy analytically and discuss the effects of disorder in the coupling constants . in section [ sec : corners ] , we study the role of the sharp corners in the subsystem boundary in the generation of entanglement by exactly calculating the second order renyi entropy for an arbitrary direction of the initial magnetization . we then discuss the effects of breaking the integrability , which we treat perturbatively , in section [ sec : non - integrable ] . we conclude the paper in section [ sec : conclusion ] . consider the toric code hamiltonian in a field @xcite @xmath10 the hilbert space of the above hamiltonian is given by spins living on the bonds of a square lattice on a torus and the sums are over all plaquettes , stars and bonds respectively . the plaquette and star operators are defined as follows @xmath11 where @xmath12s are the pauli matrices . for the usual kitaev model the coupling constants @xmath13 and @xmath14 are independent of the plaquettes and the stars but here we keep the @xmath15 and @xmath16 indeces to allow for disorder in the couplings . note that for the model in zero magnetic field ( @xmath17 ) , the ground state is independent of the actual values of the coupling constants @xmath13 and @xmath14 as long as they are all positive . one can study the dynamics of this model by changing the coefficients @xmath13 , @xmath14 and @xmath18 in time with some protocol . in this section , we consider a quantum quench where the system is initially in a fully magnetized state and then evolves with the integrable toric code hamiltonian , namely , we are initially in the ground state for @xmath19 and @xmath20 for all stars and plaquettes . we then turn off the field @xmath18 at @xmath21 ( which does not change the quantum state ) and turn on the toric code part of the hamiltonian as follows : @xmath22 and @xmath23 ( @xmath24 ) . consider for now , the case where @xmath25 ( @xmath26 ) . the ground state of the usual model ( without disorder in the couplings ) exhibits a topological phase transition : for @xmath27 the system is a paramagnet and for @xmath28 , it is topologically ordered . this topological phase is robust against weak disorder in the couplings . let us call the initial state @xmath29 . in the case where the initial magnetization is in the @xmath2 direction @xmath30 in the @xmath31 basis . the initial state is a tensor product of single spin states and thus has no entanglement . in the paramagnetic initial state we have @xmath32 for all spin operators . before addressing the question of the entanglement entropy evolution , let us consider the simple problem of what happens to the @xmath33 expectation value . as a function of time , we have @xmath34 where @xmath35 is the evolution operator given by @xmath36 since all star and plaquettes commute with one another , we have dropped the time ordering operator . if we consider a sudden quench , we have @xmath37 and @xmath38 . let us focus on sudden quenches for a more compact notation . at the end of the calculation we can replace @xmath39 and @xmath40 with the corresponding integrals over time - dependent couplings for a more general protocol . the expectation value of @xmath41 , a spin between two nearest - neighbor plaquettes @xmath15 and @xmath42 is then given by @xmath43 we have @xmath44 . therefore we get @xmath45 in the completely clean case ( a theoretical abstraction ) , the average of @xmath46 over time is equal to @xmath47 but any amount of disorder will eventually lead to a state which unlike the initial paramagnet has a vanishing time - averaged expectation value for @xmath33 . we next consider the main problem of this section , namely the evolution of the entanglement entropy . the density matrix of the system can be written as @xmath48 where @xmath35 is the time evolution operator and @xmath49 is a generic initial state . by inserting the resolution of identity twice , we can write , @xmath50 and @xmath51 . any star or plaquette operator acts on four spins which may all belong to @xmath52 , all belong to @xmath51 or some belong to @xmath52 and some to @xmath51 . in this last case , the star or plaquette operator must lie at the boundary of the two subsystems . we can then write @xmath53 where @xmath54 and @xmath55 only act on spins in @xmath52 and @xmath51 respectively and @xmath56 only depends on star and plaquette operators at the boundary of the two subsystems . the three operators @xmath54 , @xmath55 and @xmath56 commute with one another . each basis state can be written as the tensor product of states for spins in @xmath52 and @xmath51 , @xmath57 . we now integrate out the spins in @xmath51 to find the reduced density matrix for subsystem @xmath52 , @xmath58 and obtain , @xmath59 by @xmath60 in the above expression . this amounts to calculating the trace in a rotated basis . we then substitute eq . ( [ eq : u ] ) into eq . ( [ eq : rho_a ] ) above and obtain latexmath:[\[\rho_a(t)=\sum_{\alpha_a , \sigma_a , \beta_b } \langle \alpha_a \beta_b | u_a u_\partial|\omega\rangle \langle \omega | u_a^\dag u_\partial^\dag | \sigma_a \beta_b \rangle that only acts on @xmath51 spins has no effect on the time evolution of the reduced density matrix for subsystem @xmath52 . in order to calculate the entanglement entropy between the two subsystems we need to find @xmath62 . we give a replica index to @xmath63 , @xmath64 and @xmath12 and drop the subsystem index for brevity . states @xmath65 are in subsystem @xmath52 and @xmath66 in @xmath51 . using @xmath67 we obtain , @xmath68 with @xmath69 once again because traces are basis - independent we can replace @xmath65 by @xmath70 in the above equation and we obtain an expression for @xmath62 that only depends on @xmath56 and not the whole @xmath35 . let us define the following commuting operators @xmath71 where @xmath72 ( @xmath73 ) is the set of all boundary plaquettes ( stars ) which act on both @xmath52 and @xmath51 spins . we then obtain @xmath74 with @xmath69 . we now focus on the simple case where initially all spins are aligned in the @xmath2 direction . in this case , @xmath75 is an eigenstate of @xmath76 and we can therefore substitute @xmath77 in eq . ( [ eq : tr_rho_a_2 ] ) above . also @xmath78 and @xmath79 are chosen in the @xmath33 basis so the action of @xmath80 on a basis state is to flip the four spins around the plaquette @xmath15 . for concreteness , we assume that the partition @xmath51 contains all spins which lie inside a simply connected @xmath81 rectangular region . we then have @xmath82 boundary plaquettes . let us consider the action of @xmath83 on a state @xmath84 . since @xmath85 , we have @xmath86 notice that @xmath87 only if the state the state @xmath88 is obtained by acting with a certain number of @xmath72 plaquette flips on @xmath89 . let us consider then ising spin variables living at the center of @xmath72 plaquettes . a plaquette is flipped for @xmath90 and not flipped for @xmath91 . notice that , on a torus , flipping a certain set of plaquettes @xmath92 and flipping all other plaquettes that do not belong to @xmath92 have the same effect on the spins , i.e. they describe the same element in the group of plaquette flips . however , since we have eliminated @xmath54 and @xmath55 from the formulation of the problem by a basis rotation , we can uniquely specify a state with nonvanishing @xmath93 by a set of flipped _ boundary _ plaquettes . let the state @xmath94 be labeled by @xmath95 for @xmath96 . if the partitions are not too small , it is not possible for a set of boundary plaquette flips acting on @xmath97 to create a state with the same @xmath51 spins and different @xmath52 spins or vice versa . therefore @xmath98 and we obtain using eq . ( [ eq : tr_rho_a_2 ] ) @xmath99 using eq . ( [ eq : upm ] ) we find that @xmath100 is a product of terms that are equal to @xmath101 for each unflipped plaquette with @xmath102 and @xmath103 for each flipped plaquette with @xmath104 . we can then write @xmath105 switching the order of the sum and product after plugging eq . ( [ eq : matrix - element ] ) into eq . ( [ eq : trace - rho - n - x ] ) , we obtain @xmath106 we have used the fact that for @xmath107 , @xmath108 . summing over @xmath109 for each boundary plaquette simply gives @xmath110.\ ] ] by analytic continuation of @xmath111 , we write the entanglement entropy as @xmath112 and we finally obtain , @xmath113.\ ] ] for the clean kitaev model , the entanglement entropy is an oscillatory function of time whose maximum is proportional to the number of boundary plaquettes or the perimeter of the subsystem . for this initial condition , the star term in the toric code hamiltonian has no effect on the evolution of the entanglement entropy because the initial state is an eigenstate of that term . similarly if initially all spins are in the @xmath3 direction , the plaquette term will have no effect . the completely clean system will never reach a steady state . if there is some disorder in the coupling constants however , the entanglement entropy per unit boundary length will saturate to a steady state value in the thermodynamic limit . suppose the number of boundary plaquettes @xmath0 is very large . as @xmath114 , due to the presence of some disorder in couplings @xmath115 , the phases @xmath116 are equally likely to take any value between @xmath117 and @xmath118 modulo @xmath118 . therefore , taking the thermodynamic limit first we can write @xmath119 with @xmath120 where @xmath121 is a random variable between @xmath117 and @xmath122 drawn from a uniform distribution . the average entanglement entropy is then given by @xmath123 we have @xmath124 which implies that for any realization of disorder @xmath125 indicating that in the thermodynamic limit , the entanglement per unit area length saturates to a steady - state value @xmath126 . notice that these results , namely the survival of area law at all times and the oscillatory behavior in the entanglement entropy in the absence of disorder are generic features of integrable quenches where the integrals of motion are operators with local support . the entanglement entropy per unit boundary length relaxes in the thermodynamic limit to a steady state value in the presence of disorder in the coupling constants . because the system can reach a steady state , it is natural to ask if the steady state of a subsystem has the properties of a thermalized state described by a gibbs or a generalized gibbs ensemble . @xcite for such thermalized states however , we expect the entanglement entropy of a small subsystem to scale as its volume in the limit of long times . the survival of the area law at all times which follows trivially from having local integrals of motion indicates the absence of thermalization even to a generalized gibbs ensemble . let us now consider a more generic case where the initial density matrix is not diagonal in either @xmath33 or @xmath127 basis and both terms in the kitaev hamiltonian affect the entanglement dynamics . we find that even though in the thermodynamic limit the area law still applies , the physics of the out of equilibrium entanglement is much richer . we find that here there are order - one corrections to the entanglement entropy from the concave and convex corners of the subsystem . subleading corrections to the entanglement entropy arising form the sharp corners in the boundary were found before for _ static _ entanglement entropy in the logarithmic corrections to the area law in a class of conformal quantum critical points . @xcite here we find that corners can also make subleading contributions to the _ dynamical _ entanglement . we find that when subsystem @xmath51 consists of spins inside a region with @xmath5 convex and @xmath6 concave corners as seen in fig . [ fig : corners ] , the entanglement entropy after a quantum quench behaves as @xmath128 where @xmath0 is the perimeter of the subsystem and @xmath7 , @xmath8 and @xmath9 are functions of time that do not depend on the subsystem geometry . for a subsystem with a mostly smooth boundary and only a few sharp corners , the leading term is the @xmath129 but for a subsystem with a corrugated boundary , the entanglement entropy can be different in the thermodynamic limit . notice that the number of concave and convex corners are not independent . suppose subsystem @xmath51 consists of spins inside of an area with @xmath130 connected regions , each with a certain number of handles such that the total number handles is @xmath131 . as an example in fig . [ fig : corners ] , we have @xmath132 and @xmath133 . we argue that @xmath134 this is because for a rectangular simply connected region , we have @xmath135 and @xmath136 and transforming the shape by adding or subtracting plaquettes on the square lattice does not change @xmath137 . therefore each simply connected region makes a contribution of @xmath138 to @xmath139 and by a similar argument , each hole contributes @xmath140 . to see how the contribution from sharp corners comes in , consider an initial state with all the spins in the positive @xmath141 direction . let us work in the the @xmath142 basis with @xmath143 , @xmath144 and @xmath145 . the action of a star or plaquette operator in this basis is then to flip the corresponding four @xmath142 spins modulo a phase factor of @xmath146 from @xmath80 . we will see later that this sign has no effect on the entanglement entropy . we assume for simplicity that there is a minimum distance of @xmath147 times the lattice spacing between all corners . consider a convex corner as in the right hand side of fig . [ fig : ycorner ] . the @xmath64 spins live in the dashed blue ( color online ) bonds and the @xmath63 spins on the solid black bonds . flipping the corner plaquette represented by a red square has the same effect on @xmath64 spins as flipping the two boundary stars represented by red crosses . these operations of course have a different effect on @xmath63 spins . because we assumed a minimum distance between the corners , these corner star and plaquette operators uniquely correspond to one corner . similarly as in the left hand side of fig . [ fig : ycorner ] , flipping the corner star at a concave corner has the same effect on the @xmath63 spins as flipping the two corresponding plaquettes . we see from this discussion that the effects of stars and plaquette flips mix at the corners , which leads to a different entanglement generation than along a smooth boundary . the details of the derivation will be given in appendix [ app_a ] . let us write for now the result for the entanglement entropy for a subsystem @xmath51 consisting of the bonds inside an @xmath148 rectangular region . a boundary star or plaquette that is not located at a corner is refereed to as a regular star or plaquette . as seen in fig . [ fig : ycorner ] , at each convex corner , one plaquette and two stars are special . for our @xmath149 subsystem , we have four convex corners , @xmath150 regular boundary plaquettes and @xmath151 regular boundary stars and the entanglement entropy for a clean system after a sudden quench is given by @xmath152 where the contributions from regular plaquettes ( @xmath153 ) , regular stars ( @xmath154 ) and convex corners ( @xmath155 ) are given by @xmath156 \end{aligned}\ ] ] with @xmath157 the above solution conforms to the generic form eq . ( [ eq : entropy - corners ] ) for @xmath158 , @xmath135 , @xmath136 , @xmath159 and @xmath160 . in appendix [ app_a ] , we give a derivation of the above result in the more general setting of an arbitrary subsystem geometry . if instead of a sudden quench we turn on the coupling constants with some protocol @xmath161 and @xmath162 , we just need to replace @xmath39 and @xmath40 by their integrals over time . using the partitions of ref . , shown in fig . [ fig : partitions ] , the topological entanglement entropy is given by @xmath163 from the form of eq . ( [ eq : entropy - corners ] ) , we can then see that although there are order - one corrections to the area law , the topological entanglement entropy remains exactly zero at all times independently of how the toric code hamiltonian is turned on . it seems plausible that for _ any _ initial state , as long as the integrability is not broken ( with a field @xmath18 for example ) , evolution with the kitaev hamiltonian does not change the topological entanglement entropy for any time dependence of the coupling constants . this is because this integrable evolution does not allow propagation of information . the von neumann entropy is a special case ( limit of @xmath164 ) of the order @xmath111 renyi entropy defined as @xmath165 hereafter we refer to the second order ( @xmath166 ) renyi entropy as the renyi entropy . the renyi entropy @xmath167 is simpler to calculate than the von neumann entropy and is an equally valid measure of entanglement . it has been shown that the topological entanglement entropy calculated by any order renyi entropy contains the same information in the abelean topological phases @xcite such as the toric code . for an integrable quench from a paramagnetic state with magnetization in an arbitrary direction , we exactly calculate the renyi entropy @xmath168 . we consider the clean case with position - independent coupling constants . with disorder in the couplings , the entropy saturates to the time - average of the clean case similarly to the simpler initial conditions considered before . in the @xmath169 basis , an arbitrary paramagnetic state can be written as @xmath170 the calculation is rather involved and is done by a transfer matrix method explained in appendix [ app_b ] . we find the same general features for an arbitrary initial magnetization as for the initial magnetization in the @xmath141 direction . let us summarize these features below . the renyi entropy can be written as a sum of contributions from the perimeter of the subsystem as well as the convex and concave corners . these contributions are oscillatory functions of time in the absence of disorder in the couplings which at long times , saturate to the values given by the time - average of the clean case if we have some disorder in the couplings . the saturation values are independent of the final coupling constants and the protocol used for turning them on . they do however depend on the initial state . for a large subsystem with only a few sharp corners , the leading term for the renyi entropy is the first term in eq . ( [ eq : entropy - corners ] ) , namely @xmath171 . the saturation value of the renyi entropy per unit boundary length is plotted in fig . [ fig : plot ] as a function of the @xmath172 and @xmath173 angles which determine the initial state through @xmath174 and @xmath175 . notice that this saturation value is a unique function of @xmath176 and @xmath177 . to study the effects of breaking the integrability , we focus on one simple case . namely , we start with a state that has all spins aligned in the @xmath2 direction . as opposed to the integrable case where the magnetic field is turned off before turning on the integrable toric code hamiltonian , the the toric code hamiltonian is turned on in the presence of a small magnetic field @xmath18 . we can not perform exact analytical calculations in this case . we can however discuss the effects of the magnetic field by calculating a formal expansion for the ( renyi ) entanglement and topological entropies in powers of @xmath18 . the expansion can only be considered as a perturbative treatment if the toric code coupling constants are quickly pushed above @xmath18 and the time scales considered are not too large . two main insights are derived from this analysis . first we find indications as to how breaking the integrability can eventually lead to volume scaling in the entanglement entropy . the physical picture is that in the expansion in powers of @xmath18 , the contribution to the entanglement entropy comes from ring - shaped areas around the subsystem boundary whose thickness grows at higher and higher orders in @xmath18 . eventually , at a high enough order , these correlated regions meet and the terms in the expansion no longer respect the area law . even for a small @xmath18 , these high order terms become important at long times making volume scaling of the entanglement entropy possible as expected of thermalized states . secondly , we find that in this expansion in powers of @xmath18 , the topological entanglement entropy remains zero at low orders until we reach an order proportional to the system size , suggesting that a small integrability - breaking term is not enough for generating topological entanglement . a useful notion that we introduce in this section is the entanglement entropy in the interaction picture . we are used to the shcrdinger , heisenberg and interaction pictures for calculating the expectation values of observables , which are of course independent of the picture . if one defines the entanglement entropy as a property of the wave function alone , then since the wave function does not change in the heisenberg picture , calculating the entanglement entropy in that picture leads to the nonsensical result that time evolution does not change the entanglement of a quantum state . an interesting alternative is to try and define the entanglement entropy entirely in terms of the correlation functions of appropriate operators . we do not pursue this path here . we argue instead that as explained below , the entanglement entropy calculated only from the wave function in the interaction picture can be a useful concept . let us first consider a _ non - interacting _ hamiltonian @xmath178 such as a paramagnet . consider a partitioning of the system into subsystems @xmath52 and @xmath51 . the evolution operator for @xmath178 can be written as @xmath179 where @xmath180 ( @xmath181 ) only acts on the degrees of freedom in @xmath52 ( @xmath51 ) . notice that acting with @xmath182 on any wave function does not change its entanglement properties . now consider a hamiltonian @xmath183 with the corresponding evolution operator @xmath184 . the wave functions at time @xmath185 in the schrdinger and interaction pictures are respectively given by @xmath186 now since acting with @xmath187 or @xmath188 on any wave function does not change its entanglement , the interaction picture wave function has the same entanglement entropy as the schrdinger picture . note however the calculation may be much simpler in the interaction picture . in the discussion above , we made use of the interaction picture entanglement entropy because we had a hamiltonian @xmath178 whose evolution operator did not change the entanglement entropy . at the next level of complexity , where using the interaction picture wave function may be useful for calculating the entanglement entropy , the evolution operator of the _ unperturbed _ hamiltonian ( @xmath187 ) may have another known property . for example it may only create entanglement entropy scaling with the subsystem area or it could be incapable of changing the topological entanglement . this is the case for the integrable toric code hamiltonian . now if we are interested in finding whether a perturbation can generate entanglement entropy with volume scaling or whether it can generate topological entanglement entropy , we can perform the calculation for the wave function @xmath189 in the interaction picture . let us go back to our quench problem and write the wave function in the interaction picture with @xmath190 and @xmath191 starting with the initial state @xmath192 with all the spins in the positive @xmath2 direction , the wave - function at time @xmath185 in the interaction picture is given by @xmath193 where @xmath194 is given by @xmath195 for @xmath196 for the simplicity of notation , we focus on the sudden quench and write @xmath197 as @xmath39 , bearing in mind however that for a more general protocol , we can substitute the integral at the end of the calculation . we can explicitly calculate @xmath198 and the result is @xmath199 with the operators @xmath200 , @xmath0 and @xmath201 defined as follows @xmath202{p1.eps}}\end{picture}+\begin{picture}(22,0)\put(2,-6){\includegraphics[width=0.035\textwidth]{p2.eps}}\end{picture}+ \begin{picture}(22,0)\put(2,-6){\includegraphics[width=0.035\textwidth]{p3.eps}}\end{picture}+\begin{picture}(22,0)\put(2,-6){\includegraphics[width=0.035\textwidth]{p4.eps}}\end{picture}\;\right ) \nonumber \\ d & = & \sum _ { d}\;\begin{picture}(22,0)\put(2,-6){\includegraphics[width=0.07\textwidth]{d.eps}}\end{picture}\qquad.\end{aligned}\ ] ] the sums are over all bonds , plaquettes and dominos and the graphical notation represents a product of pauli matrices . for example if the bonds in a plaquette are numbered clockwise from @xmath203 to @xmath204 with the top bond being number @xmath203 we have @xmath205{p1.eps}}\end{picture}=\sigma_y^1\sigma_z^2\sigma_z^3\sigma_z^4 , \qquad \begin{picture}(22,0)\put(2,-6){\includegraphics[width=0.035\textwidth]{p2.eps}}\end{picture}=\sigma_z^1\sigma_y^2\sigma_z^3\sigma_z^4 , \qquad \cdots\ ] ] the closed form in eq . ( [ eq : vhat ] ) is simply derived from the repeated substitution of the following commutation relations @xmath206= 2 i p , \qquad \left[\sum_p b_p,\;p \right]= -4 i ( c+d)\\ & & \left[\sum_p b_p,\;c+d \right]= 4 i p\end{aligned}\ ] ] in the baker - hausdorff expansion of @xmath207 @xmath208\right.\\ & & \left . + \frac{(-i \lambda t)^2}{2\:!}\left[\sum_p b_p,\left[\sum_p b_p,\sum_i \sigma^x_i \right]\right]+\cdots\right ) \end{aligned}\ ] ] and resumming the series . the renyi entropy for the interaction picture wave function @xmath189 is given by @xmath209 where summation over repeated indices is implied and as before , the @xmath63 and @xmath64 spins belong to the @xmath52 and @xmath51 subsystems respectively . first let us consider the first order term in the @xmath210 ( the expression for the renyi entropy without the log ) . we use the shorthand notation @xmath211 and similarly for the hermitian conjugate . the first order term vanishes as seen below @xmath212 the second order contribution to @xmath210 is given by @xmath213.\end{aligned}\ ] ] let us assume that the whole system is on a torus and the total number of bonds ( spins ) is @xmath214 . the number of plaquettes @xmath215 and the number of dominos in the system is @xmath216 . let us introduce another shorthand notation for integrals involving the functions @xmath217 , @xmath15 and @xmath218 , @xmath219 so we have for example @xmath220 and @xmath221 . we denote the number of plaquettes ( dominos ) with all bonds in subsystem @xmath52 by @xmath222 ( @xmath223 ) and similarly by @xmath224 ( @xmath225 ) for subsystem @xmath51 . we then have @xmath226 the factors of @xmath227 come from the fact that each term in @xmath0 is a sum of four operators that flip the spins around a plaquette . noting that because @xmath228 is obviously symmetric under the exchange @xmath229 , we have @xmath230 and similarly for @xmath231 and @xmath232 . we then obtain @xmath233 which using @xmath234 gives @xmath235 this expression demonstrates the physical picture for building volume entanglement which we alluded to before . we observe from the above expression that at this order the contribution to the entropy is from a ring - shaped area of thickness two lattice spacings around the subsystem boundary . at higher orders , the thickness of this correlated region , which contributes to the entanglement entropy , grows until the ring closes in on itself , making volume scaling possible . notice that in contrast to exact area law which is possible in a finite system , since the renyi entropy is symmetric under exchanging the subsystems , volume scaling can only hold exactly for a finite subsystem in an _ infinite _ system and as a good approximation in the limit when a subsystem is much smaller than the whole system . it is a simple exercise in geometry to count @xmath236 and @xmath237 of the two subsystems for partitions shown in fig . [ fig : partitions ] and verify directly that no topological entanglement is generated at order @xmath238 . the details of this calculation will be given in appendix [ app_c ] . we will also show in that appendix that at the next order with a non - vanishing contribution to the renyi entropy ( @xmath239 ) , the topological term will still vanish due to an exact cancellation . similarly to the volume scaling , topological entanglement is non - perturbative and does not show up in an expansion in powers of @xmath18 in the thermodynamic limit . for a finite system , we expect non - vanishing contributions to the topological entropy to appear at orders that scale with the system size . let us comment here that on physical grounds , in the case of a sudden quench , we do not expect any topological entanglement to form even at infinitely long times . a trend toward forming volume entanglement observed at low orders suggests that in the limit of large times , the generic thermalization behavior is expected . although we can not give a mathematical proof , the fragility of topological order to thermal fluctuations @xcite and the extensive excess energy in the system suggests that the non - equilibrium state , after a sudden quantum quench , will likely bear no signature of the equilibrium topological order of the phase . we emphasize that the expansion used here can not be treated as a perturbative treatment in the adiabatic limit where the toric code coupling constants are turned on slowly because in that case @xmath240 for short times . we know that in the case of adiabatic evolution , we approximately remain in equilibrium , i.e. in the ground state manifold , and topological order emerges after a time of order system size . we studied the quench dynamics of the entanglement entropy in the kitaev toric code model . in the case of the integrable quench we found the entropy exactly as a function of time . the leading order contribution was found to scale as the subsystem perimeter with an oscillatory behavior that gets suppressed in the presence of some disorder in the coupling constants leading to the saturation of the entanglement entropy to disorder - independent values respecting the area law . we obtained results for the generated entanglement entropy as a function of an initial paramagnetic state with arbitrary magnetization . we studied the effect of the sharp corners in the subsystem boundary and found that in analogy with the static entanglement discussed in the literature , they make subleading contributions to the dynamically generated entanglement entropy as well . we showed that the topological entanglement entropy remains zero in this quantum quench . the results of the integrable quench as well as the analysis of the integrability - breaking perturbations strongly suggest that the non - equilibrium state after a quench into the topological phase does not develop any of the topological structure present in the equilibrium phase . we used the concept of the entanglement entropy in the interaction picture which simplifies perturbative approaches to the calculation of the entanglement entropy . this allows us to tease apart in this particular problem the effects of the time evolution with an unperturbed hamiltonian that can not generate either entanglement with volume scaling or topological entanglement and focus only the part of the evolution that can possibly lead to volume scaling or topological entanglement . by studying the effects of an integrability - breaking perturbation , we found a mechanism where volume scaling of the generated entanglement entropy can emerge at high orders in perturbation theory as expected of thermalized states . the notion of thermalization after a quench naturally ties in with topological order through the fragility of topological order to thermal fluctuations . we are grateful to c. castelnovo , a. hamma , m. haque , a. polkovnikov and m. rigol for helpful discussions . this work was supported by in part by the doe grant de - fg02 - 06er46316 . we explicitly calculate in this appendix the contributions to the entanglement entropy from regular boundary stars and plaquettes as well as the convex and concave corners when the initial magnetization is in the @xmath141 direction . first let us find the number of regular boundary stars ( @xmath241 ) and plaquettes ( @xmath242 ) in terms of the subsystem perimeter and the number of convex and concave corners . we define the perimeter @xmath0 as the total number of boundary plaquettes or boundary stars ( they are equal ) . we then have @xmath243 once again we use eq . ( [ eq : tr_rho_a_2 ] ) which generally holds for any initial state @xmath49 . the only intermediate states ( in the @xmath142 basis ) which contribute to the @xmath62 are the ones connected to @xmath244 by some star or plaquette flips . we label these states by variable @xmath245 and @xmath246 living on the boundary plaquettes and stars respectively . a negative @xmath247 ( @xmath248 ) indicates that the corresponding plaquettes ( star ) is flipped with respect to @xmath249 . @xmath250 where @xmath251 , @xmath252 , @xmath253 and @xmath254 label regular plaquettes , regular stars , convex corner plaquettes and concave corner stars respectively . for @xmath255 , @xmath256 ( @xmath257 ) are variables living on the two stars ( plaquettes ) corresponding to the convex ( concave ) corners . notice unlike regular star variables the two stars at a convex corner are labeled by the corner plaquette and similarly the two plaquettes at a concave corner are labeled by a corner star since a convex ( concave ) corner uniquely corresponds to a corner plaquette ( star ) . now suppose two such states have the same @xmath64 spins . all regular and concave corner plaquette and star flips must be the same for these two states . as seen in fig . [ fig : ycorner ] however , at a convex corner the @xmath247 or @xmath248 variables for these two states could be different if both the convex corner plaquette and the corresponding two corner stars simultaneously have opposite flips . then for two states @xmath88 and @xmath258 ( which can be labeled as eq . ( [ eq : labeling ] ) because they contribute @xmath62 ) all their @xmath247 and @xmath248 variables must be the same except for the ones corresponding to convex corners , namely @xmath259 and @xmath260 which follow the relation @xmath261 with auxiliary variable @xmath262 . similarly we can introduce variables @xmath263 for concave corner and for two states @xmath88 and @xmath264 we must have @xmath265 introducing a replica index @xmath266 on @xmath267 and @xmath268 , we can write @xmath269 let us for the moment focus on the renyi entropy @xmath270 for simplicity . the calculation for @xmath271 will be very similar . first we can see from the form of the the states above that the sign factors from @xmath80 acting on states in the @xmath142 basis have no effect . a @xmath80 plaquette flip can only give a different sign for the matrix elements involving @xmath94 and @xmath272 at a corner . but then the plaquette flip at that corner would also give a different sign for the matrix elements involving @xmath273 and @xmath274 and the overall sign for the product of the four matrix elements appearing in @xmath275 will be positive . the calculation is then similar to the simpler cases with the initial magnetization in the @xmath2 and @xmath3 directions except for each convex or concave corner there is an additional auxiliary variable to sum over . again , each matrix element can be written as a product similar to eq . ( [ eq : matrix - element ] ) . by simply bringing together terms we can write @xmath275 in the following form : @xmath276 where the summation is over all @xmath247 , @xmath248 , @xmath277 and @xmath278 variables for both replicas . we can take the sum eq . [ eq : pre - factorization ] inside the products by switching their order and perform the sum for individual terms . this gives the factorized form below @xmath279 where we will compute the @xmath280s below . assuming no disorder and taking the log of both sides , we immediately obtain our main result for the entropy @xmath281 which conforms to the general expression eq . ( [ eq : entropy - corners ] ) . for regular stars and plaquttes , the terms appearing in the above eq . ( [ eq : pre - factorization ] ) are identical to what we obtained when the initial spins were in the @xmath2 or @xmath3 direction . for example @xmath282 let us now define the following function for a more compact notation @xmath283 we can then write @xmath284 ^ 2,\qquad f_{s_r}=\left[g(\psi_{s_r},\mu_{s_r}t)\right]^2.\ ] ] at the corners , we have a more complicated structure and for example , @xmath285 , in addition to @xmath286 , also depends on @xmath287 and @xmath288 for @xmath255 . notice that for the renyi entropy @xmath289 and therefore @xmath290 . we can explicitly write @xmath291 it can be easily shown that @xmath292 using the above identity we can write @xmath293 a similar expression with @xmath294 , @xmath295 and @xmath296 gives @xmath297 . we can now perform the summations and we obtain @xmath298 with @xmath166 for the renyi entropy and similarly for @xmath299 with @xmath295 . to calculate @xmath300 we first perform the sum over @xmath247s and @xmath248s and then over @xmath277s and we obtain after some algebra @xmath301 with @xmath166 for the renyi entropy and @xmath302 , @xmath303 , @xmath304 and @xmath305 . we can similarly find @xmath306 for concave corners by switching @xmath295 . the generalization to @xmath307 is similar and leads to the appearance of a different @xmath111 in eq . ( [ eq : convex_f ] ) and eq . ( [ eq : regular_f ] ) . in this appendix , we discuss the transfer matrix method used in calculating the renyi entropy for an arbitrary initial magnetization in the positive @xmath308 direction . the initial state can be written as @xmath309 where @xmath310 and @xmath311 are the eigenstates of @xmath33 . let us begin by writing an explicit equation for @xmath312 . using @xmath166 in eq . ( [ eq : tr_rho_a_2 ] ) we can write @xmath313 consider the first of the four matrix elements appeasing in the equation above , namely @xmath314 . we represent the @xmath94 spin configuration in a basis where boundary spins , i.e. spins belonging to a boundary star or plaquettes , are in the @xmath169 basis and all other spins in the @xmath315 basis . in a schematic notation explained below , we can then write the matrix element as @xmath316\end{aligned}\ ] ] where in @xmath317 and @xmath318 we can more generally make the substitution @xmath319 . the notation in the above equation needs some explanation . as before we have introduced ising variables @xmath320 on the @xmath0 boundary plaquettes . for a given configuration of these variables @xmath92 , @xmath321 and @xmath322 are the number of flipped and unflipped boundary plaquettes respectively and @xmath323 and @xmath324 are the number of up and down _ boundary _ spins in the state obtained from @xmath325 after flipping the plaquettes represented by @xmath92 . we have introduced a new notation where @xmath63 denotes only the boundary spins in subsystem @xmath52 that belong to a boundary plaquette and the other boundary spins in a , which only belong to a boundary star , are denoted by @xmath326 . similarly boundary spins in subsystem @xmath51 that belong to a boundary star are labeled @xmath64 and the rest @xmath327 as seen in fig . [ fig : boundary_spins ] . assuming that the total number of boundary spins is @xmath51 , we can relate @xmath328 and @xmath329 to the spins variables above as follows @xmath330 and @xmath331 in the sums above each @xmath63 and @xmath327 spin is multiplied by the ising variable @xmath247 of the boundary plaquette it belongs to . each @xmath64 spin belongs to two boundary plaquettes and is therefore multiplied by the two corresponding @xmath247 variables . in the notation explained above , we now write the term @xmath332 as follows @xmath333 where @xmath334 like the one - dimensional classical ising model in a magnetic field , we can now integrate out the @xmath247 ising variables using the transfer matrix method . the terms containing @xmath247 are promoted to @xmath335 transfer matrices where each transfer matrix @xmath336 is between two adjacent boundary plaquettes and @xmath337 the other degrees of freedom can be systematically integrated out by repeatedly using the following trace identity @xmath338.\ ] ] repeated use of the identity eq . ( [ eq : identity ] ) leads to enlarged transfer matrices as discussed below . tracing out all degrees of freedom will lead to @xmath339 transfer matrices the trace of whose product gives the renyi entropy . notice that at concave and convex corners , we get a different @xmath335 transfer matrix . the @xmath335 transfer matrices at the corners will be given at the end of this appendix for completeness . we discuss below the procedure for obtaining the final @xmath339 transfer matrices from the @xmath335 ones along a smooth boundary . at the sharp corners , we can obtain the @xmath339 transfer matrices from the corresponding convex and concave corner @xmath335 matrices essentially in the same way . let us now explicitly write @xmath336 in eq . ( [ eq : two - by - two ] ) along a smooth boundary . for a bond @xmath340 between two adjacent boundary plaquettes as show in the top left corner of fig . [ fig : transfer_all ] , we have @xmath341 \times } \\ & & \left ( \begin{array}{ll } h^{\frac{1}{2}}a^{\frac{1}{2}({\beta^\prime}^i+\alpha^i)}&0 \\ 0&h^{-\frac{1}{2}}a^{-\frac{1}{2}({\beta^\prime}^i+\alpha^i ) } \end{array } \right ) \left(\begin{array}{ll } a^{\frac{1}{2}\beta^i}&a^{-\frac{1}{2}\beta^i } \\ a^{-\frac{1}{2}\beta^i}&a^{\frac{1}{2}\beta^i } \end{array } \right).\end{aligned}\ ] ] the next step is to consider @xmath342 we can now use the trace identity eq . [ eq : identity ] to bring together terms with the same @xmath64 and @xmath327 and trace out these degrees of freedom to get @xmath343 where the @xmath344 transfer matrix above is given by @xmath345 t_1\ : t_2 \ : t_3 \ : t_4\ ] ] with the following @xmath346 matrices @xmath347 using the trace identity above to bring together the degrees of freedom via the tensor product of matrices is the main strategy for calculating the renyi entropy . these calculations look tedious but are greatly simplified using mathematica . the next step is to bring together the same degrees of freedom in @xmath348 using the trace identity which results in @xmath349 matrices that only depend on @xmath63 and @xmath326 degrees of freedom for both replicas . notice that so far we have integrated out all @xmath64 and @xmath327 degrees of freedom ( for both replicas ) . now since the @xmath326 degrees of freedom do not interact we can do the sum directly . the @xmath63 degrees of freedom have an ising type interaction so the transfer matrix method can be applied again promoting functions of @xmath63 ( which appear in @xmath349 matrices ) to @xmath350 matrices ( because there are @xmath351 replicas ) . this operation can be done very simply in mathematica . at the end of the day the renyi entropy will be given by the trace of the product of @xmath0 ( the number of boundary plaquettes ) @xmath339 transfer matrices . if the boundary of the subsystem is largely smooth , i.e. only has a few sharp corners , the different transfer matrices at the corners will give an order - one contribution to the renyi entropy and the leading order area law term is @xmath352 where @xmath353 is the largest eigenvalue of the smooth boundary @xmath339 transfer matrix constructed above . an exact calculation of the entanglement entropy is needed for evaluating the topological entropy . therefore corrections from concave and convex corners must be taken into account by using different transfer matrices at the corners in the boundary . there is a small complication due to the the fact that a convex ( concave ) corner plaquette has an additional @xmath63 ( @xmath64 ) spin . let us review our method : we first consider two matrix elements each given by the trace of a product of @xmath335 matrices , we then bring together the same degrees of freedom and explicitly sum over @xmath64s to get the @xmath350 transfer matrices . now having two @xmath64s to sum over does not change anything except we have to do a few more summations to get @xmath354 . with our method of first integrating out the @xmath51 spins and then the @xmath52 spins , the additional @xmath63 spin at a convex corner gives a jump in the indices of the transfer matrices at the end of the day but this can be easily accounted for by inserting a matrix @xmath355 between the transfer matrices and calculating the trace of their product as usual . let us now write the @xmath335 matrices for the corners . first consider the transfer matrix @xmath356 for entering a convex corner as seen in the top right corner of fig . [ fig : transfer_all ] . each transfer matrix is written for a @xmath64 bond and the @xmath63 spins before ( @xmath357 ) and after @xmath358 that bond . as mentioned above the discontinuity in the @xmath63 index ( @xmath358 for a transfer matrix is not equal to @xmath357 for the next transfer matrix ) can be taken care of by inserting the matrix @xmath359 ( @xmath350 because we have two replicas ) at the end of the day between the @xmath360 transfer matrices . for exiting a convex corner plaquette , with the notation of the bottom left corner of fig . [ fig : transfer_all ] , we can write a @xmath361 transfer matrix @xmath362 and similarly a transfer matrix @xmath363 for a concave corner ( bottom right corner of fig . [ fig : transfer_all ] ) . in terms of the spin variables shown on the corresponding corner of fig . [ fig : transfer_all ] , these matrices are given by @xmath364\times \left ( \begin{array}{ll } h^{\frac{1}{2}}a^{\frac{1}{2}({\beta^\prime}^i+\alpha^i_{\rm b})}&0 \\ 0&h^{-\frac{1}{2}}a^{-\frac{1}{2}({\beta^\prime}^i+\alpha^i_{\rm b } ) } \end{array } \right ) \left(\begin{array}{ll } a^{\frac{1}{2}\beta^i}&a^{-\frac{1}{2}\beta^i } \\ a^{-\frac{1}{2}\beta^i}&a^{\frac{1}{2}\beta^i } \end{array } \right ) \left(\begin{array}{ll } a^{\frac{1}{2}\alpha^i_{\rm a}}&0 \\ 0&a^{-\frac{1}{2}\alpha^i_{\rm a } } \end{array } \right ) \nonumber\\ t^{r>}_{2\times 2}&= & a^ { \frac{1}{2 } { \alpha^\prime}^i } \exp \left [ i \mu t\ : { \alpha^\prime}^i \beta^i \alpha^i_{\rm b } \alpha^i_{\rm a } \right ] \times \left ( \begin{array}{ll } h^{\frac{1}{2}}a^{\frac{1}{2}\alpha^i_{\rm b}}&0 \\ 0&h^{-\frac{1}{2}}a^{-\frac{1}{2}\alpha^i_{\rm b } } \end{array } \right ) \left(\begin{array}{ll } a^{\frac{1}{2}\beta^i}&a^{-\frac{1}{2}\beta^i } \\ a^{-\frac{1}{2}\beta^i}&a^{\frac{1}{2}\beta^i } \end{array } \right)\nonumber\\ t^<_{2\times2}&= & \exp \left [ { i \mu t\ : \alpha^i_{\rm b } \alpha^i_{\rm a } \beta^i \beta^{i+1 } } \right ] \times \left ( \begin{array}{ll } h^{\frac{1}{2}}a^{\frac{1}{2}(\beta_i+{\beta^\prime}^i+\alpha^i_{\rm b})}&0 \\ 0&h^{-\frac{1}{2}}a^{-\frac{1}{2}(\beta_i+{\beta^\prime}^i+\alpha^i_{\rm b } ) } \end{array } \right ) \left(\begin{array}{ll } a^{\frac{1}{2}\beta^{i+1}}&a^{-\frac{1}{2}\beta^{i+1 } } \\ a^{\frac{1}{2}\beta^{i+1}}&a^{-\frac{1}{2}\beta^{i+1 } } \end{array } \right).\end{aligned}\ ] ] notice that for a concave corner , the transfer matrix is written between two neighboring boundary plaquettes which in this case do not have a common edge because the plaquette in the middle lies entirely in subsystem @xmath51 and is therefore not a boundary plaquette . the procedure for going from the initial @xmath361 transfer matrix to the final @xmath360 is analogous to the smooth boundary case discussed above and a detailed discussion would not be illuminating . we have explicitly verified using the transfer matrix method described in this appendix that the order - one correction to the area law is entirely from the convex and concave corners and the topological entanglement entropy remains identically zero . we stated in section [ sec : non - integrable ] that eq . ( [ eq : second - order ] ) gives zero topological entanglement at order @xmath238 . consider the partitions of fig . [ fig : partitions ] with the spins inside ( but not on the boundary of ) the blue regions belonging to subsystem @xmath51 and all other spins to @xmath52 . for partition i for example , the outer and inner boundaries of the subsystem @xmath51 are squares of side @xmath366 and @xmath367 respectively . the whole system is on an @xmath368 torus . in terms of these dimensions , we can explicitly find the quantities @xmath369 , @xmath370 , @xmath371 and @xmath372 for the four different partitions and we observe that the value of each quantity for partitions ii and iii , which are obviously equal , is the average of those for partitions i and iv . this immediately implies that the topological entanglement entropy identically vanishes at second order in @xmath18 . for completeness we include the results of this simple geometry exercise for partitions i and iv below @xmath373 where we have assumed @xmath374 . next , we discuss the fourth order term . the combinatorics get more complicated at higher orders but we can calculate the terms in the expansion by direct enumeration . this gives semi - analytical results with the time - dependence entering analytically through the multiple integrals depending on functions @xmath217 , @xmath15 and @xmath218 and the geometry dependence computed for a system of given dimensions by computer enumeration . we have performed these calculation for several systems . here we present the details for an @xmath375 system shown in fig . [ fig : partition8by8 ] . if @xmath376 , we have @xmath377 let us assume we have coefficients @xmath378 and @xmath379 for partition @xmath380 . we can then write the fourth order term in the topological renyi entropy as @xmath381 . \end{aligned}\ ] ] to find the fourth order term in @xmath382 , in addition to the @xmath7 coefficients that we found before , we need to calculate the coefficients @xmath383 , @xmath384 and @xmath385 for the different partitions . looking at the structure of the eq . ( [ eq : renyi - explicit ] ) , we observe that at fourth order we need to insert four @xmath386s into the four bins @xmath387 which give a total of @xmath388 terms of five different categories . there is one term of type @xmath389 ( with one @xmath386 in each @xmath390 ) and @xmath391 @xmath392 , @xmath393 @xmath394 , @xmath391 @xmath395 and @xmath204 @xmath396 terms . first consider the @xmath389 term . we define a matrix @xmath397 such that @xmath398 and obtain @xmath399.\ ] ] now since one @xmath386 term acting on @xmath400 can only change the state by flipping either one plaquette or one domino , the @xmath84 states with a non - vanishing @xmath401 can be enumerated and the matrix @xmath397 can be constructed . the elements of the matrix @xmath397 contain integrals @xmath402 , @xmath403 and @xmath404 depending on which term in @xmath386 contributes to the matrix element . we then obtain for example the following expression for partition 1 of the @xmath375 system @xmath405}.\end{aligned}\ ] ] we can similarly calculate @xmath406 for other partitions . the next two categories are @xmath407 and @xmath394 . the expression for @xmath408 and @xmath409 are given below . @xmath410 and @xmath411 computing these expressions in terms of the functions @xmath217 , @xmath218 and @xmath7 with direct enumeration yields the following expression for the topological renyi entropy . @xmath412\\ & & + ( d)^4 - 4(d , d)^2 -16\left[(p , d)+(d , p)\right]^2\\ & & -48(p)^4 + 256(p)^2(p , p)-320(p , p)^2 \big{\rbrace } . \end{aligned}\ ] ] notice that the last category @xmath413 is independent of the partitions and does not contribute to the topological entanglement entropy at fourth order in @xmath18 . we can show that the expression in eq . ( [ eq : renyi - topo ] ) above identically vanishes using @xmath414 , @xmath415 and the following integral identity @xmath416 m. greiner _ et al_. , nature ( london ) * 419 * , 51 ( 2002 ) ; a. k. tuchman _ et al_. , phys . a * 74 * , 051601 ( 2006 ) ; t. kinoshita , t. wenger , and d. s. weiss , nature ( london ) * 440 * , 900 ( 2006 ) ; s. hofferberth _ et al_. , _ ibid_. * 449 * , 324 ( 2007 ) . a. polkovnikov and v. gritsev , nat . phys . * 4 * , 477 ( 2008 ) ; r. barankov and a. polkovnikov , phys . * 101 * , 076801 ( 2008 ) ; c. de grandi , r. a. barankov , and a. polkovnikov , phys . rev . lett . * 101 * , 230402 ( 2008 ) . m. levin and x .- g . wen , phys . lett . * 96 * , 110405 ( 2006 ) . s. trebst , p. werner , m , troyer , k shtengel and c. nayak , phys . * 98 * , 070602 ( 2007 ) . j. vidal , s. dusuel and k. schmidt , phys . b * 79 * , 033109 ( 2009 ) ; j. vidal , r. thomale , k. schmidt and s. dusuel , phys . b * 80 * , 081104 ( 2009 ) . i. s. tupitsyn , a. kitaev , n. v. prokofev , and p. c. e. stamp , 82 , 085114 ( 2010 ) . m. rigol , v. dunjko , v. yurovsky and m. olshanii , phys . lett . * 98 * 050405 ( 2007 ) .

using the thermal quasiparticle random phase approximation , we study the process of neutrino and antineutrino capture on hot nuclei in supernova environments . for the sample nucleife andge we perform a detailed analysis of thermal effects on the strength distribution of allowed gamow - teller transitions which dominate low - energy charged - current neutrino reactions . the finite temperature cross sections are calculated taking into account the contributions of both allowed and forbidden transitions . the enhancement of the low - energy cross sections is explained by considering thermal effects on the gt strength . forfe we compare the calculated finite - temperature cross sections with those obtained from large - scale shell - model calculations .

1507.06872

it is well known that core - collapse supernova simulations require a detailed description of neutrino transport including all potentially important neutrino reactions @xcite . neutrinos are the mediators of the energy transfer from the core to the outer stellar layers and their luminosities and spectra are a crucial ingredient for the supernova explosion mechanism . despite significant progress in our understanding of the core - collapse mechanism , many supernova simulations fail to produce explosion ( see e.g. @xcite and references therein ) . one of the possible reasons for this failure could be the incomplete and inaccurate treatment of the neutrino - nucleus processes occurring in supernova environments . it was pointed by haxton @xcite that neutral- and charged - current neutrino reactions on nuclei involving the excitation of the giant resonances might help reenergize the explosion and they should , therefore , be taken into account in supernova simulations . such simulations were performed by bruenn and haxton @xcite and it was shown that inelastic neutrino - nucleus scattering plays an important role in equilibrating neutrinos with matter . however , the simulations did not confirm haxton s suggestion that neutrino - nucleus reactions might help to revive the stalled shock wave . in @xcite , the nuclear composition of the core was approximated by a single representative nucleus , @xmath0fe , and the relevant cross sections and rates were evaluated considering only allowed ( gamow - teller ) and first - forbidden upward transitions from the nuclear ground state . however , in the hot supernova environment with temperatures @xmath3 mev nuclei exist as a thermal ensemble , i.e. , nuclear excited states are thermally populated according to the boltzmann distribution . as was first realized in ref . @xcite , downward transitions from excited states remove the reaction threshold and can significantly enhance the reaction cross section at low neutrino energies . from a microscopic point of view , there are two ways how to treat neutrino reactions with hot nuclei . the first one involves a state - by - state summation over boltzmann - weighted , individually determined contributions from nuclear ground and excited states . to apply this method , one needs to know the strength distribution of electro - weak transition operators for thermally populated nuclear states . the second method is based on a statistical formulation of the nuclear many - body problem . in this approach rather than computing individual strength distributions one determines an `` average '' temperature - dependent strength function . for iron group nuclei ( @xmath4 ) the first approach was developed in @xcite on the basis of large - scale shell - model ( lssm ) diagonalization calculations . modern high - performance compute capabilities combined with state - of - the - art diagonalization approaches are able to provide detailed strength distributions for both charge - neutral and charge - changing gamow - teller transitions that dominate neutrino reactions with @xmath5-shell iron group nuclei at low neutrino energies ( @xmath6 ) . however , at temperatures @xmath7 an explicit state - by - state summation over all thermally populated states presently remains computationally prohibitive . to overcome this problem the `` brink hypothesis '' is applied , i.e. , it is assumed that gt distributions on nuclear excited states are the same as for the ground state . to account for thermal effects , the giant resonances built on the ground and low - lying daughter states in the inverse reaction are considered . such excited states are called back - resonances and their importance arises from the large nuclear matrix elements and increased phase space . thus , within lssm calculations the temperature dependence of the cross sections is comprised solely in the back - resonance terms . within the shell model approach it is assumed that the gt distributions on thermally excited states are the same as for the ground state . this is not likely the case as the vanishing of pairing correlations and smearing of the fermi surface with increasing temperature should affect the distribution centroid and move it slightly down in energy . this conjecture is confirmed by shell model monte carlo studies performed at finite temperature @xcite . in addition , with present compute capabilities the shell - model diagonalization can not be applied for nuclei beyond the @xmath5-shell because of the huge configurational spacse involved . these shortcomings can be avoided in a thermal quasiparticle random - phase approximation ( tqrpa ) . in @xcite , the tqrpa was proposed as a method to study the response of hot nuclei to weak external perturbations in the framework of a statistical approach . being based on the thermo - field dynamics ( tfd ) formalism @xcite , the tqrpa enables the computation of temperature - dependent strength function avoiding the assumption of brink s hypothesis . in @xcite , the tqrpa was applied to study thermal effects on neutral - current inelastic neutrino scattering on @xmath8fe and neutron - rich germanium isotope @xmath1ge . it was shown that the tqrpa reveals the same thermal effects on the cross sections as the lssm approach . namely , the reaction threshold for inelastic neutrino - nucleus scattering is removed at finite temperature and the cross section for low - energy neutrinos is significantly enhanced . it was found , however , that within the tqrpa the enhancement is caused by both de - excitation of nuclear excited states and thermally unblocked low - energy gt transitions . the latter do not appear within the lssm due to application of brink s hypothesis . moreover , it was shown in @xcite that , unlike in the lssm approach , the principle of detailed balance is not violated within the tqrpa and it results in a larger strength for downward gt transitions from excited states . as a consequence , at low neutrino energies the finite temperature cross sections calculated within the tqrpa turn out to be several times larger than those obtained within the shell - model calculations . in the present paper , we apply the tqrpa method to study thermal effects on charged - current neutrino - nucleus reactions @xmath9 and @xmath10 occurring in the supernova environment . here we would like to mention that the tqrpa was already applied for charge - changing reactions with hot nuclei when studying thermal effects on electron capture in supernovae @xcite . the paper is organized as follows : in sec . [ formalism ] we review the basics of the tfd formalism and outline how to treat charge - changing transitions in a hot nucleus within the tqrpa . for more details of the approach , the reader is referred to refs . in addition , in sec . [ formalism ] the expressions necessary to calculate cross - sections for @xmath11- and @xmath12-absorption on hot nuclei are given . the results of the numerical calculations are presented and discussed in sec . [ results ] for the sample nuclei @xmath0fe and @xmath1ge . for @xmath0fe we compare the calculated ground - state and finite - temperature cross sections with available results from other approaches . conclusions are drawn in sec . [ conclusion ] . in an appendix we prove in a model independent way that the principle of detailed balance is valid for charge - changing transitions if hot nuclei in the supernova environment are treated in the grand canonical ensemble . during the core - collapse phase of a supernovae explosion the temperature in the iron core is sufficiently high ( a few @xmath13 k ) to establish an equilibrium of reactions mediated by the strong and electro - weak interaction @xcite . neglecting weak - interaction reactions , we can consider nuclei as open quantum systems in thermal equilibrium with heat and particle reservoirs and , hence , they can be described as a thermal grand canonical ensemble with temperature @xmath14 and proton and neutron chemical potentials @xmath15 and @xmath16 , respectively . in tfd , such ensemble is represented by the thermal vacuum @xmath17 , which is a temperature - dependent state in the extended hilbert space . the thermal vacuum is determined as the zero - energy eigenstate of the thermal hamiltonian , @xmath18 , and it satisfies the thermal state condition @xmath19 in the above equations @xmath20 is the original nuclear hamiltonian ( proton and neutron chemical potentials are included in @xmath20 ) and @xmath21 is its tilde counterpart acting in the auxiliary hilbert space ; an operator @xmath22 acts in the physical hilbert space , @xmath23 is its tilde partner , and @xmath24 is a phase factor . the thermal state condition guarantees that the expectation value @xmath25 is equal to the grand canonical average of @xmath22 . in this sense , relation is equivalent to the kubo - martin - schwinger condition for an equilibrium grand canonical density matrix @xcite . external perturbations mediated by weak interaction induce transitions from the thermal vacuum to nonequilibrium states . within the tdf such nonequilibrium states are given by the eigenstates of the thermal hamiltonian @xmath26 . as follows from the definition of @xmath27 , each of its eigenstates with positive energy has a counterpart the tilde - conjugate eigenstate with negative but the same absolute value of energy . this gives the possibility to describe both endoergic and exoergic neutrino reactions with hot nuclei . in what follows we will refer to positive energy eigenstates as non - tilde states , and to negative energy eigenstates as tilde states . in @xcite we have introduced the proton - neutron tqrpa method which allows for a treatment of charge - changing transitions in hot nuclei induced by ( anti)neutrino absorption . for the sake of completeness , let us briefly recall the method . within the tqrpa , nonequilibrium states of a hot nucleus caused by an external perturbation are treated as phonon - like excitations on the thermal vacuum @xmath28 where we denote @xmath29 . phonon excitations are considered as the normal modes of the thermal hamiltonian : @xmath30 while the thermal vacuum @xmath17 itself is the vacuum for the annihilation operators @xmath31 , @xmath32 . thus , within the tqrpa the problem of finding the excitation spectrum of a hot nucleus is reduced to the diagonalization of the thermal hamiltonian in terms of phonon operators such that the respective phonon vacuum obeys the thermal state condition . for charge - changing multipole transitions the phonon operators are defined as a linear superposition of creation and annihilation operators of proton - neutron thermal quasiparticle pairs ^j_m$ ] denotes the coupling of two single - particle angular momenta @xmath33 to the total angular momentum @xmath34 . ] @xmath35^j_m + % \widetilde\psi^{j i}_{j_pj_p}[\widetilde\beta^\dag_{\overline{\jmath_p } } \widetilde\beta^\dag_{\overline{\jmath_n}}]^j_m \\ + % i\eta^{j i}_{j_pj_n}[\beta^\dag_{j_p } \widetilde\beta^\dag_{\overline{\jmath_n}}]^j_m % + i\widetilde \eta^{j i}_{j_pj_n}[\widetilde\beta^\dag_{\overline{\jmath_p } } \beta^\dag_{j_n}]^j_m \\ + % \phi^{j i}_{j_pj_n}[\beta_{\overline{\jmath_p}}\beta_{\overline{\jmath_n}}]^j_m % + \widetilde\phi^{j i}_{j_pj_n}[\widetilde\beta_{j_p } \widetilde\beta_{j_n}]^j_m \\+ % i\xi^{j i}_{j_pj_n}[\beta_{\overline{\jmath_p } } \widetilde\beta_{j_n}]^j_m % + i\widetilde\xi^{j i}_{j_pj_n}[\widetilde\beta_{j_p } \beta_{\overline{\jmath_n}}]^j_m \bigr\}.\end{gathered}\ ] ] in turn , thermal quasiparticles are normal modes of the pairing part of the thermal hamiltonian @xmath36 and their vacuum is the thermal vacuum in the bcs approximation . in the expression above , the notation @xmath37 implies a summation over neutron ( @xmath38 ) or proton ( @xmath39 single particle states only . thermal quasiparticles are connected with bogoliubov quasiparticles by so - called thermal transformation @xmath40 the @xmath41-coefficients as well as the @xmath42-coefficients of the bogoliubov transformations are found from the finite - temperature bcs equations ( see @xcite for more details ) . in accordance with the bcs theory @xcite , the numerical solution of these equations yields vanishing pairing correlations above a certain critical temperature . to clarify the physical meaning of different terms in , we note that the creation of a negative energy tilde thermal quasiparticle corresponds to the annihilation of a thermally excited bogoliubov quasiparticle or , which is the same , to the creation of a quasihole state ( see @xcite for more details ) . therefore , at finite temperature , single - particle charge - changing transitions involve excitations of three types : 1 ) two - quasiparticle excitations described by the operator @xmath43 and having energy @xmath44 , 2 ) one - quasiparticle one - quasihole excitations described by the operators @xmath45 , @xmath46 and having energies @xmath47 and @xmath48 , respectively , and 3 ) two - quasihole excitations described by the operator @xmath49 and having energy @xmath50 . the last two types are possible only at @xmath51 . therefore , due to single - particle transitions involving annihilation of thermally excited bogoliubov quasiparticles , the phonon spectrum at finite temperature contains negative- and low - energy states which do not exist at zero temperature and these `` new '' phonon states can be interpreted as thermally unblocked transitions between nuclear excited states . to find the energy and the wavefunctions of thermal phonons we apply the equation of motion method @xmath52|\rangle = \omega(t)\langle|[\delta q , q^\dag]|\rangle\ ] ] under two additional constraints : ( a ) the phonon vacuum obeys the thermal state condition , and ( b ) phonon operators obey bose commutation relations . the first constraint yields the following relations between phonon amplitudes @xmath53 while the last assumption is equivalent to averaging with respect to the bcs thermal vacuum in the equations of motion and leads to an orthonormality condition for the amplitudes @xcite . thus , the phonon energies @xmath54 in eq . as well as the amplitudes @xmath55 in eq . are the solution of the proton - neutron tqrpa equations . positive energy solutions correspond to non - tilde phonons in , while negative energy solutions correspond to tilde ones . since both the energies of thermal quasiparticles and the interaction strengths between them are temperature dependent , the spectrum of thermal phonons turns out to be temperature dependent . however , it is significant that in the zero - temperature limit the described method reduces into the standard proton - neutron qrpa . for a given multipole charge - changing transition operator @xmath56 the transition probabilities ( strengths ) from the thermal vacuum to thermal one - phonon states are given by the following reduced matrix elements @xmath57 here the symbol @xmath58 refers to neutron - to - proton transitions ( @xmath59 channel ) , while @xmath60 refers to proton - to - neutron transitions ( @xmath61 channel ) . it can be shown that if the transition operators @xmath62 and @xmath63 differ only by the isospin operator , i.e. , @xmath64 , the transition probabilities to tilde and non - tilde phonon states are connected as @xmath65 in @xcite , a similar relation is obtained for charge - neutral transitions and it is referred to as the principle of detailed balance . however , in contrast to the case of charge - neutral transitions , the detailed balance relation includes a phonon energy rather than a transition energy . the latter is the energy transferred to the parent nucleus and for charge - changing transitions is given by , the transition energy corresponds to the final state energy measured from parent - nucleus ground state . ] @xmath66 where @xmath67 , and @xmath68 is the difference between neutron and proton chemical potentials in the parent nucleus and @xmath69 mev is the neutron - proton mass splitting . it is obvious that non - tilde ( tilde ) charge - changing phonon states do not necessarily correspond to upward ( downward ) transitions : due to @xmath70 , some non - tilde ( tilde ) phonon states may have negative ( positive ) transition energy . with eq . , relation is rewritten as @xmath71 thus , for each @xmath72 ( @xmath73 ) upward transition with energy @xmath74 there is an inverse downward transition @xmath73 ( @xmath72 ) with energy @xmath75 and the respective transition probabilities are connected by . in the appendix we show that in this form the detailed balance for charge - changing transitions in the ensemble of hot nuclei can be derived in a model independent way . in the derivation of the ( anti)neutrino absorption cross section for a hot nucleus under supernova conditions we follow the walecka - donnelly formalism @xcite which is based on the standard current - current form of the weak interaction hamiltonian . after a multipole expansion of the weak nuclear current the temperature - dependent differential cross section for a transition from the thermal vacuum to a thermal one - phonon state is given by @xmath76 here and below , the upper ( lower ) sign refers to the neutrino ( antineutrino ) cross section . in the above expression , @xmath77 is the fermi constant for the weak interaction , @xmath78 is the cabbibo angle , and @xmath79 and @xmath80 are the energy and momentum of the outgoing electron or positron . in eq . , the contributions , @xmath81 , for the coulomb and longitudinal components , and @xmath82 , for the transverse electric and magnetic components , are written as @xmath83 \\ \times2\,\mathrm{re}\langle j i\| \hat l_j\|0(t)\rangle\langle j i\| \hat m_j\|0(t)\rangle^*,\end{gathered}\ ] ] @xmath84\mp\bigl[\frac{e_\nu+e^i_e}{q}(1-a\cos\theta)-c\bigr ] \\ \times 2\ , \mathrm{re } \langle j i\| \hat t^\mathrm{mag}_j\|0(t)\rangle\langle j i\| \hat t^\mathrm{el}_j\|0(t)\rangle^*,\end{gathered}\ ] ] where @xmath85 is the lepton scattering angle and the notation @xmath86 is used to denote both the non - tilde and the tilde thermal one - phonon states . in the former case the transition energy @xmath87 , while in the latter case @xmath88 . thus , the energy of the outgoing lepton is @xmath89 . the parameters @xmath90 and @xmath91 are obtained from the relations @xmath92 and the absolute value of the three - momentum transfer @xmath93 is given @xmath94 the multipole operators @xmath95 , @xmath96 , @xmath97 , and @xmath98 denote the charge , longitudinal , and transverse electric and magnetic parts of the hadronic current , respectively , as definedvin @xcite . for the vector , axial - vector , and pseudoscalar form - factors which describebthe internal structure of the nucleon we use parametrization from ref . @xcite ( see also ref . @xcite ) . for charged - current reactions , the cross section must be corrected for the distortion of the outgoing lepton wave function by the coulomb field of the residual ( daughter ) nucleus . the cross section can either be multiplied a fermi function @xmath99 ( see , e.g. , ref . @xcite ) , or , at higher energies , the effect of the coulomb field can be described by the effective momentum approximation ( ema ) @xcite . in the present study , by following the prescription from @xcite , we choose an energy point in which both approaches predict the same values . then the fermi function is used below the point and the ema is adopted above it . furthermore , in the supernova environment nuclei are surrounded by a nearly degenerate electron gas . thus , neutrino absorption can be strongly reduced by pauli blocking of the final electron states . the blocking factor @xmath100 in eq . accounts for this effects , where @xmath101 is the fermi - dirac distribution with temperature @xmath14 and the chemical potential @xmath102 . the positron distribution is defined in the same way with @xmath103 . therefore , we can neglect the blocking factor for antineutrino absorption . the total cross section @xmath104 , as a function of temperature and incoming ( anti)neutrino energy , is obtained from the differential cross sections by integrating over the scattering angle and summing over all possible final thermal one - phonon states @xmath105 here , the total cross section is split into two parts : @xmath106 describes the endoergic neutrino absorption and includes only upward transitions ( @xmath107 ) , while @xmath108 corresponds to the exoergic process associated with downward transitions ( @xmath109 ) . the contribution @xmath108 dominates the cross section for vanishing neutrino energies , @xmath110 . it is apparent that for beta - stable nuclei the exoergic absorption is only possible at finite temperatures and due to transitions from thermally excited states . for low - energy ( anti)neutrinos , i.e. , in the long wavelength limit @xmath111 , the allowed @xmath112 multipole operator reduces to the gamow - teller form @xmath113 where @xmath114 @xcite is the axial weak coupling constant . considering only gamow - teller transitions and taking into account detailed balance according to , the cross section can be written as @xmath115 here for simplicity the electron rest mass has been neglected , i.e. @xmath116 , and the blocking factor for the outgoing lepton has been dropped . the matrix elements @xmath117 denote the transition strength of the gt@xmath118 operator . the first term in eq . implies summation over non - tilde @xmath112 states . within the shell model approach this term reduces to the ground - state contribution ( see eq . ( 2 ) in ref . @xcite ) . however , in the tqrpa this term appears to be temperature dependent due to violation of brink s hypothesis . the second term accounts for transitions to tilde states and it is an analog of the back - resonance contribution within the shell - model approach . here , we would like to stress that we can not associate transitions to non - tilde ( tilde ) states with the endoergic ( exoergic ) absorption . as was mentioned above , due to @xmath70 some non - tilde ( tilde ) states may correspond to downward ( upward ) transitions and , hence , contribute to @xmath119 ( @xmath120 ) . we employ the theoretical framework presented above to study thermal effects on the @xmath9 and @xmath10 cross sections for two sample nuclei , @xmath121 and @xmath122 . the iron isotope is among the most abundant nuclei at the early stage of the core - collapse , while the neutron - rich germanium isotope can be considered as the average nucleus at later stages @xcite . to describe charge - changing transitions in a hot nucleus we use the same phenomenological nuclear hamiltonian as in @xcite . the hamiltonian consists of spherically symmetric woods - saxon potentials for protons and neutrons , bcs pairing interactions , and separable multipole and spin - multipole residuals interaction in the particle - hole channel . we neglect particle - particle interactions except for the bcs pairing forces . this hamiltonian is usually referred to as the quasiparticle - phonon model ( qpm ) @xcite . for the two nuclei considered , the parameters of the qpm hamiltonian are fixed in the same manner as in @xcite . here we just mention that solving the bcs equations at zero temperature we get the following proton and neutron pairing gaps : @xmath123 mev for @xmath0fe and @xmath124 mev for @xmath1ge . thus , the critical temperature ( @xmath125 ) when the pairing phase transition occurs is @xmath126 mev for the iron isotope and @xmath127 mev for the germanium isotope . a separable form of the residual interaction allows us to reduce the tqrpa equations to a secular equation . the explicit form for charge - changing excitations as well as expressions for transition strengths are given in @xcite . moreover , it has been proved in ref . @xcite that the tqrpa fulfills the ikeda sum rule for allowed fermi and gt transitions . as in @xcite , before proceeding to discuss thermal effects on neutrino - nucleus absorption cross sections , we consider the ground - state ( @xmath128 ) cross sections for @xmath129 , @xmath130 , @xmath131 , and @xmath132 reactions and compare the our results with those available from other theoretical studies . in fig [ gs_crsect ] , the calculated total cross sections are shown as functions of the incident ( anti)neutrino energy @xmath133 . we find that for @xmath134 the reactions considered are dominated by the multipole transitions @xmath135 , while contributions from higher multipoles are only a few percent of the total cross sections . for @xmath0fe the cross sections increase sharply as @xmath133 approaches the reaction threshold @xmath136 mev for @xmath11-absorption and @xmath137 mev for @xmath12-absorption ( @xmath138 , where @xmath139 are the masses of the parent and daughter nuclei ) . as the neutron number increases , the threshold energy for @xmath11-absorption decreases . for the neutron - rich nucleus @xmath122 the @xmath140-value becomes negative ( @xmath141 mev ) allowing neutrino absorption for all energies . contrary to this , for the @xmath10 reactions the @xmath142-values become less favorable with increasing neutron excess . the antineutrino has to overcome a noticeable threshold ( @xmath143 mev ) to be absorbed by @xmath122 and the corresponding cross section is considerably lower in comparison to the neutrino one . in addition to the total cross sections , contributions from the @xmath112 multipole channel are shown in fig . [ gs_crsect ] . referring to the plots in fig . [ gs_crsect ] for the @xmath129 , @xmath130 , @xmath131 reactions the cross sections are dominated by allowed @xmath112 transitions for energies up to @xmath144 mev , with contributions from other multipoles being much smaller . in the @xmath132 cross section , however , the @xmath112 contribution is negligible and forbidden transitions dominate the reaction . we find that @xmath145 and @xmath146 transitions mainly contribute to the cross section for @xmath147 mev . in fig . [ gs_crsect ] we also analyze the effect of the full @xmath93-dependent @xmath112 transition operator instead of its long - wavelength limit . as was mentioned above , in the latter case the @xmath112 transition operator reduces to the gamow - teller operator and the ground - state cross section is given by the first term in eq . . it should be noted that to make a comparison with the shell - model calculations @xcite more transparent , we use the same quenching factor for the axial weak coupling constant , @xmath148 . by comparing in fig . [ gs_crsect ] the @xmath112 and gamow - teller contributions to the cross sections , we conclude that for energies @xmath149 mev the application of the gt operators instead of the @xmath93-dependent @xmath112 operator is fully justified . therefore , we conclude that the low - energy ground - state cross sections for the @xmath129 , @xmath130 , @xmath131 reactions are completely dominated by gt transitions . in contrast , all gt@xmath150 transitions of are essentially blocked in @xmath1ge . for nuclei with @xmath151 and @xmath152 blocking occurs because the valence protons are in the @xmath5 shell , while the valence neutrons occupy already the next major shell ( @xmath153 shell ) . in the next section it will be shown , however , that thermal effects unblock gt@xmath150 transitions in @xmath1ge and for typical supernova temperatures the low - energy @xmath132 cross sections are also dominated by the gt contributions . .total @xmath129 cross sections for selected neutrino energies @xmath133 . the present qrpa results ( second column ) are compared with those from @xcite and with the hybrid approach results @xcite . the cross sections are given in units of @xmath154 , with exponents given in parentheses . [ cols="^,^,^,^",options="header " , ] [ 56fe_antineutrino ] in tables [ 56fe_neutrino ] and [ 56fe_antineutrino ] , we compare the calculated ground - state cross sections for ( anti)neutrino absorption by @xmath0fe to those obtained from self - consistent qrpa calculations with skyrme forces @xcite and with the hybrid approach ( large - scale shell - model calculations for gt contributions and rpa for other multipoles ) @xcite . as is seen , the three models , although based on different microscopic pictures , predict rather similar neutrino absorption cross sections . we note , however , that for low energies ( @xmath155 mev ) the present results are closer to the hybrid approach results than those of ref . it is evident that the reason why the qrpa and the hybrid approach cross sections are systematically lower than those of the skyrme based calculations at @xmath155 mev is caused by differences in the gt@xmath156 strength distributions . although the authors of ref . @xcite do not provide the gt@xmath156 distribution in @xmath0fe , it seems that the skyrme based calculations result in a larger strength below the gt@xmath156 resonance compared to the other two approaches . other possible reasons for the discrepancy could be a lower energy of the gt@xmath156 resonance and a larger total gt strength obtained from the calculations with the skyrme interaction . experimental results for neutrino absorption cross sections are rather limited . the karmen collaboration has measured the flux averaged @xmath157 cross section for the neutrino spectrum from the muon decay at rest and obtains @xmath158\times10^{-42}\,\mathrm{cm}^2$]@xcite . our result , @xmath159 , is in excellent agreement with the experimental value . note , that the hybrid approach predicts @xmath160 @xcite , while the qrpa calculations with skyrme forces gives @xmath161 @xcite . to the cross section for the @xmath162 reaction at @xmath163 mev . the present qrpa results are compared with those obtained from rqrpa @xcite and skqrpa @xcite calculations . ] for @xmath0fe a detailed demonstration of the most important multipole contributions to neutrino absorption cross section is presented in fig . [ partial ] at three neutrino energies , @xmath164 mev . as expected , at relatively low neutrino energies ( @xmath165 mev ) the dominant contribution to the cross section originates from @xmath112 transitions . with increasing @xmath133 , however , contributions from other multipole transitions become important . in particular , at @xmath166 mev , the dominant contribution comes from @xmath145 transitions , but other multipoles , e.g. , @xmath112 , @xmath146 , @xmath167 , and @xmath168 , also play an important role . in fig . [ partial ] , we also compare the calculated multipole decomposition of the cross section for @xmath0fe with those from the skqrpa calculations @xcite and the relativistic qrpa ( rqrpa ) calculations @xcite . as is evident from the figure , the latter model predicts somewhat smaller cross section at high neutrino energies ( @xmath169 mev ) , whereas we observe an excellent agreement between the results of the present qrpa and the skqrpa . specifically , in accordance with ref . @xcite , we find that @xmath170 allowed transitions only marginally contribute to the @xmath129 reaction . this finding is true for other three reactions and for finite temperatures as well . for this reason , in the discussion below , we will always imply @xmath112 multipole channels when referring the allowed transitions . in discussing @xmath11 and @xmath12 absorption reactions under supernova conditions we will follow the line of our recent work @xcite and consider first thermal effects on the strength distribution of gt transitions which dominate the reactions at low energies . in fig . [ gt_56fe_temp ] , we display on a logarithmic scale the gt@xmath156 and gt@xmath150 distributions in @xmath0fe calculated for the ground - state ( @xmath128 ) and at three stellar temperatures above the critical one , @xmath171 , @xmath172 , and @xmath173 . these temperatures roughly correspond to three stages in the collapse evolution @xcite . we emphasize that the distributions are plotted with the bare gt operators @xmath174 as functions of the transition energy . for each temperature we show the value of @xmath70 in eq . as well as the total transition strengths @xmath175 and @xmath176 . note that @xmath175 and @xmath176 values calculated for the ground - state satisfy the ikeda sum rule @xmath177 ( a small deviation is caused by the incompleteness of our single - particle basis ) but noticeably overestimate the experimental data ( @xmath178 @xcite , @xmath179 @xcite ) . this overestimation is common for any qrpa calculations of gt strength and is remedied by a quenched value for the axial weak coupling constant @xmath180 . one might notice that our qrpa calculations fairly well reproduce the experimental centroid energies for both gt@xmath150 @xcite and gt@xmath156 @xcite distributions in @xmath0fe . in this respect the present calculations are consistent with the large - scale shell - model calculations @xcite . of course , in contrast to the lssm approach , the qrpa can not recover all nuclear correlations needed to reproduce the fragmentation of the gt strength . ( left panels ) and gt@xmath150 ( right panels ) strength distributions for @xmath0fe vs transition energy . the solid ( dashed ) lines refer to transitions to non - tilde ( tilde ) thermal one - phonon states . the arrows indicate the ground - state reaction thresholds for neutrino ( @xmath136 mev ) and antineutrino ( @xmath137 mev ) absorption . ] for @xmath0fe , due to a relatively small absolute value of @xmath70 , the major part of the upward ( downward ) gt strength corresponds to transitions to non - tilde ( tilde ) phonon states . according to our qrpa calculations , the principal contribution to the gt@xmath156 ( gt@xmath150 ) strength in @xmath0fe comes from the @xmath181 neutron - to - proton ( proton - to - neutron ) single - particle transition which forms a resonance peak at @xmath182 mev ( @xmath183 mev ) . because the brink hypothesis is not valid within the tqrpa , we observe an evolution of the gt resonances with temperature . namely , when the temperature is increased to @xmath184 mev , the transition energy is lowered by @xmath185 mev for the gt@xmath156 resonance and @xmath186 mev for the gt@xmath150 resonance . this decrease in energy is mainly attributed to the vanishing of pairing correlations : at temperatures above the critical one no extra energy is needed to break a proton ( neutron ) cooper pair and as a consequence the gt@xmath150 ( gt@xmath156 ) resonance moves to lower energies . however , not only pairing effects lead to downward shift of the gt resonances . it can be easily seen from the structure of the tqrpa equations @xcite that quasiparticle thermal occupations factors will appear which screen the interaction term . due to the thermal blocking of the proton - neutron repulsive residual interaction , a further increase in temperature to @xmath187 mev decreases the gt@xmath156 and gt@xmath150 resonances in @xmath0fe by @xmath188 mev and @xmath189 mev , respectively . as was mentioned in the introduction , the observed downward shift of the gt strength is not present in large - scale shell - model calculations which are partially based on brink s hypothesis . in contrast , the finite - temperature relativistic qrpa calculations in ref . @xcite and shell - model monte - carlo calculations in ref . @xcite show similar features for the changes of the gt resonance energy . finite temperature also affects the low - energy gt@xmath156 strength in @xmath0fe : due to the gradual disappearance of pairing it partially shifts below the ground - state reaction threshold and becomes more fragmented . the fragmentation arises from the thermal smearing of the nuclear fermi surface that unblocks low - energy particle - particle and hole - hole gt@xmath156 transitions . here , particle ( hole ) denotes a state above ( below ) the fermi level . moreover , in accordance with detailed balance , the temperature rise exponentially increases the strength of negative - energy downward transitions . referring to the figure , the dominant part of the downward gt@xmath156 strength around @xmath190 mev originates from a transition inverse to the gt@xmath150 resonance , that is due to the transition from the thermally populated neutron orbit @xmath191 to the lower lying proton orbit @xmath192 . in contrast , transitions inverse to the gt@xmath156 resonance are suppressed by the energy - dependent exponent in eq . and the gt@xmath150 downward strength is dominated by transitions inverse to low - energy gt@xmath156 transitions . it should be mentioned that although thermal effects unblock some new gt transitions , the total gt@xmath118 strength in @xmath0fe slightly decreases with temperature . nevertheless , the tqrpa preserves the ikeda sum rule . it is obvious that the violation of brink s hypothesis should affect the downward gt strength . in @xcite , this influence was studied thoroughly for charge - neutral gt transitions in @xmath0fe by comparing the gt running sums derived with and without brink s hypothesis . applying the same method , we find that both the shift of the gt@xmath150 resonance to lower energies and the thermal unblocking of low - lying gt@xmath156 strength enhance the gt@xmath156 and gt@xmath150 downward strengths in @xmath0fe . but for @xmath1ge . due to pauli blocking of allowed @xmath73 transitions , the total gt@xmath156 strength is many times larger than the gt@xmath150 strength and @xmath193 . ] [ gt_82ge_temp ] shows the temperature evolution of the gt strength function for the neutron - rich nucleus @xmath1ge . the distributions are displayed at the same temperatures as those for @xmath0fe in fig . [ gt_56fe_temp ] . as evident from the figure , some gt strength associated with non - tilde ( tilde ) phonon states appears at negative ( positive ) transition energies which is due to a large value of @xmath70 . the temperature rise affects the gt@xmath156 and gt@xmath150 distributions in a different way . for the gt@xmath156 distribution , a temperature increase essentially affects only the downward part , yielding some new strength below @xmath194 mev . by comparing the left and right panels in fig . [ gt_82ge_temp ] , we conclude that this thermally unblocked strength corresponds to transitions inverse to the gt@xmath150 resonance . the gt@xmath156 resonance is weakly sensitive to thermal effects . by increasing the temperature to @xmath195 mev , the excitation energy gets only lowered by @xmath189 mev . since there are no neutron pairing correlations in @xmath1ge , this lowering is solely caused by the softening of the residual interaction . in contrast , thermal effect are significant for the gt@xmath150 resonance and they clearly demonstrate a violation of the brink hypothesis within the tqrpa . namely , with rising temperature the resonance peak shifts by @xmath196 mev to lower energies and the total gt@xmath150 strength shows a nonmonotomic dependence : after an initial decrease , it gradually increases . in @xcite , a detailed investigation within the tqrpa approach was performed of thermal effects on the gt@xmath150 strength in the neutron - rich nucleus @xmath197ge . the conclusions remain valid for @xmath1ge , as well . briefly , for neutron - rich nuclei with @xmath152 and @xmath151 the independent particle model predicts that all gt@xmath150 transitions of valence protons are pauli blocked due to the complete occupation of the @xmath5 neutron orbits . however , both thermal excitations and pairing correlations promote protons to the @xmath153 shell and make possible @xmath198 particle - particle transition in @xmath1ge possible . the important point is that the transition energy drastically decreases with temperature . namely , at @xmath199 , when pairing correlations are important , the transition energy is given by @xmath200 mev , while at @xmath201 , when thermal effects become significant , it is given by @xmath202 mev . in addition , the transition strength below the gt@xmath150 resonance becomes increasingly unblocked with temperature . in particular , a strong peak due to the @xmath203 hole - hole transition appears at @xmath204 mev . both unblocking mechanisms are severely suppressed in the vicinity of the critical temperature ( @xmath127 mev in @xmath1ge ) , i.e. when pairing correlations vanish while thermal effects are not yet sufficiently strong to significantly occupy the @xmath205 proton orbit or unblock the @xmath191 neutron orbit . therefore , the total gt@xmath150 strength decreases at @xmath206 . the thermal effects on the gt@xmath150 strength in neutron - rich nuclei discussed here were predicted in @xcite and also confirmed in @xcite by calculations based on the finite - temperature relativistic qrpa . we also note that , due to the large @xmath70 , downward gt@xmath150 transitions appears to be highly suppressed in @xmath1ge . fe and @xmath1ge at three different temperatures relevant for core collapse . for comparison , the ground - state cross sections are also shown . ( middle panels ) contributions from forbidden transitions to the cross sections . ( lower panels ) temperature dependence of the ratio of exoergic absorption to the reaction cross section . ] let us now illustrate the influence of the thermal effects discussed above on the neutrino and antineutrino absorption by @xmath121 and @xmath122 . in fig . [ nnb ] , we compare the @xmath207 ground - state cross sections with those calculated at the three core - collapse temperatures . for energies @xmath208 , which are typical for supernova neutrinos , the cross sections are dominated by allowed gt@xmath156 transitions at all temperatures . this is verified in the middle panels of fig . [ nnb ] , where the overall contribution of forbidden transitions to the finite temperature cross sections is given . although the forbidden cross section increases with temperature , it contributes less than 30% all the way up to @xmath209 . the lower panels of fig . [ nnb ] show the ratio of exoergic absorption to the reaction cross section @xmath210 where @xmath108 only accounts for negative - energy downward transitions induced by neutrino absorption . when considering the @xmath211 cross section , we observe that thermal effects are unimportant for @xmath212 mev , i.e. when incoming neutrinos have sufficiently large energy to excite the gt@xmath156 resonance . note , that a downward shift of the gt@xmath156 resonance only marginally affects the cross section at such high neutrino energies . thermal effects become pronounced though for lower neutrino energies since finite temperature removes the reaction threshold and significantly enhances the cross section . for energies @xmath213 mev , the ratio @xmath214 and , hence , the observed enhancement is mostly caused by downward gt@xmath156 transitions from nuclear excited states . moreover , for @xmath215 the thermally unblocked downward transitions completely dominate the reaction , increasing the cross section by more than an order of magnitude when the temperature rises from 0.86 mev to 1.72 mev . the role of the exoergic absorption decreases with increasing neutrino energy and for intermediate energies , @xmath216 , thermal effects on the gt@xmath156 resonance and its low - energy tail become important for the cross section enhancement . referring to the right panels in fig . [ nnb ] it is shown that the @xmath217 cross section to a much lesser extent depends on temperature than that for the @xmath211 reaction . this result can be understood as follows . in @xmath122 , due to the negative @xmath140-value , the downward gt@xmath156 transitions dominate the ground - state reaction up to @xmath218 mev . as discussed above , the gt@xmath156 distribution in @xmath122 is little affected by the temperature rise , which merely causes some additional strength at @xmath219 mev . this thermally unblocked downward strength becomes competitive with the ground - state contribution only at rather low energies ( @xmath220 mev ) . as a result , thermal effects on the @xmath217 reaction are noticeably milder than in the previous case and a temperature rise from @xmath128 to 1.72 mev enhances the low - energy cross section only by a factor of about four . this observation is in line with ref . @xcite , where it was shown that the thermal enhancement of the neutrino absorption cross section is reduced the more neutron - rich the nucleus is . , but for the antineutrino absorption reaction . ] the results for @xmath12-absorption are shown in fig . [ annb ] . for the @xmath221 reaction the cross section demonstrates the same trend as discussed above for the neutrino absorption by @xmath121 : ( i ) the gap in the cross section disappears and the low - energy cross section increases with temperature . ( ii ) with increasing @xmath133 the cross sections at different temperatures converge , i.e. , thermal effects become unimportant . from the lower - left panel we conclude that a significant enhancement of the low - energy cross section relative to ground - state calculations comes from the increasing contribution of downward gt@xmath156 transitions from nuclear excited states . the ratio @xmath222 gradually reduces at @xmath223 mev , since at those energies the excitation of the gt@xmath150 resonance becomes possible . however , at @xmath195 mev the exoergic component of the cross section appears to be comparable with the endoergic one up to @xmath224 mev . of the @xmath225th negative - energy gt state to the finite - temperature cross section at @xmath226 . ] it should be noted that although the calculated @xmath227 and @xmath221 cross sections show the same trend , the observed thermal enhancement at low energies is caused by different types of downward transitions . to show that , we have calculated the relative contribution @xmath228 of different negative - energy gt states to the cross section at @xmath226 . the results are depicted in fig . [ partial_t ] . as indicated in the figure , the @xmath229-absorption is completely dominated by the thermally unblocked transition inverse to the gt@xmath150 resonance . for the @xmath230-absorption , however , the main contribution to the reaction is given by transitions inverse to the gt@xmath156 low - energy strength , while the `` inverse '' gt@xmath156 resonance is suppressed by the boltzmann factor in eq . . only at @xmath195 mev , the `` inverse '' gt@xmath156 resonance gives a noticeable contribution to the absorption of low - energy antineutrinos . as illustrated in the right panels of fig . [ annb ] , the low - energy @xmath231 cross section at finite temperature is also predominantly mediated by gt@xmath150 transitions . although the cross section remains substantially smaller as compared with the other three reactions considered , it demonstrates a strong temperature dependence for antineutrinos with @xmath232 mev , i.e. below the ground - state reaction threshold . since the downward gt@xmath150 transitions are suppressed in @xmath122 , the cross section enhancement reflects the thermal unblocking of the upward gt@xmath150 strength . because of the thermal unblocking , the energy below which the gt@xmath150 contribution is more than half of the total cross section shifts to higher values : at @xmath233 mev this energy about 15 mev , at @xmath234 mev it is about 20 mev , and at @xmath195 mev it is about 25 mev . for higher energies forbidden transitions dominate the process like at @xmath128 and the cross - section depends only weakly on temperature . , but now including the outgoing electron blocking ] as was mentioned above , during the core contraction neutrino absorption by nuclei is hindered by pauli blocking of the phase space for the outgoing electron . to study this effect within the tqrpa , following ref . @xcite , we have calculated the @xmath229-absorption cross sections by introducing a blocking factor @xmath100 defined at three different sets of temperature and chemical potential ( @xmath235 in mev ) : @xmath236 . the results are shown in fig . [ nb ] for the @xmath227 and @xmath237 reactions . as can be seen , the neutrino absorption cross sections are drastically reduced due to electron blocking in the final state . moreover , as the chemical potential increases faster than the temperature the cross sections decrease with temperature . it is obvious that the absorbtion due to de - excitation of thermally excited states is less affected by the pauli blocking since the outgoing electron gains energy . this is clearly demonstrated by the plots in the lower panels of fig . [ nb ] where the relative contribution of negative - energy transitions is shown . we thus conclude that owing to blocking effect negative energy transitions dominate the cross sections up to neutrino energies @xmath238 . another consequence of the blocking is that downward @xmath239 forbidden transitions become important with increasing @xmath235 . without pauli blocking their contribution in negligible due to a small boltzmann weight . however , as shown in the middle panels of fig . [ nb ] , for @xmath240 their contribution is comparable or even larger than those of the allowed transitions . according to our calculations this downward forbidden contribution is mainly due to @xmath241 transitions . fe calculated within the tqrpa and the lssm approach ( see ref . @xcite , fig . 1 ) . the cross sections in the upper ( lower ) panel are derived without ( with ) the final - state electron blocking . the respective temperatures and chemical potentials are given in mev . ] in fig . [ comparison ] , the @xmath129 cross sections are displayed together with the results of the large - scale shell - model calculations ( see fig . 1 in ref . even if the general behaviour of the cross sections as functions of the neutrino energy and temperature is in agreement in both approaches , the tqrpa results are noticeable larger than the lssm ones but the discrepancy reduces with increasing neutrino energies . to understand the cause of the discrepancy , let us consider first the cross sections calculated neglecting the final - state electron blocking . for @xmath213 mev the tqrpa cross sections exceed the lssm values by two to three orders of magnitude . for such low energies , in both approaches , the neutrino absorption on @xmath0fe is dominated by gt@xmath156 downward transitions from thermally excited states . the differences in the description of such transitions explain the differences between the tqrpa and shell - model results . as the tqrpa is based on the grand canonical ensemble , the upward and downward gt@xmath2 strength are connected by the detailed balance . as shown in fig . [ partial_t ] , the absorption of low - energy neutrinos on @xmath0fe is dominated by the downward gt@xmath156 transition inverse to the gt@xmath150 resonance . this corresponds to an excitation energy @xmath242 mev in the boltzmann factor of eq . . within the shell - model calculations , downward transitions are included by back - resonances , that is by inverting the gt@xmath150 strength distribution of the daughter nucleus . in @xcite , the back - resonances are built on the lowest states of @xmath0co and the bulk of the back - resonance strength in @xmath0fe is located at an excitation energy of @xmath243 mev . this excitation energy is somewhat higher than @xmath244 . moreover , within the tqrpa the downward strength concentrates in a single state , while the shell model gt@xmath150 strength for @xmath0co is highly fragmented owing to multinucleon correlations @xcite . it is clear that both these factors suppress the contribution of downward transitions within the lssm . to see whether the tqrpa reliably predicts the strength of negative - energy transitions , one must go beyond the tqrpa . for a separable residual interaction used here this can be done following the method developed within the qpm , that is by taking phonon coupling into account . on the other hand we should note that due to violation of brink s hypothesis some back - resonances built on high - lying excited states of @xmath0co may be located at the same energies as those built on the nuclear ground and low - lying states . due to an increasing density of states , the contribution of such back - resonance states may be substantial and , therefore , their inclusion into the shell - model calculations may improve the agreement between the tqrpa and shell - model results . for @xmath245 mev , thermally unblocked low - energy gt@xmath156 transitions come into play , and since such transitions do not appear within the shell - model based calculation they also cause the excess of the tqrpa cross sections over the lssm ones . at @xmath246 mev , @xmath229-absorption is dominated by the strong transition involving the gt@xmath156 resonance . with increasing @xmath133 , the cross section becomes insensitive to the energy dependence of the gt distribution and depends only on the total gt@xmath156 strength . as a result we observe excellent agreement between the results of both approaches . comparing the cross sections calculated with pauli blocking for the outgoing electron ( see the lower panel of fig . [ comparison ] ) , we note that in @xcite , only allowed gt@xmath156 transitions are taken into account when calculating neutrino absorption cross section . as is shown in fig . [ nb ] , if the pauli blocking is taken into account , forbidden downward transition become important with increasing temperature and electron chemical potential . it is apparent that such transitions along with the above discussed reasons lead to larger values for the tqrpa cross sections as compared to the shell - model results . in this work , we have studied thermal effects on the ( anti)neutrino absorption for hot nuclei in supernova environments . for this purpose , we have employed the proton - neutron qrpa extended to finite temperatures within the tfd formalism . as an example , cross sections were calculated for @xmath0fe and @xmath1ge in the temperature range from @xmath128 to @xmath187 mev by taking into account the relevant charge - exchange transitions @xmath247 , and @xmath248 . a detailed analysis of thermal effects was performed for allowed gt transitions which dominate the cross sections for @xmath249 mev neutrinos . since the tqrpa does not support the brink hypothesis , new peaks appear in the gt@xmath118 strength function at finite temperature due to transitions from the excited states . moreover , thermal effects shift the gt resonance centroids to lower energies and this effect appears more strongly for the gt@xmath150 strength in @xmath1ge . the downward transitions from nuclear excited states were included in our calculations through detailed balance . the validity of detailed balance for charge - exchange transitions is a consequence of the grand canonical treatment of hot nuclei . we have found that thermal effects on the gt strength enhance the absorption cross sections for low - energy ( anti)neutrinos by several orders of magnitude . this enhancement is mainly due to increasing contributions of downward transitions from excited states . however , in the supernova environment the electron chemical potential increases more rapidly than temperature . as a result , if the electron blocking in the final state is taken into account , the neutrino cross sections are drastically reduced . although our calculations reveal the same thermal effects as the shell - model calculation , the calculated low - energy cross sections for @xmath0fe exceed the shell - model values by two to three orders of magnitude . one of the possible reason for this discrepancy is that the tqrpa underestimates multinucleon correlations which are responsible for the gt strength fragmentation . on the other hand , the inclusion of back - resonances built on highly excited daughter states into shell - model calculations may also improve the agreement between the tqrpa and shell - model results . since the tqrpa is not restricted by iron - group nuclei , it has some advantages over shell - model calculations . to enhance its reliability and predictive power several improvements could be made . first of all , to account for multiconfigurational effects , the coupling of thermal charge - exchange phonons with more complex ( e.g. , two - phonon ) configurations should be included into the approach . at zero temperature this problem was considered within the qpm @xcite by exploiting a separable form of the residual interaction . the other direction of the improvement is to combine the tqrpa method with self - consistent qrpa calculations based on either the relativistic or skyrme nuclear energy density functionals . recently , such calculations were performed at zero temperature @xcite . with a separable approximation for the skyrme interaction @xcite it will be possible calculate the phonon coupling at @xmath51 within a self - consistent theory . this is planned for the future . here we show that the detailed balance condition in the form can be derived in a model independent way . when considering a grand canonical ensemble of hot nuclei , the probability to find the @xmath225th excited state of a nucleus with @xmath250 protons and @xmath251 neutrons is given by @xmath252 where @xmath253 is the partition function of the grand canonical ensemble and @xmath254 is the angular momentum . notice that the excitation energies @xmath255 are counted from the energy of noninteracting nucleons , i.e. , @xmath256 is a ground - state binding energy and the chemical potentials @xmath257 do not include nucleon rest mass . let us now introduce the temperature - dependent strength function for charge - exchange transitions as a thermal average of all transition strengths ( probabilities ) from states in the parent nucleus to states in the daughter nucleus : @xmath258 here @xmath259 and @xmath260 is the reduced transition strength between states @xmath225 and @xmath261 in the parent and daughter nuclei , respectively . in the above equations the upper sign corresponds to @xmath72 transitions , while the lower sign refers to @xmath73 transitions . the transition energy @xmath74 can be both positive and negative . for the transition operators @xmath262 and @xmath263 , which differ only by the isospin operator , the respective transition strengths , @xmath264 and @xmath265 , are connected by detailed balance through @xmath266 combining this result with eq . , we get the following relationship between the strength functions for @xmath72 and @xmath73 transitions in the thermal grand canonical ensemble @xmath267 this relation is exactly the same as derived within the tqrpa approach for charge - changing transitions in hot nuclei . 52ifxundefined [ 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 \doibase http://dx.doi.org/10.1016/j.physrep.2007.02.002 [ * * , ( ) ] , @noop ( ) , @noop ( ) , link:\doibase 10.1103/physrevlett.60.1999 [ * * , ( ) ] link:\doibase 10.1086/170316 [ * * , ( ) ] link:\doibase 10.1086/170317 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0370-2693(01)00620-7 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0370-2693(01)01502-7 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/j.nuclphysa.2004.09.005 [ * * , ( ) ] link:\doibase 10.1103/physrevc.56.3079 [ * * , ( ) ] link:\doibase 10.1142/s0218301309013695 [ * * , ( ) ] link:\doibase 10.1142/s0217979296000817 [ * * , ( ) ] http://books.google.ru/books?id=ra7vaaaamaaj[__ ] ( , ) \doibase http://dx.doi.org/10.1016/0003-4916(81)90058-0 [ * * , ( ) ] link:\doibase 10.1103/physrevc.89.035805 [ * * , ( ) ] link:\doibase 10.1134/s1063778811080059 [ * * , ( ) ] link:\doibase 10.1134/s1063778809080079 [ * * , ( ) ] link:\doibase 10.1103/physrevc.81.015804 [ * * , ( ) ] link:\doibase 10.1143/jpsj.12.570 [ * * , ( ) ] link:\doibase 10.1103/physrev.115.1342 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0375-9474(81)90557-1 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0375-9474(83)90411-6 [ * * , ( ) ] in http://books.google.ru/books?id=l9ic3gt578gc[__ ] , ( , ) p. [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0920-5632(02)01761-9 [ * * , ( ) ] link:\doibase 10.1103/physrevc.86.035804 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0375-9474(00)00131-7 [ * * , ( ) ] link:\doibase 10.1103/physrevc.57.2004 [ * * , ( ) ] link:\doibase 10.1103/physrevc.62.015501 [ * * , ( ) ] link:\doibase 10.1103/physrevc.77.024608 [ * * , ( ) ] in http://books.google.ru/books?id=i1bjqgaacaaj[__ ] , ( , ) p. \doibase http://dx.doi.org/10.1016/0375-9474(84)90673-0 [ * * , ( ) ] http://books.google.ru/books?id=ly3_c3zg3quc[__ ] ( , ) link:\doibase 10.1103/physrevc.63.025802 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0375-9474(01)00992-7 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/j.nuclphysa.2007.06.005 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0146-6410(98)00024-6 [ * * , ( ) ] link:\doibase 10.1103/physrevc.84.047305 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0375-9474(83)90632-2 [ * * , ( ) ] link:\doibase 10.1103/physrevc.49.3128 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0375-9474(99)00240-7 [ * * , ( ) ] link:\doibase 10.1103/physrevc.83.045807 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0370-2693(99)00363-9 [ * * , ( ) ] http://stacks.iop.org/0305-4616/10/i=11/a=007 [ * * , ( ) ] http://stacks.iop.org/0305-4616/11/i=5/a=007 [ * * , ( ) ] link:\doibase 10.1103/physrevc.57.1204 [ * * , ( ) ] link:\doibase 10.1007/bf01323673 [ * * , ( ) ] http://stacks.iop.org/0953-8984/24/i=22/a=225304 [ * * , ( ) ]

we present detailed semi - analytical models for the formation of disk galaxies both in a universe dominated by dark matter ( dm ) , and in one for which the force law is given by modified newtonian dynamics ( mond ) . we tune the models to fit the observed near - infrared tully - fisher ( tf ) relation , and compare numerous predictions of the resulting models with observations . the dm and mond models are almost indistinguishable . they both yield gas mass fractions and dynamical mass - to - light ratios which are in good agreement with observations . both models reproduce the narrow relation between global mass - to - light ratio and central surface brightness , and reveal a characteristic acceleration , contrary to claims that these relations are not predicted by dm models . both models require sn feedback in order to reproduce the lack of high surface brightness dwarf galaxies . however , the introduction of feedback to the mond models steepens the tf relation and increases the scatter , making mond only marginally consistent with observations . the most serious problem for the dm models is their prediction of steep central rotation curves . however , the dm rotation curves are only slightly steeper than those of mond , and are only marginally inconsistent with the poor resolution data on lsb galaxies .

astro-ph9912004

the most successful cosmological model to date consists of a universe dominated by cold dark matter ( cdm ) , possibly with a non - zero cosmological constant . however , this paradigm faces several serious shortcomings . first is the high central densities predicted for virialized dark halos . navarro , frenk & white ( 1996 , 1997 ) have claimed the existence of a universal density profile for virialized halos , with @xmath0 at small radii ( hereafter referred to as the nfw profile ) . more recently , higher resolution simulations have argued that cdm halos should have even steeper cusps ( fukushige & makino 1997 ; moore et al . 1998 ; moore et al . 1999b ; but see kravtsov et al . several studies have argued that these steep cusps are inconsistent with the observed rotation curves of dark matter dominated systems , such as dwarfs and low surface brightness galaxies ( flores & primack 1994 ; moore 1994 ; burkert 1995 ; navarro 1998 ; mcgaugh & de blok 1998a ; stil 1999 ) . secondly , klypin et al . ( 1999 ) and moore et al . ( 1999a ) have indicated an additional problem for the cdm paradigm : simulations predict that galaxies should have many more satellite systems than observed . although the reason for this large discrepancy may be related to observational biases , feedback from supernovae , or an ionizing background , it might also indicate that the cdm power spectrum has too much power on small scales . both these problems may ( partially ) be solved with a mixture of hot and cold dark matter . however , the dark matter scenario is certainly starting to loose its appealing character , if we have to rely on a hybrid mix of baryons , cold dark matter , hot dark matter , and a non - zero cosmological constant in order to explain the observations . the problems with fine - tuning become more and more severe , and it might therefore be worthwhile to explore some alternatives to the dark matter theory . although the flat rotation curves of galaxies are generally interpreted as evidence for the existence of dark matter , they could also result from our theory of gravity ( or inertia ) being in error , such that the dynamical masses inferred from the usual newtonian equations are overestimated . several groups have tried to eliminate the need for dark matter by modifying the theory of gravity ( e.g. , milgrom 1983a ; sanders 1986 ; mannheim & kazanas 1989 ; liboff 1992 ; moffat & sokolov 1996 ) . so far , the modified newtonian dynamics ( mond ) proposed by milgrom ( 1983a ) has been remarkably successful , and in this paper we compare several predictions from mond with those based on dark matter ( hereafter dm ) . disk galaxies are , from a dynamical point of view , fairly simple systems in which the discrepancies between the luminous and the inferred dynamical masses ( the `` mass discrepancies '' ) are least ambiguous . in addition , they have been extensively studied , with data available for systems that span several orders of magnitude in both mass and surface brightness . they are thus the ideal test - beds to compare predictions based on mond and dm with observations in the hope to discriminate between the two alternatives . for both mond and dm to be considered viable theories , they should be able to explain the following properties of disk galaxies : 1 . the slope , scatter , and zero - point of the tully - fisher relation ( tully & fisher 1977 ) . in particular , the fact that both high and low surface brightness disks ( hsbs and lsbs , respectively ) follow the same relation without any significant offset ( sprayberry et al . 1995 ; zwaan et al . 1995 ; hoffman et al . 1996 ; tully & verheijen 1997 ) . the shape of the observed rotation curves . in particular , the systematic variation of these curves with surface brightness and luminosity ( rubin , thonnard & ford 1980 ; burstein et al . 1982 ; casertano & van gorkum 1991 ; persic , salucci & stel 1996 ) . the increase of mass discrepancies with decreasing luminosity and surface brightness ( e.g. , salucci & frenk 1989 ; persic & salucci 1988 , 1990 ; kormendy 1990 ; broeils 1992 ; forbes 1992 ; persic , salucci & stel 1996 ; mcgaugh & de blok 1998a ) . this is directly linked to item 2 . the increase of gas mass fractions with decreasing luminosity and surface brightness ( gavazzi 1993 ; gavazzi , pierini & boselli 1996 ; mcgaugh & de blok 1997 ) . the sharp cut - off in the distribution of central surface brightnesses at the bright end , close to the value of the freeman ( 1970 ) law ( allen & shu 1979 ; mcgaugh , bothun & schombert 1995 ; de jong 1996a ; mcgaugh & de blok 1998a ) . the presence of a characteristic acceleration in the rotation curves of disk galaxies ( mcgaugh 1998 ) . in a series of papers , mcgaugh & de blok ( 1998a , b ; hereafter mb98a and mb98b ) have used several of these observational facts to argue against the dm hypothesis and in favor of mond . in particular , they argue that the dm hypothesis faces severe fine - tuning problems . in this paper we revisit these issues using more sophisticated models for the formation of disk galaxies under both the dm and mond hypotheses . we show that , contrary to the claims made by mb98a and mb98b , the dm model can explain virtually all aforementioned observations with a minimal amount of fine - tuning . furthermore , we show that mond needs a similar amount of fine - tuning , therewith loosing its apparent advantage over the dm scenario . this paper is organized as follows . in [ sec : diskform ] we discuss the models , both for the dm and the mond models . in [ sec : tf ] we use the near - infrared tully - fisher relation to constrain the free parameters of the models . in [ sec : comp ] we compare the models to data on gas mass fractions , dynamical mass - to - light ratios , and characteristic accelerations . [ sec : rcs ] compares the detailed rotation curve shapes from the dm and mond models . we summarize our conclusions in [ sec : concl ] . throughout this paper we adopt a hubble constant of @xmath1 in all our models . any quantity that depends on the distance scale is written in terms of @xmath2 . in van den bosch ( 1999 ; hereafter paper i ) we presented detailed models for the formation of disk galaxies in the context of a universe dominated by cdm . here we briefly summarize the ingredients of the models , and refer the reader to paper i for details . in our models , disks form by the settling of baryonic matter in virialized dark halos , which are described by the nfw density profile . we assume that while the baryons radiate their binding energy , they conserve their specific angular momentum , thus settling into a disk with a scale length that is proportional to the angular momentum and size of the dark halo ( see also fall & efstathiou 1980 ; dalcanton spergel & summers 1997 ; mo , mao & white 1998 ; van den bosch 1998 ) . the angular momentum , @xmath3 , is thought to originate from cosmological torques and can be characterized by a dimensionless spin parameter , @xmath4 , the distribution of which is well constrained from both analytical and numerical studies ( e.g. , barnes & efstathiou 1987 ; warren et al . 1992 ) . for each separate model galaxy , we randomly draw a value for @xmath4 from this distribution . the disk is assumed to be a thin exponential with surface density @xmath5 . adiabatic contraction is taken into account to describe how the density profile of the dark matter changes due to the cooling of the baryonic matter ( blumenthal et al . 1986 ; flores et al . 1993 ) . a global stability criterion is used to investigate if the disks that form are stable . if unstable , part of the disk material is transformed into a bulge component , until the disk reaches stability ( van den bosch 1998 ) . once the density distribution of the baryonic material is known , we compute the fraction of baryons converted into stars . kennicutt ( 1989 ) has shown that the star formation rate ( sfr ) in disk galaxies follows a schmidt law ( schmidt 1959 ) , but is abruptly suppressed below a given threshold density . the critical density is given by toomre s ( 1964 ) stability criterion as : @xmath6 here @xmath7 is a dimensionless constant near unity , @xmath8 is the velocity dispersion of the gas , and @xmath9 is the epicycle frequency , which can be derived from the rotation curve . setting @xmath10 and @xmath11 these star formation thresholds were found to coincide with the radii where @xmath12 . following an earlier suggestion by quirk ( 1972 ) , we therefore consider the fraction of disk mass with surface densities in excess of @xmath13 to be eligible for star formation . if one further considers the empirical schmidt law determined by kennicutt ( 1998 ) for a large sample of disk galaxies , it is found that over the typical lifetime of a galaxy virtually all the mass which is eligible for star formation does in fact turn into stars . this is consistent with the lack of major star formation in the inter - arm regions in spirals and yields gas mass fractions that are in good agreement with observations ( paper i ) . in what follows we therefore assume that the entire bulge mass and the entire disk mass with @xmath14 are transformed into stars , giving a total stellar mass of @xmath15 . the total @xmath16-band luminosity of the disk - bulge system is then calculated using @xmath17 , where @xmath18 is the stellar mass - to - light ratio . since @xmath19 depends only weakly on the age and metallicity of a stellar population ( see discussion in paper i ) , we assume a constant stellar mass - to - light ratio of @xmath20 . with the mass in stars known , the total amount of energy produced by sn can be calculated , once we make assumptions about the energy per sn ( @xmath21 ) and the number of sn per solar mass of stars formed ( @xmath22 ) . we then calculate the total baryonic mass that is prevented from becoming part of the disk / bulge system due to sn feedback as @xmath23 here @xmath24 corresponds to the circular velocity of the halo at the radius @xmath25 , inside of which the average density equals 200 times the critical density for closure , and @xmath26 and @xmath27 are two free parameters that describe the efficiency with which the sn energy is used to prevent gas from settling in the disk / bulge system ( see paper i ) . throughout we adopt a value of @xmath28 which corresponds to a scalo ( 1986 ) initial mass function . an iterative procedure is used to compute the masses in the various phases in a self - consistent manner , i.e. , we ensure that @xmath29 , with @xmath30 the total mass within @xmath25 ( before sn blow - out ) , and @xmath31 the disk mass that is not turned into stars . we now construct similar models but under the hypothesis of mond , which assumes that the force law changes from conventional newtonian form when the acceleration of a test particle is smaller than a characteristic acceleration @xmath32 ( which is a universal constant ) . the true gravitational acceleration @xmath33 is related to the newtonian acceleration @xmath34 as @xmath35 ( @xmath36 ) . the interpolation function @xmath37 is not specific , but is required to have the asymptotic behavior @xmath38 ( the newtonian regime ) and @xmath39 ( the mond regime ) . with the rotation law given as usual , @xmath40 , the circular velocity of a finite bounded mass @xmath41 in the limit @xmath42 becomes @xmath43 with @xmath44 the gravitational constant . mond thus predicts asymptotically flat rotation curves . first we need to set @xmath32 and the interpolation function @xmath37 . we follow previous studies of mond , and set @xmath45 and adopt @xmath46 for these choices , mond yields good fits to the observed rotation curves of disk galaxies ( kent 1987 ; milgrom 1988 ; begeman , broeils & sanders 1991 ; sanders 1996 ; de blok & mcgaugh 1988 ) . with @xmath37 given by equation ( [ mufunc ] ) , the circular velocity as function of radius becomes @xmath47^{1/4},\ ] ] where @xmath48 is the circular velocity derived for pure newtonian dynamics . unfortunately , there is currently no mond equivalent to the theory of structure formation in a dm universe . in particular , we do nt know the formation epochs or distribution of angular momenta of galaxies that form in a mond universe . we therefore follow an empirical procedure in which we assume that galaxy formation under mond produces disk galaxies with the masses and dimensions as observed . as in the dm scenario , disks are assumed to be thin exponentials . we randomly draw a total ( baryonic ) mass from the interval @xmath49 . we initially set the disk mass , @xmath50 , equal to @xmath51 , and randomly draw a central disk surface density from the interval @xmath52 . the scale length of the disk is then @xmath53 . the first step is to investigate whether the disk is stable against global instabilities . to that extent we use the same stability criterion as for the dm scenario , i.e. , a disk is considered stable if @xmath54 where @xmath55 is the circular velocity of the galaxy at three scale lengths ( cf . christodolou , shloshman & tohline 1995 ; efstathiou , lake & negroponte 1982 ; van den bosch 1998 ) . in the case of newtonian dynamics with no dm halos , @xmath56 . however , in the case of mond , @xmath57 , at least if the acceleration at @xmath58 is sufficiently small ; i.e. , the mond force law helps to stabilize disks ( milgrom 1989 ) . unlike the dm scenario , for which numerical simulations have shown that @xmath59 ( christodolou et al . 1995 ) , no equivalent study exists based upon mond . we therefore consider @xmath60 a free parameter . once the density distribution of the gas is known , we calculate the mass in stars , in the same fashion as for the dark matter models discussed above . note , however , that in the calculation of the epicycle frequency , @xmath9 , we now use the circular velocity based on the modified force law . the @xmath16-band luminosity is again calculated from @xmath61 adopting a constant mass - to - light ratio @xmath19 , which we consider a free parameter . finally , we include sn feedback , using the same approach as above for dm . the hot gas mass , @xmath62 is calculated using equation ( [ hotmass ] ) , but with @xmath24 replaced by @xmath63 ( equation [ [ velmond ] ] ) . we use an iterative technique to ensure self - consistency , i.e. , @xmath64 . the slope , scatter and zero - point of the tully - fisher ( hereafter tf ) relation contain information about the structure and formation of disk galaxies . here we use the observed characteristics of this fundamental scaling relation to set the parameters of our disk formation models described in the previous section . the slope and scatter of the empirical tf relation depend strongly on the luminosity and velocity measures used , with the relation becoming shallower going towards bluer passbands ( aaronson , huchra & mould 1979 ; visvanathan 1981 ; tully , mould & aaronson 1982 ; wyse 1982 ; pierce & tully 1988 ; gavazzi 1993 ; verheijen 1997 ) . therefore , when trying to infer constraints on the structure and formation of disk galaxies from the tf relation , it is absolutely essential that one extracts the same luminosity and velocity measures from the models ( see discussions in courteau 1997 , verheijen 1997 , and paper i ) . the empirical tf relation which most directly reflects the mass in stars and the total dynamical mass of the halo is the @xmath16-band tf relation of verheijen ( 1997 ) , which uses the flat part of the rotation curve as a characteristic velocity : @xmath65 the slope of this tf relation is @xmath66 , and the observed scatter is @xmath67 mag . this empirical tf relation is shown in the upper left panel of figure [ fig : tf ] . open circles correspond to the galaxies in verheijen s sample of ursa major spirals , and the thick solid line is the best - fitting tf relation ( equation [ [ tffund ] ] ) . within the dm framework , simple dynamics predict a tf relation of the form @xmath68 ( dalcanton et al . 1997 ; white 1997 ; mo et al . 1998 ; syer , mao & mo 1999 ; van den bosch 1998 ; paper i ) . here @xmath69 is the galaxy formation efficiency , which describes what fraction of the baryonic mass inside the dm halo ultimately ends up in the disk / bulge system , @xmath70 is the baryonic mass fraction of the universe ( @xmath71 ) , and @xmath72 is the @xmath16-band mass - to - light ratio of the galaxy ( disk plus bulge ) . from equation ( [ velmond ] ) it is immediately apparent that in the case of mond , the tf relation becomes @xmath73 ( see e.g. , milgrom 1983b ; mb98b ) . thus whereas dm predicts a tf relation with a slope of @xmath74 , mond predicts @xmath75 , in agreement with the empirical relation of equation ( [ tffund ] ) . in the case of dm , extra physics are required to steepen the predicted tf relation to the observed slope . mb98a have presented a detailed discussion in which they argue that this results in a serious fine - tuning problem for dm . on the other hand , the almost perfect agreement between the observed tf slope and the pure - gravitational value for mond ( equation [ [ tfmond ] ] ) presents an anti - fine - tuning problem . there is very little room for any other galaxy characteristic to vary systematically with mass , without changing the tf slope . the above suggests that simple dynamical arguments can be used to predict the existence of an underlying tf relation for both dm and mond . however , to accurately predict the slope , zero - point , and scatter of the tf relationship , it is necessary to incorporate the more detailed physics discussed in [ sec : diskform ] . we do so in the following two sections for the dm and mond hypotheses , respectively . as mentioned in [ sec : tfobs ] it is essential to extract the same velocity measure from the models as the one used in the data , and we therefore extract @xmath76 from our model galaxies ( see paper i for details ) . in paper i we showed that taking the stability - derived star formation threshold densities into account ( with @xmath77 ) steepens the tf relation from @xmath78 ( as predicted by pure dynamics ) to @xmath79 . we argued that to further steepen the tf relation to its observed slope , feedback from sn is required . fine - tuning the two parameters that describe the efficiency with which sn energy is used to prevent gas from settling in the disk / bulge system ( @xmath26 and @xmath27 ) , we were able to obtain a tf relation with a slope , zero - point , and an amount of scatter that are all in excellent agreement with the empirical relation of equation ( [ tffund ] ) . this model is referred to as model l5 in paper i , and we adopt the same convention here . this model is based on a cosmology with @xmath80 , @xmath81 , @xmath82 and @xmath83 ( see paper i ) , and its parameters are listed in table [ tab : param ] . model l5 predicts an intrinsic tf scatter of @xmath84 mag , with no offset in normalization between hsb and lsb galaxies , as observed . the tf relation of model l5 is shown in the upper right panel of figure [ fig : tf ] . although simple dynamics predict a tf relation for mond with a slope @xmath75 , consistent with observations , this ignores the fact that @xmath85 varies with mass ; gas mass fractions are known to increase with decreasing mass ( see [ sec : gasfrac ] ) , and taking this into account steepens the tf relation . it therefore remains to be seen whether mond does not actually predict tf relations that are too steep . we can investigate this using the models described in [ sec : mondform ] . as with the dm models , we set @xmath77 . the stellar mass - to - light ratio in the @xmath16-band , @xmath19 , is set by fitting the normalization of the tf relation , yielding @xmath86 . this value is consistent with estimates based on stellar population models , which in itself can be considered a success of mond . to set the free parameter @xmath60 , with the help of equation ( [ vcircmond ] ) the stability criterion ( equation [ [ stabalpha ] ] ) can be written as @xmath87^{-1/2 } \ ; { \>{\rm m_{\odot}}}{\rm pc}^{-2}.\end{aligned}\ ] ] the stability parameter @xmath88 thus directly translates into a maximum surface density of disks . from the fact that the brightest observed disks have central surface brightnesses of @xmath89 mag arcsec@xmath90 ( e.g. , de jong 1996b ; tully et al . 1996 ) we infer @xmath91 ( for @xmath92 ) . if we were to take @xmath93 ( as for the dm models ) , no disks with central surface brightnesses brighter than @xmath94 mag arcsec@xmath90 would be allowed , clearly inconsistent with observations . since in general not all the gas mass is transformed into stars , the limit of @xmath95 is conservative . we find that with @xmath77 and @xmath96 a value of @xmath97 yields @xmath98 mag arcsec@xmath90 , and we adopt this value throughout . none of the results presented here , however , depend on this choice : @xmath60 merely sets the maximum central surface density for disks with a given bulge - to - disk ratio , and is set such that we obtain disks with the full range of observed values of @xmath99 . in mimicking the sample selection of verheijen ( 1997 ) we only consider model galaxies with a bulge - to - disk ratio @xmath100 ( see also paper i ) . we compare the dm model ( l5 ) with two different mond models , m1 and m2 . the former neglects sn feedback , whereas the latter considers blow - out by sn with @xmath101 and @xmath102 ( see table [ tab : param ] ) . the motivation for these particular feedback parameters will be discussed in [ sec : ups ] . the tf relations for models m1 and m2 are shown in the lower panels of figure [ fig : tf ] . the star formation recipe with @xmath77 steepens the mond tf relation from @xmath103 ( equation [ [ tfmond ] ] ) to @xmath104 ( model m1 ) , and brings it in excellent agreement with the empirical tf relation of verheijen . the predicted , intrinsic amount of scatter is only @xmath105 mag , once again in excellent agreement with observations . the introduction of sn feedback steepens the tf relation further , to @xmath106 for model m2 , and increases the scatter to @xmath107 mag . model m2 thus yields a tf relation that is too steep . in addition , the amount of scatter is only marginally consistent with the data , i.e. , verheijen ( 1997 ) derived @xmath108 mag at 95 percent confidence level . having tuned the parameters by fitting the near - infrared tf relation , we now compare the resulting models to other independent observations , to test , amongst others , the validity of the star formation and feedback recipes used . as shown in paper i , within the cdm context one can construct many physically different models that all fit the empirical tf relation . for these models additional constraints , such as the ones described below , are required to descriminate between the models . our calculations yield magnitudes and surface brightnesses in the @xmath16-band and rotation velocities for a collection of model galaxies . ideally , we would therefore compare our models to data that consists of @xmath16-band photometry combined with full hi rotation curves . unfortunately , such data is currently scarce , and we therefore use @xmath109-band photometry , for which ample data is available . although a constant stellar mass - to - light ratio is a reasonable assumption for the @xmath16-band , @xmath110 is expected to reveal a systematic trend with mass as well as a significant amount of scatter owing to variations in dust content , and the ages and metallicities of the stellar populations . we convert the @xmath16-band luminosities and surface brightnesses of our model galaxies to the @xmath109-band adopting an empirically determined @xmath111 color magnitude relation ( see appendix a ) . this , at least , takes account of the systematic trend of @xmath111 with @xmath112 . unfortunately , the scatter is relatively large , and we predict errors in @xmath113 and @xmath114 , owing to the color conversion , of @xmath115 and @xmath116 mag , respectively . a more detailed comparison of our models against data has to await accurate near - infrared photometry of a large sample of disk galaxies for which full hi rotation curves are available . we compiled the following data for comparison with our models : * @xmath112 , @xmath99 , @xmath117 , and @xmath118 for a sample of spirals in the ursa major cluster from the thesis of verheijen ( 1997 ) . we restrict ourselves to the twenty - two galaxies of the `` unperturbed sample '' on which the tf relation of equation ( [ tffund ] ) is based . as for our models , we transform @xmath112 and @xmath99 to the @xmath109-band , using the @xmath111 color magnitude relation presented in appendix a. data from this sample are plotted as open circles . * @xmath113 , @xmath114 , @xmath117 , and @xmath118 for a sample of thirty - two galaxies compiled by mb98a . these data are plotted as solid circles . * @xmath113 , @xmath114 , @xmath117 , and @xmath119 for ten dwarf galaxies from the sample of van zee et al . the velocity measure @xmath120 is virtually identical to the velocity at the last measured point of the hi rotation curves , and as such compatible with the velocity measure determined in our models ( see paper i ) . these data are plotted as solid squares . * @xmath113 , @xmath114 , and @xmath121 for a sample of one hundred disk galaxies of type sb or later , compiled by mcgaugh & de blok ( 1997 ) . these data are plotted as open squares . all data have been converted to @xmath122 , using the distance scales quoted by the various authors . note that this combined set of data is by no means complete nor homogeneous . in compiling the data we have aimed for a maximum coverage in both luminosity and surface brightness of disk galaxies with measured hi rotation curves . therefore , it is important to keep in mind that upon comparing models to data , one should only focus on global properties , such as the areas of parameter space occupied by the data and the models . our models are not constructed to correctly predict the local density distribution of data points in parameter space ( i.e. , the masses of our model galaxies are sampled uniformly rather than from a press - schechter formalism or a luminosity function ) . in addition , the incompleteness and possible observational biases in the data samples , combined with the errors associated with the conversion of @xmath16-band models to the @xmath109-band , prevents us from testing our models in too much detail . one of the empirical characteristics of disk galaxies is the increase of their gas mass fractions with decreasing luminosity and surface brightness ( i.e. , item 4 in [ sec : intro ] ) . this is shown in the upper left panels of figures [ fig : gasmag ] and [ fig : gassb ] , where we plot the logarithm of the hi mass - to - light ratio , @xmath123 , as function of @xmath113 and @xmath114 , respectively . the remaining panels plot the results for each of the three models . the solid lines in figures [ fig : gasmag ] and [ fig : gassb ] have no physical meaning , but are plotted to facilitate a comparison of the models with the data . the increase of @xmath124 with decreasing magnitude and surface brightness is remarkably well reproduced by each of the three models . note also that the range spanned by the model galaxies is in good agreement with the data . the hi mass - to - light ratios derived from the models scale with @xmath2 , whereas the measured ratios are independent of distance ( the hi mass is computed directly from the observed flux ) . the main parameter in the models that sets @xmath125 is the star formation parameter @xmath7 . it is reassuring that for @xmath126 and for the value of @xmath7 that was empirically determined by kennicutt ( 1989 ) , the models yield gas mass fractions that are in good agreement with observations . dynamical mass - to - light ratios are seen to vary systematically with mass and surface brightness . dwarfs and lsb galaxies have consistently higher mass - to - light ratios than bright hsb galaxies . these mass - to - light ratios are higher than can be explained by differences in stellar populations alone , and indicate an increasing `` mass discrepancy '' with decreasing luminosity and surface brightness ( i.e. , item 3 in [ sec : intro ] ) . in this paper we define a characteristic mass - to - light ratio as the global mass - to - light ratio inside four disk scale lengths . if we assume that @xmath127 and that disks are pure exponentials , our definition of @xmath128 is identical to that of mb98a , apart from a factor @xmath129 . ] @xmath130 in figure [ fig : upsmag ] we plot @xmath131 as function of @xmath113 for the data and our three models . the thin line is plotted to facilitate a comparison , and is chosen to roughly outline the region that reveals a clear deficit of galaxies ; low luminosity galaxies always have high mass - discrepancies , whereas bright disk galaxies can have values of @xmath128 that span over an order of magnitude . this absence of faint galaxies with low mass discrepancies is well reproduced by model l5 , and owes largely to the effect of sn feedback . the efficiency with which sne can blow gas out of a dark halo is larger in less massive galaxies . this causes most of the baryonic mass to be expelled from low - mass halos , resulting in higher values of @xmath128 for fainter galaxies . a possible shortcoming of model l5 seems to be the fact that it predicts virtually no galaxies with @xmath132 , whereas four of the sixty - four galaxies in the combined data set have values of @xmath128 below this value . however , given the scatter in the @xmath111 color magnitude relation , the inhomogeneous nature of our comparison sample , and the typical errors associated with the data , this apparent deficit is not significant . in the mond model without feedback ( model m1 ) , there is no reason for fainter galaxies to have systematically higher values of @xmath128 . indeed model m1 does not reproduce the observed absence of dwarf galaxies with low mass discrepancies , and it does a poor job in reproducing the data . we therefore included feedback , and tuned the parameters @xmath133 and @xmath27 such that the observed ` region of avoidance ' is reproduced , yielding model m2 . note that model m2 predicts , similarly to model l5 , an absence of model galaxies with @xmath132 . a similar result is shown in figure [ fig : sbvel ] , where we plot the @xmath109-band central surface brightness versus @xmath134 . the data reveal a clear absence of hsb disk galaxies with low rotation velocities . in other words , disk galaxies with @xmath135 always have low surface brightness . again , our dm model reproduces this behavior remarkably well . model m1 predicts no such deficit , and is in clear conflict with the data . model m2 , however , in which the two sn parameters have been tuned to yield the observed absence of dwarf galaxies with low mass discrepancies , is in excellent agreement with the data . note that the observed regions of avoidance in both the @xmath113 - @xmath128 and the @xmath118 - @xmath114 planes are not due to observational biases : there is no reason why dwarf galaxies with high central surface brightnesses should be missed , whereas their lsb counterparts are not ( except for a possible problem due to star - galaxy separation for compact systems at large distances ) . this therefore seems to be a strong indication that sn feedback plays an important role in the formation of ( disk ) galaxies , at least for the less massive ones . mb98a have shown that @xmath128 is strongly correlated with central surface brightness . this is confirmed from our combined data set ( which includes the data of mb98a ) plotted in the upper left panel of figure [ fig : upssb ] . the thin solid line corresponds to @xmath136 , and is plotted with arbitrary normalization for comparison . here @xmath137 is the central surface brightness of the luminous , stellar disk in @xmath138 . the data clearly indicates that @xmath139 with only little scatter . mb98a and mb98b argue that this @xmath128@xmath140 `` conspiracy '' follows naturally from mond , whereas it requires a problematic fine - tuning in the case of dm . as shown by zwaan et al . ( 1995 ) , @xmath141 is required to obtain a tf relation in which hsb and lsb galaxies follow the same relation with a slope of @xmath75 . mb98a discuss a number of physical processes that may lead to the observed @xmath128 @xmath140 `` conspiracy '' under the hypothesis of dm , none of which they render feasible . the upper right panel of figure [ fig : upssb ] , however , shows that our dm model reproduces the observed `` conspiracy '' to a good degree of accuracy . given that the same model yields a tf relation with a slope close to @xmath142 and with little scatter , this is not too surprising , since the @xmath128@xmath140 conspiracy and the tf relation are closely connected ( see discussions in zwaan et al . 1995 and mb98a ) . we thus disagree with mb98a , and conclude that the @xmath128@xmath140 relation is in fact fairly easily reproduced . the lower two panels of figure [ fig : upssb ] show that the two mond models are also in good agreement with the data , something which is not too surprising given that pure mond dynamics predict that @xmath136 ( mb98b ) . milgrom ( 1983b ) defined the characteristic acceleration parameter @xmath143 which is proportional to the ratio of the characteristic acceleration @xmath144 of a disk and the characteristic mond acceleration @xmath32 . as shown by mb98b , for an exponential disk mond predicts a tight correlation between @xmath145 and surface brightness of the form @xmath146 . the upper left panel of figure [ fig : xi ] plots @xmath147 as function of @xmath114 for the data . the thin solid line corresponds to @xmath148 , and is plotted with an arbitrary normalization . clearly , the data reveals a narrow correlation between @xmath145 and central surface brightness , with a slope close to that predicted by mond . mb98b considered this yet another success of modified newtonian dynamics . however , as we show in appendix b , the dark matter hypothesis predicts the same relation , i.e. , @xmath149 , and the observed relation between @xmath145 and @xmath114 thus does not provide a useful test to discriminate between the mond and dm hypotheses . this is also evident from figure [ fig : xi ] , which shows that each of our three models yields a relation between @xmath145 and @xmath114 that is in excellent agreement with the observations . it is important , though , to note that under the dm hypothesis , the amount of scatter can be significantly larger than observed . it is because of the sn feedback , the efficiency of which decreases with increasing halo mass , that the amount of scatter in model l5 is as small as it is ( see appendix b for a detailed discussion ) . additional support for mond was recently presented by mcgaugh ( 1998 ) , who showed that current observations hint to the presence of a characteristic acceleration , the main assumption on which mond is based . the upper panels of figure [ fig : acc ] plot data , kindly provided by stacy mcgaugh , of the cumulative mass discrepancy @xmath150 against radius @xmath151 ( panels on the right ) , against orbital frequency @xmath152 ( middle panels ) , and against centripetal acceleration @xmath153 ( panels on the right ) . here @xmath150 is defined as @xmath154},\ ] ] and corresponds to the ratio of total mass ( based on the newtonian equations ) to luminous mass ( stars plus cold gas ) within radius @xmath151 . each data point represents one resolved measurement in the rotation curve of a disk galaxy . the galaxies in the sample span a wide range in luminosities and surface brightnesses and were taken from the compilation of sanders ( 1996 ) and de blok & mcgaugh ( 1998 ) . the plot of @xmath150 versus @xmath151 reveals that the mass discrepancy sets in at a wide range of different radii ; individual rotation curves are easily discerned . this scatter is reduced when plotting @xmath150 versus the circular frequency , and becomes minimal when plotted versus the acceleration @xmath155 . the data clearly reveal that in each disk galaxy , the mass discrepancy starts to become significant below an acceleration of @xmath156 , as predicted by mond . the lower two rows of panels of figure [ fig : acc ] show the same plots for the two mond models . here we have randomly selected 40 galaxies from models m1 and m2 , and for each model galaxy , we computed @xmath150 at 15 radii , sampled uniformly between @xmath157 and the radius at which the projected hi column density is equal to @xmath158 . the two mond models reproduce the data extremely well , which is to be expected , given the presence of the characteristic acceleration @xmath32 and the fact that mond was specifically designed to fit rotation curves . more remarkably , model l5 also yields a characteristic acceleration ( see panels in second row of figure [ fig : acc ] ) . although the agreement with the data is not quite as good as in the case of models m1 and m2 , the main characteristic of the data , namely the minimization of scatter when @xmath150 is plotted versus @xmath155 , is reproduced well . most importantly , this is achieved without any fine - tuning of our model ; after setting the feedback parameters to fit the slope of the empirical tf relation , the same model reveals a characteristic acceleration . this is in clear contradiction with mcgaugh ( 1998 ) , who argued that this phenomenology can not be reproduced with dm models . not only is this a remarkable success of our dm model , it is also an argument against an argument in favor of mond . model l5 predicts mass discrepancies which are slightly too large in the range @xmath159 . the origin of this small deficit is discussed in the following . why can our dm models easily account for the @xmath128@xmath140 `` conspiracy '' , whereas mb98a have argued that no feasible solution exists ? combining equations [ [ massdiscr ] ] and [ [ xi ] ] , and taking into account that the total luminosity of an exponential disk scales as @xmath160 ) , one obtains @xmath161 . since , as shown in appendix b , we have @xmath162 one thus automatically predicts that @xmath163 ; i.e. , _ there is no conspiracy_. however , the fact that the scatter around the @xmath128@xmath140 relation is so small is not trivial , as the scatter in the @xmath145@xmath140 relation can be fairly large . it is only because of our particular feedback model , which we tuned to reproduce the empirical tf relation , that the scatter in the @xmath145@xmath140 relation , and therewith in the @xmath128@xmath140 relation , is so small ( see appendix b ) . we consider this another major succes for our feedback model . from the scaling relations listed above one derives that @xmath164 . we thus expect a narrow correlation between the characteristic accelerations and mass - to - light ratios of disk galaxies . therefore , the fact that our dm model is succesful in reproducing the narrow correlation between _ local _ accelerations and mass - to - light ratios is not too surprising . the rotation curves ( hereafter rcs ) of disk galaxies are the most direct indicators of a discrepancy between the luminous mass and the inferred dynamical mass for disk galaxies . the shapes of rotation curves therefore provide a clean laboratory for comparing models of gravitational dynamics . historically , astronomers have explained the detailed rcs of spiral galaxies by invoking the presence of non - luminous matter ( see reviews in sancisi & van albada 1987 , and ashman 1992 ) , while preserving the @xmath165 force law of newtonian gravity . more recently , some astronomers have used mond to explain the observed kinematics without requiring the presence of dark matter ( e.g. kent 1987 ; lake 1989 ; begeman et al . 1991 ; sanders 1996 ; de blok & mcgaugh 1998 ; sanders & verheijen 1998 ) . both approaches have found nearly equal success over a wide range of luminosities and surface brightnesses , well beyond the luminous , high - surface brightness galaxies for which the theories of dm and mond were developed . recently , an important problem for the cdm picture has emerged from hi studies of dwarf galaxies . the slowly rising rotation curves observed in these systems suggest a ( close to ) constant density core in their dark matter halos , inconsistent with the steeply cusped density profiles predicted for cdm ( flores & primack 1994 ; moore 1994 ; burkert 1995 ; burkert & silk 1997 ; stil 1999 , but see kravtsov et al . 1998 ) . whether mond is consistent with the rcs of dwarf galaxies , is an issue that is currently still under debate ( lake 1989 ; milgrom 1991 ; snchez - salcedo & hidalgo - gmez 1999 ) . recent studies have shown that more massive lsb disk galaxies reveal rotation curves that are similar to those of dwarf galaxies ( e.g. , de blok , mcgaugh & van der hulst 1996 ; van zee et al . 1997 ; pickering et al . 1997 ) , suggesting that the problem with the density profiles of dark matter halos is not limited to low - mass systems . indeed , mb98a have argued that the observed rotation curve of lsb galaxies are inconsistent with the nfw profile . mond , however , has been shown to yield remarkably good fits to the rcs of lsb galaxies ( de blok & mcgaugh 1998 ; sanders & verheijen 1998 ) . it is important to note , however , that the majority of the data on the rcs of the more massive lsb galaxies is severely affected by beam smearing ( see e.g. , swaters 1999 ) . van den bosch , robertson & dalcanton ( 1999 ) have shown that once these effects are taken into account , the rcs of the more massive lsb galaxies are in excellent agreement with dark matter halos with steep central cusps . given the unique opportunity to discriminate between dm and mond based on the rc shapes of disk galaxies , we now compare the rcs predicted by our dm and mond models in some detail . using our models which best fit the @xmath16-band tf relation ( l5 and m1 ) , we have selected model galaxies that are photometrically similar ( same scale length , same absolute magnitude , and same central surface brightness ) , and computed their rcs . in figure [ fig : rc ] we compare four sets of model galaxies that span four magnitudes in both surface brightness and mass ( see table [ tab : rcgal ] for the parameters of each of the eight galaxies ) . for the high surface brightness galaxies , ( d1 , m1 , d3 , and m3 ) the rcs for the mond and the dm models are very similar . for the lsb galaxies , however , the galaxies from the dm model predict rcs that are significantly steeper in the inner parts . this is the reason why dark halo fits to several lsb galaxies have been claimed to fail , and owes to the steep central cusp ( @xmath166 for the nfw profile used in our dm models ) of the dark halo . this also explains why @xmath167 is slightly overestimated for accelerations at around @xmath168 ( see figure [ fig : acc ] ) as compared to the observations . note , however , that the data on which figure [ fig : acc ] is based suffers strongly from the beam - smearing effects eluded to above , and it remains to be seen whether it is actually inconsistent with the predictions from our dm model ( see van den bosch et al . 1999 ) . the steepness of the central part of the rcs can be quantified by the parameter @xmath169 , which is defined as the radius at which the circular velocity equals 75 percent of @xmath134 . mb98a obtained @xmath169 for a number of disk galaxies with reasonably well resolved hi rotation curves , and found @xmath170 to strongly increase with decreasing luminosity and surface brightness . they compared this to predictions based on the models of dalcanton et al . ( 1997 ; hereafter dss97 ) , which are similar to the models presented here . according to mb98a , the models of dss97 predict an opposite behavior from that observed , with @xmath170 _ decreasing _ with decreasing magnitude , and they use this as strong evidence against dm models . to further test this , we compute @xmath170 for the model galaxies in each of our three models . results are plotted as function of @xmath113 and @xmath114 in figure [ fig : rcrad ] . the first characteristic to notice is that none of the models reveals a significant increase of @xmath170 with decreasing magnitude . the mond models reveal a modest increase of @xmath170 with decreasing surface brightness , but they reach a maximum at @xmath171 . the galaxies of model l5 never reach values of @xmath170 in excess of @xmath172 . both models are therewith in clear contradiction with the observed rcs presented by mb98a , which reach values as high as @xmath173 . the most likely explanation for this discrepancy is , once again , beam smearing , which tends to over - predict @xmath170 ( see e.g. , blais - ouellette , carignan & amram 1998 ) . data of higher spatial resolution is required to test this further . however , given the modest difference between the predictions from mond and the dm models , it seems unlikely that @xmath170 will allow to discriminate between the two scenarios not only are the results presented here inconsistent with data , they also disagree with the curves plotted by mb98a which they claim to represent the model predictions from dss97 . however , we do not understand how mb98a have obtained these predictions . part of the answer may lie in the fact that mb98a quote that they used models with @xmath174 `` as predicted by dss97 '' . however , the models of dss97 predict @xmath175 . nevertheless , even with this erroneous interpretation of dss97 s models , we are unable to reproduce the predictions of mb98a . the results presented here suggest that neither mond nor the dm hypothesis predict a strong dependence of @xmath170 on either luminosity or surface brightness . together with the fact that the mond and dm models yield results that differ only modestly , we conclude that they both are equally ( in)consistent with the data . the observed values of @xmath170 can not be used to simply rule against the dm hypothesis only . we have presented detailed models for the formation of disk galaxies in both a cdm and a mond universe . in the case of dm , the structure of the disk is governed by the mass and angular momentum of the proto - galaxy . for mond , however , the distribution of angular momenta of proto - galaxies is not known , and we have instead assigned scale lengths to the disks that are in agreement with observations . in addition to these recipes that determine the structure and dynamics of the disks , we include recipes that describe how ( part of ) the gas is transformed into stars over the lifetime of the galaxy . we take account of a stability related threshold density for star formation and feedback from supernovae . the models have been tuned to fit the observed @xmath16-band tf relation , and compared to numerous observations . both models do remarkably well in reproducing a wide variety of observations of disk galaxies that span several orders of magnitude in both luminosity and surface brightness . to more easily compare the different models a summary of the results is given in table [ tab : results ] , where we indicate , for each of the observational constraints discussed in this paper , whether the models are consistent with the data or not . in the case of dm ( model l5 ) , sn feedback is required to yield a tf slope as steep as observed . although the mond model without feedback ( m1 ) is in excellent agreement with the empirical tf relation , it does not reproduce the observed deficit of hsb dwarf galaxies with small mass - discrepancies . this can be remedied by introducing sn feedback ( model m2 ) , but at the cost of a tf relation which is too steep and reveals an amount of scatter that is only marginally consistent with the data . this is a serious problem for mond ; since pure dynamics already predicts a tf relation as steep as observed , it leaves virtually no room for any other galaxy characteristics to vary systematically with mass . the dm model is consistent with all observational constraints against which we have tested it . consequently , we strongly disagree with the picture that emerges from the literature ( for example , mb98a , mb98b , and mcgaugh 1998 ) , that the dm hypothesis suffers from numerous serious fine - tuning problems , which do not seem to have a clear - cut solution , whereas mond is free from such problems and capable of fitting virtually everything . we have shown here that once the dm model is tuned to fit the slope of the tf relation , it automatically passes all the tests devised by mb98a and mcgaugh ( 1998 ) to argue against it . in particular , our dm model reproduces * a close correlation between global mass - to - light ratio and surface brightness , such that hsb and lsb galaxies follow the same tf relation without a systematic offset . * the presence of a characteristic acceleration as observed . * a close relation between the characteristic acceleration , @xmath145 , and central surface brightness of the form @xmath176 . this is a remarkable result : there is no obvious reason why the dm model would reveal a characteristic acceleration , unlike in the case of mond , where it is integral to the theory . furthermore , it is encouraging that the same feedback parameters that yield the correct tf slope , result in an amount of scatter around the @xmath145@xmath140 relation that is in excellent agreement with observations , and in addition explains the observed absence of hsb dwarfs with small mass - discrepancies . of the list of observational facts about disk galaxies presented in [ sec : intro ] , the upper limit to the observed surface brightness of disk galaxies ( item 5 ) has not been addressed by our models . several studies have shown , however , that in the dm picture the presence of a maximum central surface brightness of disk galaxies is related to stability arguments ( dss97 ; mo et al . 1998 ; scorza & van den bosch 1998 ) . in [ sec : tfmond ] we have shown that the same stability argument also yields an upper limit on the central surface brightness of disks under the hypothesis of mond ( cf . milgrom 1989 ) . henceforth , both the dm and the mond models are consistent with the observational constraint of item 5 , as long as disk stability is taken into account . the main problem for cdm is related to rotation curve shapes . we have compared predicted rotation curve shapes within the context of both mond and cdm . only in the case of lsb systems do the two scenarios yield rcs that are significantly different . several studies have pointed out that dark halos with a steep cusp are inconsistent with the central rotation curves of lsb and dwarf galaxies . this problem is also evident from the fact that at accelerations of @xmath177 the dm model predicts mass - to - light ratios that are slightly too high . however , it is important to realize that most data on the more massive lsb disks is severely affected by beam - smearing . this tends to underestimate the central gradients of rotation velocities , especially in galaxies that have a central hole in their hi distribution . when beam smearing is taken into account , the hi rotation curves of massive lsb galaxies are consistent with dark halos that follow a nfw density profile ( van den bosch et al . further indications that beam smearing plays an important role comes from a comparison of the ratio @xmath170 . both the dm and the mond models predict ratios that are lower than observed , which is most likely due to beam smearing . therefore , in order to discriminate between mond and dm , high resolution rotation curves of lsb disk galaxies are required . such data is currently only available for dwarf galaxies , which , because of their relative proximity , have been observed with high spatial resolution . it has been demonstrated convincingly that the rotation curves of these low - mass systems are inconsistent with centrally cusped dark matter halos . it is currently still under debate whether mond can fit these rotation curves . if it can , this is where mond has a clear advantage over cdm , unless ( close to ) constant density cores can be produced in dark halos that form in a cdm universe ( see e.g. , navarro , eke & frenk 1996 , kravtsov et al . 1998 , and bullock et al . 1999 for possible solutions ) . one might argue that within the dm scenario one can fit basically anything as long as there are a sufficient number of free parameters . in that respect it is important to realize that model l5 has only a very limited number of truly free parameters . where possible , we have used parameters that have either empirically determined values , or that are otherwise constrained : the dark halo properties are taken from high resolution numerical simulations combined with the press - schechter formalism , @xmath19 is constrained by stellar population models , the star formation recipe uses values for @xmath7 and the schmidt law that have been determined empirically , and @xmath88 is constrained by numerical simulations . this leaves only @xmath133 and @xmath27 as real free parameters . tuning these two parameters to obtain a tf relation with the observed slope of @xmath178 , the model predicts gas mass fractions , characteristic accelerations , an @xmath128 - @xmath140 `` conspiracy '' , and global mass - to - light ratios which are in excellent agreement with observations , without additional tuning of the parameters . furthermore , we note that while mond is often presented as being nearly free of fine - tuning , in order to match the systematic properties of disk galaxies it is necessary to adjust the mond feedback parameters to the same degree as required for dm . probably the most amazing aspect of the models presented here , is that the dm and mond models are so very similar . however , both mond and dm were constructed to fit the rotation curves of disk galaxies . the fact that both theories correctly predict many other properties , which are themselves closely related to the internal dynamics ( i.e. , tf relation , gas mass fractions that are set by stability related threshold densities , sn feedback whose efficiency depends on the escape velocity , etc ) , should therefore not be seen as too remarkable . it has often been argued that even if mond turns out to not be correct , one should provide an explanation as to why it fits the properties of galaxies so well . the demonstrations in this paper suggest that _ any _ theory which yields stable disks and fits their rotation curves , would probably perform as well as any of the models presented here . this work has benefited greatly from discussions with george lake . we are grateful to stacey mcgaugh for sending us his data in electronic format , and to the anonymous referee for his suggestions that helped to improve the paper . fvdb was supported by nasa through hubble fellowship grant # hf-01102.11 - 97.a awarded by the space telescope science institute , which is operated by aura for nasa under contract nas 5 - 26555 . in order to compare our models , which yield @xmath16-band magnitudes and surface brightnesses , to @xmath109-band data , we require an estimate of @xmath111 colors for disk galaxies . to determine an empirical @xmath111 color magnitude relation we compiled apparent magnitudes in @xmath109 and @xmath16 as well as distances for a total of 139 spiral galaxies of type sb or later from the following sources : forty - two ursa - major spirals from the sample of verheijen ( 1997 ) , sixty - one galaxies from the sample of de jong & van der kruit ( 1994 ) , and thirty - six galaxies from the sample of edge - on spirals of dalcanton et al . absolute magnitudes were computed assuming @xmath179 and using the distances as quoted by these authors . finally , we used the reddening maps of schlegel , finkbeiner & davis ( 1998 ) , with the @xmath180 extinction law of cardelli , clayton & mathis ( 1989 ) and odonnell ( 1994 ) , to correct for external extinction . the resulting @xmath111 color magnitude is shown in the upper left panel of figure [ fig : colors ] . the solid line is the best linear fit with @xmath181 the upper right panel plots the predicted @xmath109-band magnitude , @xmath182 versus the observed value ( for @xmath183 ) . the data covers over eight magnitudes in @xmath113 and reveals that @xmath184 is a fairly good estimate of the observed @xmath109-band magnitude ( with a standard deviation of @xmath185 mag ) . the lower left panel of figure [ fig : colors ] plots the @xmath111 color of the central surface brightness of the disks , versus @xmath99 for the data of verheijen ( 1997 ) and de jong & van der kruit ( 1994 ) . unfortunately , at this time no central surface brightnesses are available for the data of dalcanton et al . note that @xmath111 and @xmath186 are generally not the same , since the disk scale lengths derived from the @xmath109 and @xmath16-band data can differ considerably ( see e.g. , de jong 1996b ; verheijen 1997 ) . the lower right panel of figure [ fig : colors ] plots the predicted central surface brightness in the @xmath109-band , @xmath187 , versus the observed value . here @xmath188 is calculated from @xmath189 and the color - magnitude relation of equation ( [ colmag ] ) . note that the agreement is reasonable , but that the scatter is , with @xmath190 mag , slightly higher than for the absolute magnitudes . in [ sec : acc ] we show that both the dm and mond models yield a narrow correlation between the central surface brightness of the disk and the parameter @xmath145 defined by equation ( [ xi ] ) . here we investigate the reason for this narrow relation . in the case of mond , @xmath191 , and , if we assume that the galaxy is a pure exponential disk , it is straightforward to show that @xmath192 ( see mb98b ) , with @xmath193 the total mass - to - light ratio of the galaxy . in order to obtain a similar expression under the hypothesis of dm , we make the simplifying assumption that dark halos are represented by isothermal spheres , and that @xmath194 . upon writing @xmath195 ( cf . equation [ [ tfdm ] ] ) , and since @xmath196 we obtain @xmath197 if we now set @xmath198 and @xmath199 ( as for model l5 , see paper i ) , we can write @xmath200 under the dm hypothesis one thus expects a similar relation between @xmath145 and @xmath201 as under mond , apart from an offset which depends on @xmath24 and the galaxy formation efficiency @xmath69 . since disks with the same central surface brightness can be embedded in halos with different rotation velocities ( cf . figure [ fig : sbvel ] ) , one expects a reasonable amount of scatter ; a variation in @xmath24 of a factor five , yields a scatter of @xmath202 in @xmath147 . however , this can be significantly reduced when sn feedback is important ; the feedback efficiency will be larger , leading to smaller values of @xmath69 in systems with lower values of @xmath24 , thus reducing the amount of scatter . indeed , the scatter revealed by model l5 ( see upper panel of figure [ fig : xi ] ) is somewhat larger than in the case of mond , but significantly smaller than @xmath202 and therewith consistent with the data . we thus conclude that the observed relation @xmath203 does not discriminate between the dm and the mond hypotheses . however , the small amount of scatter observed may be considered evidence for efficient sn feedback , at least within the context of dm . llcclrcc l5 & cdm & @xmath204 & @xmath205 & @xmath206 & @xmath207 & @xmath208 & @xmath209 + m1 & mond & @xmath204 & @xmath210 & @xmath211 & @xmath211 & @xmath212 & @xmath213 + m2 & mond & @xmath204 & @xmath210 & @xmath206 & @xmath214 & @xmath215 & @xmath216 + cccrcrccc d1 & l5 & @xmath217 & @xmath218 & @xmath219 & @xmath220 & @xmath221 & @xmath222 & @xmath223 + m1 & m1 & @xmath217 & @xmath218 & @xmath219 & @xmath224 & @xmath225 & @xmath226 & @xmath226 + & & & & & & & & + d2 & l5 & @xmath217 & @xmath227 & @xmath228 & @xmath229 & @xmath230 & @xmath231 & @xmath232 + m2 & m1 & @xmath217 & @xmath233 & @xmath228 & @xmath234 & @xmath235 & @xmath226 & @xmath226 + & & & & & & & & + d3 & l5 & @xmath236 & @xmath237 & @xmath238 & @xmath239 & @xmath240 & @xmath241 & @xmath242 + m3 & m1 & @xmath236 & @xmath237 & @xmath243 & @xmath244 & @xmath245 & @xmath226 & @xmath226 + & & & & & & & & + d4 & l5 & @xmath236 & @xmath232 & @xmath246 & @xmath247 & @xmath248 & @xmath249 & @xmath250 + m4 & m1 & @xmath236 & @xmath232 & @xmath251 & @xmath252 & @xmath253 & @xmath226 & @xmath226 + cccccc near - ir tf relation & 1 & 1 & @xmath254 & @xmath254 & @xmath255 + @xmath256 & 4 & 2,3 & @xmath254 & @xmath254 & @xmath254 + @xmath128 _ vs. _ @xmath113 & 3 & 4 & @xmath254 & @xmath255 & @xmath254 + @xmath128 _ vs. _ @xmath114 & 3 & 6 & @xmath254 & @xmath254 & @xmath254 + @xmath114 _ vs. _ @xmath134 & - & 5 & @xmath254 & @xmath255 & @xmath254 + @xmath145 _ vs. _ @xmath114 & - & 7 & @xmath254 & @xmath254 & @xmath254 + @xmath257 & 6 & 8 & @xmath258 & @xmath254 & @xmath254 + rc shapes & 2 & 9,10 & @xmath258 ? & @xmath254 ? & @xmath254 ?

landau level quantization in graphene reflects the dirac nature of its quasiparticles and has been found to exhibit an unusual integer quantum hall effect . in particular the lowest landau level can be thought as shared equally by electrons and holes and this leads to characteristic behaviour of the magneto - optical conductivity as a function of frequency for various values of the chemical potential . particular attention is paid to the optical spectral weight under various absorption peaks and its redistribution as is varied . we also provide results for magnetic field as well as chemical potential sweeps at selected fixed frequencies which can be particularly useful for possible measurements in graphene . both diagonal and hall conductivity are considered .

0705.3783

recent experimental studies @xcite of the dynamics of electrons and holes in graphene ( a single atomic layer of graphite @xcite ) have revealed unusual behaviour related to the dirac nature of its quasiparticles . two dimensional graphene has a honeycomb lattice structure with two atoms per unit cell . its band structure consists of two inequivalent pairs of cones with apex at the brillouin zone corners . for zero chemical potential the lower energy cones are completely filled and the upper empty . in a graphene device an applied gate voltage can be used to introduce electrons in the upper band or , by voltage reversal , holes in the lower band ( cones ) . the quasiparticles in graphene obey dirac @xcite rather than the schrdinger equation and this has profound implications for their dynamics . the unconvetional quantum hall effect was expected theoretically @xcite and recently observed @xcite to have half ( divided by spin and valley degeneracy ) rather than integer filling factors . the predicted phase shift of @xmath3 in the de haas van alphen @xcite and shubnikov de haas @xcite oscillations was also seen experimentally @xcite . another feature related to the dirac - like character of the carriers in graphene seen in the dc measurements is a finite effective cyclotron mass for the massless dirac quasiparticles which varies as the square root of the number of carriers @xcite . in this paper we consider the magneto - optical conductivity of graphene . work without a magnetic field includes the calculations of ando , zheng and suzuura @xcite who considered the effect on the frequency dependent conductivity of short- and long - range scatterers in a self consistent born approximation . more recent work @xcite describes several anomalous properties of the microwave conductivity of graphene with , as well as without magnetic field . these properties are directly related to the dirac nature of the quasiparticles . several analytic formulas for the longitudinal as well as hall ac conductivity are given in the paper @xcite . they also present extensive results for dc properties and preliminary data on the real part of @xmath4 vs @xmath0 in the optical region . another extensive work by peres et al . @xcite on the ac conductivity in graphene treats localized impurities in a self - consistent fashion as well as extended edge and grain boundaries including also effects of electron - electron interactions and self - doping . in this paper we follow most closely ref . @xcite which we extend in several directions . an aim is to provide simpler analytic formulae which should prove useful in the analysis of experiment and check on their accuracy . another is to consider magnetic field as well as chemical potential sweeps possible in graphene field effect transistor devices @xcite . recent work by li et al . in organic metals @xcite has demonstrated that the ac measurements are also possible in such devices . consistent with our aim , we consider impurities only in simplified scattering rate model and neglect real part renormalizations although these could easily be included if wished . while the renormalization effects beyond those included in this simplified model might become important for the interpretation of future experiment , we point out that so far the free quasiparticle model with associated transport lifetime has been remarkably successful in understanding the dc results of refs . @xcite . in sec . [ sec : cond - general ] we relate the magneto - optical conductivity tensor to the dirac fermionic green s function through kubo formula . the general formulas obtained can be greatly simplified and closed form expressions are obtained in two cases . in general the fermionic self - energy can depend on energy and landau level index as well as temperature and value of the external magnetic field @xmath2 . under the assumption that variations with landau level index @xmath5 can be neglected , the sum over transitions between neighboring landau levels can be carried out explicitly and a closed form expression is obtained for conductivity in terms of digamma function . a single integral over an internal frequency remains . the expression obtained is suitable for calculations of the ac conductivity for any value of temperature , chemical potential and magnetic field . its microwave frequency limit has been used to describe properties of graphene in ref . @xcite . in the case when energy dependence of the fermionic self - energy can be neglected , the internal integration over energy can be done and what remains is the sum over the landau level index @xmath5 of lorentzian forms multiplied by thermal factors and algebraic weighting factors . the weighting factors depend on the landau level energies as well as the excitonic gap ( see e.g. refs . @xcite ) should one wish to include this possibility . when we compare numerical results obtained from lorentzian model and from the previous more general expressions in the limit of constant scattering rate , we find good quantitative agreement between the two . this provides support for the analysis of experimental data @xcite . in sec . [ sec : low - b ] we consider the low field limit for the lorentzian model derived in sec . [ sec : lorentzian ] and establish its correspondence with previously known results . in sec . [ sec : cond - diagonal ] we present the numerical results for the real part of the diagonal conductivity as a function of photon energy @xmath0 for fixed value of magnetic field and various values of chemical potential . we also provide results for fixed photon energy while sweeping either chemical potential or external magnetic field which should prove useful in comparing with experiments . the effect of opening of an excitonic gap on the absorption lines is described . in sec . [ sec : weight ] a discussion of optical spectral weight redistribution by the magnetic field is given . section [ sec : cond - hall ] is structured in parallel to sec . [ sec : cond - diagonal ] but deals with the absorptive part of the off - diagonal hall magneto - optical conductivity . discussion and conclusions are given in sec . [ sec : concl ] . some of the algebra needed in this work is found in an [ sec : a ] . the optical conductivity tensor is calculated using the kubo formula @xmath6 where @xmath7 is the retarded current - current correlation function which in the bubble approximation is given by @xmath8 , \quad i = 1,2\ ] ] where @xmath9 is the fermi distribution function @xmath10 $ ] , @xmath11 not only takes care of the @xmath12 @xmath13 matrices [ @xmath14 , but includes also the summation over flavour ( spin ) index . here @xmath15\ ] ] is the spectral function ( decomposed over landau levels ) associated with the translation invariant part of the dirac fermion green s function in an external magnetic field @xmath16 applied perpendicular to the plane along the positive @xmath17 direction ( see e.g. refs . @xcite ) . in eq . ( [ spect - fun - magfield ] ) @xmath18 , \quad f_{2 } ( { \mathbf{k}})=4v_f{\mathbf{k}}{\pmb{\gamma}}l_{n-1}^{1}\left ( \frac{2c{\mathbf{k}}^{2}}{|eb| } \right)\ ] ] with @xmath19 being projectors and @xmath20 the generalized laguerre polynomials . by definition , @xmath21 and @xmath22 . the energies of the relativistic landau levels in eq . ( [ spect - fun - magfield ] ) are @xmath23 where the energy scale associated with the magnetic field expressed in the units of temperature reads @xmath24 where @xmath25 is the fermi velocity in graphene given in m / s and the field @xmath2 is given in tesla . in the following we set @xmath26 , and in some places @xmath27 , unless stated explicitly otherwise . for the numerical calculations we use the value @xmath28 @xcite which leads to the relationship @xmath29 . we consider relatively low fields @xmath30 , where spin splitting is unresolved @xcite , so that we can assume that the above mentioned summation over flavor index simply gives @xmath31 in all expressions below . the landau level energies ( [ m_n ] ) include also an excitonic gap @xmath32 . the physical meaning of this singlet excitonic gap is directly related to the electron density imbalance between the a and b sublattices of the bi - particle hexagonal lattice of graphene @xcite and there are strong indications @xcite that it was indeed observed in recent experiments @xcite . we will see here that optical measurements done on graphene can be very useful in investigations of the excitonic gap . finally , the scattering rate @xmath33 is expressed via the retarded fermion self - energy , @xmath34 which in general depends on the energy , temperature , field and the landau levels index @xmath5 . this self - energy , which in general has also a real part , has to be determined self - consistently from the schwinger - dyson equation . this equation can be solved analytically @xcite and numerically as was done in refs . @xcite . in our paper we consider @xmath35 as a phenomenological parameter for the two cases : ( i ) @xmath36 is independent of the landau level index @xmath5 and ( ii ) @xmath37 is independent of the energy @xmath38 . under these assumptions the optical conductivity can be studied analytically . assuming that @xmath33 is independent of the landau level index , i.e. @xmath39 , one can calculate the sum over landau levels and express the diagonal ac conductivity in the closed form @xcite @xmath40{\rm re}\left\{\frac{2b}{\delta^2- ( \tilde\omega+i\gamma)^2}\left[\xi_{1}(-b)-\xi_{2}(-b ) \right ] \right.\\ \fl+ \left . \left[\xi_{1}(-b)+\xi_{1}(+b)-\xi_{2}(-b)-\xi_{2}(+b ) \right]\psi\left(\frac{\delta^2-(\tilde \omega+i\gamma)^2}{2b}\right)+(\tilde \omega\leftrightarrow \tilde \omega^\prime , \gamma\leftrightarrow\gamma^\prime ) \right\ } \nonumber . \ ] ] here @xmath41 is the digamma function , we denoted @xmath42 , and also included the renormalization of energy caused by the real part of self - energy @xmath43 , @xmath44 , @xmath45 and introduced the following short - hand notations @xmath46[\tilde\omega+\tilde\omega^\prime+i(\gamma+\gamma^\prime)]\pm 2b } , \nonumber \\ \fl \xi_{2}(\pm b ) \equiv \xi_{2}(\tilde\omega,\tilde\omega^\prime,\gamma,\gamma^\prime,\pm b)= \frac{(\tilde\omega^\prime - i\gamma^\prime)(\tilde\omega+i\gamma)-\delta^2 } { [ \tilde\omega-\tilde\omega^\prime+i(\gamma+\gamma^\prime)][\tilde\omega+\tilde\omega^\prime+ i(\gamma-\gamma^\prime ) ] \pm 2b}. \ ] ] the advantage of eq . ( [ optical - diagonal ] ) is that the @xmath41 function contains the contribution to conductivity from _ all _ transitions between neighboring landau levels . an infinite number of these transitions need to be taken into account when the limit @xmath47 is considered , so that the zero field limit is easily treatable @xcite on the base of eq . ( [ optical - diagonal ] ) . another important feature of eq . ( [ optical - diagonal ] ) is that we kept the frequency dependent impurity scattering rate @xmath48 which allows us to investigate its influence on the shape of the drude peak @xcite . although in our work we will mostly use eq . ( [ optical - diagonal ] ) for the numerical computations of the diagonal conductivity , there is a possibility to derive simple approximate expressions for the diagonal and hall conductivities which turn out to be very useful when one is interested in the resonance peaks of these conductivities in the infrared region . for analyzing experimental data it is very useful to have a magneto - optical lorentzian model for the complex conductivity , @xmath49 due to inter- and intraband landau level transitions @xcite . for the case of dirac fermions it is considered in [ sec : a ] , whereof we obtain the complex diagonal conductivity @xmath50 + [ n_f(-m_{n+1 } ) - n_f(-m_{n})]\right ) \right . \fl\left.\times\left(\frac{1}{m_n - m_{n+1}+\omega + i ( \gamma_n + \gamma_{n+1})}-\frac{1}{m_n - m_{n+1}-\omega -i ( \gamma_n + \gamma_{n+1})}\right)\frac{1}{m_{n+1}-m_n}\right.\nonumber \\ \fl + \left.\left(1+\frac{\delta^2}{m_nm_{n+1}}\right)\left([n_{f}(-m_n ) - n_f(m_{n+1 } ) ] + [ n_f(-m_{n+1 } ) - n_f(m_{n})]\right)\frac{1}{m_{n+1}+m_n}\right . \nonumber \\ \fl \times\left.\left(\frac{1}{m_n + m_{n+1 } + \omega+ i(\gamma_n + \gamma_{n+1 } ) } -\frac{1}{m_n + m_{n+1 } -\omega- i(\gamma_n + \gamma_{n+1})}\right)\right\ } , \nonumber \ ] ] and the complex hall conductivity @xmath51 - [ n_f(-m_{n+1 } ) - n_f(-m_{n})]\right)\nonumber \\ \fl \times \left\{\left(1-\frac{\delta^2}{m_nm_{n+1}}\right)\frac{1}{m_{n+1}-m_n } \right . \nonumber \\ \fl \left.\times\left(\frac{1}{m_n - m_{n+1}+\omega + i ( \gamma_n + \gamma_{n+1})}+\frac{1}{m_n - m_{n+1}-\omega -i ( \gamma_n + \gamma_{n+1})}\right)\right . \fl -\left(1+\frac{\delta^2}{m_nm_{n+1}}\right)\frac{1}{m_{n+1}+m_n}\nonumber \\ \fl \times\left . \left(\frac{1}{m_n + m_{n+1 } + \omega+ i(\gamma_n + \gamma_{n+1 } ) } + \frac{1}{m_n + m_{n+1 } -\omega- i(\gamma_n + \gamma_{n+1})}\right)\right\}. \ ] ] in deriving eqs . ( [ sigma_xx - complex - corrected ] ) , ( [ sigma_xy - complex - corrected ] ) we assumed that the impurity scattering rate is independent of the energy @xmath38 , but kept its dependence on the landau level index @xmath5 . this assumption allows us to eliminate the integration over @xmath38 which is present in eq . ( [ optical - diagonal ] ) . to preserve the landau index dependence , the sum over transitions between neighboring landau levels is retained in eqs . ( [ sigma_xx - complex - corrected ] ) , ( [ sigma_xy - complex - corrected ] ) . based on eqs . ( [ sigma_xx - complex - corrected ] ) , ( [ sigma_xy - complex - corrected ] ) one can easily write down separate expressions for @xmath52 , @xmath53 , @xmath54 , @xmath55 and verify that diagonal and off - diagonal conductivities satisfy kramers - kronig relations . a big advantage of eqs . ( [ sigma_xx - complex - corrected ] ) , ( [ sigma_xy - complex - corrected ] ) is that they are more suitable for numerical computations and it is sufficient to include only a few terms in the sum even in relatively low magnetic field . also they are useful for the description of the resonance peaks when the landau level index dependence is more important than the energy dependence of the scattering rate which is included in eq . ( [ optical - diagonal ] ) . all figures in this paper were computed for @xmath56 , so that for calculation of the diagonal conductivity we used eq . ( [ optical - diagonal ] ) and checked that it gives almost identical results to eq . ( [ sigma_xx - complex - corrected ] ) . since the corresponding full expression for the optical hall conductivity derived in ref . @xcite is rather difficult for numerical calculations , in the present paper we compute all results for the hall conductivity using eq . ( [ sigma_xy - complex - corrected ] ) . in the case of landau level index independent width , @xmath57 , eqs . ( [ sigma_xx - complex - corrected ] ) , ( [ sigma_xy - complex - corrected ] ) acquire an even simpler form @xmath58 + [ n_f(-m_{n+1 } ) - n_f(-m_{n})]}{(m_{n+1}-m_n)^2-(\omega+2i\gamma)^2}\frac{1}{m_{n+1}-m_n } \right . \nonumber \\ \fl + \left.\left(1+\frac{\delta^2}{m_nm_{n+1}}\right)\frac{[n_{f}(-m_n ) - n_f(m_{n+1 } ) ] + [ n_f(-m_{n+1 } ) - n_f(m_{n})]}{(m_{n+1}+m_n)^2-(\omega+2i\gamma)^2}\frac{1}{m_{n+1}+m_n}\right\ } \nonumber \ ] ] and @xmath59 - [ n_f(-m_{n+1 } ) - n_f(-m_{n})]\right ) \\ \fl \times \left\{\left(1-\frac{\delta^2}{m_nm_{n+1}}\right ) \frac{1}{(m_{n+1}-m_n)^2-(\omega+2i\gamma)^2 } \right . \fl \left . + \left(1+\frac{\delta^2}{m_nm_{n+1}}\right)\frac{1}{(m_{n+1}+m_n)^2-(\omega+2i\gamma)^2 } \right\}. \nonumber \ ] ] one can see that the conductivity @xmath60 is even function of @xmath1 while @xmath61 is an odd one . now we check that eqs.([sigma_xx - complex - corrected1 ] ) , ( [ sigma_xy - complex - corrected1 ] ) reproduce correctly the limit @xmath47 . introducing the continuum variable @xmath38 instead of @xmath62 given by eq . ( [ m_n ] ) and replacing the sum over @xmath5 by the integral , we obtain @xmath63 \nonumber\end{aligned}\ ] ] and @xmath64 , \nonumber\end{aligned}\ ] ] where we restored planck constant @xmath65 in the overall prefactor . here the terms with the factor @xmath66 are intraband and the terms containing the factor @xmath67 are interband . the expressions ( [ sigma_xx - lowb ] ) and ( [ sigma_xy - lowb ] ) are obtained under the condition @xmath68 . the intraband term of eq . ( [ sigma_xx - lowb ] ) can be written in the familiar drude form @xmath69 which for @xmath70 reduces to eq . ( 2 ) of ref . @xcite . the whole expression ( [ sigma_xx - lowb ] ) which includes the interband term , also reduces to the other limiting cases considered in refs . @xcite . in particular , an unusual feature of graphene is that in the high frequency limit the interband contribution is a constant , @xmath71 here to rely on the linearized dirac approximation we assumed that @xmath0 is still well below a large band edge . the real part of the hall conductivity ( [ sigma_xy - lowb ] ) for @xmath72 reduces to the expression @xmath73 which is in agreement with eq . ( 4.3 ) of ref . @xcite . on the other hand , in the high frequency limit eq . ( [ sigma_xy - lowb ] ) gives @xmath74 , \qquad \omega \to \infty.\ ] ] this behavior also follows from eq . ( [ sigma_xy - complex - corrected ] ) which is valid in an arbitrary field @xmath2 . interestingly , expression ( [ hall - high - frequency ] ) is sensitive to the relationship between @xmath75 and @xmath32 and this feature can be used for the gap detection [ see the discussion of figs . [ fig:4 ] and [ fig:9 ] below ] . using eqs . ( [ diagonal - high - frequency ] ) and ( [ hall - high - frequency ] ) we obtain the weak field optical hall resistivity @xmath76 , \qquad \omega \gg \mu,\delta , t.\end{aligned}\ ] ] accordingly , the high frequency optical hall coefficient for @xmath77 and @xmath78 @xmath79 contains the information on the value of the fermi velocity in graphene . finally , the imaginary part of @xmath80 which follows from eq . ( [ sigma_xy - lowb ] ) is given by the expression @xmath81\end{aligned}\ ] ] which in high frequency limit becomes @xmath82 the last equation shows that in this limit real part of @xmath83 given by eq . ( [ hall - high - frequency ] ) is the leading term . in fig . [ fig:1 ] we show the results based on a numerical evaluation of the full equation ( [ optical - diagonal ] ) for the real part of the longitudinal conductivity @xmath84 in units @xmath85 as a function of frequency in units of @xmath86 . except for the long dashed ( red ) curve which was obtained in the limit of vanishing external magnetic field ( namely @xmath87 ) and is included for comparison , the other three curves are for @xmath88 . they differ in value of the chemical potential @xmath1 . in all cases the temperature @xmath89 , the impurity scattering @xmath90 , and the excitonic gap @xmath77 . for reference in scrutinizing the curves , the frequency of the @xmath91 landau level , @xmath92 $ ] , @xmath93 $ ] and @xmath94 $ ] , so that for the dash - dotted ( black ) curve @xmath1 falls below the energy of the @xmath91 level ( see left side of fig . [ fig:2 ] ( c ) ) , to the energy in kelvins is @xmath95 = 1.4387 \mbox{k } \cdot \mbox{cm } \omega[\mbox{cm}^{-1}]$ ] . ] for the solid ( blue ) curve it falls between @xmath91 and @xmath96 levels ( see middle of fig . [ fig:2 ] ( c ) ) , and for the short dashed ( green ) it is between @xmath96 and @xmath97 ( right side of fig . [ fig:2 ] ( c ) ) . the energies of the peaks are @xmath98 , @xmath99 , @xmath100 , etc . and @xmath101 , @xmath102 , etc . note that the position and the intensity of the last two peaks in fig . [ fig:1 ] ( largest frequency @xmath0 ) remains the same for all chosen values of the chemical potential @xmath1 . when @xmath1 falls between @xmath103 and @xmath104 [ short dashed ( green ) curve ] the intensity in the third last peak has dropped to half the value it has in solid ( blue ) curve , while the fourth last peak has merged into a low intensity background , as has the fifth last , which is seen only in the dash - dotted curve . also a new peak has appeared at @xmath105 which was not present in the dash - dotted curve . similarly for @xmath1 between @xmath98 and @xmath103 [ solid ( blue ) curve ] the intensity of the fourth highest energy peak has dropped to half the intensity it has in the dash - dotted curve and the peak at @xmath98 is entirely missing , having merged into the low intensity background . a new peak has appeared at @xmath106 . finally when @xmath1 is below @xmath98 ( dash - dotted curve ) there is no peak below @xmath98 and the line at @xmath98 always has full intensity . _ whatever the value of @xmath1 this line will never be seen to half its intensity and this is the hallmark of the dirac nature of the quasiparticles . _ a schematic which helps us understand the behaviour of the absorption lines that we have just described is given in fig . [ fig:2 ] ( c ) . on the left of the figure we show the energies of the landau levels @xmath107 of eq . ( [ m_n ] ) as solid ( red ) dots along with their quantum numbers @xmath108 . the dirac cones which come in pairs , with the positive and negative energies and exist at @xmath109 and @xmath110 points in graphene brillouin zone are also shown . the three values of chemical potential considered in fig . [ fig:2 ] ( c ) are shown in heavy solid ( violet ) horizontal lines . the possible optical transitions in each case are indicated as vertical arrows and connect levels @xmath5 to @xmath111 only . moving from left to right we see first a single transition from @xmath112 to @xmath113 , then a pair of interband from @xmath114 to @xmath115 and @xmath116 to @xmath117 followed by another pair from @xmath118 to @xmath119 and @xmath120 to @xmath121 etc . for the middle set of lines the first is an intraband transition from @xmath117 to @xmath122 followed by a single interband from @xmath123 to @xmath122 and then a pair from @xmath118 and @xmath119 and @xmath120 to @xmath121 etc . finally in the set of transitions at the right of the figure there is an intraband from @xmath122 to @xmath119 followed by a single interband from @xmath124 to @xmath119 and a pair from @xmath120 to @xmath125 and @xmath126 to @xmath119 etc . this is precisely the pattern we have seen in fig . [ fig:1 ] . we note one more aspect of the anomalous line at the far left of the figure for the transition form @xmath127 to @xmath91 landau level . it is the only line which can not be unambiguously assigned to inter or intraband because the sate at @xmath127 falls at the apex of the dirac cones , where positive and negative energy cones meet and hence they share this state equally . this is further illustrated in figs . [ fig:2 ] ( a ) and ( b ) , where two sets of cones at @xmath109 and @xmath110 are shown separately for the case of @xmath77 in frame ( a ) and finite excitonic gap in frame ( b ) . in this second case we see clearly that points @xmath109 and @xmath110 react differently under a finite magnetic field . for the cone on the left @xmath127 level has moved to energy @xmath128 and for the cone on the right it has moved to @xmath129 . note that for the value of chemical potential shown as a solid horizontal ( violet ) line , the @xmath130 to @xmath131 transition ( vertical arrow ) on the left is now definitely interband and the @xmath129 to @xmath131 on the right is intraband . the ambiguity in designation of the @xmath127 level present in the top frame ( a ) of fig . [ fig:2 ] is lifted when the gap becomes finite . the rather complicated pattern of behaviour just described can be understood simply from the mathematics of the previous section in the limit @xmath132 and @xmath133 . taking @xmath77 , @xmath134 , and @xmath135 one obtains from eq . ( [ lorentz - full ] ) @xmath136 \frac{\delta(m_n+m_{n+1}-\omega)}{m_n+ m_{n+1 } } \right . \\ \left . \lo+ [ n_f(m_n)-n_f(m_{n+1 } ) ] \frac{\delta(m_n - m_{n+1}+\omega)}{m_{n+1}-m_n } \right\}. \ ] ] for @xmath137m_n , m_{n+1}[$ ] the @xmath138 thermal factor @xmath139 while @xmath140 the line in @xmath52 with the frequency @xmath141 occurs only for @xmath142 with the others missing . further the line with @xmath143 has half of the weight it would have if it occurred in another case , namely , @xmath144 , while lines for @xmath145 have full weight @xmath146 . note that the lines with the frequency equal to the difference in landau level energies have weight @xmath147 . these lines are always present except for the case when @xmath1 falls below @xmath98 when the single line at @xmath98 has weight @xmath148 . in summary , we have seen in the above discussion that as @xmath1 moves through higher and higher values of @xmath149 the lines below @xmath150 disappear into the background with a new line appearing at @xmath151 . further the line at @xmath143 loses half its intensity while the others remain unaltered except for the special case when chemical potential falls below the landau level @xmath98 . the pattern would be quite different if instead of @xmath152 the landau level quantization was schrdinger - like @xmath153 with @xmath154 being the cyclotron frequency . in this case the position of the line corresponding to the difference in landau level energies ( intraband ) never shifts in energy . further , the lines corresponding to the sum of the landau level energies ( interband ) fall at regular energies intervals @xmath155 , namely @xmath156 , @xmath157 , @xmath158 , . furthermore as @xmath1 increases through the energies of the various landau levels all lines half their intensity before fading into the background and for @xmath1 below the first level there is no line at @xmath154 . there is also no ambiguity about whether a line is intra or interband as no level is shared between upper and lower cone . in field effect devices the chemical potential in a graphene sheet can be changed by applying a gate voltage and this may be an ideal way to observe the effects just described . however , there is an alternative way to see the same effects . for any fixed value of chemical potential , the external magnetic field can be selected in such a way that @xmath1 falls below the first landau level or in between @xmath91 and @xmath96 etc . for appropriate choices of @xmath2 @xcite the curves for @xmath84 can be made to behave exactly as in fig . [ fig:1 ] . to see this it is important that @xmath0 be divided by @xmath159 , so that the lines remain fixed in normalized frequency and the vertical scale be divided by the same factor and multiplied by @xmath160 to keep the dimensions the same . when this is done , the same pattern as seen in fig . [ fig:1 ] emerges for this configuration corresponding to fixed @xmath1 with several well chosen values of @xmath2 . we have found that the curves of fig . [ fig:1 ] change very little as the chemical potential is varied within the limited range @xmath149 to @xmath161 . we have also verified that for the parameters used here @xmath162 , @xmath163 the crossover from half intensity in the line @xmath164 to near zero , i.e. merging into the background occurs rather abruptly in an energy range set by @xmath165 and/or @xmath160 . this is illustrated in fig . [ fig:3 ] where we consider how the complete disappearance of the peak at @xmath166 ( @xmath167 ) and the depletion of the next higher peak at @xmath168 towards half its initial value proceeds as the chemical potential @xmath1 crosses through the energy of the @xmath91 landau level [ @xmath169 . the long dashed curve ( red ) is for @xmath170 chosen to be well away from the crossover point of @xmath167 and is shown for comparison . the dash - dotted curve ( black ) is for @xmath171 slightly below the crossover point , the solid curve ( blue ) is for @xmath172 just at the crossover energy and the short dashed curve ( green ) is for @xmath173 which spans @xmath174 on either side of @xmath175 which is much less than the level width of @xmath176 . as @xmath1 increases through @xmath177 we note the growth of the peak at @xmath178 , the depletion towards zero value of the peak at @xmath179 and towards half its initial value of the next higher peak at @xmath168 . the complete transfer of spectral weight between the various peaks is completed for a rather small region of chemical potential variation about @xmath180 ( @xmath167 ) of order @xmath181 ( not shown in the figure ) . after the crossover is complete , the pattern of the spectral weight distribution will remain unchanged until the chemical potential becomes close to the energy of the next landau level . we would like to stress that although fig . [ fig:3 ] is plotted using the full expression ( [ optical - diagonal ] ) , we have verified that the results obtained using eqs . ( [ lorentz - full ] ) and ( [ sigma_xx - complex - corrected ] ) are practically identical . returning to eqs . ( [ sigma - t = g=0 ] ) to ( [ inter ] ) we note that the optical spectral weight under a given inter- or intraband line varies as the square root of @xmath2 ( see eq . ( [ weight - b - final ] ) and the discussion associated with it ) . this has been verified in the recent experiment of sadowski _ et al . _ @xcite on ultrathin epitaxial graphite samples @xcite . the effect of an excitonic gap @xmath32 on the optical spectral weight distribution in a one tesla magnetic field is illustrated in fig . [ fig:4 ] . here as in the previous figures @xmath182 and @xmath162 . the chemical potential is set at @xmath183 in all cases . the long dashed ( red ) , dash - dotted ( black ) , solid ( blue ) and short dashed ( green ) curves are for @xmath184 , @xmath185 and @xmath186 respectively . the curve for @xmath77 is for reference . we note that for finite @xmath187 the line at @xmath188 splits into two peaks [ dash - dotted ( black ) curve ] . the lower peak is at energy @xmath189 , while the upper peak is at @xmath190 . these two transitions for that value of chemical potential are illustrated in fig . [ fig:2 ] middle frame ( b ) ( see arrows ) . additional transitions not shown in this frame can , of course , occur but these will have higher energy . for @xmath191 the @xmath127 state shown on the right hand cone of fig . [ fig:2 ] ( b ) is occupied with a probability @xmath192 at @xmath138 , @xmath193 and a transition from @xmath91 lower cone to @xmath127 upper cone is possible as is from @xmath127 upper cone to @xmath91 of the same cone and each must be weighted by a factor @xmath192 . the first has energy @xmath194 , while the second has energy @xmath195 . there is an additional transition coming from the second cone on the left hand side of the figure of energy @xmath196 . thus the spectral intensity of the lower energy line in the solid curve ( blue ) is lower than that of the higher energy line by a factor of @xmath197 . finally for @xmath198 @xmath199 the energy of the lowest transition possible in both left and right side cones are the same equal to @xmath196 and so there is only one line in the dashed ( green ) curve . this pattern of behaviour should allow one to measure the occurrence of an excitonic gap catalyzed by the magnetic field . we note that the splitted peak in fig . [ fig:4 ] with finite @xmath32 acquire a slight asymmetry because of the factor @xmath200 of eq . ( [ lorentz - full ] ) which is different for inter and intraband transitions . so far we have shown the results for fixed values of chemical potential and magnetic field as a function of photon energy . it is also of interest to fix the photon frequency and sweep either chemical potential ( @xmath1 ) or field ( @xmath2 ) @xcite . we begin with the first of these two possibilities and this is illustrated in fig . [ fig:5 ] . four well chosen frequencies are selected , viz . @xmath201 long dashed curve ( red ) , @xmath202 dash - dotted curve ( black ) , @xmath203 solid curve ( blue ) , and @xmath204 short dashed curve ( green ) . taking these in order , we note that the long dashed curve ( red ) is near zero until @xmath1 reaches the value of @xmath205 at which point it increases rapidly reaching a plateau at @xmath206 [ the height being set by the height of the peak in the solid ( blue ) curve in fig . [ fig:1 ] ] after which it drops rapidly as @xmath1 goes through @xmath207 . it does not drop down all the way to zero however as it continues to sample parts of the peaks at the difference in frequencies @xmath208 until these move below the sampling optical frequency set at @xmath209 . the next optical frequency chosen at @xmath210 short - long dashed ( black ) curve begins by sampling the equivalent curve in fig . [ fig:1 ] where the peak has height @xmath211 . however when @xmath1 crosses the energy of the first landau level at @xmath167 this peak disappears and the curve drops to zero . no other level crosses this frequency again . the short dashed curve is for @xmath212 . it shows very much the same behaviour as the previous case , but its plateau height is a little smaller because the peak ( dash - dotted curve of fig . [ fig:1 ] ) is sampled not at its center , but rather slightly below its maximum value . the final solid ( blue ) curve is for @xmath213 . in this case a first plateau is seen at small @xmath1 with height @xmath214 which is the height of the second peak in the dash - dotted ( black ) curve of fig . [ fig:1 ] . it drops to half of this value as @xmath0 crosses @xmath167 and then near zero as @xmath0 crosses @xmath207 . all features of these curves can be traced to the corresponding behaviour of the curves of fig . [ fig:1 ] . an additional feature of these curves in now described . note that as the chemical potential gets small the black ( dash - dotted ) curve increases slightly , while the green ( dotted ) curve drops . as we have described in connection with fig . [ fig:3 ] the line shapes do depend on the value of @xmath1 if it falls within order @xmath160 and/or @xmath165 of a landau level energy @xmath62 . for @xmath1 near zero , the first peak height in the black ( dash - dotted ) curve of fig . [ fig:1 ] increases by @xmath215 and its width is also slightly narrowed . this leads to a slight increase in peak height at @xmath216 monitored in the black ( dash - dotted ) curve of fig . [ fig:5 ] and a reduction in the green ( dotted ) curve which monitors the height of the curve slightly off the peak at @xmath217 . another set of useful curves when considering possible experimental configurations is to fix the frequency of the light as well as the chemical potential and sweep the magnetic field . results are shown in fig . [ fig:6 ] for three cases @xmath218 long dashed ( red ) curve , dash - dotted ( black ) curve for @xmath219 and solid ( blue ) for @xmath220 . in all cases @xmath162 , @xmath163 , @xmath221 and the excitonic gap @xmath32 is taken to be zero . on the horizontal axis we plotted @xmath222 , the inverse square root of the magnetic field in tesla , so that fields below one tesla fall above one on this scale . the pattern of oscillations is perhaps more complex than in previous curves , but can easily be traced out from a knowledge of such curves for @xmath52 at different values of @xmath2 . as an example consider the case @xmath223 . the value of @xmath224 at @xmath225 is just that of the dash - dotted ( black ) curve of fig . [ fig:1 ] read at @xmath223 . as @xmath2 is decreased from its value @xmath226 , the first peak in dash - dotted curve in fig . [ fig:1 ] will move to lower frequency and so increases significantly the value of longitudinal conductivity at @xmath218 . its peak will cross this reference energy for @xmath227 and this produces the first peak in the long dashed ( red ) curve of fig . [ fig:6 ] . as @xmath2 is decreased further below one tesla , the second peak in the dash - dotted curve of fig . [ fig:1 ] will also move through @xmath223 . this occurs for @xmath228 , where a second lower intensity peak is seen in the long dashed ( red ) curve of fig . [ fig:6 ] . the other curves of fig . [ fig:6 ] can be traced out from similar considerations based on fig . [ fig:1 ] with attention paid to the evolution of these curves with changing value of @xmath2 . an interesting quantity to consider is the optical spectral weight that falls between @xmath229 and @xmath230 with @xmath231 a variable upper limit in the integral @xmath232 for the case @xmath233 we have @xcite @xmath234 with @xmath31 this leads to @xmath235 for finite @xmath2 , assuming for simplicity @xmath236 we get from eq . ( [ sigma - t = g=0 ] ) @xmath237 accordingly for @xmath77 @xmath238 for large values @xmath231 the maximum @xmath150 that contributes to ( [ weight - b ] ) can be estimated from @xmath239 and so @xmath240 . but @xmath241 and so @xmath242 which agrees with eq . ( [ weight - b=0 ] ) as we would expect . we note that as we have stated , the area under each line goes as @xmath243 , but the sum over the lines up to @xmath231 gives another factor @xmath244 which means that @xmath245 is independent of @xmath2 . in fig . [ fig:7 ] we show numerical results for @xmath246 in units of @xmath86 as a function of @xmath231 is @xmath86 . the long dashed ( red ) curve is computed on the base of the first equality in eq . ( [ weight - b=0 ] ) for @xmath221 , @xmath247 , and @xmath248 and is given for comparison with the two other curves . these curves are obtained by numerical integration of eq . ( [ weight - def ] ) with @xmath52 given by eq . ( [ lorentz - full ] ) for the case @xmath249 , @xmath250 , @xmath251 and finite @xmath252 . the dash - dotted ( black ) curve is for chemical potential @xmath221 , while solid ( blue ) is for @xmath253 . for the first case @xmath1 is below @xmath98 at @xmath254 and for the second is above . at small values of @xmath231 both curves start at zero and remain small until @xmath231 goes through the first landau peak in fig . [ fig:1 ] at which point it rises sharply and subsequently exhibits a plateau before showing a next sharp rise . for the dash - dotted curve ( black ) the first rise is at @xmath255 ( interband line ) , but for the solid ( blue ) curve it is at the intraband line ( @xmath256 ) . as @xmath231 increases to include several peaks in @xmath4 the curves start to follow more closely the @xmath248 result [ long dashed ( red ) curve ] . this behaviour can be understood better from a consideration of eq . ( [ weight - b ] ) . the magnetic field mainly readjusts the available optical spectral weight among the landau levels below @xmath231 . to be specific we have considered explicitly in this section only the case @xmath257 . for @xmath137m_{n},m_{n+1}[$ ] we can show that the missing spectral weight in the lines @xmath258 is all to be found in the single intraband line at @xmath259 . from eq . ( [ sigma - t = g=0 ] ) and noting eqs . ( [ intra ] ) and ( [ inter ] ) as @xmath133 , the optical spectral weight lost in units of @xmath260 is @xmath261 the first term in the left hand side is the spectral weight from all the lines that have completely disappeared from @xmath127 to @xmath262 . the second term is from the reduction in intensity by a factor of @xmath192 of the line at @xmath143 . the quantity in the right hand side is the optical weight of the intraband line which has picked up all of the lost intensity . next we consider the absorptive part of the transverse hall conductivity . the calculations are based on the equation ( [ sigma_xy - complex - corrected ] ) . in fig . [ fig:8 ] we show results for @xmath263 in units of @xmath264 as a function of @xmath0 in @xmath86 . in all cases considered temperature @xmath89 , impurity scattering rate @xmath163 and the excitonic gap @xmath77 . the magnetic field @xmath88 and three values of chemical potential are considered . the long dashed ( red ) curve is for @xmath221 below the energy of the @xmath91 landau level , dash - dotted ( black ) @xmath265 between @xmath91 and @xmath96 , and solid ( blue ) @xmath266 between @xmath96 and @xmath97 as in fig . [ fig:1 ] . we note a single peak in the long dashed ( red ) curve at @xmath98 in contrast to two in the other two curves . for the dash - dotted ( black ) curve the peaks are at @xmath267 and @xmath268 , respectively , and for the solid ( blue ) curve they are at @xmath105 and @xmath269 . these features can be easily understood from eq . ( [ sigma_xy - complex - corrected ] ) for @xmath270 when we take its imaginary part in the limit @xmath77 , @xmath271 which is for @xmath272 , @xmath273 @xmath274\\ \lo\times \left[\frac{\delta(\omega - m_{n+1 } + m_n)}{m_{n+1}-m_n } + \frac{\delta(\omega - m_{n+1 } - m_n)}{m_{n+1}+m_n } \right ] . \ ] ] taking the limit of zero temperature , @xmath138 in eq . ( [ im_sigma_xy - gamma=0 ] ) only the first two thermal factors in the square bracket survive . for @xmath137m_0,m_1[$ ] we get @xmath275 while for @xmath137m_{n},m_{n+1}[$ ] with @xmath276 we get @xmath277.\ ] ] thus for @xmath137m_0,m_1[$ ] the transverse hall conductivity exhibits a single peak at @xmath278 with weight @xmath148 , in units of @xmath279 and for @xmath137m_{n},m_{n+1}[$ ] with @xmath276 it has two peaks at @xmath280 and @xmath259 with weight @xmath281 and @xmath282 , respectively . thus the case @xmath137m_0,m_1[$ ] is different from all others . this distinguishes dirac from a classical landau level quantization , for which , all cases would have two peaks . also the area under the peaks in @xmath283 is @xmath284 in analogy to what we found for @xmath285 . next we consider the effect of a finite excitonic gap on the imaginary part of the hall conductivity . in this case even for @xmath271 , @xmath272 , @xmath273 equation ( [ sigma_xy - complex - corrected ] ) is slightly more complicated than ( [ im_sigma_xy - gamma=0 ] ) because of the additional factors @xmath286 which are only equal to @xmath287 when @xmath77 . nevertheless to understand physically the results given in fig . [ fig:9 ] this extra complication is not needed and we can use eq . ( [ im_sigma_xy - gamma=0 ] ) as a guide . what is most important is to look at the thermal factors . we will consider only the case when the chemical potential falls between @xmath127 and @xmath91 landau level in energy . the relevant thermal factor is @xmath288 . all four curves in fig . [ fig:9 ] have @xmath289 . the long dashed ( red ) curve is for @xmath290 and is included for reference . it shows a single peak ( in @xmath291 ) at @xmath179 as we know from our previous discussion . for the dash - dotted ( black ) curve @xmath292 which falls below the value of chemical potential ( see fig . [ fig:2 ] ( b ) ) . in this case the thermal factors in eq . ( [ im_sigma_xy - gamma=0 ] ) at @xmath138 give @xmath287 and the delta functions correspond to @xmath293 and @xmath294 . thus , the peak in the long dashed curve has split into two . further such peak has approximately , but not exactly the same optical spectral weight as that under the single peak of long dashed line . this arises because of the weighting factors @xmath295 of eq . ( [ sigma_xy - complex - corrected ] ) not shown explicitly in eq . ( [ im_sigma_xy - gamma=0 ] ) . for @xmath296 [ solid ( blue ) curve ] the two peaks remain at @xmath297 ( note the small shift which corresponds to the slightly different values of the gap between solid and short - long dashed curve ) . however , the thermal factor @xmath298 now equals @xmath192 rather than @xmath287 for the two previous cases , so that this feature on its own reduces the optical spectral weight of these peaks by half . finally for the short dashed ( green ) curve the gap @xmath32 is larger than is the chemical potential and the thermal factor @xmath288 is zero , so that no peak is seen . in fig . [ fig:10 ] we show results for the change in @xmath263 in units of @xmath264 as a function of @xmath1 in kelvin at fixed optical frequency . this figure is the analog of fig . [ fig:5 ] . the same parameters @xmath162 , @xmath182 , @xmath88 and @xmath251 are chosen as well as @xmath0 , namely , long dashed ( red ) curve , @xmath299 . the absorptive hall conductivity for this curve is near zero until the first landau level energy @xmath300 is crossed , where it drops below @xmath301 after which it shows a plateau till the next level is crossed at @xmath302 , where the second step is seen , etc . the dash - dotted ( black ) curve is for @xmath303 . in this case the first plateau is at @xmath304 till the energy of the @xmath91 landau level is crossed in which case it drops to near zero value . the short dashed ( green ) curve is for @xmath305 and follows the dash - dotted ( black ) curve except the plateau is at @xmath306 as we expect . finally the solid ( blue ) curve is for @xmath307 . it starts at zero till @xmath1 crosses @xmath167 at which point it shows a step down , remains nearly constant and finally steps back to near zero value at @xmath308 as expected from consideration of fig . [ fig:8 ] . the final fig . [ fig:11 ] shows results for the variation of @xmath309 in units of @xmath264 vs the inverse square root of the magnetic field , @xmath310 . here temperature @xmath162 , impurity scattering rate @xmath311 and the chemical potential @xmath312 with excitonic gap @xmath77 . three optical frequencies are chosen : long dashed ( red ) curve @xmath313 , dash - dotted ( black ) curve @xmath314 and solid ( blue ) curve @xmath315 . these curves are similar to those of fig . [ fig:6 ] for the longitudinal case , but here exhibit only a single peak as is expected from the curves shown in fig . [ fig:8 ] . in this paper we have extended the calculation of the ac conductivity of graphene presented in ref . @xcite in several ways . one emphasis has been on presenting results for different values of the chemical potential , @xmath1 . in a field - effect device , @xmath1 can be varied within the range of a few thousands of kelvin by changing the gate voltage @xcite . this implies that for fields of the order of @xmath226 , @xmath1 can be made to sweep over several landau levels . recently @xcite infrared spectroscopy has been successfully applied to study fet devices based on poly(3-hexylthiophene ) . we find a rich pattern of behaviour for the @xmath0 dependence of the real ( absorptive ) part of the diagonal conductivity , @xmath316 as a function of @xmath0 in a fixed external magnetic field oriented perpendicular to the graphene sheet . peaks are seen in @xmath317 corresponding to the possible transition between landau levels from the hole to particle band ( interband ) or within a given cone ( intraband ) . for @xmath1 between the @xmath127 and @xmath91 levels there are lines at @xmath318 , @xmath268 , @xmath319 ( interband ) , etc . , where @xmath320 is the position in energy of the @xmath321-th landau level . there is no peak below @xmath98 . the relative optical spectral weight between the various levels decay approximately as @xmath322 with @xmath0 evaluated at the center of each peak . when the chemical potential falls between the energy of the @xmath91 and @xmath96 level the peak at @xmath323 fades into the background and a new line appears at the lower frequency @xmath101 ( intraband ) . in addition , the intensity of the line at @xmath99 drops to half its value , while all other lines remain the same . as the chemical potential crosses higher and higher landau level , say falls between @xmath149 and @xmath161 , the low energy peak has shifted to @xmath324 ( intraband ) after which all peaks previously seen have disappeared into background except for the one at @xmath325 which has half its previous intensity . again all peaks above this energy remain unaltered . the peculiar behaviour of the peak at @xmath318 which is either present with full intensity or completely absent is the hallmark of dirac as oppose to schrdinger behaviour . this peak can be classified either as inter or intra when the excitonic gap @xmath77 , an ambiguity that is lifted for @xmath326 . for landau levels based on the schrdinger equation the interband lines are evenly spaced while the intraband line is fixed in energy , and all interband lines first half their intensity before disappearing entirely as @xmath1 is increased . we have found that the transition from one configuration of absorption lines to another occurs for a small change in chemical potential near a given landau level energy with the scale for the incremental change in @xmath1 set by temperature and/or level broadening @xmath160 . away from these special values of @xmath1 the curves do not change significantly . in anticipation of experiments @xcite we also provide scans of the the behaviour of @xmath327 at fixed @xmath0 as a function of chemical potential or of magnetic field . the pattern of behaviour found is traced back to that just described for @xmath327 vs @xmath0 at different values of field and chemical potential . parallel results for the absorptive part of the transverse hall conductivity are also presented . the possibility that an excitonic gap may open in graphene under high magnetic field has been considered by many authors @xcite and may even have been observed in recent experiments @xcite . we have considered its effect on the absorption peak seen in both diagonal and hall conductivity . our specific predictions are that , a given peak can split into two , can disappear entirely or can simply shift to higher energy without splitting depending on the value of the chemical potential . in some circumstances the optical spectral weight under the splitted peak can differ from each other by a factor of order 3 . many of the results shown in the figures were obtained on the basis of general formulas for the conductivity , but it was found that simplified versions which can be more easily used to interpret experimental results are surprisingly accurate in the cases considered . they involve sums over landau level indices of lorentzian forms . recently li _ _ @xcite and sadowski _ et al . _ @xcite used related forms to analyze their results . early magnetoreflectance data @xcite in magnetic fields in the range @xmath287 to @xmath328 carried out in graphite assigned lines to the @xmath329-point and some of these lines were found to follow a square root of @xmath2 law as expected for graphene . recent data of ref . @xcite at higher fields up to @xmath330 , however , found a conventional linear in @xmath2 behaviour . very recent infrared transmission data @xcite in ultrathin epitaxial graphite @xcite in fields up to @xmath331 do show @xmath243 law and these authors took this to be evidence that their carbon sheets are sufficiently decoupled to behave like graphene . however , this interpretation also requires that they assume that various sheets carry different charges , i.e. have different values of chemical potential . nevertheless , the main lines seen in these data were assigned to the first three dirac interband transitions and the @xmath332 intraband transition on the basis of their position in energy . they were also seen to vary in optical spectral weight as the square root of the magnetic field in good agreement with the findings in this work . many other detailed predictions made in this paper have yet to be verified and should help in firmly establishing the special characteristic of quasiparticles in graphene . we thank d. basov for sharing ref . @xcite prior to publication and e.j . nicol for discussion . the work of v.p.g . was supported by the scopes - project ib7320 - 110848 of the swiss nsf and by ukrainian state foundation for fundamental research . and s.g.sh . were supported by the natural science and engineering research council of canada ( nserc ) and by the canadian institute for advanced research ( ciar ) . we consider the complex conductivity @xmath334 . substituting the spectral function ( [ spect - fun - magfield ] ) in eq . ( [ electric_cond ] ) after evaluating @xmath11 and integrating over momentum @xmath335 [ see appendix a of ref . @xcite ] one obtains @xmath336,\ ] ] where the functions @xmath337 and @xmath338 are : @xmath339 with @xmath340 and @xmath341},\qquad b_n(\omega)=\frac{\gamma_n } { \pi[(\omega+m_n)^2+\gamma_n^2]}.\ ] ] one can easily check that @xmath342 using this symmetry one can rewrite eq . ( [ sigma_complex ] ) in the form @xmath343 \psi_1(\omega,\omega^\prime)\right.\nonumber \\ \lo\pm\left.{\rm sgn}(eb)\left[\frac{1 } { \omega-\omega^\prime+\omega+i0}-\frac{1}{\omega-\omega^\prime-\omega - i0}\right ] \psi_{2}(\omega,\omega^\prime)\right\}. \ ] ] now using eq . ( [ psi1 - 2topsi - n - m ] ) we obtain @xmath344 \left(\delta_{n , m-1}+\delta_{m , n-1}\right ) \right . \lo \pm { \rm sgn}(eb)\left[\frac{1 } { \omega-\omega^\prime+\omega+i0}-\frac{1}{\omega-\omega^\prime-\omega - i0}\right ] \left(\delta_{n , m-1}-\delta_{m , n-1}\right)\right\}. \ ] ] for a typical integral over @xmath345 we have @xmath346 if the thermal factor @xmath347 is absent , further integration over @xmath38 would give an exact result : @xmath348 in case we integrate with a smooth function @xmath349 we can approximately write @xmath350 similarly , taking the complex conjugate and changing @xmath351 , we have @xmath352 hence we arrive at @xmath353 + [ n_f\left(-m_{n+1}\right)-n_f\left(-m_{n}\right)]\right]\right.\nonumber \\ \lo \times\left.\left[\frac{1}{m_n - m_{n+1}+ \omega+i(\gamma_n+\gamma_{n+1})}+\frac{1}{m_n - m_{n+1}-\omega - i(\gamma_n+\gamma_{n+1})}\right ] \right.\nonumber \\ \lo-\left.\left(1+\frac{\delta^2}{m_nm_{n+1}}\right)\left[[n_f\left(-m_{n+1}\right)-n_f\left(m_{n}\right ) ] + [ n_f\left(-m_{n}\right)-n_f\left(m_{n+1}\right)]\right]\right.\nonumber\\ \lo \times\left.\left[\frac{1}{m_n+m_{n+1}- \omega - i(\gamma_n+\gamma_{n+1})}+\frac{1}{m_n+m_{n+1}+\omega+i(\gamma_n+\gamma_{n+1})}\right ] \right.\nonumber \\ \fl \pm{\rm sgn}(eb)\left.\left[\left(1-\frac{\delta^2}{m_nm_{n+1}}\right)\left[[n_f\left(m_n\right)-n_f\left(m_{n+1}\right ) ] -[n_f\left(-m_{n+1}\right)-n_f\left(-m_{n}\right)]\right]\right.\right . \nonumber \\ \lo \times\left.\left.\left[\frac{1}{m_n - m_{n+1}+ \omega+i(\gamma_n+\gamma_{n+1})}-\frac{1}{m_n - m_{n+1}-\omega - i(\gamma_n+\gamma_{n+1})}\right ] \right.\right.\nonumber \\ \lo+\left.\left.\left(1+\frac{\delta^2}{m_nm_{n+1}}\right)\left[[n_f\left(-m_{n+1}\right)-n_f\left(m_{n}\right ) ] -[n_f\left(-m_{n}\right)-n_f\left(m_{n+1}\right)]\right]\right.\right.\nonumber \\ \lo\times\left.\left.\left[\frac{1}{m_n+m_{n+1}- \omega - i(\gamma_n+\gamma_{n+1})}-\frac{1}{m_n+m_{n+1}+\omega+i(\gamma_n+\gamma_{n+1})}\right]\right]\right\ } , \nonumber \ ] ] where only the sum over @xmath5 remained . we verified that @xmath52 computed for @xmath354 from the more approximate eq . ( [ lorentz - full ] ) agrees quantitatively with the results obtained from a full eq . ( [ optical - diagonal ] ) . this agreement is best for the resonance peaks and only small deviations are seen for @xmath355 . nevertheless , eq . ( [ lorentz - full ] ) has a few drawbacks due to approximations made in eqs . ( [ approx1 ] ) , ( [ approx2 ] ) . in particular , the drude form can not be recovered in @xmath356 limit , while it can be obtained @xcite from an exact representation ( [ optical - diagonal ] ) . moreover , the imaginary parts of the diagonal conductivity , @xmath357 and the hall conductivity , @xmath358 are divergent in the limit @xmath359 and do not satisfy kramers - kronig relations with the corresponding real parts found from eq . ( [ lorentz - full ] ) . to correct these problems we move the term @xmath322 under the sum replacing it by its value at the pole of the corresponding denominator in eq . ( [ lorentz - full ] ) and arrive at eqs . ( [ sigma_xx - complex - corrected ] ) and ( [ sigma_xy - complex - corrected ] ) . novoselov k s , geim a k , morozov s v , jiang d , zhang y , dubonos s v , grigorieva i v , and firsov a a 2004 _ science _ * 306 * 666 ; novoselov k s , jiang d , booth t , khotkevich v v , morozov s m , and geim a k 2005 _ proc . . sc . _ * 102 * 10451 the fact that the phase shift of @xmath3 in graphite is related to nonzero berry s phase was discussed in mikitik g p and sharlai yu v 1999 _ phys . lett . _ * 82 * 2147 , _ phys rev . b _ 2006 * 73 * 235112 , see also kosevich a m 2004 _ low temp . _ * 30 * 97 [ _ fiz . _ 2004 * 30 * 135 ] for a review .

we have used the japanese vlbi array vera to perform high - precision astrometry of an ho maser source in the galactic star - forming region ngc 281 west , which has been considered to be part of a 300-pc superbubble . we successfully detected a trigonometric parallax of 0.355.030 mas , corresponding to a source distance of 2.82.24 kpc . our direct distance determination of ngc 281 has resolved the large distance discrepancy between previous photometric and kinematic studies ; likely ngc 281 is in the far side of the perseus spiral arm . the source distance as well as the absolute proper motions were used to demonstrate the 3d structure and expansion of the ngc 281 superbubble , pc in size parallel to the galactic disk and with a shape slightly elongated along the disk or spherical , but not vertically elongated , indicating the superbubble expansion may be confined to the disk . we estimate the expansion velocity of the superbubble as km s both perpendicular to and parallel to the galactic disk with a consistent timescale of myr .

0806.4635

a paradigm shift has been occurring in recent years in the observational study of the interstellar medium ( ism ) in the galaxy and in external galaxies , where the ism is now recognized as being of fundamental importance not only in the life cycle of stars but also in galactic evolution . the present understanding of the ism draws a dynamic and violent picture of the gas in the vast interstellar space , rather than a static one envisioned previously . recent observations with higher angular resolution have revealed the existence of dynamic and diverse gas structures in the galaxy that can not be explained by the traditional evolutionary cycle of the ism , requiring new insights toward a comprehensive understanding of the ism ( e.g. , reynolds 1997 , 2002 ; dickey 2001 ) . extreme activity of the ism is indicated by large - scale circular or arc - like structures ( up to a few kiloparsecs in scale ) called superbubbles or supershells , many of which were identified earlier by heiles ( 1979 , 1984 ) based on h 21-cm line surveys by weaver and williams ( 1973 ) and heiles and habing ( 1974 ) . kiss , mor , and tth ( 2004 ) later identified 145 large - scale loop- or arc - like structures in the 2nd galactic quadrant ( i.e. for galactic longitudes @xmath5 ) on _ iras_-based far - infrared maps . recent results for the 1st , 3rd , and 4th galactic quadrants by knyves ( 2007 ) increased the number of identified loops to 462 . similar objects are also found in nearby galaxies , including the local group spirals andromeda ( m31 ; brinks & bajaja 1986 ) and m33 ( deul & den hartog 1990 ) , local group irregulars including the large magellanic cloud ( lmc ; chu 1993 , 1995 ; kim 1998 , 1999 ) , the small magellanic cloud ( smc ; staveley - smith 1997 ; stanimirovi 1999 ) and ic 10 ( shostak & skillman 1989 ) , large spiral galaxies such as m101 ( kamphuis 1991 ) and ngc 891 ( howk & savage 1997 , 2000 ) , and dwarf galaxies such as m82 ( matsushita 2000 ; matsumoto 2001 ) , ic 2574 , ddo 47 ( walter & brinks 1999 , 2001 ) and holmberg ii ( hoii ; puche 1992 ) . ( 86mm,110mm)figure1.eps the most widely accepted theory for the origin of superbubbles suggests that supernova explosions and strong stellar winds from ob associations in a galactic disk blow the surrounding gas up into the halo , carving out extensive cavities in the ism ( see tenorio - tagle & bodenheimer 1988 ) . superbubbles are therefore believed to play a vital role in disk - halo interactions , serving as a conduit for matter and energy from the disk into the halo in the form of galactic fountains ( shapiro & field 1976 ) or chimneys ( norman & ikeuchi 1989 ) . in the galaxy , only a small number of chimneys ( or also called worms ) or fragmenting superbubbles / supershells ( on several hundred to one thousand parsec scales ) are known and well - studied : e.g. , the orion - eridanus superbubble ( cowie 1979 ; reynolds & ogden 1979 ; boumis 2001 ; welsh 2005 ) , the cygnus superbubble ( cash 1980 ) , the stockert chimney ( mller 1987 ; kundt & mller 1987 ; forbes 2000 ) , the aquila supershell ( maciejewski 1996 ) , the scutum supershell ( callaway 2000 ; savage 2001 ) , gsh 277 + 00 + 36 ( mcclure - griffiths 2000 ) , gsh 242 - 03 + 37 ( mcclure - griffiths 2006 ) , the w4 chimney / superbubble ( normandeau 1996 ; reynolds 2001 ; terebey 2003 ; madsen 2006 ; west 2007 ) , the ophiuchus superbubble ( pidopryhora 2007 ) , and the ngc 281 superbubble ( megeath 2002 , 2003 ; sato 2007 ) . superbubbles are also a key target of research for understanding their impact on the history of star formation in the galactic disk , including but not limited to superbubble - triggered star formation ( e.g. , ngc 281 as explained later in the section ) . in either the galaxy or external galaxies , very little is known about the dynamics and energetics of superbubbles / chimneys or about their influence on the formation of the halo or on the star formation history in the galactic disk . a difficulty in revealing the structure and kinematics of superbubbles stems from the limited information obtainable from two dimensions of sky positions and velocity information only along the line of sight . a direct observation of the three - dimensional ( 3d ) motion and structure of a superbubble has not been easy due to the high astrometric accuracy required . however , sato ( 2007 ) reported the first vlbi ( very long baseline interferometry ) observations that obtained direct evidence for expanding motions of a superbubble by measuring the proper motions . in sato ( 2007 ) , we observed the h@xmath0o maser emission in ngc 281 west over 6 months with vera ( vlbi exploration of radio astrometry ) , a new japanese radio telescope array dedicated to phase - referencing vlbi astrometry ( e.g. , honma 2000 , 2005 , 2007 ; kobayashi 2003 ; imai 2006 , 2007 ; hirota 2007 , 2008a , 2008b ; nakagawa 2008 ; kim 2008 ; choi 2008 ) . these observations revealed the systemic motion of the ngc 281 superbubble away from the galactic plane at a velocity of 20@xmath630 km s@xmath4 . at its estimated heliocentric distance of [email protected] kpc , ngc 281 most likely originated in the galactic plane and has been blown out by a superbubble expansion ( see @xmath71.1 and sato 2007 ) . a precise and direct determination of the distance to ngc 281 is desirable and essential for a better understanding of the region , including further description of the 3d structure and motion of the superbubble to reveal the origin , the energetics and the timescale of the superbubble . these are the primary reasons why we decided to complete our vera observations to directly measure the parallactic distance to ngc 281 via an h@xmath0o maser source . the astrometric determination of the distance to ngc 281 is also important for studying early - type ( o@xmath6b ) high - mass ( @xmath8 ) stars and stellar evolution in combination and in comparison with photometric studies of the ob stars in ngc 281 ( e.g. , henning 1994 ; guetter & turner 1997 ) . in this paper , we present the results of our parallax measurements with vera over 18 months , which enables us to further discuss the 3d structure , expanding motion , the origin and the timescale of the ngc 281 superbubble . the ngc 281 region ( @xmath900@xmath1052@xmath11 , @xmath1256@xmath13 or @[email protected] , @xmath156.@xmath1331 ) provides an excellent laboratory for studying in detail the cycle of the ism and its impact on both star formation and galactic disk - halo interaction through superbubbles . at its estimated heliocentric distance of 3 kpc , this region is remarkably located far above ( @xmath16 pc ) the midplane of the perseus arm of the galaxy . of special interest in the ngc 281 region is a possibility of triggered star formation occurring on two different scales : large - scale ( @xmath2300 pc ) supperbubble - triggered formation of the first ob stars ( megeath 2002 , 2003 ) , and the sequential and ongoing triggered star formation ( @xmath21@xmath610 pc ) in an adjoining giant molecular cloud ( ngc 281 west ) through interaction with an h region ( the ngc 281 nebula ) excited by the first ob stars ( elmegreen & lada 1978 ; megeath & wilson 1997 ) . figure [ fig : optical ] is an optical image of the region , which clearly shows the red nebulosity of the h region ngc 281 ( also known as sharpless 184 ) of diameter 20 arcmin ( @xmath17 pc ) , in which an early - type cluster called ic 1590 is embedded ( guetter & turner 1997 ) . the brightest member of the cluster ic 1590 is an o - type trapezium - like system hd 5005 ( or also called ads 719 ) , whose component stars hd 5005ab ( unresolved ) , hd 5005c , and hd 5005d have spectral types of o6.5 v , o8 v , and o9 v , respectively ( walborn 1973 ; abt 1986 ; guetter & turner 1997 ) . the southwestern quadrant of the ngc 281 nebula is obscured ( as seen in figure [ fig : optical ] ) by the adjoining molecular cloud ngc 281 west in front of the nebula . ongoing star formation in the ngc 281 west cloud is indicated by the h@xmath0o maser emission and _ iras _ sources within this cloud near the clumpy interface between the h region and the cloud . this generation of stars may have been triggered by interactions with the h region ( elmegreen & lada 1978 ; megeath & wilson 1997 ) . the ngc 281 molecular cloud complex that surrounds the h region ( including the ngc 281 west and east clouds ) was mapped in co emission lines by lee and jung ( 2003 ) . the central radial velocity of the ngc 281 west molecular cloud , @xmath18 km s@xmath4 ( lee and jung 2003 ) , agrees well with that of the h@xmath0o maser emission in the cloud ( sato 2007 ) . on a larger scale , megeath ( 2002 , 2003 ) identified the co molecular cloud complex as being on an h loop extending over 300 pc away from the galactic plane toward decreasing galactic latitude . in an @xmath19 vs. @xmath20 diagram ( i.e. a plot of galactic longitude vs. observed radial velocity ) , the co molecular clouds appear to be part of a broken ring of diameter 270 pc , expanding at a velocity of 22 km s@xmath4 parallel to the galactic plane ( megeath 2002 , 2003 ) . megeath ( 2002 , 2003 ) suggested that these clouds were formed in a fragmenting superbubble , which triggered the formation of the first ob stars , and that these ob stars then ionized the surrounding gas which subsequently triggered ongoing star formation in the neighboring cloud . in sato ( 2007 ) , we derived the absolute proper motions of the h@xmath0o maser features in ngc 281 west by phase - referencing vlbi observations with vera over 6 months . we found 10 maser features , highly likely excited by two spatially distinct young stellar objects ( ysos ) , are systematically moving toward southwest , in agreement with the direction of the proper motion of hd 5005 in the h region measured with larger uncertainties by _ hipparcos _ ( perryman 1995 , 1997 ) . our results with vera revealed the systemic motion of the ngc 281 region away from the galactic plane , most likely having originated in the galactic plane and been blown out by the expansion of the ngc 281 fragmenting superbubble . the uncertainties in our velocity estimates are largely due to the large distance uncertainty of ngc 281 , despite many measurements , e.g. , photometric distances of 2 kpc ( sharpless 1952 , 1954 ) , 3.68 kpc ( cruz - gonzlez 1974 ) , 3.5 kpc ( henning 1994 ) and [email protected] kpc ( guetter & turner 1997 ) and kinematic distances of 2.2 kpc ( georgelin & georgelin 1976 ; roger & pedlar 1981 ) and 3 kpc ( lee & jung 2003 ) . in sato ( 2007 ) , we adopted a value of 2.9 kpc as the distance to ngc 281 , based on a recent estimate of [email protected] kpc derived by guetter and turner ( 1997 ) through photometry of 279 individual stars in and about the young cluster ic 1590 embedded in the ngc 281 nebula . the most reliable and direct way to determine the distance ( @xmath21 ) of an astronomical object is to measure the trigonometric parallax ( @xmath22 ) of the object due to the earth s motion around the solar system barycenter , where the distance is given by @xmath21(pc ) @xmath23 ( ) . the astrometric determination of the distance to ngc 281 provides valuable information for studying the formation and evolution of early - type ( o@xmath6b ) high - mass ( @xmath8 ) stars in comparison with photometric studies and distances , for example , by guetter & turner ( 1997 ) . in the present study , we report on our successful determination of the parallactic distance to ngc 281 , which reveals the 3d structure of the ngc 281 fragmenting superbubble , and investigate the origin and timescale of the formation of such a large - scale ( @xmath2650 pc ) structure . the vera observations were carried out over 13 epochs spaced approximately at monthly to bimonthly intervals . the first 6 epochs reported in sato ( 2007 ) were on 2006 may 14 , july 21 , august 3 , september 5 , october 25 , and november 18 ( days of year 134 , 202 , 215 , 248 , 298 , and 322 , respectively ) . seven new observations were conducted on 2006 december 11 , 2007 january 22 , march 19 , may 1 , july 27 , september 30 , and october 28 ( days of year 345 , 387 , 443 , 486 , 573 , 638 , 666 as counted from 2006 january 0 ) . note that the dates stated here are universal time ( ut ) start times , while for the data analyses in the following section we use more precisely the median hour of each observation , which lasted 7@xmath69 hours . the observational procedures are as described in sato ( 2007 ) and as follows . we observed two sources simultaneously in the dual - beam mode of vera for phase referencing ( e.g. , honma 2003 ) , and the real - time instrumental phase difference data between the two beams were taken with an artificial noise source in each beam and recorded for calibration ( kawaguchi 2000 ; honma 2008a ) . we simultaneously observed the h@xmath0o maser source in the ngc 281 west cloud with one of the two extragalactic position - reference quasars j0047 + 5657 ( [email protected] separation ; at @xmath900@xmath1047@xmath1100.@xmath24428805 , @xmath1256@xmath13 ) and j0042 + 5708 ( [email protected] separation ; at @xmath900@xmath1042@xmath1119.@xmath24451727 , @xmath1257@xmath13 ) . positions are from beasley ( 2002 ) . we alternated between the two quasars typically every 10 minutes . we observed a bright calibrator source hourly : j0319 + 4130 ( @xmath25 3c 84 ) at the first epoch ( 2006 may 14 ) and j2232 + 1143 ( @xmath25 cta 102 ) for the other 12 epochs . the data correlation was performed with the mitaka fx correlator ( chikada 1991 ) with frequency and velocity resolutions for the maser lines of 15.625 khz and 0.21 km s@xmath4 , respectively . the observed frequencies of the maser lines were converted to radial ( line - of - sight ) velocities with respect to the local standard of rest ( lsr ) , @xmath26 , using a rest frequency of 22.235080 ghz for the h@xmath0o 6@xmath27 - 5@xmath28 transition . visibility calibration and imaging were performed in a standard manner with the nrao astronomical image processing system ( aips ) package . the visibilities of all h@xmath0o maser channels at each epoch were phase - referenced to each of the position - reference sources j0047 + 5657 and j0042 + 5708 independently by applying the phase solutions from fringe fitting ( aips taks fring ) for each position - reference source to the h@xmath0o maser channels for the corresponding observation time and frequencies . the instrumental phase difference between the two beams was also corrected using the recorded phase - calibration data described above . we also calibrated the error in the visibility phase caused by the earth s atmosphere based on gps measurements of the atmospheric zenith delay due to tropospheric water vapor ( honma 2008b ) . after calibration , we made spectral - line image cubes using the aips task imagr around the maser spots with 512 pixels @xmath29 512 pixels of size 0.05 mas . the synthesized beam had an fwhm beam size of 1.3 mas @xmath29 0.8 mas with a position angle of @xmath643@xmath13 east of north . least - squares fittings were performed to simultaneously solve for the sinusoidal parallax curve and linear proper motion in the directions of right ascension ( ra ) and declination ( dec ) . ( 150mm,300mm)figure2.eps lcccccc & reference & @xmath22 [ mas ] & @xmath21 [ kpc ] & @xmath30 [ mas ] & @xmath31 [ mas yr@xmath4 ] & @xmath33 [ mas yr@xmath4 ] + & ( 2)&(3)&(4)&(5)&(6)&(7 ) + & j0047 + 5657 & 0.346 ( 0.052 ) & 2.89 ( 0.43 ) & 0.113 & @xmath34 ( 0.07 ) & + ( ra ) & j0042 + 5708 & 0.334 ( 0.047 ) & 2.99 ( 0.42 ) & 0.104 & @xmath35 ( 0.07 ) & + & combined & 0.340 ( 0.034 ) & 2.94 ( 0.30 ) & 0.106 & @xmath36 ( 0.05 ) & + & j0047 + 5657 & 0.400 ( 0.095 ) & 2.50 ( 0.59 ) & 0.188 & & @xmath37 ( 0.12 ) + ( dec ) & j0042 + 5708 & 0.412 ( 0.092 ) & 2.43 ( 0.54 ) & 0.183 & & @xmath38 ( 0.11 ) + & combined & 0.406 ( 0.065 ) & 2.46 ( 0.39 ) & 0.181 & & @xmath39 ( 0.08 ) + & combined & 0.355 ( 0.030 ) & 2.82 ( 0.24 ) & 0.148 & & + figure [ fig : parallax ] shows the position measurements of the brightest maser spot at @xmath40 km s@xmath4 ( feature # 4 of 10 features identified in sato 2007 ) using j0047 + 5657 ( blue ) and j0042 + 5708 ( red ) as position - reference sources independently . figures [ fig : parallax]a and [ fig : parallax]b show the positional offsets of the maser spot in the eastward ( @xmath41 , i.e. ( ra ) cos(dec ) ) and northward ( @xmath42 , i.e. dec ) directions , respectively , relative to the reference position at the origin at @xmath900@xmath1052@xmath1124.@xmath2470081 for @xmath41 and @xmath1256@xmath13 for @xmath42 , as a function of time ( day of year , as counted from 2006 january 0 ) . the constant offsets of @xmath43 mas seen in each of @xmath41 and @xmath42 between the two measurements using different position - reference sources j0047 + 5657 and j0042 + 5708 are due to the uncertainties in the absolute positions of these reference sources ( 0.64 mas for j0047 + 5657 and 0.82 mas for j0042 + 5708 ; beasley 2002 ) and are solved for in the multi - epoch parallax and proper - motion fittings . the best - fit models are also plotted with gray lines for proper motions and black curves for the sum of sinusoidal parallax and the linear proper motions . table 1 summarizes the fitted distances and proper motions . figures [ fig : parallax]c is a plot of the spot position in @xmath42 vs. @xmath41 and its movement during the observation epochs . the obtained values of the parallax in @xmath41 are : @xmath44 mas ( corresponding to a distance of @xmath45 kpc ) using j0047 + 5657 and @xmath46 mas ( corresponding to a distance of @xmath47 kpc ) using j0042 + 5708 . combining these independent measurements using two different position - reference sources , we have obtained as a final value in @xmath41 ( ra ) direction : @xmath48 mas ( @xmath49 kpc ) . similarly , we have obtained the parallax in @xmath42 to be : @xmath50 mas ( @xmath51 kpc ) using j0047 + 5657 and @xmath52 mas ( @xmath53 kpc ) using j0042 + 5708 . combining these two measurements yields the final value in @xmath42 ( dec ) direction to be : @xmath54 mas ( @xmath55 kpc ) . here we have estimated the associated uncertainties from the uniformly - weighted standard deviation @xmath30 from the least - squares fitting in @xmath41 and @xmath42 , which were @xmath56 mas and @xmath57 mas , respectively ( see table 1 for the individual value of the standard deviation in each measurement ) . random errors in individual position measurements estimated with the aips gaussian - fit task jmfit caused by the noise ( which are approximated by the beam hwhm divided by the signal - to - noise ratio ) were @xmath58 mas , which is thus not the main cause of the deviations @xmath59 and @xmath60 from the fits . we will discuss the possible error sources in @xmath61 in more detail . however , those errors are difficult to measure quantitatively , and we therefore estimated the astrometric errors by the standard deviations of the residuals from the fits and plotted the standard deviations as the error bars of each point in figure [ fig : parallax ] . weighting by @xmath59 and @xmath60 , we simultaneously fitted the position measurements of the maser spot in both @xmath41 and @xmath42 altogether , including all data using j0047 + 5657 and j0042 + 5708 , to obtain the final result for the parallax measurements to be : @xmath62 mas , which corresponds to a distance to ngc 281 of @xmath63 kpc . we have therefore successfully determined the parallactic distance to ngc 281 as far as @xmath64 kpc with a precision better than 10% . our distance estimate of ngc 281 , @xmath63 kpc , agrees well with the recent photometric distance of [email protected] kpc by guetter and turner ( 1997 ) , but excludes the kinematic distance of 2.2 kpc by georgelin and georgelin ( 1976 ) and by roger and pedlar ( 1981 ) and most distance estimates in earlier studies . [ cols="^,^,^,^,^ " , ] + as listed in table 1 , we fitted the position measurements of 13 epochs simultaneously for both the parallax @xmath22 and proper motions ( @xmath31 , @xmath33 ) ( in ra and dec , respectively ) of the peak maser spot of feature 4 ( the brightest feature in the maser component c3 , which is one of the two spatial components c1 and c3 with maser emission detected in sato 2007 ) . for the proper motions , we obtained ( @[email protected] @[email protected] ) mas yr@xmath4 using j0047 + 5657 , and ( @[email protected] , @[email protected] ) mas yr@xmath4 using j0042 + 5708 . the error - weighted mean proper motions of this spot ( which are the same as unweighted mean in this case of equal errors ) are : ( @xmath31 , @xmath33)@xmath25(@[email protected] , @[email protected] ) mas yr@xmath4 . we consider this mean as revised values ( called modified hereafter ) of the error - weighted mean proper motions ( denoted by 4w ) of feature 4 reported in sato ( 2007 ) . table 2 shows a comparison of the absolute proper motions derived for feature 4 between the revised values in the present study and the previous values in sato ( 2007 ) . in sato ( 2007 ) , we derived absolute proper motions of feature 4 ( in c3 ) and features 9 and 10 ( in c1 , the other one of the two spatial maser components ; see sato 2007 ) assuming the distance to ngc 281 to be @xmath65 kpc ( adopting the photometric distance by guetter and turner 1997 ) and subtracting the expected parallax in the position measurements during the first 6 observation epochs . the revised values of @xmath31 for feature 4 in the present study agree with the previous values in sato ( 2007 ) within the margin of errors ( @xmath66 ) , however , for @xmath33 we find deviations of the revised values from the previous ones larger than the errors ( @xmath67 ) estimated in sato ( 2007 ) . this means in sato ( 2007 ) we overestimated @xmath33 of feature 4 and underestimated the errors in the linear least - squares fittings ( after subtracting the assumed parallax ) . as seen in figure [ fig : parallax ] based on the present longer - time observations , the overestimate of the previous @xmath33 values for feature 4 is due to a large position deviation from the best - fit at the first ( @xmath68 ) epochs , which yielded a larger slope . the underestimate of the errors occured mostly due to non - gaussian errors larger than 0.1 mas ( larger than thermal errors by an order of magnitude ) in the position measurements , owing to astrometric error sources that are discussed in more detail in @xmath61 . however , as will be described in @xmath69 , the scientific conclusions remain essentially unchanged after the modifications to the proper motions . by adding the mean relative motion of all 8 features ( including feature 4 ) in c3 with respect to the reference feature 4 , ( @xmath70 , @xmath71)@xmath25([email protected] , [email protected] ) mas yr@xmath4 as given in sato ( 2007 ) , to this modified absolute motion of feature 4 , the mean absolute proper motion of these features in c3 ( denoted by c3 m in sato 2007 ) is then also revised to be : ( @xmath72 , @xmath73)@xmath74(@[email protected] , @[email protected] ) mas yr@xmath4 . our new distance estimate of ngc 281 by parallax measurements , @xmath75 kpc , also introduces a slight modification to the measured absolute proper motions of features 9 and 10 on the same data of 6 epochs ( as these features were not persistent through the total 13-epoch observations with high enough signal - to - noise ratios for the parallax and proper motion fittings ) . the modified absolute proper motions ( @xmath31 , @xmath33 ) of feature 9 are ( @[email protected] , @[email protected] ) mas yr@xmath4 using j0047 + 5657 , and ( @[email protected] , @[email protected] ) mas yr@xmath4 using j0042 + 5708 . similarly we obtained the absolute proper motion ( @xmath31 , @xmath33 ) of feature 10 to be : ( @[email protected] , @[email protected] ) mas yr@xmath4 using j0047 + 5657 , and ( @[email protected] , @[email protected] ) mas yr@xmath4 using j0042 + 5708 . here the associated uncertainties in @xmath31 and @xmath33 were estimated from the standard deviations from the linear least - squares fits in @xmath41 and @xmath42 positional offsets ( after the parallax for @xmath75 kpc is subtracted ) , respectively . the error - weighted mean proper motions of each of these features in c1 ( denoted by 9w and 10w in sato 2007 ) are ( @xmath31 , @xmath33)@xmath25(@[email protected] , @[email protected] ) mas yr@xmath4 for feature 9 and ( @xmath31 , @xmath33)@xmath25(@[email protected] , @[email protected] ) mas yr@xmath4 for feature 10 . then the unweighted mean of the absolute motions of these two features in c1 ( denoted by c1 m in sato 2007 ) is also slightly modified to be ( @xmath72 , @xmath73)@xmath76(@xmath63.41 , @xmath62.84 ) mas yr@xmath4 . in the present work , we measured the annual parallax of ngc 281 west to be @xmath77 mas , which corresponds to a distance of @xmath78 kpc . in this section , we first discuss possible error sources in our parallax and proper - motion measurements . as already mentioned in @xmath79 , the true uncertainties in the measurements were estimated from standard deviations of the post - fit residual from the least - squares fits ( as listed in table 1 ) , because thermal errors due to noise in individual position measurements ( @xmath58 mas ) are far smaller than the deviations from the fits . the likely sources for such deviations are difficult to measure quantitatively , which led to our error underestimate in the derived absolute proper motions of feature 4 in sato ( 2007 ) as described in @xmath80 . the reference sources did not show any resolved structure such as additional components or structural change , and we obtained similar results . therefore we do not consider the contribution of the reference source as a dominant error source . one of the likely causes for position uncertainties of @xmath56 mas and @xmath57 mas ( see table 1 ) is mis - modeling of tropospheric zenith delay , as also reported in previous astrometric measurements with vera ( e.g. , hirota 2007 ; nakagawa 2008 ; honma 2008b ) . the water vapor in the troposphere introduces optical path differences through the atmosphere between the target maser source and reference sources because of the elevation angle difference between the sources . therefore , even after the phase referencing calibration , there remain residuals of tropospheric zenith delay due to the water vapor , which are difficult to measure precisely and are one of the most serious error sources in the vlbi astrometry in the 22-ghz band . nakagawa ( 2008 ) and honma , tamura & reid ( 2008b ) give detailed discussion on astrometric errors caused by such residuals of tropospheric zenith delay calibration with vera and report probable errors , @xmath59 and @xmath60 , of 0.05-mas and 0.10-mas levels , respectively , for source declination of 60@xmath13 . our parallax determination with vera is no exception in having astrometric errors larger in declination ( @xmath42 ) than in right ascension ( @xmath41 ) ( i.e. @xmath81 ) as the same tendencies reported in other observations with vera ( e.g. , hirota 2007 ) . larger errors in @xmath42 ( dec ) than in @xmath41 ( ra ) are explained by the tendency of tropospheric zenith delay errors to have more severe effects in the @xmath42 direction ( although depending of the separation angle between the target and reference sources and the declination of the sources ) and are as expected by the simulation in honma , tamura and reid ( 2008b ) . however , we do not consider the residuals of tropospheric zenith delay as the dominant cause of errors in our parallax measurements . as the simulations by honma , tamura and reid ( 2008b ) clearly show , the astrometric errors due to the zenith delay residuals for the ngc 281 case at its high declination ( @[email protected] ) and with separation angle ( sa ) smaller than 1@xmath13 for j0047 + 5206 ( [email protected] ) are expected to be as good as @xmath83 @xmath84as ( i.e. @xmath85 mas ) in each of @xmath41 and @xmath42 directions , which do not account for all the standard deviations @xmath56 mas and @xmath57 mas in our measurements . also , as figure [ fig : parallax ] clearly demonstrates , using two different reference quasars with different separation angles ( [email protected] for j0047 + 5206 and [email protected] for j0042 + 5708 ) did not result in remarkable differences in the astrometric errors for the two cases . this certainly excludes the simple zenith delay error effect as the main source of our astrometric errors . the most likely major source of errors in the present position measurements is a variation of the intrinsic maser structure . most maser spots are unresolved in vlbi observations , however evidence has been shown for intrinsic maser structure smaller than 1 au on submilliarcsecond scales in previous maser studies ( e.g. , fish & sjouwerman 2007 ) . a maser spot likely consists of both unresolved substructure and observed larger structure which are indistinguishable ( e.g. , fish & sjouwerman 2007 ) , and thus a variation in maser substructure can lead to fluctuation of the peak position , limiting the accuracy in position measurements of a maser spot . this effect of the maser spatial structure is also seen for other observations in parallax measurements with up to 10% errors , independently of the source distance , for example , in orion kl by hirota ( 2007 ) . errors due to the maser structure can be reduced by using multiple maser features for derivation of a parallax . in our case of ngc 281 west , however , only one bright maser feature ( above 10 jy beam@xmath4 ) that was used in the present study was found with a good signal - to - noise ratio for the precise parallax measurements , even persistent for the all observation epochs . according to discussions by honma ( 2007 ) , other possible astrometric errors in vera observations that arise from the uncertainties in the antenna - station positions , delay models , and path length differences due to ionosphere are smaller by at least an order of magnitude than the uncertainties due to tropospheric zenith delay residuals . therefore , we conclude that the main error sources of our astrometric measurements is the variability of maser feature structure , and the contribution by tropospheric zenith delay residuals is also nonnegligible . ( 140mm,80mm)figure3.eps using our new results from parallax and proper - motion measurements of the h@xmath0o maser source in ngc 281 west , we now revise the motion of the ngc 281 region with respect to the galactic rotation as reported in sato ( 2007 ) . in an analogous manner to sato ( 2007 ) , we show the modified systemic motion of the ngc 281 region away from the galactic plane , traced by our h@xmath0o maser observations , as a red arrow in figure [ fig : hilv]a , relative to a frame rotating with the galactic rotation . figures [ fig : hilv]a , b and c are an image of the ngc 281 superbubble plotted as @xmath86 vs. @xmath19 ( velocity - integrated for the perseus - arm line - of - sight velocity range , @xmath87 to @xmath88 km s@xmath4 ) and @xmath20 vs. @xmath19 ( latitude - integrated for the galactic latitude range indicated by a cyan rectangle in figure [ fig : hilv]a ) diagrams after megeath ( 2003 ) and based on h 21-cm line data by hartmann and burton ( 1973 ) and @xmath89co(@xmath90 ) line data by dame , hartmann , and thaddeus ( 2001 ) . at the position of ngc 281 west , the galactic plane lies almost parallel to the east - west ( ra ) direction with a position angle of 90.@xmath132 , so that we can regard the motions in @xmath41 ( ra ) and @xmath42 ( dec ) directions as those parallel ( @xmath19 ) and perpendicular ( @xmath86 ; with only small projection effect at @xmath156.@xmath1331 ) to the galactic plane , respectively . subtracting the expected apparent motions of ngc 281 due to the ( simulated ) galactic rotation ( with respect to the lsr ) and the solar motion effect , ( @xmath91 , @xmath92)@xmath25(@xmath63.27 , @xmath60.29 ) mas yr@xmath4 ( including the apparent motion arising from the nonzero galactic latitude of ngc 281 ; @xmath156.@xmath1331 ) and ( @xmath93 , @xmath94)@xmath25(0.84 , @xmath60.53 ) mas yr@xmath4 , respectively , from the measured mean proper motion of the h@xmath0o maser source in ngc 281 west yields the systemic motion of the region with respect to the rotation of the galaxy . here we have adopted our new distance estimate of ngc 281 , 2.82 kpc , and the solar motion relative to the lsr based on _ hipparcos _ data ( dehnen & binney 1998 ) , and also assumed the distance from the sun to the galactic center , @xmath95 , to be 8.0 kpc ( reid 1993 ) , the rotational velocity of the galaxy at the local standard of rest ( lsr , at @xmath95 ) , @xmath96 , to be 236 km s@xmath4 ( reid & brunthaler 2004 ) , and a flat rotation curve of the galaxy ( i.e. a differential rotation with a nearly constant rotational velocity @xmath97 for the outer galaxy ) . subtracting these two effects of the galactic rotation and the solar motion on the apparent motion of ngc 281 from the observed proper motions , we obtain the modified mean absolute proper motions of the maser features in c3 and c1 with respect to the galactic rotation to be : ( @xmath31 , @xmath33)@xmath98(0.26 , @xmath60.62 ) mas yr@xmath4 and ( @xmath31 , @xmath33)@xmath99(@xmath60.98 , @xmath62.02 ) mas yr@xmath4 , respectively . the mean of these two spatial components is expected to trace the systemic motion of the ngc 281 region and obtained to be : ( @xmath31 , @xmath33)@xmath100(@xmath60.36 , @xmath61.32 ) mas yr@xmath4 the direction of this systemic motion with respect to a frame rotating with the galactic rotation is shown as a red arrow in figure [ fig : hilv]a . at a distance of 2.82 kpc , a proper motion of 1 mas yr@xmath4 corresponds to a transverse velocity of 13.4 km s@xmath4 , and therefore our velocity estimate for the systemic motion of ngc 281 is : ( @xmath101 , @xmath102)@xmath25(@xmath64.8 , @xmath617.7)@xmath103(@xmath65 , @xmath618 ) km s@xmath4 in the directions toward increasing @xmath19 and @xmath86 , respectively ( the minus signs here thus indicate the motions toward decreasing @xmath19 and @xmath86 ) . as mentioned in sato ( 2007 ) , if we adopt the iau standard values for the galactic rotation of @xmath1048.5 kpc and @xmath105220 km s@xmath4 ( kerr & lynden - bell 1986 ) instead of @xmath1048.0 kpc and @xmath105236 km s@xmath4 , and for the solar motion relative to the lsr of the velocity @xmath10619.5 km s@xmath4 toward @xmath107271.@xmath130 , @xmath10829.@xmath130 , then the resulting values vary by up to @xmath210@xmath109 in the @xmath86 direction : ( @xmath110 , @xmath111)@xmath25(@xmath62.92 @xmath60.25 ) mas yr@xmath4 , ( @xmath112 , @xmath113)@xmath25(1.25 , @xmath60.61 ) mas yr@xmath4 and ( @xmath114 , @xmath115)@xmath100(@xmath61.12 , @xmath61.28 ) mas yr@xmath4 , yielding a systemic velocity estimate of ( @xmath116 , @xmath117)@xmath103(@xmath615 , @xmath617 ) km s@xmath4 . in the @xmath19 direction , we did not discuss in sato ( 2007 ) the velocity component @xmath101 parallel to the galactic plane due to its larger dependence of this direction on the galactic rotation model as can be seen in the two cases above . the velocity component @xmath101 toward decreasing galactic longitude obtained above differs from what one would expect from a simple expansion of the ngc 281 superbubble if the origin ( e.g. , supernova explosions ) of the superbubble , expected to be around @xmath118 from the ring center in the @xmath20 vs. @xmath19 diagram of figures [ fig : hilv]b and [ fig : hilv]c ( see next section ) , was rotating with the galaxy . the galactic rotation velocity that would yield @xmath119 ( toward increasing galactic longitude ) is @xmath120 km s@xmath4 for @xmath121 kpc ; thus the negative velocity component @xmath101 could be explained by a faster rotational velocity of the galaxy . however , it is likely that this motion @xmath101 of the region relative to the galactic rotation is real , reflecting the motion of the superbubble as a whole in a large - scale phenomenon such as velocity jumps over a spiral shock in the perseus arm of the galaxy ( e.g. , xu 2006 ; see @xmath122 ) . in the @xmath86 direction , we find the ngc 281 region at a distance of @xmath123 pc ( at @xmath75 kpc ) from the galactic plane and moving away from the galactic plane with a velocity component of @xmath124 km s@xmath4 . assuming the origin of the ngc 281 superbubble to lie within the galactic plane ( at @xmath125 ) , the dynamical timescale of the superbubble estimated from a simple ballistic motion of @xmath102 is : @xmath126 mas yr@xmath127 myr , where @xmath128 is the derived @xmath42 motion of the maser source in ngc 281 west relative to the galactic rotation . this timescale , 17 myr , is consistent with our timescale estimate of the superbubble expansion in the direction ( almost ) parallel to the galactic plane to be discussed in the next section . here we also simply estimate the kinetic energy of the ngc 281 region in the direction perpendicular to the galactic plane , and in the next section we will estimate the expansion velocity , timescale , and energy parallel to the plane as well . according to the mass estimates by megeath ( 2002 ) of the clouds in the region ( indicated by a cyan rectangle in figure [ fig : hilv]a ) , the velocity - integrated mass in atomic and molecular gas are @xmath129 and @xmath130 , respectively . megeath ( 2002 ) estimated the total kinetic energy of @xmath131 ergs for the ring of clouds indicated by a cyan ( dashed ) ellipse in the ( @xmath19 , @xmath20 ) diagram of figure [ fig : hilv]b , expanding at a velocity of @xmath13222 km s@xmath4 . using the same mass estimate as given by megeath ( 2002 ) and our new estimate of the velocity component @xmath133 km s@xmath4 away from the galactic plane , we estimate the region to also have kinetic energy of @xmath134 ergs in the direction perpendicular to the plane , which requires the energy of multiple supernovae . ( 130mm,120mm)figure4a.eps ( 110mm,110mm)figure4b.eps our determination of the parallactic distance to ngc 281 enables us to discuss the 3d structure and expansion of the ngc 281 superbubble . in the ngc 281 superbubble , there exists another h@xmath0o maser souce in the star - forming region iras 00420 + 5530 ( @xmath900@xmath1045@xmath11 , @xmath1255@xmath13 or @[email protected] , @xmath135 ) , whose position is indicated by green dots in figure [ fig : hilv ] . moellenbrock , claussen , and goss ( 2007 ) measured the parallactic distance of the h@xmath0o maser source in iras 00420 + 5530 with the very long baseline array ( vlba ) to be @xmath136 kpc . combining the two parallax results , one can examine the size , structure and the expansion of the superbubble three - dimensionally . figures [ fig:3d ] illustrates the 3d positions of the two sources , ngc 281 and iras 00420 + 5530 , in ( a ) edge - on ( as a cross section at a galactic longitude @xmath137 ) and ( b ) face - on ( as seen from the north galactic pole ) views of the galaxy . as mentioned previously , megeath ( 2002 , 2003 ) found an expanding ring of molecular clouds ( called megeath s ring hereafter ) , and estimated from @xmath138 ( i.e. the ring size in galactic longitudes @xmath19 ) to be 270 pc . from @xmath139 ( i.e. the ring size in line - of - sight velocities @xmath20 ) of @xmath14022 km s@xmath4 and from @xmath141 the dynamical age is @xmath142 myr . however , since the heliocentric distances to ngc 281 and iras 00420 + 5530 are now determined by the parallax measurements to be 2.82 kpc and 2.17 kpc , respectively , these clouds ( shown as red and green points in figure [ fig : hilv ] ) are separated from each by 650 pc ( @xmath1290 pc ) along the line of sight . this then implies the major axis of megeath s ring is along the line of sight and a size of approximately @xmath143 pc in diameter ( as derived geometrically from figure [ fig : hilv]b ; the radius of megeath s ring can be approximated by the distance between ngc 281 and iras 00420 + 5530 ) . however , it remains unclear whether such an elongated shape of the molecular ring is plausible , compared to the apparent ring size of @xmath16 pc ( in part of the ngc 281 superbubble with apparent size of @xmath144 pc ) as seen in the ( @xmath19 , @xmath86 ) plane ( figure [ fig : hilv]a ) . we thus suggest a smaller ring of clouds indicated by a pink ( dashed ) ellipse ( called our ring hereafter ) in figure [ fig : hilv]c , excluding a cloud at @xmath145 km s@xmath4 from megeath s ring , as another possible interpretation of the ( @xmath19 , @xmath20 ) diagram . in this interpretation , ngc 281 and iras 00420 + 5530 are close to the far side and near side of the expanding ring , yielding a smaller ring diameter of @xmath146 pc along the line of sight , a lower ( @xmath270% ) expansion velocity of the ring , @xmath147 km s@xmath4 , and thus a lower ( @xmath250% ) kinetic energy of @xmath148 ergs than for megeath s ring . the sum of this kinetic energy of the ring expansion ( almost ) parallel to the galactic plane and the kinetic energy perpendicular to the plane ( estimated to be @xmath134 ergs in the previous section ) yields total kinetic energy of @xmath149 ergs , again requiring multiple supernovae . we also estimate the timescale of the ring expansion parallel to the plane , @xmath150 pc@xmath151 myr , which is in good agreement with the timescale @xmath152 myr independently estimated in the previous section from the velocity component @xmath102 perpendicular to the galactic plane . note that in either interpretation of these two rings , the center of the ring ( denoted by @xmath153 in figures [ fig : hilv]b and [ fig : hilv]c ) is at the same galactic longitude , @xmath118 ( as indicated by yellow dotted lines in figure [ fig : hilv ] ) . considering the ring center as the origin of the superbubble , we can compare the ( observed ) line - of - sight velocities of the origin ( i.e. the ring center in figures [ fig : hilv]b and [ fig : hilv]c ) , ngc 281 and iras 00420 + 5530 with those expected from the galactic rotation model , expressed by @xmath154 where @xmath155 and @xmath156 are the galactocentric distance and rotational velocity of the galaxy at the source position ( @xmath19 , @xmath86 ) in the galactic coordinates , and those for the lsr are @xmath95 and @xmath96 . here we also included @xmath157 , the source velocity component perpendicular to the galactic plane ( toward the north galactic pole ) , in order to take into account the projection effects on the line - of - sight velocities for sources at non - zero galactic latitudes . however , we neglect this projection - effect term because @xmath158 for ngc 281 and iras 00420 + 5530 and also because our measurement gives @xmath159 km s@xmath4 of ngc 281 smaller than the galactic rotation ( @xmath156 and @xmath96 ) by an order of magnitude . using the heliocentric distance to the source , @xmath21 , its projection onto the galactic plane , @xmath160 , and the cosine formula , @xmath155 is expressed by @xmath161 again we adopt @xmath162 km s@xmath4 , @xmath163 kpc , and then the line - of - sight velocities expected from the galactic rotation model are : @xmath164 km s@xmath4 for ngc 281 , @xmath165 km s@xmath4 for iras 00420 + 5530 , and @xmath166 km s@xmath4 for megeath s ring center ( expected to be at the same heliocentric distance @xmath75 kpc as ngc 281 from figure [ fig : hilv]b ) , and @xmath167 km s@xmath4 for our ring center , assumed to be at @xmath168 kpc . the observed line - of - sight velocities are : @xmath169 km s@xmath4 for ngc 281 ( in @xmath89co lines ; lee & jung 2003 ) , which is redshifted from the model by @xmath170 km s@xmath4 , and for iras 00420 + 5530 , on the other hand , @xmath171 km s@xmath4 ( in c@xmath172s(3@xmath62 ) and other lines ; brand et al . 2001 ) , which is blueshifted from the model by @xmath173 km s@xmath4 . this large blueshift of iras 00420 + 5530 is likely due to the expansion of the molecular ring in the superbubble . the redshift of ngc 281 is capable of two different interpretations using the megeath s ring model and ours . for megeath s ring , the line - of - sight velocity of the ring center ( in figure [ fig : hilv]b ) , @xmath174 km s@xmath4 , is also redshifted from the model by @xmath175 km s@xmath4 , suggesting that not only ngc 281 but also the molecular ring as a whole has a redshifted line - of - sight velocity component relative to the galactic rotation . then the ring expansion velocity is @xmath176 km s@xmath4 . for our ring , on the other hand , the line - of - sight velocity of the ring center ( in figure [ fig : hilv]c ) , @xmath177 km s@xmath4 , is blueshifted from the model velocity by @xmath178 km s@xmath4 , indicating the redshift of ngc 281 is due to the ring expansion . then the ring expansion velocity is obtained to be @xmath179 km s@xmath4 , and the timescale of expansion is roughly : @xmath150 pc@xmath180 myr , in good agreement with our independent timescale estimates above for ring expansion along the galactic disk ( @xmath181 myr ) and for vertical expansion out of the disk ( @xmath182 myr ) . as mentioned in the previous section , the superbubble as a whole seems to have small systemic velocity deviations ( @xmath183 km s@xmath4 ) from the galactic rotation , not only along the line of sight ( toward blueshift ) but also in the direction toward decreasing galactic longitude . this motion relative to the galactic rotation model is plotted as blue dotted arrows in figures [ fig:3d]a and [ fig:3d]b with expected expanding motion as green dashed arrows and the resulting observed velocity of ngc 281 as red arrows . the deviation of the superbubble motion from the galactic rotation , in particular the motion toward decreasing galactic longitude ( against the expected expanding motion toward increasing longitude ) , might be reflecting the peculiar motion of the perseus spiral arm , which may be explained by velocity jumps over a spiral shock in the perseus arm . peculiar motion in the perseus arm is also reported by xu ( 2006 ) for the massive star - forming region w3(oh ) ( the position indicated in figure [ fig:3d]b ) to be rotating slower than the galactic rotation by @xmath184 km s@xmath4 . for the ngc 281 superbubble , this lends support to the likely peculiar motion of the whole superbubble inward the galactic rotation and toward the sun . also in figure [ fig:3d]b , we have plotted star - forming regions in the solar neighborhood and in the perseus arm of the galaxy with recent parallactic distances precisely measured with vera and the vlba : orion kl ( hirota 2007 ; menten 2007 ; kim 2008 ) , ngc 1333 ( hirota 2008 ) , iras 16293 - 2422 in @xmath185 oph east ( imai 2007 ) and w3(oh ) ( xu 2006 ; hachisuka 2006 ) , demonstrating the high capability of the vlbi techniques to reliably trace out the spiral arms of the galaxy . our parallactic distance to ngc 281 is likely to trace the far side of the perseus arm , and we expect to reveal the detailed structure and motion of the perseus arm with future parallax measurements with vera . the parallactic distances to ngc 281 and iras 00420 + 5530 revealed the size of the superbubble to be above @xmath186 pc in diameter along the line of sight , almost parallel to the galactic plane . compared to the superbubble size @xmath187 perpendicular to the galactic plane , @xmath188 pc , the superbubble has a structure likely elongated along the plane , or at least spherical ( i.e. by taking the lower limit of 360 pc in size along the line of sight ) and not elongated vertically in the @xmath187 direction . this is contrary to expectation , i.e. one would expect a superbubble to expand and blowout in the @xmath187 direction with a vertical density gradient and with a long timescale of @xmath189 myr ( see e.g. , tomisaka & ikeuchi 1986 and mac low 1989 for superbubble models with vertical density stratifaction ) . it is thus indicated that the expansion of the ngc 281 superbubble is confined to the galactic disk . this may be explained by tomisaka ( 1998 ) s 3-dimensional mhd simulations which show that the superbubble expansion can be confined to the disk by the magnetic field of the disk , and our estimated 3d structure of the ngc 281 superbubble shows a very analogous shape , size and timescale to those of model a ( size @xmath190 pc perpendicular to the plane , timescale @xmath191myr ) of tomisaka ( 1998 ) , which assumes a uniform density of the ism , @xmath192 @xmath193 and a uniform magnetic field , @xmath194 g , aligned to one direction within and along the disk . therefore , the shape of the ngc 281 superbubble sugessts that the superbubble expansion may be confined to the disk due to the magnetic field of the disk . within the disk , the tomisaka ( 1998 ) model predicts the elongation of the superbubble to be along the direction of the magnetic field of the disk , which could be in disagreement with the ngc 281 case if the magnetic field lies along the perseus spiral arm ( e.g. , han 2006 ) . the density decrease of the perseus spiral arm might be also responsible for the structure formation at the position and heliocentric distance of ngc 281 , thus yielding a possible implication on the width of the perseus spiral arm . the further details of the 3d structure of the ngc 281 superbubble and its relation to the perseus arm should be revealed by future parallax measurements with vera of multiple maser sources in the superbubble . ( 160mm,150mm)figure5.eps our astrometric determination of the distance to ngc 281 also provides important information toward better understanding of early - type ( o@xmath6b ) high - mass ( @xmath8 ) stars and stellar evolution in combination with photometric studies of the ob stars in ngc 281 ( e.g. , henning 1994 ; guetter & turner 1997 ) . in the 1990s , the _ hipparcos _ astrometry satellite measured the trigonometric parallaxes of more than 100 000 stars in the solar neighborhood with a precision of about 1 mas , including direct distance estimates of 20 853 stars with a precision better than 10 percent and of 49 399 stars ( in total ) with a precision better than 20 percent ( perryman 1995 , 1997 ) . figure [ fig : hr ] is the h - r ( hertzsprung - russell ) diagram or the color - absolute magnitude diagram based on the _ hipparcos _ data of stars ( shown as black dots ) with parallactic distances determined with precisions better than 10% . the achievements of _ hipparcos _ marked a significant milestone in the history of astronomy and brought a new era of astrometry , since before _ , parallax measurements were limited to the nearest stars and thus direct distance estimates of only several hundred nearby stars were known with precisions better than 20 percent . nevertheless , the parallax measurements by _ hipparcos _ were restricted only to within a few hundred parsecs from the sun , which is much smaller than the size of the galaxy ( e.g. , @xmath215 kpc in disk radius ) . due to the low local space density of ob stars , there is a lack of early - type high - mass stars with absolute magnitudes reliably derived from astrometrically measured distances , as seen on the left side ( i.e. the high - mass end ) of figure [ fig : hr ] . alternatively , most of the distances and thus absolute magnitudes of ob stars are estimated by a comparison of photometric colors and magnitudes with an empirical standard of zero - age main sequence ( zams ) stars in the color - magnitude diagram , derived from other well - studied objects such as hyades ( see anthony - twarog 1982 ) . such zams comparisons assume compositional differences in clusters can be ignored , however , the metalicity dependence of zams stars is important when used as distance scale and is the cause of difficulty in a direct comparison between theoretical models and empirical zams curves ( e.g. , anthony - twarog 1982 ; zinnecker & yorke 2007 ) . therefore , precise and direct determination of distances , and thus absolute magnitudes , are desired and essential for further studies of stellar evolution sensitivity to metalicity . as mentioned earlier , extremely high astrometric accuracy can now be achieved with vlbi techniques out to as kiloparsec distances with precisions better than 10 percent ( e.g. , xu 2006 ; hachisuka 2006 ; menten 2007 ) . in particular , recent results of parallax measurements with vera are reported in hirota ( 2007 , 2008a , 2008b ) , honma ( 2007 ) , nakagawa ( 2008 ) , kim ( 2008 ) , and choi ( 2008 ) , clearly demonstrating vera s high capability for the galaxy - scale astrometry . using our new distance determination of ngc 281 by parallax measurements with vera , @xmath195 kpc , to derive absolute magnitudes , we have plotted the photometric data of zams of early - type stars in ngc 281 by guetter & turner ( 1997 ) in figure [ fig : hr ] as red dots , which give improved coverage of the high - mass end of the h - r diagram . in figure [ fig : hr ] , the empirical zams curves are also plotted for reference : turner ( 1976 ) in blue ; blaauw ( 1963 ) in pink ; and schmidt - kaler ( 1982 ) in green ( dashed ) . orange dotted ( lower ) and dashed ( upper ) curves in figure [ fig : hr ] show zams curves that would yield distances of ngc 281 to be 2.2 kpc and 3.5 kpc ( i.e. the distance discrepancy of ngc 281 from previous studies ) , made by additionally shifting the zams by turner ( 1976 ) , from which guetter & turner ( 1997 ) obtained a photometric distance to ngc 281 of 2.94 kpc . figure [ fig : hr ] clearly shows the ngc 281 data agree well with the empirical zams curves , and that resolving the large distance discrepancy of ngc 281 from previous studies was important . the agreement of the parallactic distance with the photometric distance is important for confirming the accuracy of the correction for extinction or reddening in photometric studies such as guetter & turner ( 1997 ) and for enhancing the precision of the h - r diagram . our determination of the parallactic distance to ngc 281 therefore contributes to the extension of the h - r diagram for the study of high - mass stars and , furthermore , opens up new possibilities of using the h - r diagram with parallax measurements for galaxy - scale studies , for example , investigating the formation of the galaxy via the difference in the h - r diagram for components such as the bulge and spiral arms . further contributions are expected with the vlbi techniques and with vera for direct parallax determination far beyond the solar neighborhood toward a comprehensive understanding of the galaxy . in this paper , we have presented the results of our multi - epoch phase - referencing observations with vera over 18 months of an h@xmath0o maser source in the galactic star - forming region ngc 281 west , associated with a fragmenting superbubble @xmath196 pc above the perseus spiral arm . the primary results are summarized as follows : 1 . we detected a trigonometric parallax of [email protected] mas , corresponding to a distance of [email protected] kpc to ngc 281 . our parallactic distance agrees well with the photometric distance of [email protected] kpc derived by guetter & turner ( 1997 ) , allowing for improved study of the absolute magnitutudes of high - mass ob stars , and resolved the large distance discrepancy of ngc 281 from previous photometric and kinematic studies . we revised the absolute proper motions of the h@xmath0o maser features measured in sato ( 2007 ) , and using our new determination of the parallactic distance , we derived more precisely the velocity component of the ngc 281 region @xmath197 km s@xmath4 perpendicular to the galactic plane . this yields a timescale of @xmath182 myr and kinectic energy of the region to be @xmath134 ergs in the direction perpendicular to the plane . we demonstrated the 3d structure and expansion of the ngc 281 superbubbles in comparison of our parallactic distance to ngc 281 with the parallactic distance derived by moellenbrock ( 2007 ) for another h@xmath0o maser source , iras 00420 + 5530 in the superbubble . our new parallactic distance revealed the structure of the superbubble @xmath2650 pc in size parallel to the galactic disk and with a shape slightly elongated along the disk or spherical , but not vertically elongated in the @xmath187 direction . therefore , the superbubble expansion may be confined to the disk , possibly due to the magnetic field of the disk . we suggested a new possible interpretation of an expanding molecular ring parallel to the galactic plane in the longitude - velocity diagram by megeath ( 2002 , 2003 ) . in either interpretation , the ring center , i.e. the likely origin of the superbubble lies at @xmath118 . the velocity deviation of the superbubble from the galactic rotation is estimated to be @xmath183 km s@xmath4 inward the galactic rotation and toward the sun from our proper motion measurements and from the longitude - velocity diagram , which might be due to velocity jumps over a spiral shock in the perseus spiral arm . we also estimated the velocity component and timescale of the ring expansion to be @xmath198 km s@xmath4 and @xmath199 myr , respectively , parallel to the galactic plane , which are in good agreement with those independently estimated for the direction perpendicular to the galactic plane , @xmath197 km s@xmath4 and @xmath182 myr . the kinetic energy estimate of the region parallel to the plane is @xmath148 ergs , and the total kinetic energy of both perpendicular and parallel to the plane is then estimated to be @xmath149 ergs , requiring multiple supernovae . we are deeply grateful to the referee dr . felix j. lockman for his invaluable comments and suggestions that improved both scientific discussions and presentation of the manuscript . we would like to thank dr . mark j. reid for his tremendous help with carefully reading and improving the manuscript and for his insightful comments on scientific issues . we are very grateful to prof . kohji tomisaka for illuminating and insightful discussions . we also greatly appreciate the optical image for figure 1 that was kindly provided by dr . travis a. rector and the national optical astronomy observatory ( noao ) . we wish to thank all the support staff at vera for their efforts and continuous support for our observations . m. sato gratefully acknowledges financial support by a research fellowship from the japan society for the promotion of science ( jsps ) for young scientists . this research has made use of the simbad database , operated at cds , strasbourg , france and of nasa s astrophysics data system . megeath , s. t. , biller , b. , dame , t. m. , leass , e. , whitaker , r. s. , & wilson , t. l. 2002 , in asp conf . 267 , hot star workshop : the earliest stages of massive star birth , ed . p. a. crowther ( san francisco : asp ) , 257 moellenbrock , g. a. , claussen , m. j. , & goss , w. m. 2007 , in proc . iau symposium 242 : astrophysical masers and their environments , ed . j. m. chapman & w. a. baan ( cambridge : cambridge univ . press ) , 168

we analyse two - dimensional near - irband spectra from the inner 660pc of the seyfert galaxy ngc7582 obtained with the gemini gnirs ifu at a spatial resolution ofpc and spectral resolving power r . the nucleus harbors an unresolved source well reproduced by a blackbody of temperature tk , which we attribute to emission by circumnuclear dust located closer than 25pc from the nucleus , with total mass ofm . surrounding the nucleus , we observe a ring of active star formation , apparently in the galactic plane , with radius ofpc , an age ofmyr and total mass of ionized gas ofm . the radiation of the young stars in the ring accounts for at least 80% of the ionization observed in the br emitting gas , the remaining being due to radiation emitted by the active nucleus . the stellar kinematics was derived using the co absorption band at m and reveals : ( 1 ) a distorted rotation pattern in the radial velocity field with kinematic center apparently displaced from the nuclear source by a few tens of parsecs ; ( 2 ) a high velocity dispersion in the bulge ofkms ; ( 3 ) a partial ring ofkms , located close to the br emitting ring , but displaced bypc towards the nucleus , interpreted as due to stars formed from cold gas in a previous burst of star formation . the kinematics of the ionized gas shows a similar rotation pattern to that of the stars , plus a blueshifted component with velocitieskms interpreted as due to an outflow along the ionization cone , which was partially covered by our observations . the mass outflow rate in the ionized gas was estimated as , which is one order of magnitude larger than the accretion rate to the active galactic nucleus ( agn ) , indicating that the outflowing gas does not originate in the agn , but is instead the circumnuclear gas from the host galaxy being pushed away by a nuclear outflow . the flux distribution and kinematics of the hot molecular gas , traced by the h.22 m emission line , suggests that most of this gas is in the galactic plane . an excess blueshift along pa , where a nuclear bar has been observed , can be interpreted as an inflow towards the nucleus . we thus conclude that the h kinematics traces the feeding of the agn , while the ionized gas kinematics traces its feedback via the outflows . an agn - starburst connection in the nucleus of ngc7582 is supported by the ratio between the mass accretion rate and the star formation rate in the circumnuclear region of.26% , which is close to the expected relation between the mass of the smbh and that of the host galaxy bulge in galaxies ( the magorrian relation ) . [ firstpage ] galaxies : seyfert infrared : galaxies galaxies : ngc7582 ( individual ) galaxies : kinematics galaxies : outflows galaxies : starburst

0811.2327

many recent studies on the stellar population in the vicinity of active galactic nuclei ( hereafter agn ) report an excess of young to intermediate age stars in the inner few hundred parsecs when compared to non - active galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? this excess supports the existence of the so - called agn - starburst connection @xcite , which can be understood as due to the fact that both circumnuclear star formation and nuclear activity feed on gas inflowing towards the nuclei of galaxies . the gas infllow , if massive enough , in its way to the centre , may trigger circumnuclear star formation , and , reaching the nucleus , also the nuclear activity . alternatively , there may be a delay in the triggering of the nuclear activity , which would be fed via mass loss from the evolving stars . the agn - starburst connection is consistent and somewhat expected within the presently favored scenario of co - evolution of galaxies and their nuclear supermassive black holes ( smbh ) , supported by the m@xmath18 relation @xcite : if the bulge grows via the formation of new stars ( starbursts ) , and the smbh grows via mass accretion from its surroundings , one could expect a relation between the nuclear mass accretion rate and the star - formation rate in circumnuclear starbursts . the mechanism through which the gas makes its way from the circumnuclear regions of star formation which in many cases have a ring - like morphology to the agn are nevertheless still elusive . in @xcite , using optical images obtained with the hubble space telescope ( hst ) of the inner few hundred parsecs of both active and non - active galaxies , the authors concluded that there is a strong correlation between the presence of dusty structures ( spirals and filaments ) and nuclear activity in early - type galaxies , proposing that these structures trace the material in its way in to feed the active nucleus . recently , this was shown to be the case for at least two low - activity agn with circumnuclear rings of star - formation : ngc1097 @xcite and ngc6951 @xcite . in these two galaxies , the authors have observed streaming motions along nuclear spirals in the h@xmath19 emitting gas . another similar case is the seyfert galaxy ngc4051 where @xcite found streaming motions along a nuclear spiral in molecular gas emitting the [email protected]@xmath15 m line . inflows towards agn are difficult to observe , particularly in ionized gas , due the predominance of emission from outflowing gas in the vicinity of the active nucleus . the inflow probably occurs mostly in `` cold '' gas , while the outflows are seen in ionized gas ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? studies aimed to investigate the kinematics and flux distributions of the molecular and ionized gas in the near - infrared ( hereafter near - ir ) bands indeed reveal that the h@xmath17 emitting gas usually has a distinct kinematics and distribution from that of the ionized gas . the h@xmath17 kinematics seems to be `` colder '' and , together with the flux distribution , suggest that the molecular gas is more restricted to the galactic plane , while the ionized gas seems to extend to higher galactic latitudes and has a `` hotter '' kinematics @xcite . these studies suggest that the feeding of agn is dominated by inflow of cold molecular gas , while the ionized gas constributes mostly to its feedback via the outflows . this work is aimed at investigating the relation between circumnuclear star formation and nuclear activity , as well as to look for the mechanisms of feeding and feedback in the nearby seyfert galaxy ngc7582 , which has a hubble type sbab and has a distance of 21.6mpc ( for @xmath21 from @xcite , with 1@xmath220 corresponding to @xmath3 105pc at the galaxy , adopting @xmath23 ) . this galaxy is an ideal candidate for this study because it has both circumnuclear star formation @xcite and a nuclear outflow observed by @xcite using the taurus fabry - perot spectrometer . the outflows , observed in the [ oiii ] emission line are oriented approximately perpendicular to the galactic plane , with velocities of up to @xmath24 , and are co - spatial with an ionization cone observed in an [ oiii]@xmath25 narrow band image @xcite . besides mapping the [ oiii ] outflow , @xcite also reported the presence of a kpc - scale rotating disc in the plane of the galaxy , observed in h@xmath19 . @xcite have studied the star formation in the circumnuclear disc using [ neii]12.8@xmath15 m narrow band images , and report the presence of embedded star clusters claiming that there are no counterparts detected at optical or near - ir wavelengths . they calculate the mass of the smbh as 5.5@xmath110@xmath26m@xmath5 using high resolution mid infrared spectroscopy . we use k - band spectroscopic observations obtained with the integral field unit ( ifu ) of the gemini near - infrared spectrograph ( gnirs ) to map the stellar and gaseous flux distributions and kinematics of the inner few hundred parsecs of ngc7582 . the k - band spectra allow the mapping of the stellar kinematics via the [email protected]@xmath15 m absorption feature , the ionized gas flux distribution and kinematics via the br@xmath7 emission line and the molecular gas flux distribution and kinematics via the h@xmath17 emission lines . this paper is organized as follows : in section 2 we describe the observations and data reduction . in section 3 we present the two - dimensional flux distributions and kinematics for both the emitting gas and stars . the results are discussed in section 4 , and in section 5 we present our conclusions . the data were obtained at the gemini south telescope with the gnirs ifu @xcite in 2005 nov 10 as part of program gs-2005b - q-65 , and comprise four individual exposures of 900s centred at @xmath27=2.24@xmath15 m . the 111 l / mm grating with the short blue camera ( 0@xmath2215/pixel ) was used , resulting in a nominal resolving power of r=5900 . the gnirs ifu has a rectangular field of view , of approximately 3@xmath222@xmath14@xmath228 , divided into 21 slices . at the detector , the slices are divided along their length into 0@xmath2215 square ifu elements . the major axis of the ifu was oriented along position angle pa=203@xmath28 , in order to partially cover the ionization cone observed in the [ oiii]@xmath29 narrow - band image by @xcite . two sets of observations were obtained , one centred at 1@xmath222 from the nucleus along pa=23@xmath28 and another centred at 0@xmath2275 from the nucleus along pa=203@xmath28 , the small offset being required to fill in the gaps between the slices . the observing procedure followed the standard object sky sky object dither sequence , with off - source sky positions since the target is extended . conditions during the observations were stable and clear , with image quality in @xmath30-band , in the range 0@xmath2245 - 0@xmath2255 , as obtained for the full width at half maximuum ( fwhm ) of the stellar profiles . the resulting spectral resolution , measured from the fwhm or the emission lines of the arc calibrations is @xmath32.9 . the data reduction was accomplished using tasks in the gemini.gnirs iraf package as well as generic iraf tasks . the reduction procedure included trimming of the images , flat - fielding , sky subtraction , wavelength calibration and s - distortion correction . the telluric bands were removed and the frames flux calibrated by interpolating a black body function to the spectrum of the telluric standard star , observed immediatelly after the galaxy . the final ifu datacube contains 840 spectra covering the spectral range from 2.14@xmath31 m to 2.33@xmath31 m , each spectrum corresponding to an angular coverage of 0@xmath2215@xmath10@xmath2215 , which translates to 16@xmath116pc@xmath2 at the galaxy . the total observed field of view 6@xmath223@xmath13@xmath220 ( obtained by mosaicing the two set of observations ) corresponds to a region of projected dimensions 660pc@xmath1315pc at the galaxy . in the top - left panel of figure[large ] we present a large scale optical @xmath32-band image of ngc7582 from @xcite , obtained with the cerro tololo inter - american observatory ( ctio ) 1.5 m telescope . the top - right panel presents a continuum image observed with the hst wide field planetary camera 2 ( wfpc2 ) using the filter f606w obtained from the hst archive ( program 8597 pi : regan , m. ) . we have overlaid on the hst image the contours ( in green ) of the narrow - band [ oiii]@xmath29 image from @xcite , which maps the ionization cone . the central rectangle shows the field - of - view of the gnirs ifu observations , which covers only a small fraction of the ionization cone observed in [ oiii ] . in the bottom - right panel we present a br@xmath7 flux distribution image reconstructed from the gnirs ifu datacube , where we have marked , in magenta , the star clusters found by @xcite . these clusters delineate an elongated ring of recent star formation , with semi - major axis of @xmath33 , which was almost entirely covered by our observations . in the bottom - left panel , we show three charateristic ifu spectra obtained within circular apertures of @xmath34 radius : the nuclear spectrum centred on the location corresponding to the peak flux in the continuum ( position n in the br@xmath7 intensity map ) , and spectra from two locations at 17 nw ( position a ) , and at 17 se of the nucleus ( position b ) , located in the star - forming ring . the spectra of the latter regions present emission lines of br@xmath35 m , h@xmath36 m and h@xmath37 m , used to map the gaseous kinematics and flux distributions , and the stellar absorption band of @xmath38co(2,0)@xmath39 m , used to obtain the stellar kinematics . the nuclear spectrum is redder than those from the ring , with a flux in the continuum 10@xmath4015 times higher than in the ring , and shows a narrow br@xmath7 emission line , on the top of what seems to be a broad component of the line . ( top ) , and h@xmath36 m ( middle ) emission lines with mean uncertainties of 6% and 11% , respectively , and reconstructed image for the @xmath41 m continuum emission ( bottom ) . the black contours are iso - intensity curves , and the green contours in the br@xmath7 flux map delimit regions where the line flux per pixel is higher than 50% of the peak flux in the region . the magenta filled circles indicate the position of the embedded star clusters observed by @xcite . the black circles surrounding the nucleus in the br@xmath7 flux map represent the apertures used to extract the spectra shown in fig.[fitting ] and the green dashed lines show the orientation of the pseudo - slits used to construct one - dimensional cuts shown in fig.[cut ] . ] flux distributions in the emission lines were obtained directly from the datacube by integrating the flux under each emission - line profile and subtracting the continuum contribution using images obtained in windows adjacent to the emission line . we have also adjusted gaussian profiles to the emission lines in order to measure the gas radial velocity ( from the central wavelength of the line ) and velocity dispersion ( from the width of the line ) . the fitting of each emission - line profile was obtained using the splot iraf task , and resulted also in an integrated flux for each line , from which we also obtained flux distributions . a comparison of the flux distributions obtained via the fitting of gaussians to the flux distributions obtained directly from the datacube showed that , although both flux distributions agree with each other within the uncertainties , the ones obtained via gaussian fitting are less noisy away from the nucleus . these flux distributions are shown in figure[flux ] for the br@xmath7 ( top panel ) and h@xmath36 m ( middle panel ) emission lines with mean uncertainties of 6% and 11% , respectively . the br@xmath7 emission is extended along the major axis of the galaxy ( pa@xmath42 ) and show flux enhancements in an arc shaped structure which approximately trace the star clusters observed by @xcite , shown again by the magenta circles . two strong br@xmath7 emission peaks in the ring are located at @xmath33nw and @xmath33se of the nucleus , and are encircled by the green contours in fig.[flux ] which correspond to the limits within which the flux in each pixel is higher than 50% of the peak flux in the region . we note that each of our encircled regions include two of the clusters identified by @xcite . while the br@xmath7 flux distribution show intensity variations and enhacements in the position corresponding to the star clusters , mapping the ring of star - formation , the h@xmath17 flux distribution ( middle panel of fig.[flux ] ) is more uniform througouth the nuclear region , showing no enhancements at the star clusters . the black regions correspond to pixels for which the s / n ratio was too low to allow a reliable measurement of the emission line . the bottom panel of figure[flux ] shows the @xmath41 m continuum image obtained from interpolation between two continuum windows adjacent to the br@xmath7 profile . the iso - intensity contours in the continuum image are assymetric , being more extended to the w sw than to the e ne , as expected , due to the strong dust absorption observed in optical images , in good agreement with the 1.6@xmath15 m continuum image presented by @xcite . . the observed spectra are shown in black , the fitted template in red and the residuals in green.,title="fig : " ] . the observed spectra are shown in black , the fitted template in red and the residuals in green.,title="fig : " ] in order to obtain the stellar line - of - sight velocity distribution ( losvd ) we fitted the @xmath38co(2,0 ) stellar absorption bandhead at 2.2935@xmath15 m in the @xmath30-band spectra using the penalized pixel - fitting ( ppxf ) method of @xcite . the algorithm finds the best fit to a galaxy spectrum by convolving template stellar spectra with a given losvd . this procedure outputs the stellar radial velocity , velocity dispersion and higher - order gauss - hermite moments for each spectrum fitted . the ppxf method allows the use of several stellar templates , and also to vary the relative contribution of the different templates to obtain the best fit , thus minimizing the problem of template mismatch . for this study , we have selected as template spectra those of the gemini spectroscopic library of late spectral type stars observed with the gnirs ifu using the grating 111 l / mm in the _ k_-band @xcite . in @xcite , we describe some properties of this library and discuss the problem of template mismatch using the co absorption bandheads with the ppxf method . in figure[fit ] we show the resulting fit for positions a ( top ) and b ( bottom ) marked in fig.[large ] . for the regions where the signal - to - noise ( hereafter s / n ) ratio was lower than @xmath43 we replaced the spectrum of each single pixel by an average including the 9 nearest pixels . even adopting this procedure , we could obtain measurements of the stellar kinematics only for the central @xmath44 region , as the s / n ratio of the spectra close to the borders of the ifu field becomes too low to obtain a reliable fit . for this same reason , we could not obtain stellar kinematic measurements also very close to the nucleus ( radius@xmath4503 ) , as the co bandhead feature becomes diluted by dust and line emission . in figure[stel ] we present 2d maps of the stellar kinematics . black pixels in this figure correspond to the masked regions where we did not obtain a reliable fit . the central cross marks the position of the nucleus , defined as the locus of the peak of the @xmath0band continuum and the dashed line shows the pa=165@xmath28 , which is the position angle of the major axis of the galaxy @xcite . the top left panel of fig.[stel ] shows the stellar radial velocity field , from which we subtracted the systemic velocity of the galaxy listed in the hyperleda database @xmath46 ( relative the local standard of rest ) , which agrees with our value for the nuclear region . the mean uncertainty in the radial velocity is @xmath47 . the radial velocity field shows bluehisfts to the se and redshifts to the nw , suggesting rotation , although with the kinematical centre displaced from the peak of the continuum emission . the velocity field has an amplitude of @xmath48 ( from @xmath24 to @xmath49 ) and seems not to be symmetric relative to the nucleus , with the highest blueshifts to the s sw being reached closer to the nucleus than the highest redshifts to the n nw . a one - dimensional cut along the galaxy major axis suggests that the turnover of the rotation curve to the s sw is reached already at @xmath3100150pc from the nucleus , but the turnover to the n nw seems to be beyond the border of the field ( see fig.[cut ] ) . in the top right panel of fig.[stel ] we present the stellar velocity dispersion ( @xmath50 ) map , with values ranging from 40 to 170kms@xmath10 , and mean uncertainties of @xmath51 . the highest values are observed to the e and s of the nucleus . a partial ring of low @xmath50 values ( @xmath52 ) is observed at @xmath53 from the nucleus , surrounding it from n to sw . the @xmath54 and @xmath55 gauss - hermite moments maps , which measure deviations of the losvd from a gaussian distribution , are presented at the bottom panels of fig.[stel ] and have mean uncertainties of 0.03 . the values of @xmath54 and @xmath55 are between @xmath400.18 and 0.18 , which are similar to typical values obtained for a large number of galaxies in similar analyses @xcite . in the left panels of figure[gas ] we present the radial velocity fields obtained from the peak wavelengths of the br@xmath7 and [email protected]@xmath31 m emission lines , with mean uncertainties of 6kms@xmath10 and 9kms@xmath10 , respectively . the systemic velocity of the galaxy has been subtracted from all the emission - line velocity maps . the velocity field of the br@xmath7 emitting gas is very similar to that of the h@xmath17 emitting gas , being also similar to the stellar velocity field , with the s se side approaching and the n nw side receeding . the largest blueshifts are observed s of the nucleus reaching values of @xmath56 , while the largest redshifts reach smaller values of @xmath57 and are observed nw of the nucleus . the gas velocity dispersion ( @xmath58 ) maps are shown in the right panels of fig.[gas ] . mean uncertainties are 7kms@xmath10 for br@xmath7 and 10kms@xmath10 for h@xmath17 . the values were corrected for the instrumental broadening ( @xmath59 ) , and range from @xmath60 to @xmath61 for br@xmath7 and from @xmath60 to @xmath62 for h@xmath17 . the br@xmath7 velocity dispersion show regions of low values ( @xmath63 ) to the n of the nucleus , and higher values to the s se . the highest @xmath58 values observed in the h@xmath17 line are located around the nucleus , and the lowest values in regions away from it . in order to better map the gaseous kinematics , and as our data has a relatively high spectral resolving power , we could `` slice '' the line profiles in a sequence of velocity channels . our goal is to better sample the gas kinematics over the whole velocity distribution , including the emission - line wings ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? the flux distributions in the velocity slices were obtained after subtraction of the continuum contribution determined as the average of the flux from both sides of the emission lines . each slice corresponds to a velocity bin of @xmath64 ( two spectral pixels ) and the resulting flux distributions are shown in figs.[slices_br ] and [ slices_h2 ] . in these figures , the flux levels are presented in logarithmic units , and the slices trace the gas from negative ( blueshifted , top left ) to positive ( redshifted , bottom right ) velocities relative to the systemic velocity . for the br@xmath7 emitting gas , the highest blueshifts of @xmath65 are observed at @xmath31sw of the nucleus , while the highest redshifts of @xmath66 are observed at @xmath67nw of the nucleus . for the h@xmath17 emitting gas , the highest blueshifts reach @xmath68 and the highest redshifts @xmath69 , observed aproximately at the same locations as for the br@xmath7 emision . while the highest blueshifts are observed displaced from the major axis , in the velocity range from @xmath70200 to 200kms@xmath10 the emission peak moves from s se to n nw essentially parallel to the major axis , suggesting that , for this range of velocities , the emission is dominated by gas rotating in the plane of the galaxy . 8 diameter ( top spectrum ) , extranuclear spectrum for a ring with inner radius of 0@xmath71 and outer radius of 0@xmath72 from the nucleus ( bottom spectrum ) and the difference between the two ( middle spectrum ) . the extranuclear spectrum was scaled by the ratio between the areas of the nuclear and extranuclear apertures . ] the flux distribution in the continuum shows an unresolved source at the nucleus ( whose location we assume to coincide with the peak of the flux distribution ) . in order to isolate the spectrum of the nuclear source , we have extracted its spectrum within an aperture of radius 0@xmath71 , which corresponds to the radial distance where the point source profile decreases to 1/3 of its peak value . in order to subtract the underlying stellar contribution to the nuclear flux we have extracted a spectrum within a ring with inner and outer radii of 0@xmath224 and 0@xmath226 , respectively , surrounding the nuclear aperture . we show the apertures corresponding to the nuclear and extranuclear spectra as circles in the top panel of fig.[flux ] . this choice of apertures was dictated by the need to isolate the nuclear spectrum from that of the circumnuclear ring of star formation . larger apertures for the nuclear and extranuclear spectra would include contribution from the star - forming ring . the resulting nuclear and extranuclear spectra are shown in fig.[fitting ] ( top and bottom spectra ) . in the same figure , we also show the difference between them ( middle spectrum ) after scaling the circumnuclear spectrum to the flux it would have if obtained through the same aperture as that of the nuclear spectrum . we note that the circumnuclear spectrum does not differ much from those of the star - forming regions a and b , suggesting that recent star formation is occurring also inside the ring . this is supported by previous studies of the stellar population in the optical @xcite and near - infrared @xcite . the nuclear spectrum seems to present both a narrow and a broad component in the br@xmath7 emission line . the fit of a gaussian to the broad component , as illustrated in the top spectrum of fig.[fitting ] , gives a fwhm@xmath33900 kms@xmath10 , with the centre of the line blueshifted by @xmath73kms@xmath10 relative to the narrow component . the presence of a broad component in br@xmath7 , with fwhm @xmath34000 kms@xmath10 has been previously reported by @xcite , in long - slit observations obtained with the very large telescope . a broad component , observed in h@xmath19 , was also reported by @xcite , although this component seems to have been produced by a transient event . these authors noted the appearance and subsequent fading of the broad h@xmath19 component as well as other optical emission lines and proposed three possible scenarios for its origin : capture and disruption of a star by a smbh , a reddening change in a dusty torus surrounding the nucleus , and a type iin supernova exploding in a compact starburst . the nuclear spectrum is redder than the extranuclear one , and the subtraction of the stellar contribution does seem to have eliminated the stellar 2.3@xmath15 m co absorption , clearly observed prior to the subtraction . the resulting spectrum shows a red continuum similar to those of other red unresolved components observed in other agn and attributed to emission from hot dust , possibly distributed in a toroidal structure around the smbh , with temperature close to the dust sublimation temperature @xcite . under the assumption that the red unresolved component is indeed due to emission by dust , we have derived its temperature and mass , by fitting a blackbody function to its spectrum . the resulting fit is shown as a dashed line in the middle spectrum of fig.[fitting ] , corresponding to a temperature @xmath74k , for which the maximuum of the blackbody curve is reached at @xmath75 m . we have estimated the mass of the hot dust producing the emission in the unresolved component using the formalism of @xcite , for dust composed by grains of graphite . the infrared spectral luminosity of each dust grain , in ergs@xmath10hz@xmath10 , can be written as : @xmath76 where @xmath77 is the grain radius , @xmath78 is its absorption efficiency and @xmath79 is its spectral distribution assumed to be a planck function for a temperature @xmath80 . the total number of graphite grains can be obtained from : @xmath81 where @xmath82 is the total luminosity of the hot dust , obtained by integrating the flux under the function fitted to the nuclear extranuclear spectrum and adopting a distance to ngc7582 of @xmath83mpc . in order to obtain @xmath84 we have adopted @xmath85k , as derived above , @xmath86 m and @xmath87 ( for @xmath88 in hz ) as in @xcite . finally , we can obtain the total mass of the emitting dust as : @xmath89 adopting @xmath90g@xmath91 for the graphite density @xcite , we obtain @xmath92 . in table[masses ] we compare this dust mass with masses of hot dust obtained for other agn in previous studies . the hot dust mass in ngc7582 is similar to those obtained for mrk766 , ngc1068 and ngc3783 , while it is smaller than the values for mrk1239 , ngc7469 and fairall9 , and higher than those derived for ngc1566 and ngc4593 . [ cols="<,<,<",options="header " , ] [ masses ] for mrk1239 @xcite and mrk766 @xcite the dust mass values were derived from long - slit spectroscopy , with extraction apertures of the nuclear spectrum corresponding to more than 100pc at the galaxies , while the spatial resolution of our data allow us to constrain the location of the hot dust as being within @xmath93pc from the nucleus . as pointed out above , the @xmath0band excess is probably due to the emission of hot dust from a circumnuclear torus which may be located even closer to the smbh . in the case of ngc1097 , for example , @xcite have found a heavily obscured starburst closer than 9pc from the nucleus , which could be located in the outskirts of the torus . in the case of ngc1068 , interferometric mid - ir observations constrained the hot dust emission to originate much closer , in a region of aproximately 1pc diameter @xcite . ngc7582 is known to present a kpc - scale h@xmath19 emitting disk which harbors recent star formation plus heavy dust obscuration in the nuclear region @xcite . the presence of dust to the ne is responsible for the assymetric contours of the continuum flux map in the bottom panel of fig.[flux ] . the recent star - formation in the dust - embedded clusters reported by @xcite ( shown as magenta circles in fig . [ flux ] ) are seen as enhancements in our br@xmath7 image , delineating a ring of major axis of @xmath33@xmath225 and minor axis of @xmath31@xmath225 . this gives an axial ratio of @xmath94 , which is similar to the axial ratio of the galaxy ( @xmath95 ) , suggesting that the star - forming ring is circular , with a radius of @xmath3190pc and is located in the plane of the galaxy . as pointed out by @xcite , these star - forming clusters had not been previously seen neither in the optical nor in the near - infrared . the near - ir ifu data from @xcite , although consistent with our results , does not have the necessary spatial resolution to resolve these clusters . in our data alone , clear peaks in the br@xmath7 flux distribution are located in the ring at @xmath96nw and @xmath96se of the nucleus , and are encircled by the green contours in fig.[flux ] , which comprise the pixels with fluxes higher than 50% of the peak flux . we will refer to each of these two regions whose physical boundaries we define as corresponding to the green contours as circumnuclear star - forming regions ( cnsfrs ) , and will use our data for these regions to characterize the star - formation in the circumnuclear ring . from the locations of the star clusters found by @xcite it can be seen that each of these cnsfrs correspond to two star clusters , thus we should bear in mind that the parameters we obtain for these regions correspond to two star clusters instead of one . following @xcite , we can estimate the mass of ionized hydrogen in each cnsfr as : @xmath97 where @xmath98 is the line flux , @xmath99 is the distance of the galaxy and we have assumed an electron temperature @xmath100k and electron density @xmath101 for the cnsfrs . we can also obtain the emission rate of ionizing photons for each cnsfr using @xcite : @xmath102=\frac{\alpha_b\,l_{h\alpha}}{\alpha^{\rm eff}_{h\alpha}\,h\nu_{h\alpha}},\ ] ] where @xmath103 is the hydrogen recombination coefficient to all energy levels above the ground level , @xmath104 is the effective recombination coefficient for @xmath105 , @xmath106 is the planck s constant and @xmath107 is the frequency of the @xmath105 line . using @xmath108 , @xmath109 and @xmath110 for case b recombination assuming an electronic temperature @xmath111 @xcite , we obtain : @xmath112}{\rm s^{-1}}\right)=7.47\times10^{13}\,\left(\frac{l_{br\gamma}}{\rm erg\,s^{-1}}\right ) . \label{q}\ ] ] finally , we can also obtain the star formation rate ( sfr ) using @xcite : @xmath113 under the same asumptions adoped to derive equation[q ] . as there is considerable obscuration in the nuclear region of ngc7582 , the emission - line fluxes used in the above equations have been corrected for reddening . we have used @xmath114 , obtained by @xcite in a spectrum extracted withing an aperture of 2@xmath12arcsec@xmath2 , adopting the extinction law of @xcite . besides calculating the above quantities for the two cnsfrs , we have also calculated them for the nucleus and for the complete ring of star formation , as follows . in order to obtain the emission rate of ionizing photons for the nucleus , we integrated the br@xmath7 emission - line flux within a circular aperture of 0@xmath224 plus that emitted by the outflowing gas , under the assumption that the outflowing gas is ionized by the active nucleus . the flux in the outflowing gas was estimated as the sum of the fluxes in the velocity slices with @xmath115kms@xmath10 . in this manner , we obtain only a lower limit for the rate of ionizing photons emitted by the nucleus , as our field of view covers only a very small portion of the outflow , as discussed in sec.[results ] . an upper limit can be obtained under the assumption that the total flux in the outflow is the one observed times the ratio between the total area covered by the outflow ( estimated from the [ oiii ] image ) and that covered by our observations , which is @xmath35 . considering that there should be a counterpart outflow to the opposite side of the galaxy plane , probably hidden by the dust in the plane , the br@xmath7 flux in the outflow may be @xmath310 times the one measured from our data the br@xmath7 flux of the star - forming ring was obtained by integrating its value over the whole ifu field and subtracting the contribution of the nuclear flux plus that of the outflowing gas . table[par ] presents the results of the above calculations , together with the area used in the integrations and the br@xmath7 emission - line luminosities corrected for reddening , for the nucleus we list both the lower limit ( including only the contribution of the observed outflow ) and the upper limit ( including the contribution of the total estimated outflow ) . we also list the equivalent widths of the br@xmath7 emission line . l c c c c parameter & @xmath96nw & @xmath96se & nucleus@xmath116 & ring + area ( arcsec@xmath2 ) & 1.48 & 1.15 & 0.5 & 18.4 + log@xmath117 [ ergs@xmath10 ] & [email protected] & [email protected] & [email protected] & [email protected] + log@xmath121 { \rm [ s^{-1})]}$ ] & [email protected] & [email protected] & [email protected] & [email protected] + @xmath123 & [email protected] & [email protected] & & [email protected] + ew(br@xmath7 ) [ ] & [email protected] & [email protected] & [email protected] & [email protected] + @xmath124 ( @xmath125 m@xmath5 ) & [email protected] & [email protected] & [email protected] & [email protected] + + + [ par ] table[par ] shows that the br@xmath7 flux in inner @xmath3400pc of ngc7582 is dominated by emission from the circumnuclear star - forming ring , which ranges from 4 to 13 times that of the nucleus plus outflow . the br@xmath7 fluxes of the two cnsfrs contribute with 50% of the total emission from the star formation ring . the emission rates of ionizing photons obtained for the two cnsfrs are in good agreement with values obtained for cnsfrs in other galaxies ( e.g. * ? ? ? * ; * ? ? ? * ) , as well as with rates derived for ngc7582 in the mid - ir @xcite . such rates correspond to about 1000 o6 stars in each cnsfr @xcite . considering that each cnsfr actually corresponds to two clusters , then this would correspond to @xmath3500 o stars per cluster . the star - formation rates for the cnsfrs of @xmath127 , caracterize a moderate star - forming regime , and are in good agreement with the values derived by @xcite for a sample of 385 cnsfrs in galaxies covering a range of hubble types . we can use our data also to estimate the age of the cnsfrs , applying the method described by @xcite , which consists of comparing the observed values of ew(br@xmath7 ) to those predicted by evolutionary photoionization models . as stressed by those authors , the real age of the cnsfrs can only be obtained if the contributions both from the underlying bulge and from older stars in the cnsfrs are taken into account . in order to obtain the contribution of the underlying bulge , we used aperture photometry in the continuum image ( fig.[flux ] ) to measure both the flux of the cnsfrs continua and that of the bulge in the surrounding regions . we find a bulge contribution of @xmath350 % of the total flux at both cnsfrs , and thus the value of ew(br@xmath7 ) increases from 15 to 30 after this correction . the presence in the cnsfrs of stars formed in previous bursts of star formation is supported by the work of @xcite , who found a large spread of age in the circumnuclear star - forming regions . in order to take this effect into account , we apply the same approach as @xcite , representing the cnsfr stellar content by three bursts of star formation , with ages of 0.1 , 10 , and 20 myr , and considering two values for the upper limit in stellar mass ( @xmath128= 100 and 30 @xmath129 ) . the resulting evolutionary behaviour of ew(br@xmath7 ) is illustrated in figure [ photo ] . for a value of ew(br@xmath7)@xmath330 , we obtain an age of @xmath35.0 myr , which is similar to the values found for the cnsfrs in ngc6951 and ngc1097 by @xcite . as observed in fig.[stel ] the stellar velocity field seems to be dominated by rotation . we have thus tried to fit the stellar velocities by a model of circular orbits in a plane subject to a plummer potential , which sucessfully reproduced the stellar kinematics of the central regions of other seyfert galaxies in previous studies ( e.g. * ? ? ? * ; * ? ? ? . nevertheless , in the case of ngc7582 this model fails , as the residuals are too large everywhere . the kinematic centre of the rotation in fig.[stel ] seems to be displaced from the nucleus by @xmath1305nw ( @xmath350pc at the galaxy ) , an effect which can also be observed in one - dimensional cuts of the velocity field shown in fig.[cut ] . this is an interesting result which should be further investigated . could this be the kinematic centre of the galaxy ? if this is the case , then the peak flux of the infrared continuum would not correspond to the centre of the galaxy but instead to the location of the smbh ( the unresolved red source ) , and the ring of star formation would be centred on the smbh instead of the galaxy centre . we believe this is unlikely , as the radius of influence of the smbh is much smaller than the radius of the ring , and thus one would expect that the ring would be centred on the galaxy nucleus . our field of view is too small for us to reach a firm conclusion , but one possibility is that the apparent displacement of the centre of the rotation field can be due to distortions in a circular velocity field . observations in the mid - ir @xcite show that ngc7582 presents both a nuclear bar and two nuclear spiral arms , which could be related to the observed distortions . the nuclear bar is oriented along pa@xmath131 extending from @xmath53e to @xmath53w from the nucleus , while the nuclear spiral arms are oriented approximately along the major axis of the galaxy and extending to @xmath132nw and se of the nucleus . we show the orientation and extents of the nuclear bar and spiral as dotted and solid white lines , respectively in fig.[stel ] . we note that the se arm seems to correspond to the location where there seems to be an excess blueshift " in the stellar radial velocity field in fig.[stel ] . the stellar velocity dispersion shows the highest values of @xmath3170kms@xmath10 to e and s of the nucleus ( fig.[stel ] ) which we interpret as representing the @xmath50 of the galaxy bulge . we use this value to estimate the mass of the smbh ( @xmath133 ) of ngc7582 from the relation @xmath134 , where @xmath135 , @xmath136 and @xmath137kms@xmath10 @xcite . for @xmath138 , we obtain @xmath139 , in good agreement with the value obtained by @xcite through modelling of the kinematics of [ neii]12.8@xmath15 m emitting gas . the @xmath50 map ( fig.[stel ] ) shows a region with low values ( @xmath140 ) delineating a partial ring around the nucleus and which seems to be embedded in the bulge . these locations are close to those of the star clusters observed by @xcite , but somewhat displaced , being @xmath350pc closer to the nucleus . the lower @xmath50 values reveal that these stars have a `` colder kinematics '' than that of the stars of the bulge . we note that these @xmath50 values are similar to those of the h@xmath17 gas , suggesting that these stars have recently formed in the circumnuclear ring from kinematically cold gas , and which have not yet `` thermalized '' with the `` hotter '' stars in the bulge . but in order to present co absorption , the stars should be at least 10@xmath26yr old @xcite , what suggests that they belong to a previous burst of star - formation and not the one which is presently ionizing the gas . this interpretation is supported by the displacement of this ring relative to that in the ionized gas , which suggests that the star - formation is propagating outwards from the nucleus . ring - like regions with low @xmath50 values have been observed also around other agn @xcite . the radial velocity maps of both the br@xmath7 and h@xmath17 emitting gas ( fig . [ gas ] ) are similar to the stellar one ( fig . [ stel ] ) . in order to better observe the possible differences , we have extracted one - dimensional cuts from the radial velocities maps which are shown in figure[cut ] . these cuts were obtained by averaging the velocities within pseudo - slits of @xmath141 width passing through the nucleus . the orientations of the slits were selected to provide a good coverage of the velocity fields , and are shown as dashed green lines in the bottom panel of fig.[flux ] . the top - left panel of fig.[cut ] shows the velocities along pa=@xmath142 , which is the orientation of the major axis of the galaxy @xcite , while in the top - right panel we present the velocities along pa@xmath143 , in the bottom - left panel along pa@xmath144 and in the bottom - right panel along pa=@xmath145 . the highest differences between the gaseous and stellar velocities are observed at 0@xmath225@xmath402@xmath146sw from the nucleus along pa@xmath147 , where the gaseous velocities reach values of up to @xmath148 , while the stellar velocities reach at most @xmath24 . for the pas @xmath149 and 240@xmath150 the gaseous velocities are still more blueshifted than the stellar but only by about 40kms@xmath10 ( to se sw of the nucleus ) . along pa=@xmath151 , to nw w of the nucleus , the velocities observed for the stars are similar to those of the br@xmath7 emitting gas , while the h@xmath17 emitting gas is blueshifted by @xmath7020kms@xmath10 . we note that this orientation coincides with that of the nuclear bar , and suggest that , as this blueshift is observed against the far side of the galactic plane , we could be observing an inflow of molecular gas towards the nucleus along the bar ( which could be in the galaxy plane ) . we have already observed similar h@xmath17 inflows towards the nucleus of another active galaxy , ngc4051 , not along a nuclear bar , but along a nuclear spiral arm @xcite . , @xmath152 , @xmath153 and @xmath154 for a pseudo - slit of @xmath141 width . ] from fig.[cut ] it can be concluded that the emitting gas is blueshifted relative to the stars to sw of the nucleus . this region is co - spatial with part of the [ oiii ] ionization cone observed by @xcite , shown by the green contours in fig.[large ] . we interpret this excess blueshift as due to the contribution of gas outflowing from the nucleus to the total gas emission . we note that we are observing only a small fraction of the outflowing gas because the field - of - view of our observations is much smaller than the dimensions of the ionization cone , along which the gas is outflowing @xcite . figs . [ stel ] , [ gas ] and [ cut ] suggest that the radial velocity field of the gas is similar to that of the stars except for the region along the ionization cone / nuclear outflow and possibly along the nuclear bar where there may be an h@xmath17 inflow . in order to isolate the outflowing and inflowing components , we have subtracted the stellar velocity field from the br@xmath7 and h@xmath17 velocity fields . these residual maps are shown in figure[res ] . in the residual br@xmath7 velocity map the velocity values to ne n nw of the nucleus are close to zero , while to the s sw w the gas is blueshifted by up to @xmath155kms@xmath10 relative to the stars . this velocity is of the order of those found by @xcite for the [ oiii ] emitting gas . in the residual h@xmath17 velocity map , the blueshift at pa@xmath3293(-67 ) can be observed in approximate alignment with the nuclear bar , whose orientation is shown as the white solid line in the bottom panel of fig.[res ] . and h@xmath17 radial velocity fields and the stellar one . the white line in the bottom panel show the orientation and extent of the nuclear bar observed by @xcite . ] even higher blueshifts of up to @xmath156 are observed to the s sw in the velocity channel maps ( fig.[slices_br ] ) . the difference between the radial velocity maps and the velocity channel maps is due to the fact that the first give the peak velocity of the emission line for a given spatial position , while the latter show how the gas with a certain velocity is distributed along the nlr . the peak wavelength will usually correspond to the kinematics of the component(s ) with the highest flux(es ) , while the channel maps will reveal also weaker components which contribute to the wings of the emission - line profiles . we thus attribute the high blueshifts seen in the velocity channel maps to low mass , high velocity outflows , originated close to the nucleus which compress and push the gas further out to the observed radial velocities of @xmath70100kms@xmath10 . these high velocity outflows do not appear in the radial velocity maps due to their low flux , appearing only as a blue wing to the profile , thus not affecting the peak wavelenght used to derive the radial velocity . we have observed similar results in a recent study of the nlr of other nearby seyfert galaxies using ifus @xcite . outflows from the nucleus of active galaxies have been observed at optical wavelengths using long - slit spectroscopy ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and from radio observations ( e.g. * ? ? ? * ) and more recently by our group using ifus @xcite . we have used the velocity channel maps of fig.[slices_br ] in order to estimate the mass outflow rate in ionized gas . inspecting these maps , it can be concluded that radial velocities of up to @xmath119200kms@xmath10 are observed to both sides fo the nucleus , suggesting that up to these velocities , we are mapping gas motions in the galaxy disk . above 200kms@xmath10 , there are mostly blueshifts and almost no redshifts , what can be interpreted as due to the presence of ouflows . our interpretation for this results is that these outflows are seen in blueshift to the s sw because they are partially oriented towards us and are projected against the far side of the galaxy disk while the counterpart redshifted outflows are not seen because they are hidden behind the near side of the dusty disk . thus , assuming that the emitting gas with blueshifts higher than 200kms@xmath10 is outflowing from the nucleus , we can estimate the mass outflow rate . in order to do this , we calculate the mass flux through a circular cross section with radius @xmath157 at @xmath158 from the nucleus , estimated from the shape of the light distribution in the velocity channels corresponding to @xmath159 and @xmath160 in fig.[slices_br ] . the mass outflow rate has been calculated as @xmath161 , where @xmath162 is the proton mass , @xmath163 is the electron density , @xmath164 is the velocity of the outflowing gas and @xmath165 is the filling factor @xcite and the factor of 2 is to take into account the likely presence of a hidden counterpart outflow to the one we observe . adopting @xmath101 and @xmath166 and adding the contribution of the two velocity bins above , we obtain @xmath167 . this outflow rate is somewhat smaller than those quoted in the literature for active galaxies , which are between 0.1 and 10@xmath168 ( e.g. * ? ? ? however , the latter values are generally obtained for objects with higher levels of activity than ngc7582 . for similar levels of activity and in comparable scales of hundreds of parsecs , @xcite have obtained similar ionized gas outflow rates of @xmath169 from 2d mapping of the gas kinematics in the central region of 6 nearby seyfert galaxies , in good agreement with the value obtained here . we now calculate the mass accretion rate necessary to produce the observed nuclear luminosity of ngc7582 . the latter can be estimated from @xmath170 , where @xmath171 is the bolometric luminosity , @xmath172 is the efficiency of conversion of the rest mass energy of the accreted material into radiation and @xmath173 is the speed of light . we adopt @xmath174 , a typical value for geometrically thin and optically thick accretion discs ( e.g. * ? ? ? * ) and use the relation @xmath175 @xcite , where @xmath176 is the hard x - ray luminosity @xcite . the result is @xmath177 . comparing the mass outflow rate with the accretion rate calculated above , it can be concluded that the former is an order of magnitude larger than the latter , a relation similar to those reported by @xcite for more luminous active galaxies , as well as to those obtained by @xcite for other nearby seyfert galaxies and to that obtained by @xcite for the seyfert galaxy ngc4151 . although the are similarities between the velocity fields observed for h@xmath17 and br@xmath7 , there are also differences in the detailed kinematics for the two gas phases , as follows . ( 1 ) the small but systematic excess blueshift in the h@xmath17 emitting gas along pa@xmath178 , suggests there could be an inflow of molecular gas along a nuclear bar , as discussed above . ( 2 ) the br@xmath179 map ( fig . [ gas ] ) shows the highest values to the s se of the nucleus , approximately at the same location where the highest blueshifts are observed ( see fig . [ res ] ) . this result supports an origin for at least part of the br@xmath7 emission of this region in disturbed gas which is not confined to the galaxy plane but is outflowing from the nucleus along the ionization cone seen in [ oiii ] . the h@xmath180 map , on the other hand , presents overall lower @xmath58 values consistent with a dominant origin in molecular gas orbiting in the plane of the galaxy . ( 3 ) the velocity channel maps shown in figs . [ slices_br ] and [ slices_h2 ] also support a larger contribution from outflowing gas to the br@xmath7 than to the h@xmath17 emission , since the blueshifted emission in br@xmath7 reaches larger velocity values and is found farther away from the line of nodes than in the h@xmath17 emitting gas . previous studies have also found that the h@xmath17 traces less disturbed kinematics than the ionized gas in the circumnuclear region of seyfert galaxies @xcite , leading to the suggestion that the molecular gas , orbiting in the plane , traces the feeding of the active nucleus via inflow of cold gas , while the ionized gas traces its feedback via the outflows . in @xcite we could trace streaming motions in the h@xmath17 emitting gas towards the nucleus along a nuclear spiral arm . in the present observations , there is also some evidence of inflow along a nuclear bar . the compactness and high inclination of the circumnuclear ring of star - formation does not favor the detection of possible additional streaming motions between the ring and the nucleus . the h@xmath17 flux distribution shows that the molecular gas is not concentrated in the star - forming ring , but is distributed over the whole nuclear region , indicating the presence of similar density of molecular gas at the ring and between the nucleus and the ring . this molecular gas could be the material which is feeding and probably will continue to feed the active nucleus for some time in the future . the total mass of hot h@xmath17 in the nuclear region of ngc7582 can be estimated as @xcite : @xmath181 where we assume a vibrational temperature of @xmath182k , which implies a population fraction @xmath183 @xcite , and a transition probability @xmath184s@xmath10 @xcite . integrating the flux of the h@xmath185 m emission line over the wole ifu field we obtain @xmath186 and thus @xmath187 . the above value is the mass of hot h@xmath17 emitting gas , which is probably only the `` hot skin '' of the molecular gas available in the nuclear region to feed the active nucleus . the total mass of molecular gas should be much higher , as it should be dominated by cold gas , which has been estimated to be 10@xmath188 to 10@xmath189 times that of the hot gas in galaxies for which both hot and cold molecular gas have been observed @xcite . thus , there should be plenty of molecular gas , not only to feed the agn but also to give origin to the recent bursts of star - formation in the ring . the molecular gas accumulated in the nuclear region was probably the fuel which gave origin to the nuclear ring of star formation , as well as to the present episode of nuclear activity . although both the star formation and nuclear activity could have been triggered simultaneously , there is also the possibility that the star formation was triggered first , and then the mass loss from stellar winds and supernova explosions in the cnsfrs have been the source of fuel for the smbh . it is thus instructive to estimate the available mass to feed the smbh under the above assumption . for solar metallicity and an age of 5myr , the total mass loss for the two cnsfrs is in the range @xmath310@xmath190m@xmath5 @xcite . assuming a value of 10@xmath191m@xmath5 , and that the duration of this process is a few 10@xmath192yr , one gets a rate of mass loss of @xmath193 m@xmath5yr@xmath10 . ( note that this value is @xmath35 times larger than the nuclear accretion rate calculated above , of @xmath177 , within the uncertainties of the calculations . ) if this mass reaches the nucleus along a nuclear bar ( as observed in ngc7582 ) or nuclear spiral arms [ as observed in ngc1097 @xcite and ngc6951 @xcite ] , with velocities of @xmath320@xmath4060kms@xmath10 , this gas would take a few @xmath194yr to reach the nucleus . this timescale alows for the coexistence of the nuclear activity and star - forming regions of 5myr under the hypothesis that the nuclear activity has been triggered by mass accretion originated from mass loss of young stars formed in the circumnuclear ring . alternatively , the triggering may have happened during a previous burst of star - formation in the ring . the presence of such previous starburst is supported by the partial ring of low @xmath50 observed in the stellar kinematics . in addition , as pointed out in sec.[nuclear ] , the extranuclear spectrum does not differ significantly from those of the star - forming ring , suggesting that there may be regions of recent star formation even closer to the nucleus , which may be contributing to the smbh feeding as well . finally , we quantify the agn - starburst connection in ngc7582 by comparing the mass accretion rate to the smbh with the star formation rate ( sfr ) in the circumnuclear region . as pointed out in the introduction , the m@xmath18 relation implies that the mass of the smbh in galaxies grows in proportion to the growth of the bulge . if the bulge grows via star formation episodes , the mass accretion rate to the smbh should be proportional to the sfr . which should be the ratio between the mass accretion rate and the sfr ? the same as the between the mass of the smbh and the mass of the host galaxy bulge , which was first obtained by @xcite , and today known as the magorrian relation @xcite , which has been revised to a ratio of 0.1% between the mass of the smbh and that of the bulge . assuming that the total sfr can be represented by that of the ring , we get for ngc7582 a ratio of 0.26% between the mass accretion rate and the sfr , a number which is close to that of the magorrian relation , considering the uncertainties . these uncertainties include , in particular , a possible underestimate of the sfr in the nuclear region , as our observation misses the top border of the star - forming ring , and a possible contribution of a starburst included within the nuclear aperture . two - dimensional near - ir @xmath0band spectroscopy from the inner 660@xmath1315pc@xmath2 of the seyfert galaxy ngc7582 , obtained with the gemini gnirs ifu at a spatial resolution of @xmath195pc@xmath2 and spectral resolution of @xmath33 , was used to map the molecular and ionized gas emission - line flux distributions and kinematics as well as the stellar kinematics . the region covered by the observations include most of a circumnuclear ring of star - formation in the plane of the galaxy plus a small part of an outflow extending to high galactic latitutes . our main conclusions are : * the nucleus contains an unresolved source whose continuum is well reproduced by a blackbody function with temperature t@xmath31050k , which we atribute to emission by circumnuclear dust heated by the agn , with an estimated mass of @xmath4m@xmath5 located within 25pc of the nucleus . the nuclear spectrum shows also a broad component in the br@xmath7 emission line with fwhm@xmath33900kms@xmath10 . * the br@xmath7 flux distribution is dominated by emission from a circumnuclear ring of star formation , previously seen only in mid - ir observations , with radius of @xmath3190pc , which contributes with @xmath38090% of the br@xmath7 emission , the rest being contributed by the nuclear source . two large star - forming regions with age @xmath35myr contribute with 50% of the br@xmath7 emission of the ring , resulting in @xmath31000 o6 stars per region and a star - forming rate of @xmath30.25m@xmath5yr@xmath10 . the total rate of ionizing photons emitted by the ring is @xmath196s@xmath10 , the total star - forming rate is @xmath31m@xmath5yr@xmath10 and the total mass of ionized gas is @xmath197m@xmath5 . * the h@xmath17 flux distribution is more uniformly distributed over the nuclear region than the br@xmath7 flux distribution and gives a mass of hot molecular gas of @xmath3620m@xmath5 . from previous studies , it can be concluded that the total mass of h@xmath17 gas ( which is dominated by cold h@xmath17 ) can be 10@xmath19810@xmath26 times larger . * the stellar velocity field shows a distorted rotation pattern whose centre appears to be displaced up to 50pc from the nucleus ( identified as the peak of the continuum emission ) . the distortion could be associated to a nuclear bar and a nuclear spiral previously seen in mid - infrared images . the velocity dispersion of the bulge is @xmath138 , implying in a mass of @xmath199 for the smbh , in good agreement with previous determinations . immersed in this bulge , there is a partial ring of lower @xmath50 ( @xmath52 ) , which is close to the star - forming ring but displaced @xmath350pc inwards . as the velocity dispersion is similar to that of the molecular gas , this ring can be interpreted as due to stars recently formed from cold gas , although in a previous burst ( age@xmath1210@xmath26yr ) , still keeping the gas kinematics as they did not have time yet to `` thermalize '' with the stars of the bulge . * the radial velocity field of the ionized gas is similar to that of the stellar component to the n nw , while to the s sw there is an additional blueshifted component ( @xmath200kms@xmath10 ) , which we attribute to outflows along the ionization cone ( previously observed in a narrow - band [ oiii ] image ) which is partially covered by our observations . velocity channel maps along the br@xmath7 emission - line profile show even higher blueshifts , reaching @xmath201kms@xmath10 in the outflow , while the maximum redshift observed is @xmath202kms@xmath10 . the velocity dispersion is enhanced in the region of the outflow . * the mass outflow rate in the ionized gas is estimated to be @xmath13 , which is an order of magnitude larger than the accretion rate to feed the agn , a ratio which is similar to those found in other agn . this large ratio indicates that the outflowing gas does not originate in the agn , but is instead the circumnuclear gas from the host galaxy being pushed away by a nuclear outflow . * the kinematics of the hot molecular gas , traced by the h@xmath17 emission , shows smaller blueshifts along the outflow as well as lower velocity dispersions , which suggests that most of the molecular gas is in the galactic plane . an excess blueshift along pa@xmath16 , where a nuclear bar has been observed , can be interpreted as an inflow towards the nucleus . we thus conclude that the h@xmath17 kinematics traces the feeding of the agn , while the ionized gas kinematics traces its feedback via the outflows . * the estimated mass loss rate from the evolving stars in the circumnuclear ring is @xmath203m@xmath5yr@xmath10 , which is @xmath35 times the nuclear accretion rate . as the estimated time for this mass to reach the nucleus is a few 10@xmath192yr , this material could be the fuel which just triggered the nuclear activity . nevertheless , as there is also a previous burst of star formation almost co - spatial with the ring , the fuel may have been available previously ( e.g. @xmath204yr ) . in addition , previous data suggest that there may be also recent star formation inside the ring , closer to the nucleus . * we conclude that the agn starburst connection in ngc7582 may have occurred in two ways : ( 1 ) there is a large molecular gas reservoir in the nuclear region which gave origin to both the star - formation in the circumnuclear ring and nuclear activity ; ( 2 ) the molecular gas has given origin first to circumnuclear star formation and the mass loss from the evolving stars have then triggered the nuclear activity . * the ratio between the mass accretion rate and the sfr in the circumnuclear region , of 0.26% is close to the value implied by the magorrian relation between the mass of the smbh and the mass of the bulges of the host galaxies . this result shows that a growth of the smbh proportional to the growth of the bulge can proceed via circumnuclear star formation around agn . we thank the anonymous referee for valuable suggestions which helped to improve the paper , as well as dr . michele cappellari for help with the ppxf routine . based on observations obtained at the gemini observatory , which is operated by the association of universities for research in astronomy , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( united states ) , the science and technology facilities council ( united kingdom ) , the national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , ministrio da cincia e tecnologia ( brazil ) and secyt ( argentina ) . this work has been partially supported by the brazilian institution cnpq .

checking the admissibility of quasiequations in a finitely generated ( i.e. , generated by a finite set of finite algebras ) quasivariety amounts to checking validity in a suitable finite free algebra of the quasivariety , and is therefore decidable . however , since free algebras may be large even for small sets of small algebras and very few generators , this naive method for checking admissibility in is not computationally feasible . in this paper , algorithms are introduced that generate a minimal ( with respect to a multiset well - ordering on their cardinalities ) finite set of algebras such that the validity of a quasiequation in this set corresponds to admissibility of the quasiequation in . in particular , structural completeness ( validity and admissibility coincide ) and almost structural completeness ( validity and admissibility coincide for quasiequations with unifiable premises ) can be checked . the algorithms are illustrated with a selection of well - known finitely generated quasivarieties , and adapted to handle also admissibility of rules in finite - valued logics .

1305.3530

the problem of checking the _ validity _ of quasiequations in finitely generated ( i.e. , generated by a finite set of finite algebras ) quasivarieties or , similarly , checking consequences from finite sets of formulas in finite - valued logics , is decidable and has been investigated extensively in the literature . in particular , uniform methods for generating proof systems to check validity such as tableaux , resolution , and multisequent calculi , have been developed , as have standard optimization techniques for these systems such as lemma generation and indexing ( see , e.g. , @xcite ) . however , checking the _ admissibility _ of quasiequations in finitely generated quasivarieties , or similarly , checking the admissibility of rules in finite - valued logics , is not so well - understood . the problem is decidable , but a naive approach leads to computationally unfeasible procedures even for small sets of small algebras . the main goal of this paper is to define uniform methods that generate computationally acceptable proof systems for checking admissibility in an arbitrary finitely generated quasivariety . intuitively , a rule is said to be admissible in a logical system if it can be added to the system without producing any new theorems . more formally , a quasiequation is admissible in a class of algebras @xmath1 if every @xmath1-unifier of the premises is a @xmath1-unifier of the conclusion , where a @xmath1-unifier of an equation @xmath2 is a substitution @xmath3 such that @xmath4 is valid in @xmath1 . admissibility plays a fundamental meta - level role in describing `` hidden properties '' of classes of algebras and logical systems . for example , establishing the completeness of a logical system with respect to some restricted class of algebras ( perhaps just one standard algebra ) often involves showing that a certain rule or quasiequation is admissible ; see , e.g. , @xcite for applications of the admissibility of rules in the context of fuzzy logics . also , the closely related problem of deciding unifiability of concepts can be a useful tool for database redundancy checking in description logics @xcite . moreover , it may be possible to automatically obtain admissible rules for classes of algebras and logics that can then be used to simplify reasoning steps or to speed up derivations for checking validity . admissibility ( in tandem with unification ) has been studied intensively in the context of intermediate and transitive modal logics and their algebras @xcite , leading in some cases to proof systems for checking admissibility @xcite . axiomatizations and characterizations have also been obtained for certain families of finite algebras and many - valued logics , in particular ukasiewicz logics ( or classes of mv - algebras ) @xcite and other fuzzy logics @xcite , fragments of the substructural logic r - mingle @xcite , and classes of de morgan algebras @xcite . however , a general theory , covering arbitrary finite algebras and finite - valued logics , has so far been lacking . the starting point for this work is the observation ( see lemma 4.1.10 of @xcite and corollary [ c : admfin ] below ) that for a finite set of finite algebras @xmath1 , admissibility in the quasivariety @xmath5 amounts to validity in the free algebra on @xmath6 generators @xmath7 , where @xmath6 is the maximum cardinality of the algebras in @xmath1 . since by birkhoff s theorem on the structure of free algebras @xcite , this algebra @xmath7 is finite , checking admissibility in @xmath5 is decidable . on the other hand , even for small @xmath6 and a small set of small algebras @xmath1 , the size of @xmath7 may be prohibitively large for checking validity . this is striking since validity and admissibility in @xmath5 may coincide , @xmath5 is then called _ structurally complete _ , or at least coincide for quasiequations with @xmath5-unifiable premises , in which case , @xmath5 is called _ almost structurally complete_. in other cases , @xmath5-admissibility may correspond to validity in other , often quite small , algebras . we provide general algorithms here that discover such algebras , or , more precisely , generate finite sets of finite algebras such that the @xmath5-admissibility of a quasiequation corresponds to validity in the quasivariety generated by these algebras . it is shown , moreover , that these are the smallest sets of algebras with this property with respect to a standard well - ordering on the multiset of their cardinalities . we proceed as follows . first , in section [ s : prelim ] , we recall some basic notions from universal algebra . then in section [ s : fingen ] , we introduce some key ideas and methods for finitely generated quasivarieties ; in particular , we apply a standard multiset well - ordering to the cardinalities of algebras in generating sets for quasivarieties , and provide an algorithm for finding the ( unique up to isomorphism ) minimal generating set of a finitely generated quasivariety . section [ s : admfree ] provides characterizations of admissibility , unifiability , structural completeness , and almost structural completeness . these characterizations are then exploited in section [ s : algorithms ] to define corresponding algorithms , and illustrated using a selection of well - known finite algebras , confirming some known results from the literature , and establishing new ones . in section [ s : logics ] , the approach is extended to finite - valued logics , where the designated values as well as the underlying finite algebra play a significant role . finally , in section [ s : concluding ] , we conclude with some remarks on future directions for this research . let us first recall some basic ideas from universal algebra , referring to @xcite for further details . given an algebraic language @xmath8 ( i.e. , without relation symbols ) , an _ @xmath9 is an algebraic structure consisting of a set @xmath10 ( the universe ) and an @xmath6-ary function @xmath11 corresponding to each @xmath6-ary function symbol @xmath12 of @xmath8 ( as usual , calling nullary functions _ constants _ ) . we call @xmath9 _ finite _ if @xmath10 is a finite set and @xmath8 consists of finitely many function symbols congruence _ on an @xmath8-algebra @xmath9 is an equivalence relation @xmath13 on @xmath10 satisfying for each @xmath6-ary function symbol @xmath12 of @xmath8 : @xmath14 implies @xmath15 . the congruences of @xmath9 form a complete lattice @xmath16 with bottom element @xmath17 and top element @xmath18 , where the meet of a set of congruences on @xmath9 is just the intersection of those congruences . given @xmath19 , the _ quotient algebra of @xmath9 by @xmath13 _ is the @xmath8-algebra @xmath20 with universe @xmath21 consisting of the equivalence classes @xmath22 for @xmath23 with functions defined for each @xmath6-ary function symbol @xmath12 of @xmath8 by @xmath24 . term algebras @xmath25 are defined over a set of variables @xmath26 in the usual way , writing just @xmath27 when @xmath26 is a fixed countably infinite set , and letting @xmath28 stand for arbitrary members of the universe @xmath29 called _ @xmath8-terms_. an _ @xmath8-equation _ is an ordered pair of @xmath8-terms , written @xmath2 . @xmath8-clause _ is defined as an ordered pair @xmath30 of finite sets of @xmath8-equations , written @xmath31 , and called an _ @xmath8-quasiequation _ if @xmath32 and an _ @xmath8-negative clause _ if @xmath33 . as usual , if the language is clear from the context , we may omit the prefix @xmath8 . let us fix @xmath1 to be a class of @xmath8-algebras , noting that often in what follows @xmath1 will consist of a finite set of @xmath8-algebras @xmath34 , and in this case we typically omit brackets . given a finite set of @xmath8-equations @xmath35 , we write @xmath36 and say that the @xmath8-clause @xmath31 is _ @xmath1-valid _ , if for every @xmath37 and homomorphism @xmath38 , @xmath39 recalling that @xmath40 . we also say that @xmath41 is _ @xmath1-satisfiable _ if @xmath42 for some @xmath37 and homomorphism @xmath38 . the class @xmath1 is said to be _ axiomatized _ by a set of @xmath8-clauses @xmath43 if @xmath1 is the class of @xmath8-algebras @xmath9 such that all @xmath8-clauses in @xmath43 are @xmath9-valid . @xmath1 is called a _ variety _ , _ quasivariety _ , or _ antivariety _ if it is axiomatized by a set of @xmath8-equations , @xmath8-quasiequations , or @xmath8-negative clauses , respectively . the variety @xmath44 , quasivariety @xmath5 , and antivariety @xmath45 _ generated by _ @xmath1 are the smallest variety , quasivariety , and antivariety containing @xmath1 , respectively . moreover , let @xmath46 , @xmath47 , @xmath48 , @xmath49 , @xmath50 , @xmath51 , and @xmath52 be the class operators of taking homomorphic images , isomorphic images , subalgebras , products , ultraproducts , non - empty ultraproducts , and homomorphic preimages , respectively . then @xmath53 , @xmath54 , and @xmath55 , and if @xmath1 is a finite set of finite algebras , these last two equivalences refine to @xmath56 and @xmath57 ( see @xcite for details ) . given a language @xmath8 and a set of variables @xmath26 such that either @xmath58 or @xmath8 contains at least one constant symbol , the term algebra @xmath25 exists and admits a congruence : @xmath59 following @xcite , we let @xmath60 and define the _ free algebra of @xmath1 over @xmath61 _ : @xmath62 then @xmath63 has the universal mapping property for @xmath1 over @xmath61 : namely , for each @xmath37 , any map from @xmath61 to @xmath10 extends to a homomorphism from @xmath63 to @xmath9 ( ( * ? ? ? * theorem ii.10.10 ) ) . if also @xmath64 , then @xmath65 ( ( * ? ? * theorem ii.10.12 ) ) . note that @xmath66 whenever @xmath67 ( also @xmath68 if @xmath1 contains at least one non - trivial algebra ) . hence we may consider for each cardinal @xmath69 , the ( unique up to isomorphism ) _ free algebra of @xmath1 on @xmath69 generators _ @xmath70 , where @xmath71 is a subalgebra of @xmath72 for cardinals @xmath73 . we note also that @xmath74 and that @xmath75 implies @xmath76 ( see ( * ? ? ? * corollary ii.11.10 ; exercise ii.11.2 ) ) . a quasivariety @xmath0 is said to be _ finitely generated _ if @xmath77 for some finite ( generating ) set @xmath1 of finite @xmath8-algebras . our first goal in this section will be to define a reasonable measure for comparing these generating sets ; we will then describe a method for obtaining a `` smallest '' generating set according to this measure . more precisely , we apply the standard multiset well - ordering defined by dershowitz and manna in @xcite to the multiset of the cardinalities of algebras in @xmath1 . we show that by decomposing finite @xmath8-algebras into their @xmath0-subdirectly irreducible components and appropriately refining the set of algebras obtained , we arrive at a ( unique up to isomorphism ) minimal set of @xmath8-algebras that still generates @xmath0 . recall that a _ multiset _ over a set @xmath78 is an ordered pair @xmath79 where @xmath80 is a function @xmath81 , and is called _ finite _ if @xmath82 is finite . for a well - ordered set @xmath83 , the _ multiset ordering @xmath84 _ on the set @xmath85 of finite multisets over @xmath78 is defined by @xmath86 if @xmath87 implies that for some @xmath88 , @xmath89 and @xmath90 . it then follows that @xmath84 is a well - ordering of @xmath85 ( see @xcite ) . as usual , we write a finite multiset of elements from @xmath78 as @xmath91 $ ] where @xmath92 may include repetitions . a set of finite @xmath8-algebras @xmath93 will be called a _ minimal generating set _ for the quasivariety @xmath94 if for every set of finite @xmath8-algebras @xmath95 : @xmath96 \leq_m [ |b_1| , \dots , |b_k|].\ ] ] there are of course many measures that could be used to compare generating sets . indeed , it may seem unreasonable to view one hundred algebras with three elements as an improvement on a single algebra with four elements . however , the order is optimal in the following sense . in general , checking a quasiequation with @xmath97 variables in a finite algebra @xmath9 requires checking @xmath98 assignments of variables to elements of @xmath9 . but then checking validity in @xmath93 will involve checking fewer assignments of variables than checking validity in @xmath95 if @xmath99 \leq_m [ |b_1| , \dots , |b_k|]$ ] for quasiequations with sufficiently many variables . to obtain minimal generating sets for @xmath5 where @xmath1 is a finite set of finite algebras , we consider representations of the algebras in @xmath1 using smaller algebras of @xmath5 . an algebra @xmath9 is called a _ subdirect product _ of algebras @xmath34 if there exist surjective homomorphisms @xmath100 for @xmath101 such that the induced homomorphism @xmath102 is an embedding . in this case , @xmath80 is called a _ subdirect representation of @xmath9 _ and @xmath34 are called _ subdirect components _ ( for this representation ) of @xmath9 . if @xmath0 is a quasivariety and @xmath103 , then @xmath9 is called a _ @xmath0-subdirect product of @xmath104 _ and @xmath80 is called a _ @xmath0-subdirect embedding_. @xmath9 is called _ @xmath0-subdirectly irreducible _ if for every @xmath0-subdirect embedding @xmath105 , @xmath9 is isomorphic to @xmath106 for some @xmath107 . we will make essential use of the following : [ lem : qdecomp ] let @xmath0 be a quasivariety and @xmath108 . then @xmath9 is a @xmath0-subdirect product of @xmath0-subdirectly irreducible members of @xmath0 . [ lem : irr_emb ] if @xmath77 for a finite set @xmath1 of finite algebras and @xmath9 is a @xmath0-subdirectly irreducible algebra , then @xmath109 . observe now that if @xmath0 is a quasivariety and @xmath108 is a @xmath0-subdirect product of @xmath104 , then each @xmath106 is in @xmath0 by definition , and @xmath9 is isomorphic to a subalgebra of a product of members of @xmath0 . hence we obtain : [ lem : reducegens ] let @xmath1 be a class of @xmath8-algebras and suppose that @xmath110 is obtained from @xmath1 by either ( a ) replacing @xmath37 with @xmath34 where @xmath9 is a @xmath5-subdirect product of @xmath104 , or ( b ) replacing @xmath111 with @xmath112 where @xmath113 . then @xmath114 . in particular , replacing each algebra @xmath9 in a finite set @xmath1 of finite algebras with the @xmath5-subdirectly irreducible algebras in some @xmath5-subdirect representation of @xmath9 , then removing any algebra that embeds into another algebra in the set , produces a minimal generating set for @xmath5 that is unique up to isomorphism . [ t : mingenset ] suppose that @xmath115 where @xmath106 is a finite @xmath0-subdirectly irreducible algebra for @xmath116 and @xmath117 for @xmath118 . then @xmath93 is the unique minimal generating set for @xmath0 up to isomorphism . let @xmath115 where @xmath106 is a finite @xmath0-subdirectly irreducible algebra for @xmath116 and @xmath117 for @xmath118 . suppose for a contradiction that @xmath119 and @xmath120 < _ m [ |a_1| , \dots , |a_n|]$ ] . without loss of generality , we can suppose that @xmath121 is @xmath0-subdirectly irreducible for @xmath122 ; otherwise , by lemmas [ lem : qdecomp ] and [ lem : reducegens ] , @xmath121 can be replaced with the @xmath0-subdirectly irreducible components of a @xmath0-subdirect representation of @xmath121 and we obtain a smaller ( according to @xmath123 ) generating set of algebras for @xmath0 . it follows that there exists a largest @xmath124 such that there are strictly more occurrences of @xmath97 in @xmath125 $ ] than in @xmath120 $ ] , and for each @xmath126 , the number of occurrences of @xmath127 in @xmath125 $ ] and @xmath120 $ ] are equal . each @xmath106 is finite and @xmath0-subdirectly irreducible , and hence by lemma [ lem : irr_emb ] , embeds into some @xmath121 where @xmath128 . if every @xmath106 of size @xmath97 embeds into , and is hence isomorphic to , a @xmath121 of size @xmath97 , then ( by the pigeonhole principle ) there must be two isomorphic algebras in @xmath129 , a contradiction hence , suppose without loss of generality that @xmath130 embeds into @xmath131 with @xmath132 and @xmath133 . but notice now that @xmath131 is also @xmath0-subdirectly irreducible and hence embeds into some @xmath106 with @xmath134 . so @xmath135 , a contradiction . finally , consider any minimal generating set @xmath136 for @xmath0 , and suppose for a contradiction that @xmath137 for some @xmath138 . then by lemma [ lem : irr_emb ] , @xmath139 properly embeds into @xmath140 for some @xmath141 . but also by lemma [ lem : irr_emb ] , @xmath140 embeds into @xmath142 for some @xmath143 . it follows that @xmath139 can be embedded into the strictly larger algebra @xmath142 . but then @xmath136 is not a minimal generating set for @xmath0 , a contradiction . in the remainder of this section , we develop some results for @xmath0-subdirect products and @xmath0-subdirectly irreducible algebras , and present an algorithm for obtaining minimal generating sets for finitely generated quasivarieties . first , let us recall the following theorem of birkhoff , which establishes a useful relationship between subdirect representations of a given algebra @xmath9 and sets of congruences on @xmath9 . [ lem : correspondence ] if @xmath9 is a subdirect product of the family @xmath144 , then there exist for @xmath145 , congruences @xmath146 such that @xmath147 and @xmath148 . conversely , for any family of congruences @xmath149 on @xmath9 , the quotient @xmath150 is a subdirect product of the family @xmath151 . the set of _ @xmath0-congruences on @xmath9 _ is defined as @xmath152 . clearly , the above lemma also holds for @xmath0-subdirect representations of an algebra @xmath9 with respect to @xmath0-congruences . note , moreover , that the number of congruences needed to obtain a subdirect representation of a finite algebra @xmath9 is at most @xmath153 , the maximal number of coatoms of the congruence lattice @xmath16 . [ cor : correspondence ] let @xmath0 be a quasivariety and @xmath108 . 1 . if @xmath9 is a @xmath0-subdirect product of the family @xmath144 , then there exist for @xmath145 , @xmath0-congruences @xmath154 such that @xmath147 and @xmath148 . conversely , for any family of @xmath0-congruences @xmath149 on @xmath9 with @xmath148 , @xmath9 is a @xmath0-subdirect product of the family @xmath151 . @xmath9 is @xmath0-subdirectly irreducible iff the bottom element @xmath155 of @xmath156 is meet - irreducible ( i.e. , if @xmath148 for @xmath157 , then @xmath158 for some @xmath145 ) . the problem of finding the congruence closure for a given equivalence relation on a finite algebra , i.e. , the smallest congruence containing this equivalence , can be solved in polynomial time . this result was used in @xcite to provide a polynomial time algorithm for calculating a subdirect representation of a finite algebra . the problem of finding the @xmath0-congruence closure of an equivalence relation on a finite algebra with respect to a finitely generated quasivariety @xmath0 appears to be much harder , however . instead , we use here the following characterization of @xmath0-subdirectly irreducible algebras as the basis for a procedure that constructs @xmath0-subdirectly irreducible components for a @xmath0-subdirect representation of a given finite algebra without needing to calculate the @xmath0-congruence lattice . [ lem : charqsubdir ] for a finite set @xmath1 of finite algebras and @xmath159 , the following are equivalent : 1 . @xmath9 is @xmath5-subdirectly irreducible . 2 . @xmath160 . for convenience , let @xmath161 ( 1)@xmath162(2 ) we proceed contrapositively . if @xmath163 , then by corollary [ cor : correspondence](a ) , @xmath9 is a @xmath5-subdirect product of algebras in @xmath164 . but also by corollary [ cor : correspondence](b ) , since @xmath165 , @xmath9 is not @xmath5-subdirectly irreducible . ( 2)@xmath162(1 ) again , we proceed contrapositively . if @xmath9 is not @xmath5-subdirectly irreducible , then combining lemma [ lem : qdecomp ] and corollary [ cor : correspondence ] , there exist @xmath166 such that @xmath148 and @xmath9 is a @xmath5-subdirect product of @xmath5-subdirectly irreducible algebras @xmath167 ( @xmath145 ) . but then also by lemma [ lem : irr_emb ] , we have @xmath168 for each @xmath145 . so @xmath169 , and hence @xmath163 . * declare * @xmath170 : set * declare * @xmath171 : list * declare * @xmath9 : algebra * declare * @xmath172 : integer @xmath173 @xmath174 @xmath175 $ ] @xmath176 [ alg : mingenset : con ] @xmath177 embeds into @xmath178 [ alg : mingenset : s1 ] @xmath179 embeds into some @xmath180 \neq { { \ensuremath{{\mathbf{a}}}}}\}$ ] remove @xmath9 from @xmath181 [ alg : mingenset : dela ] @xmath182 [ alg : mingenset : e1 ] remove @xmath9 from @xmath181 [ alg : mingenset : e2 ] we now have all the ingredients necessary to describe an algorithm mingenset ( see figure [ alg : mingenset ] ) that calculates the ( unique up to isomorphism ) minimal generating set for a quasivariety @xmath77 , where @xmath1 is a finite set of finite @xmath8-algebras . by theorem [ t : mingenset ] , it suffices to find a set of @xmath0-subdirectly irreducible algebras that generates @xmath0 , where no member of the set embeds into another member of the set . the algorithm proceeds by considering each @xmath37 in turn . first , the congruence lattice @xmath16 is generated ( line [ alg : mingenset : con ] ) by checking for all equivalence relations if they are congruences . next , the congruences @xmath183 such that @xmath20 embeds into @xmath9 or some other member of @xmath1 are collected in sets @xmath184 and @xmath185 , respectively . if @xmath186 , then @xmath9 is @xmath0-subdirectly irreducible by lemma [ lem : charqsubdir ] , so the algorithm proceeds to the next algebra in @xmath1 . otherwise @xmath187 and by lemma [ lem : charqsubdir ] , @xmath9 is not @xmath0-subdirectly irreducible . in this case , for each @xmath188 , the algebra @xmath20 is added to @xmath1 ( line [ alg : mingenset : addquot ] ) and @xmath9 is removed from @xmath1 ( line [ alg : mingenset : dela ] ) . note that since the cardinalities of the added algebras are strictly smaller than the cardinality of the removed algebra , the new set of algebras is smaller according to the multiset ordering defined in section [ s : fingen ] . hence this procedure is terminating . moreover , the resulting finite set of finite algebras must generate the quasivariety @xmath0 ( by lemma [ lem : reducegens ] ) , contain only @xmath0-subdirectly irreducible algebras , and not contain any algebra that embeds into another member of the set ( lines [ alg : mingenset : e1][alg : mingenset : e2 ] ) . hence by theorem [ t : mingenset ] , we obtain : for a finite set @xmath1 of finite @xmath8-algebras , mingenset(@xmath1 ) returns the ( unique up to isomorphism ) minimal generating set for the quasivariety @xmath5 . we remark that although the algorithm mingenset does not need to calculate the @xmath0-congruence lattice , already calculating the congruence lattice of a finite algebra can take exponential time . moreover , we make frequent use in our algorithm here of checking embeddings , which is in general an np - complete problem ( see @xcite ) . let us fix a class of @xmath8-algebras @xmath1 . a homomorphism @xmath189 is called a _ @xmath1-unifier _ of a set @xmath41 of @xmath8-equations if @xmath190 for every @xmath191 ; in this case , @xmath41 is said to be _ @xmath1-unifiable_. an @xmath8-quasiequation @xmath192 is _ @xmath1-admissible _ if every @xmath1-unifier of @xmath41 is a @xmath1-unifier of @xmath2 . in this section , we present characterizations of @xmath1-unifiability , @xmath1-admissibility , and related properties , emphasizing their close relationship to free algebras . the characterizations obtained will then be used in section [ s : algorithms ] to develop algorithms for checking these properties in the context of finitely generated quasivarieties . let us first take a closer look at @xmath1-unifiability , noting that a finite set @xmath41 of @xmath8-equations is @xmath1-unifiable if and only if the @xmath8-negative clause @xmath193 is not @xmath1-admissible ( equivalently , when @xmath1 contains a non - trivial algebra , if and only if the @xmath8-quasiequation @xmath194 with @xmath195 not occurring in @xmath41 is not @xmath1-admissible ) . we will see that for checking @xmath1-unifiability , the `` optimal '' solution is to check satisfiability in the smallest finite subalgebra ( if such an algebra exists ) of @xmath196 . [ l : unif ] the following are equivalent for any class @xmath110 of @xmath8-algebras : 1 . @xmath41 is @xmath1-unifiable iff @xmath41 is @xmath110-satisfiable . 2 . @xmath197 . recall that @xmath197 is equivalent to the condition that an @xmath8-negative clause @xmath193 is @xmath110-valid iff it is @xmath196-valid . however , @xmath193 is @xmath110-valid iff @xmath41 is not @xmath110-satisfiable and @xmath193 is @xmath196-valid iff @xmath41 is not @xmath196-satisfiable . for the equivalence of ( 1 ) and ( 2 ) , it suffices therefore to show that @xmath41 is @xmath196-satisfiable iff @xmath41 is @xmath1-unifiable . suppose first that @xmath198 satisfies @xmath41 . then any homomorphism @xmath199 defined such that @xmath200 for each variable @xmath201 is a @xmath1-unifier of @xmath41 . conversely , if @xmath202 is a @xmath1-unifier of @xmath41 , then the homomorphism @xmath198 defined by @xmath203 for each variable @xmath201 satisfies @xmath41 . [ p : unif ] let @xmath1 be a class of @xmath8-algebras and @xmath204 . 1 . @xmath41 is @xmath1-unifiable iff @xmath41 is @xmath205-satisfiable . 2 . if @xmath205 is a smallest finite subalgebra of @xmath196 and @xmath110 is a class of @xmath8-algebras such that @xmath41 is @xmath1-unifiable iff @xmath41 is @xmath110-satisfiable , then @xmath206 for each @xmath207 . \(a ) by assumption , @xmath208 , so @xmath209 . but also , since @xmath210 and @xmath211 has the universal mapping property for @xmath44 over countably infinitely many generators , we obtain a homomorphism @xmath212 defined by @xmath213 for every variable @xmath201 for some fixed @xmath214 . hence @xmath215 $ ] is a subalgebra of @xmath205 and @xmath216 . so @xmath217 and the result follows by lemma [ l : unif ] . \(b ) let @xmath205 be a smallest finite subalgebra of @xmath196 and suppose that @xmath110 is a class of @xmath8-algebras such that @xmath41 is @xmath1-unifiable iff @xmath41 is @xmath110-satisfiable . then by lemma [ l : unif ] and part ( a ) , @xmath218 . hence if @xmath219 , then clearly @xmath206 . [ ex : demorganuni ] the variety @xmath220 of de morgan algebras is generated as a quasivariety by the @xmath221-element algebra @xmath222 consisting of a distributive bounded lattice with an involutive negation defined as follows : ( -4,-0.5)(4,3 ) ( 0,0)(1,1 ) ( 1,1)(0,2 ) ( 0,0)(-1,1 ) ( -1,1)(0,2 ) ( 0,0)(-1,1)(1,1)(0,2 ) ( 0,-0.3)@xmath223 ( -1.2,1)@xmath224 ( 1.2,1)@xmath225 ( 0,2.3)@xmath226 ( 1.3,1.2)(1.5,1.4)(1.7,1)(1.5,0.6)(1.3,0.8 ) ( -1.3,1.2)(-1.5,1.4)(-1.7,1)(-1.5,0.6)(-1.3,0.8 ) ( 0,0.2)(0,1.8 ) since there are constants in the language of @xmath227 , the smallest algebra for checking @xmath220-unifiability is the @xmath228-element ground algebra @xmath229 : i.e. , the @xmath228-element boolean algebra . that is , checking unifiability amounts to checking classical satisfiability . e.g. , @xmath230 is not @xmath220-unifiable , since in the @xmath228-element boolean algebra , @xmath231 and @xmath232 . the case of the `` constant - free '' variety @xmath233 of de morgan lattices , generated as a quasivariety by @xmath234 , is not so immediate . however , there is also a smallest @xmath228-element subalgebra of @xmath235 with elements corresponding to @xmath236 and @xmath237 . so checking @xmath233-unifiability amounts again to checking classical satisfiability . let us turn our attention now to admissible quasiequations , recalling the following useful characterizations of admissibility in terms of free algebras and generated subvarieties . [ t : eqadm ] the following are equivalent : 1 . @xmath192 is @xmath1-admissible . @xmath192 is @xmath5-admissible . 3 . @xmath238 . 4 . @xmath239 . [ ex : lattices ] the following quasiequations , expressing meet and join semi - distributivity in the language of lattices with @xmath240 and @xmath241 , are satisfied by all free lattices ( see @xcite ) , and are therefore admissible in the variety of lattices : @xmath242 \{x \lor y { \approx}x \lor z\ } & { \rightarrow } & x \lor y { \approx}x \lor ( y \land z ) . \end{array}\ ] ] similarly , the quasiequations below expressing torsion - freeness in a language of groups with @xmath243 , @xmath244 , and @xmath245 are satisfied by all free groups , and are hence admissible in the variety of groups : @xmath246 given a class @xmath1 of @xmath8-algebras , we will be interested in determining when the @xmath1-admissibility of quasiequations coincides with their @xmath110-validity in another class of @xmath8-algebras @xmath110 . by theorem [ t : eqadm ] , this is the case exactly when @xmath247 . the next result provides a further useful characterization of this situation . [ p : adm ] the following are equivalent : 1 . @xmath192 is @xmath1-admissible iff @xmath248 . 2 . @xmath247 . 3 . @xmath249 and @xmath250 . \(1 ) @xmath251 ( 2 ) follows directly from theorem [ t : eqadm ] . \(2 ) @xmath162 ( 3 ) suppose that @xmath247 . then @xmath249 . moreover , @xmath252 , so @xmath250 . \(3 ) @xmath162 ( 2 ) suppose that @xmath249 and @xmath250 . then clearly @xmath253 . but also @xmath254 . that is , hence @xmath256 and @xmath257 . for some well - behaved classes of algebras , admissibility and validity coincide . more precisely , a class @xmath1 of @xmath8-algebras is said to be _ structurally complete _ if it satisfies the condition : @xmath258 the following characterization then follows almost immediately from theorem [ t : eqadm ] : [ p : bergman ] the following are equivalent : 1 . @xmath1 is structurally complete . 2 . @xmath259 . 3 . @xmath260 and @xmath261 implies @xmath262 . the variety @xmath263 of boolean algebras , generated as a quasivariety by the @xmath228-element algebra @xmath264 , is structurally complete . it suffices to observe that @xmath265 embeds into @xmath266 via a mapping that sends @xmath267 and @xmath268 to their respective equivalence classes , and hence that @xmath269 . [ ex : m5n5 ] a _ modular lattice _ @xmath270 may be characterized as a lattice satisfying the equation @xmath271 . famously , a lattice @xmath270 is non - modular if and only if the lattice @xmath272 ( below ) embeds into @xmath270 ( see ( * ? ? ? * theorem i.3.5 ) ) . but since @xmath272 is non - modular , also @xmath273 ( which must satisfy the same equations ) is non - modular . so @xmath272 embeds into @xmath273 , and @xmath274 is structurally complete . similarly , it is well - known that a lattice @xmath275 is distributive if and only if neither @xmath272 nor @xmath276 ( below ) embeds into @xmath270 ( see ( * ? ? ? * theorem i.3.6 ) ) . since @xmath276 is non - distributive and modular , also @xmath277 is non - distributive and modular . so @xmath276 embeds into @xmath277 , and @xmath278 is structurally complete . ( 0,-2)(5.94,1.5 ) ( 0.86,1.2 ) ( 0.06,0.4 ) ( 0.06,-0.4 ) ( 0.86,-1.2 ) ( 1.66,0.0 ) ( 4.66,0.0 ) ( 4.66,1.2 ) ( 4.66,-1.2 ) ( 5.86,0.0 ) ( 3.46,0.0 ) ( 0.06,0.4)(0.86,1.2 ) ( 1.66,0.0)(0.86,1.2 ) ( 0.06,0.4)(0.06,-0.4 ) ( 0.06,-0.4)(0.86,-1.2 ) ( 0.86,-1.2)(1.66,0.0 ) ( 3.46,0.0)(4.66,-1.2 ) ( 4.66,-1.2)(5.86,0.0 ) ( 5.86,0.0)(4.66,1.2 ) ( 4.66,1.2)(3.46,0.0 ) ( 4.66,-1.2)(4.66,0.0 ) ( 4.66,0.0)(4.66,1.2 ) ( 0.9,-1.6)@xmath272 ( 4.7,-1.6)@xmath276 for certain other classes , admissibility and validity coincide for quasiequations with unifiable premises . more precisely , we call a class @xmath1 of @xmath8-algebras _ almost structurally complete _ if it satisfies the condition : @xmath279 the next result provides a useful characterization of this situation : [ thm : asc ] the following are equivalent for any @xmath280 : 1 . @xmath1 is almost structurally complete . 2 . @xmath281 . 3 . @xmath282 . \(1 ) @xmath162 ( 2 ) suppose that @xmath1 is almost structurally complete . to establish @xmath281 , it suffices to show that a quasiequation @xmath192 is valid in all algebras @xmath283 for @xmath37 iff it is valid in @xmath196 . suppose first that @xmath284 . then by theorem [ t : eqadm ] , either @xmath41 is not @xmath1-unifiable or @xmath192 is @xmath1-valid . in the first case , by proposition [ p : unif ] , @xmath41 is not @xmath112-satisfiable , so @xmath192 is valid in @xmath283 for all @xmath37 . in the second case , @xmath192 is valid in @xmath285 for all @xmath37 . conversely , if @xmath192 is valid in @xmath283 for each @xmath37 , then either @xmath41 is not @xmath112-satisfiable or @xmath192 is valid in each @xmath9 in @xmath1 . in the first case , by proposition [ p : unif ] , @xmath41 is not @xmath1-unifiable , so @xmath192 is valid in @xmath196 . in the second case , @xmath192 is valid in @xmath5 and hence valid in @xmath196 . \(2 ) @xmath162 ( 1 ) suppose that @xmath286 . then whenever @xmath192 is @xmath1-admissible , it is @xmath196-valid and hence also valid in @xmath283 for all @xmath37 . if @xmath41 is @xmath1-unifiable , then , by proposition [ p : unif ] , it is @xmath112-satisfiable . i.e. , there exists a homomorphism @xmath287 with @xmath288 . for any @xmath37 and homomorphism @xmath289 with @xmath290 , define @xmath291 by @xmath292 . then , since @xmath192 is valid in @xmath283 for all @xmath37 , @xmath293 , so @xmath294 and @xmath295 . i.e. , @xmath296 . so we have shown that @xmath297 . \(2 ) @xmath162 ( 3 ) immediate . \(3 ) @xmath162 ( 2 ) suppose that @xmath282 . then also , since @xmath298 for each @xmath37 , we obtain @xmath299 . hence by proposition [ p : adm ] , @xmath281 . [ ex : luk ] consider the @xmath228-element and @xmath300-element wajsberg algebras @xmath301 and @xmath302 where @xmath303 the algebra @xmath304 embeds into @xmath305 , as illustrated in the diagram below by the terms associated to elements , and has @xmath306 as a homomorphic image , as indicated by the arrows . hence @xmath306 is almost structurally complete . note that it is not structurally complete since , for example , @xmath307 is @xmath306-admissible , but not @xmath306-valid . on the other hand , its implicational reduct @xmath308 is structurally complete , since it embeds into the free algebra @xmath309 . + ( -5,-1)(10,2.5 ) ( 0,0)(3,1)(0,3)(3,4)(0,6)(3,7)(21,0.5)(21,3.5)(21,6.5 ) ( -1,0)@xmath310 ( -1,3)@xmath311 ( -1,6)@xmath312 ( 4,1)@xmath313 ( 4,4)@xmath314 ( 4,7)@xmath315 ( 22,0.5)@xmath316 $ ] ( 22,3.5)@xmath317 $ ] ( 22,6.5)@xmath318 $ ] ( 0,0)(3,1 ) ( 0,3)(3,4 ) ( 0,6)(3,7 ) ( 0,0)(0,6 ) ( 3,1)(3,7 ) ( 21,0.5)(21,6.5 ) ( 13,0.5)(18,0.5 ) ( 13,3.5)(18,3.5 ) ( 13,6.5)(18,6.5 ) ( -7,-1)(11,-1 ) ( -7,2)(11,2 ) ( -7,5)(11,5 ) ( -7,8)(11,8 ) ( -7,-1)(-7,8 ) ( 11,-1)(11,8 ) ( -7,-2.5)@xmath319 [ s : algorithms ] in this section , we present algorithms for checking admissibility of quasiequations and related properties in finitely generated quasivarieties . we also present results on admissibility ( some new , some well - known ) for a range of examples taken from the universal algebra and non - classical logic literature . these results , obtained using a delphi implementation of the algorithms described here , are collected at the end of this section in table [ table ] . we begin by recalling birkhoff s result that the finitely generated free algebras for a finite set @xmath1 of finite algebras are themselves finite ; more precisely : [ l : finite ] for any set of finite algebras @xmath320 and @xmath321 : @xmath322 hence for the relatively easy task of checking @xmath1-unifiability , it suffices to find ( e.g. , through exhaustive search ) the ( unique up to isomorphism ) smallest subalgebra @xmath205 of the finite free algebra @xmath323 , noting that this is @xmath324 if the language contains constants . it then follows by proposition [ p : unif ] that a set of equations @xmath41 is @xmath1-unifiable if and only if @xmath41 is @xmath205-valid , and indeed that there is no smaller algebra with this property . for checking @xmath1-admissibility , we make use of a known result for finitely generated quasivarieties ( see ( * ? ? ? * lemma 4.1.10 ) ) , obtained here as a corollary of proposition [ p : adm ] : [ c : admfin ] given a finite set @xmath1 of finite @xmath8-algebras with @xmath325 : 1 . @xmath192 is @xmath1-admissible iff @xmath327 . observe first that each @xmath37 is a homomorphic image of @xmath7 . that is , define any surjective map from the @xmath6 generators of @xmath7 to @xmath10 ; this extends to a homomorphism from @xmath7 onto @xmath9 since @xmath7 has the universal mapping property for @xmath1 over @xmath6 generators . so @xmath328 and , since also @xmath329 , ( a ) and ( b ) follow by proposition [ p : adm ] . hence checking @xmath1-admissibility of quasiequations is decidable . however , even when @xmath1 consists of a small number of small algebras , free algebras on a small number of generators can be quite large . for example , the free algebra @xmath330 ( see example [ ex : demorganuni ] ) has @xmath331 elements . we therefore seek smaller algebras or finite sets of smaller algebras that also generate @xmath332 as a quasivariety . in fact , since @xmath333 is finitely generated , we may apply the multiset ordering defined in section [ s : fingen ] , and seek a minimal generating set of finite algebras for this quasivariety that is unique up to isomorphism . one strategy would therefore be to apply the algorithm mingenset directly to @xmath334 . however , this method is not feasible for large free algebras , since it involves the computationally labour - intensive task of building the congruence lattice of @xmath334 . instead , we make auxiliary use of the following immediate corollary of proposition [ p : adm ] : [ c : adm ] given a class @xmath1 of @xmath8-algebras and @xmath335 such that @xmath336 : 1 . @xmath192 is @xmath1-admissible iff @xmath248 . * declare * @xmath337 : set * declare * @xmath338 : algebra @xmath339 @xmath340 @xmath341 @xmath342 [ alg : subprehom ] @xmath343 @xmath342 add @xmath112 to @xmath344 we describe an algorithm admalgs ( see figure [ alg : admalgs ] ) which takes as input a finite set @xmath1 of finite @xmath8-algebras and outputs the ( unique up to isomorphism ) minimal generating set for @xmath332 . first , the procedure mingenset is applied to @xmath1 ( which typically would be a small set of small algebras ) to obtain the set of @xmath8-algebras @xmath345 . then for each @xmath346 , a procedure free(@xmath347 ) is invoked to produce the smallest free algebra @xmath348 having @xmath9 as a homomorphic image . ( the procedure begins by checking the smallest free algebra @xmath349 or @xmath350 , then increases the number of generators one at a time . ) the algorithm then searches for progressively smaller subalgebras of @xmath348 which have @xmath9 as a homomorphic image . more precisely , the procedure @xmath351 searches for a proper subalgebra of @xmath112 that is a homomorphic image of @xmath9 , returning @xmath112 if no such algebra exists ( line [ alg : subprehom ] ) . this process terminates with a ( hopefully reasonably small ) algebra which is added to a set @xmath352 . finally , the procedure mingenset is applied to @xmath352 . for a finite set @xmath1 of finite @xmath8-algebras , admalgs(@xmath1 ) returns the ( unique up to isomorphism ) minimal generating set for the quasivariety @xmath333 . [ ex : no_embedding ] note that given even just one algebra @xmath9 , it might not be the case that the smallest subalgebra of @xmath353 having @xmath9 as a homomorphic image is the smallest algebra that generates @xmath354 . consider the @xmath221-element algebra @xmath355 where the unary function @xmath12 and the free algebras @xmath356 are described by the following diagrams : + ( -1.5,0)(3,3 ) ( -4,7)@xmath224 ( -7.2,7)@xmath357 ( -4.2,6.5)(-4.2,4.5 ) ( -3.8,6.5)(-3.8,4.5 ) ( -4,4)@xmath225 ( -6.5,1.5)(-4.5,3.5 ) ( -7,1)@xmath358 ( -1.5,1.5)(-3.5,3.5 ) ( -1,1)@xmath359 ( 1.5,7)@xmath356 ( 2,5)@xmath360 ( 3,5)(4.5,5 ) ( 6,5)@xmath361 ( 7.5,5.2)(9,5.2 ) ( 7.5,4.8)(9,4.8 ) ( 9.5,5)@xmath362 ( 6,2.5)(6,3)(6,3.5 ) ( 2,1)@xmath363 ( 3,1)(4.5,1 ) ( 6,1)@xmath364 ( 7.5,1.2)(9,1.2 ) ( 7.5,0.8)(9,0.8 ) ( 9.5,1)@xmath365 the smallest subalgebra of @xmath366 with @xmath357 as a homomorphic image is the @xmath367-element free algebra @xmath368 . however , @xmath369 , performed at the beginning of the algorithm admalgs , discovers that @xmath357 is a @xmath370-subdirect product of ( two copies of ) the @xmath300-element free algebra @xmath371 . hence @xmath372 and this quasivariety is structurally complete . in the previous example , the fact that the algebra @xmath357 is structurally complete is discovered by the algorithm admalgs . however , structural completeness can also be checked more directly using the following result : [ prop : charsc ] the following are equivalent for any finite set @xmath1 of finite @xmath8-algebras : 1 . @xmath1 is structurally complete . 2 . @xmath373 where @xmath374 . \(1 ) @xmath162 ( 2 ) if @xmath1 is structurally complete , then , by proposition [ p : bergman ] and corollary [ c : admfin ] , @xmath375 where @xmath374 . so @xmath376 . but each @xmath377 is @xmath378-subdirectly irreducible , so by lemma [ lem : irr_emb ] , @xmath9 embeds into @xmath7 . i.e. , @xmath373 . \(2 ) @xmath162 ( 1 ) if each @xmath377 embeds into @xmath7 , then @xmath379 . so @xmath1 is structurally complete by proposition [ p : bergman ] . [ ex : pseudocomp ] proposition [ prop : charsc ] has been used to confirm known structural completeness results for the @xmath300-element positive gdel algebra @xmath380 where @xmath381 is @xmath382 if @xmath383 , otherwise @xmath268 , and the stone algebra @xmath384 where @xmath385 . moreover , a new structural completeness result has been established for the pseudocomplemented distributive lattice @xmath386 , obtained by adding a top element to the @xmath221-element boolean lattice , and calculating the pseudocomplement . similarly , we can check almost structural completeness using the following result : [ prop : almost ] the following are equivalent for any finite set @xmath1 of finite @xmath8-algebras and @xmath280 : 1 . @xmath1 is almost structurally complete . 2 . @xmath387 where @xmath374 . \(1 ) @xmath162 ( 2 ) if @xmath1 is almost structurally complete , then by theorem [ thm : asc ] and corollary [ c : admfin ] , @xmath388 where @xmath389 . in particular , @xmath390 . but each @xmath391 is @xmath378-subdirectly irreducible , so by lemma [ lem : irr_emb ] , @xmath205 embeds into @xmath7 . i.e. , @xmath387 . \(2 ) @xmath162 ( 1 ) if @xmath387 , then @xmath392 . so by theorem [ thm : asc ] , @xmath1 is almost structurally complete . [ ex : sugimono3 ] the sobociski algebra @xmath393 with operations @xmath394 is a weak characteristic matrix for the multiplicative fragment @xmath395 of the substructural logic r - mingle : @xmath314 is a theorem of @xmath395 if and only if @xmath396 . using proposition [ prop : almost ] , we establish that @xmath397 is almost structurally complete and that @xmath398 where @xmath399 . note , moreover , that the implicational reduct @xmath400 of @xmath397 is structurally complete , which can be confirmed directly using proposition [ prop : charsc ] or discovered automatically by admalgs . [ ex : sugimono4 ] consider now the algebra @xmath401 ( the reduct of a @xmath221-element algebra for r - mingle ) with a constant @xmath268 and operations @xmath402 admalgs produces a @xmath367-element algebra ( the product of a @xmath228-element algebra and a @xmath300-element algebra ) . remarkably , if we drop @xmath403 to obtain @xmath404 , admalgs produces a @xmath221-element algebra that is not isomorphic to @xmath405 . [ ex : demorgankleene ] recall from example [ ex : demorganuni ] that the algebras @xmath406 and @xmath227 generate the varieties of de morgan latices and de morgan algebras , respectively , as quasivarieties . both @xmath406 and @xmath227 are homomorphic images of the corresponding free algebras on two generators ( with @xmath407 and @xmath331 elements , respectively ) but not on one generator . in the de morgan lattice case , admalgs finds a smallest suitable subalgebra isomorphic to @xmath408 , while for de morgan algebras , the smallest suitable algebra is isomorphic to @xmath409 with additional top and bottom elements . these results were established `` by hand '' in @xcite ; our procedure here confirms them automatically . similar results were also obtained in @xcite for kleene lattices and kleene algebras ( subvarieties of de morgan lattices and de morgan algebras , respectively ) which are generated as quasivarieties by the @xmath300-element chains @xmath410 and @xmath411 where @xmath403 swaps @xmath223 and @xmath226 and leaves @xmath412 fixed . in both cases , the smallest algebra for @xmath413 and @xmath414 , found automatically by admalgs is a @xmath221-element chain . [ e : twoalgebras ] consider the @xmath228-element and @xmath300-element chains , @xmath415 and @xmath416 where @xmath403 swaps @xmath223 and @xmath226 and leaves @xmath412 fixed . individually , these algebras are structurally complete . however , applying admalgs to @xmath417 , we find that @xmath1 is not structurally complete : both @xmath418 and @xmath419 are homomorphic images of the @xmath420-element free algebra @xmath421 , and the minimal generating set for @xmath333 consists of a single @xmath221-element algebra . we remark finally that except for example [ e : twoalgebras ] , all the case studies considered in this section ( see table [ table ] ) involve quasivarieties generated by a single algebra with five elements or fewer , and produce minimal generating sets for checking admissibility also containing just one algebra . indeed in many of the cases considered , the quasivariety is shown to be either structurally complete or almost structurally complete . the main obstacle at present to extending our experimental results to ( larger sets of ) larger algebras is the size of the resulting free algebras ( in the worst case , size @xmath422 for @xmath6 generators and an algebra of size @xmath423 ) , both for computing these algebras and then for generating suitable subalgebras . nevertheless , we hope in future work to obtain procedures capable of treating larger examples by using heuristics to constrain the search for appropriate generating algebras . .algebras for checking admissibility [ cols="^,^,<,<,^",options="header " , ] [ s : logics ] the preceding characterizations of admissibility can be adapted relatively straightforwardly to finite - valued logics . however , in this setting , the designated values of the logic the `` true truth values '' must also be considered , and dealt with appropriately . here we describe a method that given a finite - valued logic @xmath275 , provides another ( hopefully small ) finite - valued logic @xmath424 such that validity in @xmath424 corresponds to admissibility in @xmath275 . the more general case , where a smallest finite set of logics is sought such that validity in all members of the set corresponds to admissibility in a logic ( or logics ) , is left as an exercise for the interested reader . recall that a _ finite - valued logic _ @xmath425 for a language @xmath8 consists of a finite @xmath8-algebra @xmath9 and a set of _ designated values _ @xmath426 . given @xmath427 , we let @xmath428 denote that for all homomorphisms @xmath38 , whenever @xmath429 \subseteq d$ ] , also @xmath430 . a term @xmath314 is _ @xmath275-valid _ if @xmath431 . we call a logic @xmath432 for a language @xmath8 a _ sublogic _ of a logic @xmath433 if @xmath130 is a subalgebra of @xmath434 and also @xmath435 . consider now a finite - valued logic @xmath425 for a language @xmath8 and a finite set of terms @xmath427 . we say that @xmath436 is _ @xmath275-unifiable _ if there exists a homomorphism @xmath437 such that @xmath438 for all @xmath439 and call @xmath3 in this case an _ @xmath275-unifier _ of @xmath436 . a rule @xmath440 is said to be _ @xmath275-admissible _ if every @xmath275-unifier of @xmath436 is an @xmath275-unifier of @xmath314 . now if we define the finite - valued logic @xmath441 where @xmath442 \in { { \ensuremath{{\mathbf{f}}}}}_{{\ensuremath{{\mathbf{a}}}}}(|a| ) : \ { \vdash_{{\mathrm{l } } } } { \ensuremath{\varphi}}\}$ ] , then we easily obtain the following analogue of theorem [ t : eqadm ] . [ l : finval ] let @xmath425 be a finite - valued logic for a language @xmath8 . then @xmath440 is @xmath275-admissible iff @xmath443 . the next result may then be understood as an analogue of proposition [ p : adm ] . let @xmath444 be a finite - valued logic for a language @xmath8 , and let @xmath445 be a sublogic of @xmath446 such that there exists a surjective homomorphism @xmath447 satisfying @xmath448 \subseteq d_a$ ] . then @xmath440 is @xmath275-admissible iff @xmath449 . if @xmath440 is @xmath275-admissible , then by lemma [ l : finval ] , @xmath443 . since @xmath424 is a sublogic of @xmath446 , also @xmath449 . conversely , suppose that @xmath449 and that @xmath3 is an @xmath275-unifier of @xmath436 . notice that if @xmath450 , then @xmath451 and @xmath452 . so @xmath3 is also an @xmath446-unifier and @xmath424-unifier of @xmath436 . but @xmath453 and therefore @xmath454 . now consider any homomorphism @xmath455 . since @xmath456 is a surjective homomorphism from @xmath112 to @xmath9 , there exists a homomorphism @xmath457 such that @xmath458 is the identity map on @xmath9 . but @xmath454 and hence @xmath459 . \subseteq d_a$ ] . so @xmath461 . the @xmath300-valued ukasiewicz logic @xmath462 and jakowski logic @xmath463 may both be presented using the @xmath300-element wajsberg algebra @xmath306 ( example [ ex : luk ] ) but with @xmath268 as designated value for @xmath462 and @xmath464 and @xmath268 as designated values for @xmath463 . that is , @xmath465 and @xmath466 . in this case , there is a smallest subalgebra of @xmath467 isomorphic to @xmath304 with a homomorphism that maps @xmath304 onto @xmath306 and sends the inherited designated values @xmath468 to @xmath268 , and @xmath469 to @xmath464 . we therefore obtain a logic @xmath470 corresponding to admissibility in @xmath462 , and another logic @xmath471 corresponding to admissibility in @xmath463 . the algorithms described in this paper have been implemented in the tool @xmath472 , a delphi xe2 implementation compiled for windows operating systems available for download online at https://sites.google.com/site/admissibility . this tool has been used in particular to obtain minimal generating sets and to check structural completeness and almost structural completeness for all @xmath300-element groupoids ( i.e. , @xmath300-element algebras with a single binary operation ) of which there are @xmath473 up to isomorphism . the cardinalities of the minimal free algebras required to check admissibility range from @xmath300 to @xmath474 , while the minimal generating algebras have at most @xmath475 elements . @xmath476 of the groupoids are structurally complete and a further @xmath477 are almost structurally complete . for further details of @xmath472 and these experiments , consult the system description @xcite . although the algorithms described here produce minimal sets of algebras for testing admissibility ( and of course have to be run only once for a given finite set of finite algebras ) , the steps involved generating a suitable free algebra and its subalgebras and checking homomorphic images may be computationally unfeasible . for this reason , our case studies have consisted so far mostly of algebras with @xmath478 or fewer elements where the appropriate free algebra is of a reasonable size ( less than @xmath479 elements , say ) . this bottleneck could perhaps be addressed by introducing heuristics such as constructing and checking small subalgebras of free algebras incrementally , rather than beginning with the whole free algebra , or calculating only some and not all of the congruences in the @xmath480 algorithm . we might also hope to obtain faster algorithms for cases where the given algebras admit certain properties such as congruence distributivity or permutability . we recall finally that one of the motivations for investigating admissibility in arbitrary finite algebras or , similarly , finite - valued logics , is to obtain admissible quasiequations or rules that can be used to improve proof systems ( shortening derivations , constraining proof search , simplifying rules , etc . ) for these algebras or logics . the work reported here provides a significant step towards this goal . the next , equally ambitious , step will be to determine which rules have the potential to be useful for a given finite algebra or finite - valued logic and to add these appropriately to corresponding proof systems .

the tully - fisher ( tf ) and fundamental plane ( fp ) relations are used to obtain two independent estimates of the relative distance between the clusters a2634 and coma . previously published studies of a2634 showed a large discrepancy between the distance estimates obtained with the tf and the d relations , questioning the reliability of redshift - independent distances obtained using these relations . because of the importance of this issue , we have obtained new distance estimates for a2634 , based on much larger samples than previously used , and selected according to rigorous membership criteria . new i band ccd photometry for 175 galaxies , new 21 cm observations of 11 galaxies , and new velocity dispersion measurements for 62 galaxies are used together with previously published data in building these samples . as part of a larger project to compare the tf and fp distance - scales , we have obtained a new fp template using for the first time i band photometry . the template is derived using a sample of 109 e and s0 galaxies that are members of the coma cluster . its parameters are in very good agreement with recent determinations of the fp obtained at shorter wavelengths . the uncertainty with which the fp can provide peculiar velocity estimates for single galaxies is.43 mag in the distance modulus , or 20% of the distance . this uncertainty is slightly larger than the typical uncertainty that characterizes tf estimates . however this disadvantage is partly compensated by the fact that the sample incompleteness bias has a less severe effect on fp cluster distance estimates than it has on the corresponding tf distance estimates . also , cluster membership is more readily established for early type objects than for spirals . after the appropriate corrections for sample incompleteness have been taken into account , we find the tf and fp distance estimates to be in good agreement , both indicating that a2634 has a negligibly small peculiar velocity with respect to the cosmic microwave background reference frame . because of the high accuracy with which the two distance estimates have been obtained , their agreement strongly supports the universality of the tf and fp relations , and therefore their reliability for the estimate of redshift - independent distances .

astro-ph9610082

the measurement of redshift - independent distances , together with that of redshift measurements , constitutes one of the basic tools of observational cosmology . while redshift surveys alone can be used to obtain maps of the distribution of luminous matter in the local universe ( see giovanelli & haynes 1991 for a recent review ) , only redshift - independent distances can dynamically constrain the total matter content ( i.e. dark plus luminous ) of the universe . in the absence of perfect `` standard candles '' , i.e. objects of well determined and invariant intrinsic properties , many different methods have been developed to climb the `` distance scale ladder '' out to cosmologically significant distances ( see for example rowan - robinson 1985 ) . the ideal method would provide high accuracy , i.e. it would exhibit small dispersion in the predicted distance , it would be applicable over a large range of distances and for a well distributed set of targets , and it would have a well understood physical base . so far no technique has been developed that satisfies all of the above requirements . among the available methods ( see jacoby _ et al . _ 1992 for a review ) , those most extensively utilized are the tully - fisher ( tf ) relation ( tully & fisher 1977 ) , which is applied to spiral galaxies , and the d@xmath0 or fundamental plane ( fp ) relation ( dressler _ et al . _ 1987 ; djorgovski & davis 1987 ) , which is applied to elliptical and lenticular galaxies . tf , fp and d@xmath0 are very similar methods , all based on the correlation between a photometric distance - dependent parameter and a spectroscopic , distance - independent quantity . the three relations have fair accuracy ( the uncertainty in the distance prediction of a single galaxy is 12 25% ) , and they can be applied over a large distance interval ( from the local group out to galaxies with redshift greater than 0.1 ; see for example franx 1993 ; vogt _ et al . _ 1993 , et al . _ 1996 , van dokkum & franx 1996 ) . although they can be generically understood in terms of virial equilibrium , the astrophysical details of these relations are obfuscated by the variance in the formation and evolution histories of galaxies . the calibration of the relations is thus a purely empirical exercise . also , because of the uncertainty on the value of the hubble constant , and because the number of nearby galaxies that could be used for an absolute calibration of the template relations is very small , these relations are most commonly used to derive only relative distances . in the following , the term distance is therefore used as an abbreviation for the more appropriate `` recession velocity corrected for peculiar motions '' . it is well known that in the absence of a perfect standard candle , distance measurements are subject to bias , arising from a variety of sources ( see teerikorpi 1984,1990,1993 ; sandage 1994a , b ; willick 1994 ; giovanelli _ et al . _ 1996b ; and references therein ) . bias corrections thus play an important role in the measurement of peculiar velocities . it is also possible that the environment might have an effect . for example , the template tf , d@xmath0 , or fp relation might depend on the local density of galaxies , and thus differ among rich and poor clusters , and between clusters and the field . yet again , the intrinsic scatter about the mean relation might be coupled with the space distribution of galaxies , complicating the bias correction procedures . it is then legitimate to ask whether the tf , d@xmath0 and fp are universal relations . some attempts have been made toward quantifying this issue ( aaronson _ et al . _ 1986 ; djorgovski , de carvalho & han 1988 ; guzmn _ et al . _ 1992 ; jrgensen , franx & kjrgaard 1996 ) , producing however conclusions of marginal clarity . an invalid or biased template relation will yield spurious peculiar velocities . it has even been argued , for example , that the very evidence suggesting the existence of a great attractor ( dressler _ et al . _ 1987 , lynden - bell _ et al . _ 1988 ) might result from bias ( silk 1989 ) . a comparison by mould _ et al . _ ( 1990 ) between cluster peculiar velocities obtained by respectively using the tf and d@xmath0 techniques shows disagreement which exceeds that expected from the reported uncertainties associated with the two methods . the case of a2634 , pointed out by lucey _ et al . _ ( 1991a ; hereafter l91a ) , is even more notorious : aaronson _ et al . _ ( 1986 , hereafter a86 ) applied the tf relation to find the cluster roughly at rest in the cosmic microwave background ( cmb ) reference frame , while l91a used the d@xmath0 relation to obtain a cluster peculiar velocity of 3400 @xmath2 . the difference between the two estimates is significantly larger than the combined uncertainty of the two separate results ( approximately a 5@xmath3 discrepancy ) , leading l91a to conclude that environmental effects in this cd dominated cluster might be responsible for systematic bias in the properties of the early - type galaxy population . in such case , _ d@xmath0 and fp would not be universal relations , and they could not be used to reliably map the peculiar velocity field_. other possible explanations for the discrepancy in a2634 should be considered . the a86 and l91a samples have very different spatial distributions , prompting the doubt that early- and late - type galaxies may not sample the same dynamical unit . the l91a sample is composed of 18 e and s0 galaxies in the core of a2634 ( all galaxies are at a radial distance of @xmath4 deg from the cluster center ) , while the a86 sample is composed of 11 spiral galaxies scattered over a large area , several degrees to the side , that includes not only a2634 but also the cluster a2666 . et al . _ ( 1995 , hereafter s95 ) have shown that only 7 galaxies in the a86 sample are proper members of a2634 , while the remaining 4 are foreground objects . in addition , both the a86 and l91a studies failed to apply bias corrections . because of the important concerns raised by the previously mentioned results , it appeared desirable to carry out a new detailed and multi pronged study of a2634 , which would satisfactorily answer the issues of sample adequacy and template relation universality . in a previous paper ( s95 ) we presented the results of the first part of our study of a2634 , which aimed at easing sample selection concerns . there , an extensive redshift survey provided a detailed kinematical and dynamical analysis of the a2634 region . the ensuing ability to characterize cluster membership accurately ( a ) suggested that the loose sample selection criteria of a86 did not seriously affect their inferred peculiar velocity , and ( b ) was used to select two samples , one of e + s0 and one of spiral galaxies , that are used here to derive new peculiar velocity estimates of a2634 . for the spiral sample , the motion of a2634 is referred to the template relation recently obtained by giovanelli _ _ ( 1996b ; hereafter g96b ) , based on 24 clusters . since such a template is not yet available for early - type galaxies , we refer the motion of a2634 inferred from the fp relation to a template obtained using the coma cluster . for that purpose , in both a2634 and coma we have obtained new photometric observations , and new velocity dispersions to supplement the data already available in the literature . in a forthcoming work ( scodeggio , giovanelli & haynes 1996 ) we will present a more accurate derivation of the fp template , based on a multi - cluster sample , and extend the comparison of the tf and fp distance - scales to a larger sample of nearby clusters . the remainder of this paper is organized as follows . in 2 we discuss cluster membership . in 3 we present our new spectroscopic and photometric observations . in 4 we briefly discuss the incompleteness bias correction . in 5 we present the new tf and fp distance estimates and in 6 we summarize the main results of this work . throughout the paper we parameterize distance dependent quantities via h@xmath5 km s@xmath6 mpc@xmath6 . reliable assignment of cluster membership is an important criterion in the definition of a tf or fp sample for a cluster of galaxies . because of the sparse character of the spiral population in clusters , tf samples often fail strict membership criteria . poor knowledge of the cluster structure or the need to restrict to the brightest objects , has led observers to draw targets from wide regions around cluster centers in order to maximize target counts . in this study we aim to overcome such limitations . the properties of coma and a2634 are well known and we summarize them in table 1 . thanks to the availability of a large body of galaxy redshifts , detailed kinematical and dynamical studies of both clusters are available . the coma cluster was first studied in detail by kent & gunn ( 1982 ) , and redshift space caustic curves that mark the separation between the gravitationally bound cluster members and the unbound `` field '' have been published , most recently by regs & geller ( 1989 ) and gavazzi , randone & branchini ( 1995 ) . the cluster is well separated in redshift space from both the foreground and the background field , and membership can be reliably assigned to objects to an angular separation of approximately 5 degrees from the cluster center . following the criterion used by giovanelli _ et al . _ ( 1996a ; hereafter g96a ) , we define as cluster members all galaxies within 4 degrees from the cluster center that fall within the radial velocity limits defined by the regs & geller ( 1989 ) caustic curves estimated for @xmath7 . the substructure within the coma cluster recently discussed by colless & dunn ( 1996 ) appears to be insignificant for our discussion . the difference in distance between the merging units , as inferred by these authors , is a negligible fraction of the distance to the coma cluster , and would produce , at most , only a small increase in the scatter of of the tf and fp diagrams for this cluster . the first in depth redshift survey of a2634 was carried out by pinkney _ et al . _ ( 1993 ) . a more detailed study of the cluster structure was carried out by s95 , which also estimated redshift space caustic lines . a2634 is located in a region of higher complexity than that containing coma . the cluster a2666 ( cz@xmath8 = 8,134 @xmath2 ) lies @xmath9 to the east of a2634 ( cz@xmath8 = 9,240 @xmath2 ) ; two more groups , one in the foreground ( cz@xmath8 = 7,546 @xmath2 ) and one in the background ( cz@xmath8 = 11,619 @xmath2 ) , were identified by s95 , projected respectively 1.1 and 0.6 degrees to the ne and to the se of a2634 . in addition , two background clusters , a2622 at cz@xmath8 = 18,345 @xmath2 , and an anonymous cluster at cz@xmath8 = 37,093 @xmath2 , are projected to within less than 1 degree from the center of a2634 . we define as a2634 members all galaxies within 2.5 degrees from the cluster center that fall within the radial velocity limits defined by the s95 caustic curves corresponding to @xmath7 . in addition to the strict cluster members , we include in our samples also a number of galaxies that we classify as `` peripheral members '' . these are galaxies with radial velocity very close to the cluster systemic velocity , but sufficiently removed from the cluster center that they do not satisfy a strict membership criterion . in the following paragraph , the term cluster members will be used to indicate both the strict cluster members and the peripheral ones . the selection of the coma tf sample is described in g96a . the sample includes 41 galaxies , of morphological type between sb and sd , and with inclination greater than 45 , that satisfy the membership criteria discussed in the previous paragraph . the a2634 tf and fp samples are based on the catalog of the redshift survey by s95 . the tf sample was selected following the same criteria as in the coma case , except that it includes two galaxies with inclination lower than 45 , and one sa galaxy . two very late spiral or irregular galaxies with serendipitously measured widths are also included in this sample , for a total of 27 galaxies . both the coma and the a2634 fp samples were selected to satisfy membership criteria . the coma sample is composed of 109 galaxies , 71 ellipticals and 38 lenticulars , while the a2634 one is composed of 55 galaxies , 22 ellipticals and 33 lenticulars . three galaxies were found to show some indication of spiral structure , and have been tentatively classified as sa in tables 3 and 4 . in addition to the data presented in g96a , measurements for 11 spiral galaxies in the a2634 tf sample are included here . velocity widths were obtained from 21 cm observations carried out between 1990 and 1994 with the arecibo 305 m radio telescope . the observational setup was as described in giovanelli & haynes ( 1989 ) . since this sample includes galaxies significantly fainter than those customarily observed in @xmath10 emission , integration times averaged 0.7 hours per object on source . the typical rms noise per channel ranged from 0.3 to @xmath11 . all observations were taken with a spectral resolution of approximately 8 @xmath2 , later reduced by smoothing by an amount dependent on the signal - to - noise ratio . the velocity width was measured using an algorithm that fits the rising sides of the galaxian profile with low order polynomials , and obtains the width at a 50% level of each of the profile horn peaks or single peak , depending on the line shape . we refer the reader to the discussion in g96a for further details . photometric parameters for 175 galaxies were derived from i band ccd images obtained with the 0.9 m telescope of the kitt peak national observatory ( kpno ) , during 3 observing runs between april 1994 and september 1995 , plus some service observing by kpno personnel in august 1993 ( a2634 only ) . the telescope was used with the f/7.5 secondary , field corrector and t2ka ccd chip ( 2048 x 2048 pixels ) , to obtain a field of view of 23 x 23 , with a spatial scale of 0.68 per pixel . all frames were obtained with 600 seconds integration time . the median seeing ( measured as the fwhm of the stellar light profile ) for the observations presented here was 1.5 . the cores of the clusters were mapped with a mosaic of frames in a regularly spaced grid , while specific pointings were used for target galaxies at the clusters periphery . observations of landolt fields ( landolt 1992 ) , both at i and at r band , were repeated many times during each night , at airmasses between 1.2 and 2.5 , to provide the photometric calibration . the mean uncertainty in the zero point calibration at i band was 0.021 magnitudes . the 1993 service observing frames were obtained in non - photometric conditions , and have been calibrated `` a posteriori '' , using overlapping sky regions between those frames and frames obtained in photometric conditions during the sept . 1994 and sept . 1995 runs . an average of 15 stars was used to calibrate each frame , with mean uncertainty of 0.028 magnitudes . the data reduction process was performed using standard iraf procedures , and the galphot surface photometry package written for iraf / stsdas by w. freudling , j. salzer , and m.p . all frames have been bias - subtracted and flat - fielded using a `` superflat '' obtained combining , with a median filter , a large number ( @xmath12 ) of frames . for each galaxy , a local sky background was determined as the mean number of counts measured in 10 - 12 regions of `` empty '' sky , and subtracted from the frame . the typical uncertainty in this mean background , determined from the differences in the sky value obtained for galaxies in the same frame , is approximately 0.2% . all pixels contaminated by the light of foreground stars or nearby galaxies , or by cosmic rays hits , were blanked , and excluded from the final steps of surface photometry . a galaxy s light distribution was fitted with elliptical isophotes using a modified version of the stsdas _ isophote _ package , maintaining as free parameters the ellipse center , ellipticity and position angle , and incrementing the ellipse semi - major axis by a fixed fraction of its value at each step of the fitting procedure . the fitted parameters yield a model of the galaxy , which in turn is used to compute integrated magnitudes as a function of semi - major axis . the outer part of the surface brightness profiles of spiral galaxies were fitted with an exponential disk . this fit was used to extrapolate the disk profile to infinity , and to define the interval of radii where the mean disk ellipticity is to be computed . the latter was then used to derive the galaxy inclination . total magnitudes were obtained adding the flux corresponding to the extrapolated part of the disk profile to the flux measured within the outermost fitted elliptical isophote , and have a median uncertainty of 0.04 magnitudes . the uncertainty on the ellipticity is a function of inclination itself , and can vary from 1 to 20% , as shown in g96a . the surface brightness profile of each early - type galaxy was fitted with a de vaucouleurs @xmath13 profile , yielding an effective radius @xmath14 and effective surface brightness @xmath15 ( the mean surface brightness within @xmath14 ) . the fit was performed from a radius equal to twice the seeing radius , out to the outermost isophotes for e galaxies ; for s0 and s0a galaxies only the central core was fitted . total magnitudes were obtained by extrapolating the @xmath13 fit to infinity ( e galaxies ) , or by extrapolating to infinity the exponential profile that fitted the outer parts of the galaxy light profile ( s0 and s0a galaxies ) , and adding the flux corresponding to the extrapolated part of the profile to the one measured within the outermost fitted galaxy isophote . the median uncertainty on the determination of @xmath14 is 5% , and 0.06 mag on that of @xmath15 . for galaxies in close galaxy pairs , and for brightest cluster members and the smaller companions embedded in their halos , an iterative subtraction procedure was used to obtain the individual galaxy light profiles . first , the brighter object was fitted , after the fainter one(s ) was blanked . the fit was then used to build a model of the brighter galaxy , which was subtracted from the original image . the subtracted image was used to fit the fainter object , and then to build a model for it ; the latter was then subtracted from the original image and the bright object fitted again . the procedure converged usually at the second iteration , producing companion - subtracted light profiles , though the uncertainty on @xmath15 was higher than in other cases . stellar velocity dispersion measurements for 62 early - type galaxies in our fp samples were obtained from moderate dispersion optical spectra . all spectroscopic observations were obtained with the hale 5 m telescope of the palomar observatory , during 4 observing runs between september 1992 and september 1994 . the red camera of the double spectrograph ( oke & gunn 1982 ) was used with a 1200 lines mm@xmath6 grating and a 2 x 128 slit to obtain spectra with a dispersion of 0.86 per pixel , and a resolution of 2.2 ( corresponding to a velocity resolution of 129 @xmath2at 5300 ) . the spectral coverage was approximately from 5000 to 5600 , centered on the mg ib lines at @xmath165175 . exposure times ranged from 15 to 90 minutes , depending on the brightness of the target galaxy , with a median value of 30 minutes . the slit was kept to an east - west orientation , unless a close galaxy pair was being observed , and both spectra were recorded simultaneously . henear lamp spectra were obtained before and after each galaxy observation , to provide wavelength calibration . late g and early k type giant stars were used as velocity standards and as templates for the velocity dispersion measurements . all spectroscopic data were reduced using standard iraf procedures . all frames were bias - subtracted and flat - fielded using normalized dome flat - fields . no correction for the slit profile was necessary . cosmic ray hits were removed interactively from all frames . because approximately half of the observations were performed in non - photometric conditions , the spectra were not flux - calibrated . the wavelength calibration was obtained in two steps . first , the henear lamp spectra frames were used to obtain a two - dimensional dispersion solution . typically 12 - 14 spectral lines , spread uniformly over the entire spectral range , were used in the fit . the r.m.s . deviations between true and fitted wavelength were between 0.06 and 0.1 . the galaxy spectra were then transformed to linear dispersion using the lamp dispersion solution . next , the two strong night - sky lines at 5460.7 and 5577.4 were used to check the absolute wavelength calibration . in a small number of cases where the original calibration was not satisfactory , the wavelength solution was rigidly shifted to bring the two sky lines in agreement with their expected wavelength . finally the global consistency of the wavelength calibration procedure was verified using the velocity standard stars . radial velocities were measured repeatedly for all stars , using in turn one star as template and all others as unknown . the r.m.s . dispersion in the differences between true and measured velocities , considering spectra obtained in all 4 observing runs , is approximately 9 @xmath2 , in good agreement with the measured uncertainty in the wavelength of the two sky lines . one - dimensional galaxy spectra , to be used for the velocity dispersion determinations , were extracted using a 6 wide window ( in the cross - dispersion dimension ) , centered on the peak of the galaxy continuum . all measurements of velocity and velocity dispersion were obtained using the cross - correlation technique of tonry & davis ( 1979 ) , implemented in the iraf task _ fxcor_. the basic assumption of this and other similar methods ( sargent _ et al . _ 1977 , franx illingworth & heckman 1989 , bender 1990 ) is that the spectrum of an early - type galaxy is well approximated by the spectrum of its most luminous stars ( k0 k1 giants ) , modified only by the effects of the stellar motions inside the galaxy . since these motions introduce only a doppler shift in the stellar spectra , the galaxy spectrum is given by the convolution of the spectrum of a k giant star with the line of sight stellar velocity distribution ( losvd ) . therefore the losvd can be obtained by deconvolving the galaxy spectrum . this task was done in the fourier transform domain , where convolution and cross - correlation reduce to a product . before the fourier transform was computed , all spectra and templates were continuum - subtracted , normalized , end - masked with a cosine bell function , and logarithmically re - binned . because of the limited signal - to - noise ratio of the spectra , and in order to facilitate comparisons with other studies , we constrained the deconvolution procedure by assuming that the losvd is gaussian , and therefore characterized by two parameters only : a mean velocity and a velocity dispersion @xmath3 . the accuracy of the procedure was determined using simulated galaxy spectra . we broadened stellar template spectra with gaussian profiles that simulate a large range of velocity dispersions , added poissonian noise to the broadened spectra , and then measured the velocity dispersion using the original spectrum as the template . these tests showed that _ fxcor _ produces an overestimate of the velocity dispersion at the 45% level , and therefore we have corrected the raw measurements to remove this effect . the final velocity dispersion values were obtained by averaging the determinations obtained with five different stellar templates , of spectral types between g9 and k2 . given the relatively large size of the extraction window , it is possible that galaxy rotation , if present , might contribute some significant broadening of the spectral lines , and bias the velocity dispersion measurements . we have obtained rough estimates of the galaxy rotation velocity , and of its contribution to the line broadening , using the original two - dimensional spectra . a five points rotation curve within the 6 extraction aperture was obtained for each galaxy , based on 1-dimensional spectra extracted using 2-pixel wide apertures positioned side by side . the differences in the radial velocity measured in each one of the four lateral spectra , and the one measured in the central one , define the projection of the rotation curve along the slit , which was arbitrarily oriented in the e w direction . the contribution from rotation to the line broadening was estimated using a very simple model . a galaxy spectrum was simulated combining 5 copies of a stellar template spectra , broadened with a gaussian of fixed width , and shifted to reproduce the 5 velocities in the rotation curve . the weights used in the combination were derived from the relative intensities in the continuum of the 5 spectra used to determine the rotation curve . the comparison of the velocity dispersion measured in the simulated spectrum with the one used to broaden the stellar templates provides an estimate of the amount of broadening due to rotation . given the accuracy with which we can measure radial velocities in our spectra , and the small number of points used to derive the rotation curve , we can not reliably measure rotation velocities smaller than @xmath125 @xmath2 . within the a2634 and coma samples presented here ( 62 galaxies ) , 21 galaxies do not show detectable rotation , while the remaining 41 have a median rotation velocity of 49 @xmath2 , with the largest measured rotation velocity being 132 @xmath2 . this rotation is responsible for an average broadening of the losvd of @xmath12% , with a maximum of @xmath17% for the largest rotation velocity . the velocity dispersions listed in table 3 and 4 are corrected for this effect . our spectroscopic sample partly overlaps those presented by l91a for a2634 , and by lucey _ et al . _ ( 1991b , hereafter l91b ) , davies _ et al . _ ( 1987 ) , and faber _ et al . _ ( 1987 ) for coma . we can therefore compare the different velocity dispersion scales , and eventually merge all spectroscopic observations in a single database . for coma , we have 6 velocity dispersion measurements in common with davies _ _ ( 1987 ) and faber _ et al . _ ( 1987 ) , and 6 measurements in common with l91b . in both cases the mean difference between measurements is @xmath11% , with a dispersion of @xmath110% , which is close to the expected value obtained combining the quoted uncertainties of the single measurements . for a2634 we have 7 velocity dispersion measurements in common with l91a . our measurements are systematically smaller , by 14% on average , although much of this discrepancy is due to just two galaxies ( 330649 and 330678 , l91a objects 1482 and 134 ) . for these two objects our velocity dispersions are 24% smaller than those published by l91a . figure 1 shows the comparison between our measurements and those found in the literature . filled symbols identify galaxies in the coma cluster , while empty symbols identify the galaxies in a2634 that we have in common with l91a . given the good agreement between the different measurements , and because the small number of objects in common with the l91a sample prevents an accurate determination of any possible offset in the l91a velocity dispersion scale , in the following we use the l91a , l91b , davies _ et al . _ ( 1987 ) , and faber _ et al . _ ( 1987 ) velocity dispersion measurements as reported in the literature , and combine them with the new measurements presented here . both a2634 and coma are included in the cluster sample of g96a and g96b . because no new tf observations are presented here for the coma cluster , we do not discuss that sample again ( all the data can be found in table 2 in g96a ) . new tf observations have been obtained for 11 galaxies in a2634 , and we present them in table 2 , together with the sample of g96a , giving a complete list of the tf sample for that cluster . table 2 is organized as follows : + col . 1 : galaxy name , using one or two galaxy identifiers . if the galaxy is listed in the ugc catalog , the ugc number is listed first ; if not , our internal coding number is given . the second identifier is the ngc or ic number , if the galaxy is listed in those catalogs , or the cgcg field and ordinal number within that field , if the galaxy is listed in the cgcg . if the galaxy is neither an ngc / ic nor a cgcg object , no second name is entered . + cols . 2 and 3 : right ascension and declination , in 1950.0 epoch . 4 : morphological type code , in the rc3 scheme ( 1 for sa , 3 for sb , 5 for sc , 10 for irr ) . morphological types were derived from visual inspection of the blue plates of the palomar observatory sky survey ( poss ) . + col . 5 : radial velocity , in @xmath2 , measured in the cmb reference frame as defined by kogut _ et al . _ ( 1993 ) . + col . 6 : angular distance from the cluster center , in degrees . + col . 7 : membership code : a2634 members according to s95 are identified by a `` c '' ; peripheral objects are identified by a `` g '' ; members of s95 foreground group at cz = 7500 @xmath2are identified by a `` 7 '' . + col . 8 : observed velocity width , in @xmath2 . velocity widths are derived either from either 21 cm observations or from optical rotation curves , as described in g96a . to that source we address the reader for details on the corrections applied for instrumental , reduction , and relativistic broadening , turbulent motion , and inclination . + col . 9 : velocity width , in @xmath2 , corrected for turbulent broadening , instrumental and data taking effects , cosmological stretch , and shape of the optical rotation curve , but not for inclination . + col . 10 : velocity width , in @xmath2 , corrected for inclination . + col . 11 : inclination , in degrees . a galaxy inclination is given by @xmath17 where @xmath18 is the mean ellipticity of the disk , corrected for the smearing effects of seeing ( see giovanelli _ et al . _ 1995 ) , and @xmath19 is the intrinsic axial ratio of spiral disks ( we assume @xmath20 for sbc and sc galaxies , and @xmath21 for all other types ) . 12 : logarithm ( base 10 ) of the corrected velocity width , and its associated uncertainty . the format 2.377(13 ) should be read as [email protected] . + col . 13 : measured apparent i band total magnitude . + col . 14 : corrected i band magnitude . total magnitudes are corrected for extinction within our galaxy ( adopting burstein & heiles 1978 prescriptions , and @xmath23 ) , for the cosmological k - correction term ( adopting han 1992 @xmath24 ) and for internal extinction . this is done following the prescriptions of g96a , with the correction to face - on value that depends on the galaxy inclination , and also on its luminosity . 15 : absolute magnitude . this is obtained assuming that the galaxy is at the distance indicated by the cluster redshift , in the cmb reference frame , for the objects labeled `` c '' in col . 7 , and at a distance indicated by the galaxy redshift , for the other objects , for h@xmath25=100 km s@xmath6 mpc@xmath6 . the magnitude uncertainty , in hundredths of a magnitude , has the same notation as the velocity width uncertainty in col . 16 : if an asterisk appears in this column , special comments on the object are available in section 3.5 . + figure 2 ( upper panels ) shows the spatial distribution of the galaxies in the a2634 and coma tf samples ( large filled and open symbols identify proper and peripheral cluster members , respectively ) . the two dashed concentric circles have respectively radii of 1 and 2 abell radii , @xmath26 . in the lower panels of figure 2 , the radial velocity is plotted versus the angular separation from the cluster center ; the dashed vertical lines are at 1 and 2 @xmath26 , the solid horizontal lines refer to the systemic velocity of the two clusters , and the solid curves mark the redshift space caustic lines , taken from s95 for a2634 and from regs & geller ( 1989 ) for coma . in both panels the small filled symbols indicate other galaxies with measured redshift located in the area . the galaxies in the fp samples of a2634 and coma are listed in tables 3 and 4 , respectively . the spectroscopic observations of the 109 coma galaxies in table 4 derive principally from a large extant body of data , namely : 51 objects from lucey _ et al . _ ( 1991b , hereafter l91b ) , 36 s0 galaxies from dressler ( 1987 ) , and 37 e galaxies jointly from davies _ et al . _ ( 1987 ) and faber _ et al . _ ( 1987 ) . the spectroscopy of the 55 galaxies listed in table 3 , in the cluster a2634 , includes our own observations , complemented by 18 objects in the l91a sample . this sample of 55 objects includes 3 possible members of the foreground group near 7,000 @xmath2and one in the background group near 11,000 @xmath2discussed by s95 . all the photometric data in tables 3 and 4 are new and obtained by us . the contents of tables 3 and 4 are organized as follows : + col . 1 : galaxy name , using one or two galaxy identifiers , as in table 2 . for coma , the second identifier lists the dressler ( 1980 ) catalog number ( in the format dnnn ) for galaxies that are not included either in the ngc / ic or the cgcg . + cols . 2 and 3 : right ascension and declination , in 1950.0 epoch . 4 : morphological type code , in the rc3 scheme ( -5 for e , -2 for s0 , 0 for s0a ) . morphological types were derived from visual inspection of the blue plates of the poss . few objects originally classified as s0 had their type revised to sa after inspection of the ccd images + col . 5 : measured apparent i band magnitude . + col . 6 : radial velocity , in @xmath2 , measured in the cmb reference frame . + col . 7 : angular distance from the cluster center , in degrees . + col . 8 : measured effective radius , in arcseconds . + col . 9 : effective radius , in arcseconds , corrected for seeing effects . this is done adopting the prescriptions of saglia _ et al . _ ( 1993 , see in particular their figure 8) . + col . 10 : uncertainty on the effective radius , in arcseconds . 11 : corrected metric effective radius , in kiloparsec , obtained assuming the galaxy is at the distance indicated by the cluster redshift and @xmath27 km s@xmath6 mpc@xmath6 , in the cmb reference frame . 12 : measured effective surface brightness , in i band magnitudes per square arcsecond . + col . 13 : corrected effective surface brightness , in i band magnitudes per square arcsecond . we corrected @xmath15 for the smearing effects of seeing , adopting the prescriptions of saglia _ et al . _ ( 1993 , see in particular their figure 8) , for extinction within our galaxy , using burstein & heiles ( 1978 ) method as described above , for the cosmological k - correction term ( which is simply @xmath28 because of the flat spectrum of early type galaxies in the far red ) , and for the @xmath29 cosmological dimming . 14 : uncertainty on the effective surface brightness , in magnitudes per square arcsecond . + col . 15 : stellar velocity dispersion , in @xmath2 . for our new measurements , the measurement uncertainty , in @xmath2 , is given between brackets : e.g. 177(11 ) is equivalent to 177@xmath2211 @xmath2 . measurements taken from the literature are given without uncertainty . see original sources for details . 16 : reference codes for the spectroscopic measurements taken from the literature . if an asterisk appears in this column , special comments on the object are available in section 3.5 + whenever multiple spectroscopic measurements were available , we have used here only the one with smaller uncertainty . therefore , if no reference is listed in col . 16 , the new measurement presented in col . 15 was used for that particular galaxy . otherwise , the measurement in the referenced source , and only that one , was used . figure 3 shows the spatial distribution ( upper panels ) and the radial velocity distribution as a function of angular distance from the cluster center ( lower panels ) for the a2634 and coma fp samples . symbol and inset curves follow the same conventions as assumed for figure 2 . it has been argued that cluster samples for tf and fp work are almost bias - free , since all galaxies are roughly at the same distance . while it is certainly true that the classical malmquist bias is avoided , a different and equally important bias affects cluster samples . this is generally referred - to as the `` cluster population incompleteness bias '' ( teerikorpi 1987 , 1990 ) . in a recent work sandage , tammann & federspiel ( 1995 ) have provided a detailed analysis of this bias and of the corrections it requires before unbiased distance estimates can be obtained with cluster tf samples . their recipes are based on the assumption that galaxy samples are magnitude - limited . since that is not the case for our samples , we have derived incompleteness bias corrections following the spirit of the sandage _ et al . _ ( 1995 ) treatment , but using monte carlo simulations to reproduce as closely as possible the completeness , measurement uncertainties , and scatter characteristic of our samples . we refer the reader to the discussion in g96b for the details on the bias correction applied to the tf samples , while here we discuss briefly the bias correction for the fp samples , since the latter is derived using a significantly different procedure ( a more detailed discussion will be given in scodeggio 1997 ) . it is customary to express the completeness of a sample relatively to the expected luminosity distribution of a galaxy population , mainly because magnitudes are relatively easy to measure , and sample selection most often is based on them . we therefore define completeness as the ratio of the number of galaxies of a given magnitude included in a sample with respect to the total number of galaxies of the same magnitude predicted by the appropriate luminosity function . the effect of sample incompleteness on the tf relation is relatively straightforward , because magnitude is one of the two parameters used to build the relation . at any given velocity width , fainter galaxies are preferentially missing from an incomplete sample , and this affects the determination of the slope , the zero - point , and the dispersion of the tf relation . in the fp case , magnitudes enter only indirectly into the fp diagram , via the existing relations between luminosity and effective radius ( fish 1964 ) , velocity dispersion ( faber & jackson 1976 ) , and effective surface brightness ( binggeli , sandage & tarenghi 1984 ) . these @xmath30 , @xmath31 , and @xmath32 relations all show rather large scatter . therefore , at any given velocity dispersion or effective surface brightness , the systematic lack of fainter galaxies is only marginally reflected into a lack of small effective radius objects , and it is not possible to translate the completeness of a sample directly in terms of an effective radius , velocity dispersion or effective surface brightness completeness . our simulations are therefore designed to reproduce as closely as possible the observed relations between @xmath33 and @xmath3 and the galaxy luminosity , and the relation between @xmath33 and @xmath15 ( kormendy 1977 ) . then we select galaxies according to their magnitude , reproducing the observed luminosity incompleteness , and measure the indirect effect of this incompleteness on the fp parameters . in practice , we first divide both a complete sample and an incomplete one into bins according to the value of the combination of @xmath3 and @xmath15 that provides an edge - on view of the fundamental plane . then we measure the mean value of @xmath33 for each bin , and compare these mean values for the complete and incomplete samples . because of the lack of low luminosity objects , the mean @xmath33 measured in the incomplete sample is systematically larger than the one measured for the complete sample , except in the two to three bins which include the most luminous galaxies . we finally use these differences to correct the observed values of @xmath33 , and reproduce a relation as close as possible to the unbiased one . [ [ tf ] ] tf ~~ although the results of s95 revealed that the a86 result was unlikely to be severely affected by substructure in a2634 , the a86 tf sample was quite small and sparse . the situation improved with the study of g96a , which included a sample of 15 strict members and two additional cluster outliers . our current sample includes 23 _ bona fide _ cluster members , and 4 peripheral members , nearly tripling the size of the a86 sample . the tf relation for our new data is shown in figure 4a . filled symbols indicate cluster members , and unfilled ones indicate peripheral members . the solid line is the tf template derived by g96b combining a sample of 555 galaxies in 24 clusters : @xmath34 the total uncertainty in the tf template zero point is 0.05 mag . this includes statistical uncertainties associated with the fit ( 0.02 mag ) , uncertainties associated with the incompleteness bias corrections ( 0.025 mag ) , which are related to the choice of parameters that describe the luminosity function of spiral galaxies , and the estimated accuracy to within which the average peculiar velocity of the cluster set approaches zero in the cmb reference frame ( 0.04 mag ) . an uncertainty of 0.05 mag translates to about 210 @xmath2 at the distance of a2634 , and 170 @xmath2 at the distance of coma . the mean scatter about the tf template relation is @xmath10.35 mag . , although this figure varies with the mean velocity width ( and luminosity ) of galaxies : it reduces to @xmath10.25 mag . at the high @xmath35 end , and expands to @xmath10.45 mag . at the low @xmath35 end of the typical range of the tf diagram . the mean offset from the template , for the a2634 sample shown in fig . 4a , is [email protected] mag . , corresponding to a peculiar velocity of + 165@xmath22250 @xmath2 in the cmb reference frame . for comparison , the peculiar velocity estimate obtained by g96b using only 15 strict cluster members is + 61@xmath22378 @xmath2 . for later comparison with the fp results , we also derive a coma - a2634 relative tf distance . figure 4b shows the tf relation for the coma sample of g96a . the convention for the symbols and the solid line are the same as in fig . the mean offset from the template for coma is [email protected] mag . , corresponding to a peculiar velocity of + 194@xmath22196 @xmath2 . the relative peculiar motion of a2634 with respect to coma is therefore -29@xmath22318 @xmath2 , and the ratio of their distances is @xmath36 . we remark also that the differential bias correction between the two clusters is only 0.002 magnitudes , and therefore the amplitude of the coma a2634 relative motion in the comoving frame is practically independent of the amplitude of the two separate bias corrections . in the comoving reference frame , it thus appear that the two clusters are at rest with respect to each other , as nearly determined as the error budget allows . the peculiar velocities of each of the two clusters also appear to be of statistically negligible amplitude . [ [ fp ] ] fp ~~ the d@xmath0 and fp relations have been obtained in the past using photometric measurements in many different bands , from b ( faber _ et al . _ 1989 ) to v ( l91a , l91b ) , @xmath37 ( djorgovski & davis 1987 ) , gunn r ( jrgensen , franx & kjrgaard 1996 ) , and kron - cousins r ( colless _ et al . _ 1993 ) . in the study of early - type galaxies there are no compelling reasons for favoring one optical band over the others . our adoption of i band for photometry was mainly dictated by the practical notion that the cluster images contain several galaxies per frame , both of early and late type , and thus can be useful for tf application , where the i band is favored for compatibility with g96a . the fp is a flat surface in the 3-dimensional space of the parameters @xmath38 , @xmath39 , and @xmath15 . we use the data set of 109 coma galaxies listed in table 4 in order to obtain a fp template relation , which we define as the plane obtained by averaging the coefficients of the 3 possible fits that can be performed using one of the 3 parameters as the dependent variable and the remaining two as the independent ones . the resulting fp template , obtained by assuming that coma is at rest in the cmb reference frame and h@xmath40 km s@xmath6 mpc@xmath6 , is given by @xmath41 ( or @xmath42 ) . no statistically significant difference is observed when only e or only s0 galaxies are considered . the dispersion around this mean relation , measured as the r.m.s . scatter in the residuals of @xmath38 , is 0.086 . this is in remarkable agreement with the result of jrgensen , franx & kjrgaard ( 1996 ) , who found a scatter in @xmath38 of 0.084 . the dispersion is equivalent to a scatter of 0.43 mag in the distance modulus ( or 20% in distance ) of a single galaxy . the statistical uncertainty on the fp zero point is of 0.008 , equivalent to a distance ( or peculiar velocity ) uncertainty for the coma cluster of 133 @xmath2 . the incompleteness bias correction for this sample was derived fitting a fermi - dirac distribution to the completeness histogram . the 50% completeness level is at @xmath43 , while the mean i band magnitude of a complete sample is @xmath44 . the mean bias correction to @xmath38 derived from the monte carlo simulations is 0.018 ( equivalent to a correction of 0.090 mag for the distance modulus ) . figure 5 shows an edge - on view of the fp for the coma sample . the solid line is the projection of the plane ( [ fp_temp ] ) . to make the figure more readable we plot separately the data points and the error - bars associated with those points . as already pointed out by djorgovski & davis ( 1987 ) , and jrgensen , franx & kjrgaard ( 1996 ) , there might be a small curvature in the plane . galaxies with the smallest and largest effective radii appear to have systematically positive residuals in @xmath38 , but the number of objects involved is too small to make this effect statistically significant . figure 6 shows the edge - on view of the fp for the a2634 sample . the solid line is the fp template derived for coma . error - bars are plotted separately in panel ( b ) . two galaxies in this sample deviate significantly ( @xmath45 ) from the main relation . one ( 330747 ) is a small companion to a much brighter galaxy ( ugc 12727 ) , so that the iterative subtraction procedure used to obtain its photometric parameters yields uncertain results . the second galaxy ( 331345 ) has a very flat photometric profile , with abnormally faint central surface brightness ( its effective surface brightness is almost a magnitude fainter than the next fainter effective surface brightness in our sample ) . we have excluded these two galaxies from the a2634 sample used to derive the cluster peculiar motion . the relative agreement between data - points and template line seen in figure 6 indicates that the difference in peculiar velocity between coma and a2634 is small . the mean offset in @xmath38 between a2634 and the coma template ( [ fp_temp ] ) is [email protected] , which corresponds to a peculiar velocity of 204@xmath22266 @xmath2 for a2634 , with respect to coma in the comoving reference frame . the ratio of the distances of the two clusters , according to this fp analysis , is @xmath46 this result is in good agreement with the estimates obtained using the completely independent spiral samples and the tf relation , and in disagreement with the result of l91a . the a2634 sample is characterized by a 50% completeness level at @xmath47 , and by a bias correction of 0.021 dex . the differential bias correction between coma and a2634 is therefore only 0.003 dex ( equivalent to 0.015 magnitudes on the distance modulus ) , and does not play a significant role in the determination of the relative motion between the two clusters . in order to identify the source of the discrepancy with the results of l91a , we have re - determined the a2634 peculiar velocity using two subsets of our sample . the first subset is composed of the 18 galaxies observed by l91a . for this computation we used our own i band photometric measurements , and the original velocity dispersion measurements of l91a . the mean residual in @xmath38 for this sample is [email protected] , corresponding to a peculiar velocity of 1324@xmath22573 @xmath2 . this is significantly bigger than the value we obtain using the whole sample , but also smaller than the value obtained by l91a using the same velocity dispersions , and their own photometric measurements . as discussed in 3.3 , from a direct comparison of velocity dispersions the l91a measurements appear to be systematically larger , by @xmath114% , than our measurements , but this difference alone is not enough to explain the discrepancy between our distance estimate for a2634 and the l91a one . unfortunately l91a published only v band isophotal diameters for their galaxies , and a comparison with our i band photometry for those same galaxies is very difficult , without accurate v - i color information . therefore we hesitate to carry the analysis of the l91a sample any further . we repeat the exercise with a second a2634 subset , composed of the 42 galaxies for which we have obtained both photometric and spectroscopic measurements ( but two galaxies are removed from the sample , as discussed above ) . the mean residual in @xmath38 for this sample is [email protected] , which corresponds to a relative peculiar velocity of @xmath48287 @xmath2 . the result is again in good agreement with the tf determination . the small uncertainties in the tf and fp determinations we have just discussed , and the good agreement between the two independent results , indicate that the peculiar velocity of a2634 is small . we conclude that the l91a result is spurious and find no reason to doubt that the fp and tf relations have universal applicability . similar conclusions have been recently obtained by lucey and coworkers , using new velocity dispersion measurements in a2634 and a2199 ( guzman 1996 ) . these new measurements confirm that the measurements reported by l91a were systematically overestimated , and therefore that the peculiar velocity of a2634 is negligible . new tf and fp measurements for the cluster a2634 and coma give compatible results for the relative distance and peculiar velocity of the two clusters . contrary to the findings of l91a , the peculiar velocity of a2634 with respect to the cmb reference frame is unlikely to exceed a few @xmath49 @xmath2 , and extremely unlikely to exceed @xmath50 @xmath2 . both tf and fp measurements suggest that the ratio of the distances to a2634 and to coma is @xmath51 , which is not too dissimilar from the ratio of systemic velocities in the cmb reference frame of 1.24 . our determinations are more accurate and reliable than those of previous work , thanks to ( a ) more accurate criteria for the assignment of cluster membership to individual galaxies and ( b ) significantly expanded samples . we have also obtained a new fp template , using for the first time i band photometry . the parameters of this template are in very good agreement with recent determinations of the fp obtained at shorter wavelengths ( gunn r ) , confirming that there is little dependence of the fp relation on the passband used for the photometric measurements . the uncertainty with which the fp can provide peculiar velocity estimates for single galaxies is @xmath10.43 mag in the distance modulus , or 20% of the distance . this uncertainty is slightly larger than the typical uncertainty that characterizes tf estimates , the latter being @xmath10.35 magnitudes . this disadvantage is however partly compensated by the fact that the sample incompleteness bias has a less severe effect on fp cluster distance estimates than it has on tf cluster distance estimates , and cluster membership is more readily established for early type objects . the original motivation for this study was provided by the desire to investigate the universality of the tf and fp relations , the discrepancy in peculiar velocity estimates reported by l91a making a2634 an extreme case study . our results restore a measure of trust in the reliability of those relations as cosmological tools . we would like to thank juan carrasco for the precious help he provided during the observing runs at palomar , and bill schoening for his assistance with the 0.9 m telescope at kpno . we are grateful to the kpno tac for the generous allocation of observing time . this research was partially supported by nsf grants ast9420505 to rg , and ast9023450 and ast9218038 to mph . 330564 : marginal cluster membership assignment . + 331234 : probably a member of an interacting system . + 330636 : extremely irregular system , with several clumps of light spread in irregular fashion . inclination unreliable . + 330663 : spectrum with poor s / n ratio ; very uncertain width . + u12721 : the discrepancy in flux among multiple sources reported by g96a is resolved by these observations . + u12755 : the magnitude reported by g96a is 12.43 ; new observations yield 12.46 ; we adopt the average value . u12716 : @xmath3 measured also by l91a ( 327 @xmath2 ) , tonry 1984 ( 348 @xmath2 ) , malumuth & kirshner 1981 ( 386 @xmath2 ) , tonry & davis 1981 ( 365 @xmath2 ) , and faber _ et al . _ 1989 ( 298 @xmath2 ) . + 331456 : @xmath3 measured also by l91a ( 218 @xmath2 ) , and by tonry 1984 ( 250 @xmath2 ) . + 330649 : @xmath3 measured also by l91a ( 285 @xmath2 ) . + 330658 : @xmath3 measured also by l91a ( 201 @xmath2 ) . + 330668 : @xmath3 measured also by l91a ( 207 @xmath2 ) . + 330678 : @xmath3 measured also by l91a ( 272 @xmath2 ) . + 331456 , 330648 , 330651 , 330658 , 330660 , 330665 , 330668 , 330678 are within the halo of the cd ugc 12716 ; their @xmath15 has high uncertainty . + 330706 : bright nearby star subtracted . + 330600 : faint outer disk visible . + 330747 : bright nearby companion ( ugc 12727 ) subtracted . surface brightness determination unreliable . + u12744 : bright nearby star subtracted . u8049 : @xmath3 also measured by l91b ( 212 @xmath2 ) , and faber _ et al . _ 1989 ( 208 @xmath2 ) . + u8065 : @xmath3 also measured by l91b ( 279 @xmath2 ) . + u8070 : @xmath3 also measured by davies & illingworth 1983 ( 255 @xmath2 ) , faber _ et al . _ 1989 ( 259 @xmath2 ) , and l91b ( 298 @xmath2 ) + u8103 : @xmath3 also measured by l91b ( 279 @xmath2 ) , faber _ et al . _ 1989 ( 245 @xmath2 ) , malumuth & kirshner 1981 ( 304 @xmath2 ) ours spectrum has poor s / n ratio , and therefore we use faber _ et al . _ measurement . + u8110 : @xmath3 also measured by oegerle & hoessel 1991 ( 381 @xmath2 ) , tonry 1985 ( 404 @xmath2 ) , tonry & davis 1981 ( 412 @xmath2 ) , davies & illingworth 1983 ( 342 @xmath2 ) , faber _ et al . _ 1989 ( 381 @xmath2 ) , faber & jackson 1976 ( 400 @xmath2 ) , and l91b ( 414 @xmath2 ) . + u8175 : @xmath3 also measured by l91b ( 275 @xmath2 ) . + 221266 : @xmath3 also measured by davies _ et al . _ 1987 ( 150 @xmath2 ) . + 221354 : @xmath3 also measured by l91b ( 147 @xmath2 ) , and faber et al . 1989 ( 164 @xmath2 ) . + 221410 : @xmath3 also measured by dressler 1987 ( 133 @xmath2 ) . + 221216 : spectrum with poor s / n ratio . also , the galaxy is in a close pair . + 221290 , 221291 , 221293 , 221298 , 221303 , 221304 , 221317 , 221323 , 221329 , 221331 , 221334 are within the halo of the d galaxy ugc 8103 , and 221354 , 221362 , 221377 , 221380 , 221382 , 221392 are within the halo of the d galaxy ugc 8110 ; their @xmath15 has high uncertainty . + u8133 : faint outer disk . + u8072/u8073 : close pair ; iterative subtraction required . + u8100 : bright nearby star subtracted . aaronson , m. , bothun , g. , mould , j. , huchra , j. , schommer , r.a . , & cornell , m.e . 1986 , , 302 , 536 ( a86 ) bender , r. 1990 , , 229 , 441 binggeli , b. , sandage , a. , & tarenghi , m. 1984 , , 89 , 64 burstein , d. & heiles , c. 1978 , , 225 , 40 colless , m. , burstein , d. , wegner , g. , saglia , r.p . , mcmahan , r. , davies , r.l . , bertschinger , e. , & baggley g. 1993 , , 262 , 475 colless , m. , & dunn , a.m. 1996 , , 458 , 435 david , l.p . , slyz , a. , jones , c. , forman , w. , vrtilek , s.d . , & arnaud , k.a . 1993 , , 412 , 479 davies , r.l . , & illingworth , g. 1983 , , 266 , 516 davies , r.l . , burstein , d. , dressler , a. , faber , s.m . , lynden - bell , d. , terlevich , r.j . , & wegner , g. 1987 , , 64 , 581 djorgovski , s. , & davis , m. 1987 , , 313 , 59 djorgovski , s. , de carvalho , r. , & han , m.s . 1988 , in `` the extragalactic distance scale '' , ed . s. van den bergh & j. pritchet , p. 329 dressler , a. 1980 , , 236 , 351 dressler , a. 1987 , , 317 , 1 dressler , a. , lynden - bell , d. , burstein , d. , davies , r.l . , faber , s.m . , terlevich , r.j . , & wegner , g. 1987 , , 313 , 42 faber , s.m . , & jackson r.e . 1976 , , 204 , 668 faber , s.m . , wegner , g. , burstein , d. , davies , r.l . , dressler , a. , lynden - bell , d. , & terlevich , r.j . , & 1987 , , 69 , 763 fish , r.a . 1964 , , 139 , 284 franx , m. 1993 , , 105 , 1058 franx , m. , illingworth , g. , & hechman , t. 1989 , , 344 , 613 gavazzi , g. , randone , i. , & branchini , e. 1995 , , 438 , 590 giovanelli , r. , & haynes , m.p . 1989 , , 97 , 633 giovanelli , r. , & haynes , m.p . 1991 , , 29 , 499 giovanelli , r. , haynes , m.p . , salzer , j.j . , wegner , g. , da costa , l.n . , & freudling , w. 1995 , , 110 , 1059 giovanelli , r. , haynes , m.p . , herter , t. , vogt , n.p . , wegner , g. , salzer , j.j . , da costa , l.n . and freudling , w. , 1996a , submitted ( g96a ) giovanelli , r. , haynes , m.p . , herter , t. , vogt , n.p . , da costa , l.n . , freudling , w. , salzer , j.j . and wegner , g. , 1996b , submitted ( g96b ) guzmn , r. 1996 , private communication guzmn , r. , lucey , j.r . , carter , d. , & terlevich , r.j . 1992 , , 257 , 187 han , m.s . 1992 , , 81 , 35 jacoby , g.h . , branch , d. , ciardullo , r. , davies , r.l . , harris , w.e . , pierce , m.j . , pritchet , c.j . , tonry , j.l . , & welch , d.l . 1992 , , 104 , 599 jrgensen , i. , franx , m. , & kjrgaard , p. 1996 , , 280 , 167 kent , s.m . , & gunn , j.e . 1982 , , 87 , 945 kogut , a. , et al . 1993 , , 419 , 1 kormendy , j. 1977 , , 218 , 333 landolt , a.u . 1992 , , 104 , 340 lynden - bell , d. , faber , s.m . , burstein , d. , davies , r.l . , dressler , a. , terlevich , r.j . , & wegner , g. 1988 , , 326 , 19 lucey , j.r . , gray , p.m. , carter , d. , & terlevich , r.j . 1991a , , 248 , 804 ( l91a ) lucey , j.r . , guzmn , r. , carter , d. , & terlevich , r.j . 1991b , , 253 , 584 ( l91b ) malumuth , e.m . , & kirshner , r.p . 1981 , , 251 , 508 mould , j.r . , et al . 1990 , , 383 , 467 oegerle , w.r . , & hoessel , j.g . 1991 , , 375 , 15 oke , j.b . , & gunn , j.e . 1982 , , 94 , 586 pinkney , j. , rhee , g. , burns . hill , j.m . , oegerle , w.m . , batuski , d. , & hintzen , p. 1993 , , 416 , 36 regs , e. , & geller , m.j . 1989 , , 98 , 755 rowan - robinson , m. 1985 , `` the cosmological distance ladder '' ( new york , freeman and co. ) saglia , r.p . bertschinger , e. , baggley , g. , burstein , d. , colless , m. , davies , r.l . , mcmahan , r.k . , & wegner , g. 1993 , , 264 , 961 sandage , a. 1994a , , 430 , 1 sandage , a. 1994b , , 430 , 13 sandage , a. , tammann , g.a . , & federspiel , m. 1995 , , 452 , 1 sargent , w.l.w . , schechter , p.l . , boksenberg , a. , & shortridge , k. 1977 , , 212 , 326 scodeggio , m. 1997 , ph . d. thesis , cornell university scodeggio , m. , solanes , j.m . , giovanelli , r. , & haynes , m.p . 1995 , , 444 , 41 ( s95 ) scodeggio , m. , giovanelli , r. , & haynes , m.p . 1996 , in preparation silk , j. 1989 , , 345 , l1 teerikorpi , p. 1984 , , 141 , 407 teerikorpi , p. 1987 , , 173 , 39 teerikorpi , p. 1990 , , 234 , 1 teerikorpi , p. 1993 , , 280 , 443 tonry , j.l . 1984 , , 279 , 13 tonry , j.l . 1985 , , 90 , 2431 tonry , j.l . , & davis , m. 1979 , , 84 , 1511 tonry , j.l . , & davis , m. 1981 , , 246 , 666 tully , r.b . , & fisher , j.r . 1977 , , 54 , 661 van dokkum , p.g . , & franx , m. 1996 , , 281 , 985 vogt , n.p . , herter , t. , haynes , m.p . , & courteau , s. 1993 , , 415 , l95 vogt , n.p . , forbes , d.a . , phillips , a.c . , gronwall , c. , faber , s.m . , illingworth , g.d . , & koo , d.c . 1996 , , 465 , l15 willick , j.a . 1994 , , 92 , 1 zabludoff , a.i . , huchra , j.p . , & geller , m.j . 1990 , , 74 , 1

in this paper we explore the idea that black holes can persist in a universe that collapses to a big crunch and then bounces into a new phase of expansion . we use a scalar field to model the matter content of such a universe near the time of the bounce , and look for solutions that represent a network of black holes within a dynamical cosmology . we find exact solutions to einstein s constraint equations that provide the geometry of space at the minimum of expansion and that can be used as initial data for the evolution of hyperspherical cosmologies . these solutions illustrate that there exist models in which multiple distinct black holes can persist through a bounce , and allow for concrete computations of quantities such as number density . we then consider solutions in flat cosmologies , as well as in higher - dimensional spaces ( with up to nine spatial dimensions ) . we derive conditions for the black holes to remain distinct ( i.e. avoid merging ) and hence persist into the new expansion phase . some potentially interesting consequences of these models are also discussed .

1701.05750

bouncing cosmologies , in which the universe collapses to a minimum of expansion before returning to expansion , have a long history @xcite . one of the interesting considerations for such a scenario is whether the matter content of the expanding phase is identical to that of the collapsing phase . in general , one might expect entropy to increase from one cycle to the next @xcite . this implies the production of radiation or the creation of black holes , which can be thought of as the objects that have the highest entropy allowed in nature @xcite . it is therefore of interest to consider whether black holes can persist through a cosmic bounce , as well as what effect they would have on the large - scale dynamics and on each other . one can divide such black holes into two classes : ( i ) those that persist in a universe that recollapses to a big crunch and then bounces into a new expansion phase ; and ( ii ) those that are generated by the high density of matter at the bounce . following ref . @xcite , these objects will be referred to as `` pre - crunch black holes '' ( pcbhs ) and `` big - crunch black holes '' ( bcbhs ) , respectively . it is interesting to consider whether there would be any observational way of distinguishing them from the primordial black holes ( pbhs ) that formed shortly after the bounce , or during the early stages of the standard hot big bang model without a bounce . it is also important to know whether black holes really can persist through a bounce in the way that is being envisaged . the precise properties of a cosmic bounce depend upon the way in which it is generated . many mechanisms have been proposed for this in the long history of the subject . these include classical effects associated with a cosmological constant @xcite , higher - derivative effects @xcite , or other modified theories of gravity @xcite . there are also semi - quantum gravitational effects associated with string theory @xcite , loop quantum gravity @xcite , quantum gravity condensates @xcite , and the `` pre - big - bang '' scenario @xcite . there have also been some claims that scale - invariant fluctuations generated during a cosmological collapse phase could turn into scale - invariant curvature fluctuations after a bounce @xcite , although in a fully quantum gravitational bounce there may be conceptual problems with these approaches @xcite . for a recent review of bouncing cosmologies , see ref . the scalar - field - driven bounces we consider here are probably the simplest models that can accommodate a bounce , and allow us to consider the problem of persistent black holes in an exact and fully relativistic way . besides the mechanism of the bounce itself , there is also the tricky mathematical issue of what is meant by a black hole in a universe that is not eternally expanding . this issue is further complicated by the lack of any asymptotic spatial infinity in most cosmological models . for a pbh that forms after the big bang , it is usually understood that the black hole singularity is part of any future big crunch singularity ( if one should occur ) . however , this can not apply to pcbhs or bcbhs if the cosmological bounce is non - singular . without knowledge of the location of the black hole singularities it is impossible to locate the event horizon , which is defined as the boundary of the past of future null infinity . a more satisfactory approach is therefore to use the concept of the black hole apparent horizon , which is defined as the outermost marginally outer trapped surface ( mots ) @xcite , and will be used throughout this study . a further matter of interest for models of this type concerns what happens to the number density of black holes as the minimum of expansion is approached . one could speculate about the maximum fraction of space that can be filled by pcbhs or about the number of bcbhs that form during the collapsing stage . these quantities would depend on the maximum density of matter that occurs at the bounce , and there are then various interesting constraints that can be imposed on the fraction of the universe s mass density that resides in black holes as a function of mass @xcite . if the black holes are randomly distributed , one would expect a process of hierarchical merging to occur , in which progressively larger apparent horizons form around groups of holes , so that the characteristic hole size steadily increases . alternatively , if the holes were distributed in a uniform fashion then pre - crunch hierarchical merger would be eliminated , and one could investigate the maximum number of black holes that could fit through a given bounce . in this paper , we study this scenario in detail by considering models that contain a regular lattice of black holes and a scalar field at the maximum compression . these restrictions are primarily motivated by mathematical simplicity , as we find that they yield a number of exact solutions to the einstein constraint equations . the associated initial data then prescribe a unique evolution in both the expansion and collapse directions . they therefore constitute a proper cosmological model . although in general the evolution will need to be determined numerically , the initial data itself can be used to analyse a number of the problems outlined above . in particular , the positions of the apparent horizons can be determined and used to calculate the distance between neighbouring black holes . this information can be used to determine the fraction of space that is filled by black holes as a function of the maximum energy density at the bounce . it also unambiguously demonstrates that black holes can persist through a cosmological bounce . the plan of this paper is as follows . in sec . 2 we derive the form of the constraint equations in a time - symmetric scalar field - dominated model , focussing particularly on the geometry of the spatial hypersurface at the bounce . in secs . 3 and 4 we consider black holes in hyperspherical and spatially flat cosmological models , respectively . in sec . 5 we consider black holes in higher - dimensional bouncing cosmologies , a possibility also emphasized in ref . @xcite . in sec . 6 we draw some general conclusions . appendix a discusses solutions for the initial data in which the scalar field is stationary at the maximum compression , these relating to multiple janis - newman - winicour - like black hole solutions . appendix b discusses solutions with a non - vanishing scalar potential . in this section we will present the einstein field equations ( efes ) in the presence of radiation and a scalar field , as relevant for studying the geometry of space - time in the vicinity of a bounce . this follows from the standard approach of using a scalar field in early - universe friedmann - lematre - robertson - walker ( flrw ) cosmology in order to model the effects of new physics . in our case , however , we will not assume anything about the existence of global symmetries ( approximate or otherwise ) . instead , we will give the general form for the constraint and evolution equations , and describe how a bouncing cosmology with black holes can be formulated as an initial value problem , with initial data prescribed at the bounce . just as in standard flrw cosmology , the scalar field may be expected to dominate at early times but become negligible compared to the energy and momentum in radiation and dust at late times . this means that if one considers a regular array of black holes and a scalar field , the late - time evolution of the space might approach that of a space just containing an array of black holes ( i.e. the space should forget that it once contained a dominant scalar field ) . this is mathematically very interesting , because for a scalar field with a negative coupling constant there exists a time - symmetric initial value problem for an array of black holes which expand away from each other as the space evolves from the time - symmetric surface . the existence of time symmetry means that the constraint equations on the initial surface can be manipulated into a linear form , allowing for the initial data for arbitrary many black holes to be constructed by simple superposition . let us consider the efes with a scalar field : [ fe ] g_ab = t^_ab + t^_ab , where @xmath0 is a coupling constant ( necessarily negative for a bounce ) and we have chosen units with @xmath1 . the quantity @xmath2 is the energy - momentum tensor of the scalar field , and is given by t^_ab = _ a _ b - g_ab _ c ^c - g_ab v ( ) , where @xmath3 is the scalar potential . the quantity @xmath4 is the energy - momentum tensor of an additional radiation fluid , which is assumed to be separately conserved . the contracted second bianchi identity then gives the scalar propagation equation as [ boxvarphi ] = , where @xmath5 is the covariant dalembertian operator . if we now choose a time - like vector field , @xmath6 , we can decompose @xmath2 into energy density , isotropic pressure and momentum density : [ rho ] ^ & & u^a u^b t^_ab = ^2 + d^a d_a + v ( ) + [ p ] p^ & & h^ab t^_ab = ^2 - d^a d_a -v ( ) + [ j ] j_a^ & & - h_a^ b u^c t^_bc = - d_a . here an over - dot denotes @xmath7 , and we have introduced @xmath8 and @xmath9 as the projection operator and the projected covariant derivative on the set of spaces orthogonal to @xmath6 . likewise , the energy density and isotropic pressure of a perfect fluid of radiation are given by ^ & & u^a u^b t^_ab , p^ h^ab t^_ab , where @xmath10 and the momentum density is @xmath11 ( by virtue of the fluid being radiation and perfect , respectively ) . using the appropriate embedding equations , together with the efes ( [ fe ] ) , we can write the hamiltonian and momentum constraint equations for the space - time as [ hc ] & & + k^2 - k_ab k^ab = 2 u^a u^b g_ab = 2 ^ + 2 ^ + [ mc ] & & d_b k^b _ a - d_a k = - h^b _ a u^c g_bc = j_a^ , where @xmath12 is the extrinsic curvature of the hypersurface we are considering , @xmath13 and @xmath14 denotes the ricci scalar of this hypersurface ( assuming @xmath6 is irrotational ) . these equations must be satisfied on the initial hypersurface , in order to have a solution to the efes . the corresponding evolution equations for the geometry will be presented in sec . [ sec : time ] below . it now remains to decompose the conservation equations for the scalar field and radiation fluids . for the radiation field this yields [ radcon ] d_a ^ + 4 ^ _ a = 0 , ^ = k ^ as the constraint and evolution equations for @xmath15 , respectively . here we have chose a @xmath6 that is comoving with the radiation fluid . for the scalar field , the propagation equation ( [ boxvarphi ] ) can be decomposed into evolution and constraint equations with respect to the vector field @xmath6 . to do this , we first define [ varphivar ] _ a d_a . we can then decompose eq . ( [ boxvarphi ] ) into the following first - order evolution equations : [ evo1 ] & = & d_a ^a+ k + _ a ^a + v^ ( ) + _ a & = & d_a + _ a + u_a ^b _ b + & = & , [ evo3 ] where the last of these defines the quantity @xmath16 . ( [ varphivar ] ) is then the only constraint equation , while eqs . ( [ evo1])-([evo3 ] ) constitute the evolution equations for @xmath16 , @xmath17 and @xmath18 . this completes our formulation of the initial value problem in terms of constraint and evolution equations , as required for this study . let us now consider time - symmetric initial data . in this case the initial data must satisfy @xmath19 . from eqs . ( [ j ] ) and ( [ mc ] ) , this immediately implies d_a = 0 , so that either @xmath20 or @xmath21 or both on the initial hyper - surface . if both are true , then the scalar field evolution equations show that @xmath22 at all points in space and time , which is the trivial case in which there is no scalar field . the case where @xmath20 is considered in the appendices . if @xmath23 , then initial data of this type can be shown to include cosmological solutions with multiple janis - newman - winicour - like black holes ( see appendix a ) . if @xmath3 is an exponential function , then exact solutions can also be found ( see appendix b ) . however , the remaining case , where @xmath21 , is a more interesting one for cosmology and will form the basis of the study that follows . if @xmath24 , then @xmath18 is constant on the initial hypersurface and we can infer from eqs . ( [ rho ] ) and ( [ p ] ) that @xmath25 and @xmath26 . time - symmetry on the initial hypersurface then requires @xmath27 and @xmath28 . as we require @xmath29 in order to have a non - vacuum solution , the time - symmetry condition implies @xmath30 , which is compatible with the scalar field evolution equation ( [ evo1 ] ) . in this case , we therefore have a solution where @xmath18 is initially homogeneous but @xmath31 is inhomogeneous and @xmath32 vanishes . we are also forced to require @xmath33 to be a constant , so that it acts like a cosmological constant . we therefore neglect @xmath33 in what follows , as the cosmological constant is not expected to play an important role near the time of the bounce . the remaining equation that needs to be satisfied in order to have a solution to the einstein - scalar system is the hamiltonian constraint ( [ hc ] ) . when @xmath34 , this can be written as [ constraintr ] = ^2 + 2 ^ . the quantity @xmath16 is not involved in the scalar field constraint equation and can therefore be specified freely . once we have a solution for the geometry of an initial hypersurface that satisfies eq . ( [ constraintr ] ) for a given @xmath16 , we have a complete solution to all the constraint equations and hence the full initial data required to determine the entire evolution . to find explicit solutions with constant conformal curvature , let us make the following ansatz for the line - element of the time - symmetric initial hypersurface : [ le ] ds^2 = ^4 d|s^2 , where @xmath35 is the line - element for a 3-space of constant curvature . standard results from conformal geometry then give [ conformal ] = - |d^2 , where @xmath36constant is the ricci scalar of the conformal space and @xmath37 is the covariant derivative on the conformal space . combining eqs . ( [ constraintr ] ) and ( [ conformal ] ) then gives [ constraint ] |d^2 - = - ^5 . if @xmath38 , so that we are considering a vacuum space - time , then this equation is linear in @xmath39 , which is promising if we are interested in looking for a multi - black hole solution @xcite-@xcite . although eq . ( [ constraint ] ) is not linear in @xmath18 , it can be made so by a suitable choice of @xmath16 . it is this equation that we will use to solve for black holes in a hyperspherical cosmology in section [ sec : sphere ] , in a spatially flat cosmology in section [ sec : flat ] and in a higher - dimensional cosmology in section [ sec : higher ] . given the efes , it can be seen that we can only obtain a bounce when the matter source violates the null energy conditions ( e.g. when @xmath40 and @xmath23 , which describes a massless scalar field with a negative energy density ) . however , bounces can be more naturally obtained with normal matter in modified gravity theories @xcite . in subsequent sections we will present exact solutions of the constraint equation ( [ constraint ] ) that can be used to develop an exact 4-dimensional dynamical solutions of the efes . these solutions , while highly idealised , describe realisations of the scenario in which multiple distinct black holes persist through a cosmological bounce . therefore , they both demonstrate the existence of solutions that could be generalised to more realistic models and provide space - time geometries in which specific computations can be made . however , in order to study the full time - evolution of these models we need to solve the full efes and not just the conservation and constraint equations . let us choose coordinates so that the full 4-dimensional space - time is given by the following line - element : [ dynmetric ] ds^2 = -a^2(t , x^ ) dt^2 + h _ ( t , x^ ) dx^ dx^. an evolving solution would be a metric of this form solving both the constraint and the evolution equations . for a dynamical model with a bounce at the time - symmetic surface at @xmath41 , the 3-metric we are interested in is given by @xmath42 ( as above ) . we therefore have [ hypmetric ] h _ ( t_0,x^ ) dx^ dx^ = ^4(x^)d|s^2 . we can then obtain an exact _ space - time _ by solving the evolution equations _ t h _ & = & -2 a k _ and _ t k _ & = & - d _ d _ a + a ( _ - 2 k _ k^ _ + k k _ ) + & & + ( p^ - ^ - 2 ^ _ - ^ ) h _ , where @xmath43 and @xmath44 is the anisotropic stress of the scalar field . as the initial value problem is well - posed , once appropriate initial values for the metric and extrinsic curvature functions have been specified , we are guaranteed a unique solution . in general , this will need to be obtained numerically @xcite-@xcite , although approximate solutions may also be possible @xcite-@xcite . at this point , to fix ideas , it may be useful to compare the inhomogeneous models we are constructing to analogous flrw models . for an isotropic and spatially homogeneous geometry , we have @xmath45 and @xmath46 , where @xmath47 is the scale factor that describes the cosmic evolution in its entirety . in the massless scalar field flrw models with dust , radiation , and spatial curvature there can of course exists a bounce , and the corresponding solution for @xmath47 can be easily obtained . in the absence of energy exchange between these fluids , the bounce at @xmath41 is generically expected to be time - symmetric . this means that at the minimum of the bounce , the geometry described by the conformal factor @xmath48 is instantaneously static . the time - symmetric initial data we considered above is the generalisation of this situation to inhomogeneous space - times that are capable of harbouring black holes . just as the flrw solution moves away from the minimum of expansion into a regular phase of expansion , we expect our initial data to evolve into a regular universe in which the scalar field dilutes , and the radiation and dust fields eventually come to dominate . that is , a set of instantaneously static initial data should not be expected to lead to a static space - time , but rather a dynamical cosmological model . this will be explained further in the sections that follow . let us start by considering hyperspherical cosmological models . we study this first not because it is more likely to be associated with a bounce , but because it is mathematically simpler when formulated in terms of an initial value problem . for a hyperspherical cosmology , we take the conformal line - element in eq . ( [ le ] ) to be [ hs ] d|s^2 = dr^2 + ^2 r ( d^2 + ^2 d ^2 ) . the scalar curvature of this space is @xmath49 , so the hamiltonian constraint ( [ constraint ] ) becomes |d^2 = ( - ( ^2 + 2 ^ ) ^4 ) . [ ham ] if we now choose the initial value of @xmath16 such that [ c ] ^2 = ( - ) - , where @xmath50 is a constant , then the constraint equation becomes [ helmholtz ] |d^2 = , so that @xmath51 if both @xmath40 and @xmath52 . this is the helmholtz equation on a 3-sphere , which has well - known solutions . it is linear in @xmath39 , so we can construct multi - black hole solutions by simple superposition . if @xmath53 it can be seen that the energy densities in the scalar field and radiation must sum to zero on the initial hypersurface . there exist solutions to eq . ( [ helmholtz ] ) of the following form : [ omi ] _ i ( r ) = _ i + _ i , where @xmath54 and @xmath55 are constants . these solutions diverge at @xmath56 , which we take to be the location of the source . in order for the geometry of the space to be smooth at all other points , we require the first derivative of @xmath39 to be single - valued . applying this condition at @xmath57 implies @xmath58 , so eq . ( [ omi ] ) gives [ cont0 ] _ i(r ) = _ i . this equation gives the contribution to the conformal factor for a single point - like source at @xmath56 . for @xmath59 sources , located at arbitrary positions on the hypersphere , we can write the corresponding solution as [ cont ] ( r , , ) = _ i=1^n _ i , where @xmath60 should be understood as the radial coordinate in eq . ( [ hs ] ) , after rotation of the hypersphere so that the @xmath61th source is located at @xmath62 . these solutions , corresponding to @xmath59 point sources , are characterized by : * complete freedom in the positions of the sources on a reference 3-sphere . * one number to be specified for each black hole ( @xmath54 ) . * a single number ( @xmath50 ) that sets the magnitude of the scalar field energy density . * a single number ( @xmath0 ) for the coupling strength of the scalar field . it now remains to determine the relationship between the constants @xmath54 , and the proper mass of each of the sources . to determine the proper mass of one of the sources , we start by taking the @xmath63 limit of eq . ( [ cont ] ) . this gives ds^2 ( + a_i ) ^4 ( dr_i^2 + r_i^2 ( d^2 + ^2 d ^2 ) ) , where a_i - + _ ji _ j and @xmath64 is the coordinate distance to the @xmath65th source from the one located at @xmath62 . we then define a new radial coordinate by @xmath66 , such that ds^2 ( 1 + ) ( dr^2_i + r_i^2 ( d^2 + ^2 d ^2 ) ) . this solution is dual to the original one , in the sense that @xmath56 and infinity have been interchanged ; this is because each black hole is represented in the initial data by an einstein - rosen - type bridge such that the dual solution is just the reflection of the original one in the apparent horizon . this new geometry can be compared with the schwarzschild metric in the limit @xmath67 , [ schwarz ] ds^2 ( 1 + ) ( dr^2 + r^2 ( d^2 + ^2 d ^2 ) ) , which gives the proper mass of the @xmath61th source as m_i = 2 _ i a_i = - + 2 _ ji _ i _ j . note that the mass of each source is a function of every @xmath68 and the position of every other source . this mass can be either positive or negative , depending on the value of constants @xmath50 , @xmath54 and @xmath64 . note that the expression above is regular at @xmath69 , as in these cases the trigonometric functions merely turn into hyperbolic ones . let us now consider a specific arrangement of masses . consider eight sources , all with equal mass , and all equidistant from each other on a 3-sphere . the positions of these masses is given in table [ table1 ] , where ( @xmath70 , @xmath71 , @xmath72 , @xmath73 ) are cartesian coordinates on a flat embedding 4-space and ( @xmath74 , @xmath75 , @xmath76 ) are hyperspherical polar coordinates on the 3-sphere . the scale factor , @xmath39 , and the energy density of the scalar field , @xmath77 , that result from this configuration are displayed in figs . [ om8 ] and [ ph8 ] . .the positions of eight regularly arranged points on a 3-sphere in the 4-dimensional euclidean embedding space with hyperspherical polar coordinates . [ cols="^ , < , < " , ] for @xmath78 , @xmath79 , @xmath80 , @xmath81 , @xmath82 , and @xmath83 regularly arranged black holes on a 3-sphere . the grey shaded region shows allowable values of @xmath50 , where black holes have positive mass and the cosmology still bounces.,width=566 ] [ results1 ] for @xmath78 , @xmath79 , @xmath80 , @xmath81 , @xmath82 and @xmath83 regularly arranged black holes on a 3-sphere . the grey shaded area shows the allowed region.,width=510 ] [ results2 ] for @xmath78 , @xmath79 , @xmath80 , @xmath81 , @xmath82 , and @xmath83 regularly arranged black holes on a 3-sphere . the grey shaded area shows the allowed region , where solutions exist.,width=510 ] [ results3 ] the results from table [ table2 ] are shown graphically in figs . [ results1]-[results3 ] . it can be seen from fig . [ results1 ] that the upper and lower bounds on @xmath50 generally increase with the number of black holes , although the lower bound decreases at @xmath84 . the grey area in this plot shows the allowed region for @xmath50 , in order that the black hole horizons have positive mass and the bounce occurs at all points along a cell edge . this band widens as @xmath59 increases but in a rather irregular way ( for the lower bound , at least ) . this irregularity is probably due to the different configurations having different shaped primitive cells , which is an unavoidable consequence of arranging the sources on the 3-sphere so they are equidistant from all their nearest neighbours . the fractional distance that the horizon can extend to the midpoint between sources , @xmath85 , is shown in fig . [ results2 ] . the grey region again shows the allowed region for each of the different values of @xmath59 . there is a clear trend , with the minimum allowed value of @xmath85 decreasing as @xmath59 increases however , the behaviour of the lower bound in fig . [ results2 ] is again a little irregular and it even increases somewhat between @xmath84 and @xmath83 . we again attribute this to the different shaped primitive cells in the different configurations . it is difficult to discern a trend in the upper bound of this plot , which stays around @xmath86 for every configuration . finally , in fig . [ results3 ] , we show the allowed values of @xmath87 . the trend in this case is clear , and increases as the number of masses in the universe is increased . in vacuum , there do not exist any time - symmetric black hole - lattice solutions with positive mass and zero spatial curvature @xcite . in the presence of a scalar field , however , these solutions do exist . to see this , we can integrate eq . ( [ constraint ] ) over a domain @xmath88 with periodic conditions and a number of masses @xmath89 . in the absense or radiation , this gives [ intflat ] _ d dv = _ d ^2 ^5 dv + 2 _ i m_i , where @xmath90 is the determinant of the metric of the conformal 3-space and @xmath91 is a volume element on that space . if we now choose the conformal space in eq . ( [ le ] ) to be 3-dimensional euclidean space , so that [ euclid ] d |s^2 = dx^2 + dy^2 + dz^2 , then @xmath92 and eq . ( [ intflat ] ) has solutions if we can ensure 16 _ i m_i = - _ d ^2 ^5 dx dy dz . for @xmath93 , we require the right - hand side of this equation to be positive , which implies @xmath29 and @xmath40 . in this section , we therefore assume the conformal euclidean geometry ( [ euclid ] ) and look for solutions to [ master ] |d^2 = - ^2 ^5 , which is what remains of the hamiltonian constraint ( [ hc ] ) . we will now study some of the solutions of this equation . to find solutions to eq . ( [ master ] ) , we choose [ piflat ] ^2 = , where @xmath94 is the ( constant ) energy density in the scalar field at the moment of time symmetry . we are free to make this choice as there is no constraint on the value of @xmath16 in the initial data ( see section [ sec : constraints ] ) . in this case , the hamiltonian constraint ( [ master ] ) becomes linear in @xmath39 , and therefore much easier to solve . to represent the point - like masses in our lattice , we add the following term to the right - hand side of eq . ( [ master ] ) : m _ n ^3 ^(3 ) ( x - l n ) = _ n ^3 e^i 2 /l . this describes a comb of dirac delta functions separated by distance @xmath95 and with magnitude @xmath96 . the locations in space where this term is non - zero will eventually correspond to points that will be excised from our geometry , with eq . ( [ master ] ) being satisfied at all remaining points . we can now look for a solution of the form = _ n _ * ^3 a_n e^i 2 /l + for some constant @xmath97 . here @xmath98 is the three - dimensional set of integers with @xmath99 removed and the @xmath100 are a set of functions that remain to be determined . we can immediately see that the zero - mode of eq . ( [ master ] ) gives @xmath101 , while all other modes give - ^2 a_n = _ 0 . a solution is therefore = - _ n _ * ^3 + , which is equivalent to [ psisol ] = - _ ( n , p , q ) _ * ^3 + . this solution has singular points at @xmath102 for all @xmath103 . it also has @xmath104 , where @xmath105 is a vector normal to the boundary @xmath106 of a primitive cell of the lattice ( again defined as the locus of all points closest to a given mass ) . these are exactly the properties required for an infinite array of regularly - arranged black holes on a euclidean 3-space . the functional form of @xmath39 is illustrated in fig . [ psiplot ] . oscillations arise in taking partial sums in eq . ( [ psisol ] ) , rather than the full ( infinite ) series . we consider this to be an artefact akin to the gibbs phenomenon that occurs when modelling a step function using fourier series . the oscillations have been smoothed out in this figure by taking the average of different partial sums . ) for a lattice of black holes in a @xmath107-dimensional euclidean conformal space . partial sums from @xmath108 in @xmath109 , @xmath110 and @xmath111 have been averaged over to remove ringing artefacts.,width=680 ] [ psiplot ] these solutions , corresponding to an infinite number of point sources on a spatially - flat lattice , are characterized by : * the distance between sources in the lattice ( @xmath95 ) . * the product of the coupling strength and energy density of the scalar field ( @xmath112 ) . * a constant in eq . ( [ psisol ] ) that sets the magnitude of @xmath39 . once again , it remains to determine the relationship between these quantities and the proper mass of each of the sources . to find the proper mass of the black holes in this space , we will again compare the geometry at a mass point with the corresponding limit of a time - symmetric slice through the schwarzschild solution . this latter is given in isotropic spatial coordinates by [ psis ] _ s = 1 + = 1 + d^3 k . if we define @xmath113 , then our solution can be written as [ psi2 ] = - _ n ^3 _ * ( k)^3 + , where @xmath114 is the difference between two successive values of @xmath115 in the sum . in the vicinity of a mass point , it is the small - scale contributions to the sum in eq . ( [ psi2 ] ) that are most important . these contributions are the terms with @xmath116 , so the sum of terms in eq . ( [ psi2 ] ) approaches the following integral expression in the vicinity of a mass point : [ psi3 ] - d^3 k + . comparison of eqs . ( [ psis ] ) and ( [ psi3 ] ) then gives @xmath117 and @xmath118 . if we now choose @xmath119 , we have m = _ 0 l^3 and the proper mass of each black hole is equal to the energy density of the scalar field times the spatial volume of the cell in the conformal space . in the conformal euclidean space . the five curves correspond to @xmath120 ( red ) , @xmath121 ( orange ) , @xmath122 ( yellow ) , @xmath123 ( green ) and @xmath124 ( blue ) . the minimal value of the area is indicated , in each case , by a black dot.,width=529 ] [ flatresults1 ] .,width=529 ] [ flatresults2 ] let us now consider the locations of the apparent horizons in these solutions . again , due to time - symmetry , these are given by minimal closed surfaces in the initial data . we estimate their position by calculating the area of a sphere of constant @xmath125 that is centred on a source , and varying the value of @xmath74 until this area has a local minimum . in this case , the area of such a sphere is a(r ) = _ 0 ^ 2 _ 0^ ^4 r^2 d d , where @xmath39 is given by eq . ( [ psisol ] ) . to find its numerical values , we take partial sums in eq . ( [ psisol ] ) and put @xmath126 . as discussed above , the partial sum results in gibbs - like ringing . we therefore take the average of different partial sums to remove this unwanted artefact . the results are displayed in fig . [ flatresults1 ] for five different values of @xmath127 . here we take the average of partial sums up to @xmath128 and @xmath129 in each of @xmath109 , @xmath110 and @xmath111 . also indicated in fig . [ flatresults1 ] are the positions of the minima . these correspond to the locations of the apparent horizon , whose size increases with @xmath127 , as expected . figure [ flatresults2 ] shows the position of the horizon as a function of @xmath127 . it can be seen that the horizons of neighbouring black holes touch when @xmath96 reaches m_max = 0.963 l . for larger values of @xmath96 , the black holes can no longer be considered as separate objects . this is therefore the minimum mass that can pass through a bounce before the black holes are forced to merge . it is very close to the number that one would obtain from a naive application of the schwarzschild radius @xcite , which is at @xmath130 in isotropic coordinates . finally , let us determine the acceleration of the scale factor along a cell edge , which is again given by eq . ( [ acc1 ] ) . it is straightforward to calculate this quantity for the flat lattice , but problems are introduced by the ringing phenomena discussed above . these are exacerbated in the present case because the acceleration of the scale factor is determined by the ricci scalar of the space ( contracted twice with a space - like unit vector ) . the derivatives required to calculate the ricci tensor make the ringing stronger than in the earlier case . one way of reducing the ringing is to increase the number of terms in the partial sum . in this section we therefore increase the maximum value of @xmath109 , @xmath110 and @xmath111 to @xmath131 in eq . ( [ psisol ] ) . the results of calculating @xmath87 along a cell edge in this case are displayed in fig . [ flatresults3 ] for @xmath132 . although the ringing is still present , and gets worse towards the centre of the cell edge , it is reasonably clear what the form of the underlying solution would be if the ringing were absent . along a cell edge in the flat lattice . the dots show the value obtained from a partial sum up to @xmath131 for @xmath109 , @xmath110 and @xmath111 . the red line that connects them shows the ringing phenomenon from the partial sum . the blue line is a fitted function , as described in the text . we have chosen @xmath132.,width=529 ] [ flatresults3 ] at the centre of a cell edge as a function of @xmath96 . the function increases monotonically with @xmath96 and is never negative while the mass is positive.,width=510 ] [ flatresults4 ] as we want to know the minimum value of @xmath87 , the ringing presents a particular problem because it is strongest at exactly the centre of a cell edge , where we expect this minumum to occur . we therefore proceed by fitting sums of sinusoidal functions to the @xmath133 of data points that are furthest from the centre of the cell edge . this allows smooth curves to be fitted to the data and values of @xmath87 at the centre of an edge to be determined . such a fit is shown by the blue curve in fig . [ flatresults3 ] , which in this case corresponds to f(x ) 0.129 - 0.044 ( 2 x ) - 0.002 ^2 ( 2 x ) . repeating this method for different values of @xmath96 gives fig . [ flatresults4 ] . this shows the minimum value of @xmath87 , obtained by fitting smooth curves to the data , and indicates that the acceleration of the scale factor increases monotonically with @xmath127 and without any bound , as far as we can determine . it also appears to be positive for all positive @xmath96 , which means that a bounce always occurs , no matter how low the mass in the black holes gets compared to the mass in the scalar field . we therefore have an upper bound on @xmath127 , above which black holes merge , but no lower bound . this is qualitatively different to the hyperspherical cosmologies considered in section [ sec : sphere ] . in some cosmological scenarios one might expect the number of spatial dimensions in the universe to increase at sufficiently high densities ( _ i.e. _ as one approaches the big bang ) . for example , if there were compactified extra dimensions , this would apply at times sufficiently early that the compactification scale exceeds the cosmological particle horizon size . this would modify both the bounce and merger conditions , so it is interesting to consider the consequences of extra spatial dimensions for the persistence of black holes through a cosmological bounce . the constraint equations ( [ hc ] ) and ( [ mc ] ) are valid for any number of extra spatial dimensions @xcite . with time - symmetric initial data , and in the absense of radiation , we therefore only require ^(n ) = 2 _ n ^ , where @xmath134 is a constant and @xmath135 is the ricci scalar of the @xmath109-dimensional space , the space - time itself being @xmath136-dimensional . if we now make a metric ansatz of the form [ higherds ] ds^2 = ^ ( dr^2 + ^2 r d |s^2_n-1 ) , where @xmath137 is the line - element of a @xmath138-dimensional unit sphere , and we neglect radiation , then the constraint equations are satisfied if [ highercon ] |d^2 = _ n , where @xmath37 is now the covariant derivative on the conformal @xmath109-sphere , @xmath139 and @xmath140 is defined by the last equality . this reduces to eq . ( [ ham ] ) if @xmath141 and @xmath52 . if we distribute our scalar field so that @xmath94 is constant , then @xmath140 is also constant . the constraint ( [ highercon ] ) can then be written in the helmholtz form : _ , rr + ( n-1 ) r _ , r = _ n . this equation is linear in @xmath39 for any value of @xmath109 and has solutions [ legendre ] _ i = _ i + _ i ^- ( r ) q^ _ ( -r ) , where @xmath54 and @xmath55 are constants of integration and -1 , ( -1 ) . the functions @xmath142 and @xmath143 in eq . ( [ legendre ] ) are the associated legendre polynomials , and we have included the additional ( constant ) denominator in the term involving @xmath142 so that we recover a non - zero result in the limit @xmath144 . due to the linearity of eq . ( [ highercon ] ) , one can sum any number of solutions of the form ( [ legendre ] ) to obtain a new solution . if either or both of @xmath54 and @xmath55 are non - zero , then the solution ( [ legendre ] ) diverges as @xmath145 . we can therefore take this to be the location of one of the sources in our higher - dimensional lattice . to ensure that the geometry is smooth away from this point , we require the spatial derivatives of @xmath146 to vanish at @xmath57 . if @xmath109 is even , this requires @xmath147 . if @xmath109 is odd , it requires @xmath148 . by rotating these solutions on the conformal @xmath109-sphere and setting the appropriate constants to zero , we therefore have initial data for @xmath59 arbitrarily located black holes in a positively - curved @xmath136-dimensional cosmology . to find the mass of each of the black holes in these solutions , we follow a similar approach to that in section [ spheremass ] . the first step is to transform coordinates so that the conformally spherical geometry in eq . ( [ higherds ] ) becomes conformally flat . such a transformation is given by @xmath149 , so the geometry of our initial surface becomes [ higherflat ] ds^2 = ( ) ^ ( d^2 + ^2 d |s^2_n-1 ) , where @xmath150 is a constant . a power series expansion of the overall conformal factor in this equation , around a point - mass at @xmath151 , then gives ds^2 = ( + + + ) ^ ( d^2 + ^2 d |s^2_n-1 ) , where @xmath97 , @xmath152 and @xmath153 are constants , and the dots denote smaller terms . finally , we perform an inversion , to exchange @xmath151 with @xmath154 . this can be achieved by the further transformation @xmath155 , where @xmath156 is a constant , to give [ hmass ] ds^2 = ( + + + ) ^ ( dr^ 2 + r^ 2 d schwarzschild - tangherlini solution @xcite , which is the generalization of the schwarzschild metric to higher dimensions : [ st ] ds^2 = ( 1 + ) ^ ( dr^2 + r^2 d|s^2_n-1 ) , where we have used isotropic coordinates @xcite . direct comparison of the first and second terms in the conformal factors of eqs ( [ hmass ] ) and ( [ st ] ) then gives the mass of the black hole at @xmath56 as [ hmass2 ] m = . some comments are required to explain the meaning of this result . if @xmath157 , then the two terms that have counterparts in the schwarzschild - tangherlini solution are not the largest terms that appear in the conformal factor of eq . ( [ hmass ] ) in the limit @xmath158 . it can , however , be verified that these larger terms all vanish when @xmath159 . this suggests that in higher - dimensional spaces , when the black hole is viewed from infinity on the far side of the einstein - rosen bridge , the leading - order part of the gravitational field comes not from the black hole but from the the scalar field , i.e. the energy in the scalar field appears to gravitate more strongly than the mass of the black hole at large distances . this situation is rather odd , and does not occur in @xmath160-dimensional einstein - scalar field systems . nevertheless , it appears to be unavoidable for higher - dimensional spaces and this makes it difficult to distinguish between the gravitational field of the black hole mass and the scalar field . operationally , we usually define the black hole mass through the leading - order part of its gravitational field at infinity , so if the leading - order part is itself due to the scalar field , this changes the meaning of the whole procedure . the best we can do is therefore to read off the coefficients in the expansion that look like the terms in the schwarzschild - tangherlini geometry and this leads to eq . ( [ hmass2 ] ) . one final point is that when the dimension of space is even , there also appear logarithmic terms in the power series expansion of the conformal factor . these also contribute to the gravitational field at large distances , but do not qualitatively change the above results , as there are no such terms in the schwarzschild - tangherlini geometry . in what follows , we use the expression ( [ hmass2 ] ) as the proper mass of each black hole in @xmath109 spatial dimensions . the values of @xmath97 and @xmath152 in each case will be calculated using the power series expansion of eqs . ( [ legendre ] ) and ( [ higherflat ] ) . we now investigate some of the properties of these higher - dimensional spaces , beginning with the acceleration of a cell edge at its midpoint ( the cell , as before , being defined as the locus of points closest to each black hole ) . the acceleration at this point is given by [ acc10 ] = - ^(n)(k , k ) , which is the same as eq . ( [ acc1 ] ) in the @xmath160-dimensional case but with the ricci tensor now being calculated using the geometry of the higher - dimensional space . this equation can be derived using the higher - dimensional evolution equations @xcite . ( red ) , @xmath162 ( orange ) , @xmath78 ( yellow ) , @xmath163 ( green ) , @xmath164 ( blue ) , @xmath79 ( purple ) , and @xmath165 ( mauve ) . the red curve is the same as in fig . [ acc8 ] , and the green and blue curves no longer have real values after @xmath166 and @xmath167 , respectively . the total mass in each configuration is @xmath168.,width=585 ] [ haccplot ] in order to evaluate this quantity for specific configurations , we consider the tilings on the @xmath109-sphere corresponding to the @xmath109-cube . these structures are obtained by placing the largest possible @xmath109-sphere inside an @xmath109-cube , so that the sphere touches the cube only at the centre of each of its square faces . if we mark these positions on the @xmath109-sphere , and then place a black hole at each of them , then each black hole is equidistant from each of its nearest neighbours . the @xmath107-dimensional example of such a lattice is given by the @xmath79-mass configuration studied in section [ sec:8mass ] . more generally , this procedure leads to a @xmath169-mass configuration on an @xmath109-dimensional hypersphere . we also choose the total mass of all of the black holes in each of these spaces to be equal to @xmath168 , so that that they can be easily compared with each other . the results of calculating the acceleration from eq . ( [ acc10 ] ) for each of the configurations are displayed in fig . [ haccplot ] . the @xmath107-dimensional results from fig . [ acc8 ] are shown in this plot as a red curve . the other six curves correspond to @xmath170 , @xmath78 , @xmath163 , @xmath164 , @xmath79 and @xmath165 . the value of @xmath140 required to make the acceleration positive , and hence cause a bounce , seems to increase with the number of dimensions . for @xmath170 , @xmath78 , @xmath79 and @xmath165 , the curves continue to increase monotonically with @xmath140 after crossing the line that corresponds to @xmath171 . unlike the @xmath141 case , the mass of each black hole ( @xmath96 ) remains positive for all values of @xmath140 . in these dimensions , we therefore only have a lower bound on the value of @xmath140 in order for a bounce to occur . the curves that correspond to dimensions @xmath172 and @xmath164 are qualitatively different . they start out with negative acceleration at low @xmath140 , just like the other cases , but the value of @xmath96 becomes zero at the point where @xmath171 . for larger values of @xmath140 , @xmath96 becomes negative above @xmath173 and @xmath167 , respectively . for values of @xmath140 larger than these bounds , we find that the value of @xmath174 is no longer real , so it seems that there are no bouncing cosmologies containing black holes of the type given in eq . ( [ higherds ] ) if there are @xmath163 or @xmath164 spatial dimensions . let us now consider the locations of the apparent horizons in the same hyper - cubic lattices , in dimensions @xmath141 , @xmath162 , @xmath78 , @xmath79 and @xmath165 . the apparent horizon is , in each case , the smallest sphere that encloses a mass point . as before , we will estimate the position of these surfaces by positioning a mass at @xmath56 , and then considering the area of spheres of constant @xmath74 . this area is given by [ hah ] a_n(r ) = ^2 ^n-1 ( r ) d_1 d_n-1 , where @xmath175 indicates the determinant of the metric of a unit @xmath138-sphere , and where the @xmath176 are the @xmath138-spherical polar coordinates on that sphere . in any given dimension , we look for the value of @xmath74 that minimizes this quantity . the numerical results we find are shown graphically in fig . [ hhorplot ] . as before , we show these results as a fraction of the distance to the midpoint between masses , @xmath177 . when a given curve crosses the line @xmath178 we therefore say that the apparent horizons of neighbouring black holes should be expected to merge . for the lower - dimensional cases , with @xmath141 and @xmath170 , it is possible to compute the integrals in eq . ( [ hah ] ) using standard deterministic techniques . however , the computational complexity of such an approach scales very disfavourably as the number of dimensions increases . for dimensions @xmath157 we therefore resort to monte carlo integration . this method places points randomly , and computes a statistical approximation to the true value of an integral . it also scales much more favourably with dimensionality , making the necessary numerical integrals possible within a feasible time - scale . the results of the monte carlo integrations are shown as crosses in fig . [ hhorplot ] . we have fitted curves to this data by performing a least squares fit to polynomial functions of order four , which are also displayed in the figure . ( red ) , @xmath162 ( orange ) , @xmath78 ( yellow ) , @xmath79 ( purple ) , and @xmath165 ( mauve ) . the red curve is the same as in fig . [ hor8 ] , and the crosses show the results from monte carlo integrals in the cases @xmath179 , @xmath79 and @xmath165 . the solid yellow , purple and mauve lines are best fitting polynomials of order four , using the method of least squares . the total mass in each configuration is @xmath168 . it is interesting to note that all the curves cross at @xmath180.,width=585 ] [ hhorplot ] we find , in each of the cases displayed , that the apparent horizons of neighbouring black holes should be expected to merge when the value of @xmath140 exceeds a certain bound . this bound is lowest for the @xmath107-dimensional case , for which we obtain @xmath181 . the bound increases by a small amount when an extra dimension is included , but subsequent extra dimensions do not appear to raise it much further ( to within the accuracy we can sensibly claim , given our statistical techniques ) . however , it appears that in every case we have @xmath182 . this is not a problem for the cases @xmath170 , @xmath78 and @xmath163 , as there are values of @xmath183 in each of these cases that can still produce a bounce ( see fig . [ haccplot ] ) . however , this is not true for the cases with @xmath184 or @xmath165 . in both of these cases the lowest value of @xmath140 required to produce a bounce is greater than the bound below which apparent horizons merge . it therefore appears that we can not produce a bounce in either of these dimensionalities without forcing neighbouring black holes to merge . taking both the requirement for a bounce and the requirement for neighbouring black holes to remain distinct consequently appears to place strong restrictions on these solutions in higher - dimensional spaces . in fact , it does not appear that we can produce a bounce if @xmath185 ( at least up until @xmath186 , which is as far as we have studied here ) . inclusion of one or two extra spatial dimensions does , however , remove the upper bound on the number density of black holes that exists in @xmath141 , and which comes from the requirement that the mass of each black hole is positive . there is no such bound in dimensions @xmath170 or @xmath78 , and so the black holes can safely be packed in tighter , all the way up until the point their horizons merge . in this paper we have considered bouncing cosmological models at the moment of maximum compression . our models contain a lattice of black holes in a universe whose energy density is dominated by a scalar field . we then obtained exact solutions for time - evolving models in which multiple distinct black holes persist through the bounce . first we considered hyperspherical cosmological models . we found that the model parameter @xmath50 must lie within a given range of values if the black holes are to have positive mass and the magnitude of the scalar field is to be large enough to ensure a bounce . a total of six configurations for multiple , regularly - arranged black holes were considered , and it was found that the upper and lower bounds on @xmath50 generally increase with the number of black holes . black holes in a spatially flat cosmology and in the presence of a scalar field were also considered , and were found to allow time - symmetric solutions with positive mass . we again found an upper bound above which black holes merge , but unlike the hyperspherical cosmologies we found no lower bound . finally , we considered the bounce and merger conditions on persistent black holes through a cosmological bounce in higher - dimensional spaces , which may be of relevance in cosmological scenarios in which the number of spatial dimensions in the universe increase at sufficiently high densities . the lower bound on the values of the model parameter @xmath140 required for a bounce seems to increase with the number of dimensions . however , it appears that that there are no bouncing cosmologies containing non - merging black holes of the type discussed in this paper if there are more than five spatial dimensions ( i.e. if @xmath185 ) . we also found that the upper bound on the number density of black holes that exists in @xmath141 , coming from the requirement that the mass of each black hole be positive , is removed when @xmath170 or @xmath78 . there are several areas in which the current analysis could be extended , and these will be the subject of future work . this includes the study of models in which the black holes persist without the bounce being time - symmetic , models in which the black holes merge on the approach to the bounce , and more detailed higher - dimensional models . beyond these more mathematical questions , there are also a number of potentially very interesting physical consequences of the models presented here , and these will be discussed elsewhere @xcite . for example , if there is more than one bounce then it is possible that there are more black holes in each subsequent cycle ( i.e. both the newly created ones and the persisting ones from previous cycles ) . in this case , the filling factor of black holes increases with each cycle , so that the bounces may at some point terminate . more speculatively , it is possible that bcbhs and pcbhs might act as the seeds for galaxy formation @xcite or even as the dark matter itself @xcite . this offer an interesting alternative to the currently popular suggestion that pbhs , formed at around a second after the big bang , might be responsible for these phenomena . let us now consider solutions with @xmath187 and @xmath188 , which contain a lattice of black holes . these solutions are perhaps of less cosmological interest than the solutions presented in secs . 3 , 4 , and 5 , but they constitute interesting generalisations of the scalar field solutions that are often referred to as janis - newman - winicour black holes @xcite ( although they were discovered by several previous authors @xcite ) . if @xmath187 , the momentum constraint ( [ mc ] ) is automatically satisfied on time - symmetric hypersurfaces . the hamiltonian constraint ( [ hc ] ) can then be written , in the absence of radiation , as = d^a d_a , which , for the hyperspherical geometry ( [ le ] ) , gives |d^2 = , where we have taken the conformal metric to be a hypersphere with @xmath49 . if we now transform variables from @xmath39 and @xmath18 to @xmath189 and @xmath190 , using = ^ ^ , = 2 , where @xmath191 is a constant and @xmath192 , then all constraints are satisfied if [ masterx ] |^2 = , = . the solutions to these equations i=1^n , = _ i=1^n , where @xmath60 is the @xmath74 coordinate for the @xmath61th source in a coordinate system rotated so that the source is at @xmath56 . the quantities @xmath54 and @xmath55 are constants to be specified for each of the @xmath59 terms in the sum . to determine the proper mass of the black holes in this solution , we consider the limit @xmath63 . the line - element of the space then becomes [ adm1 ] ds^2 + a_i ^2 ( 1- ) + b_i ^2 ( 1 + ) d|s^2 , where a_i = _ j i , b_i = _ j i , and @xmath64 is the coordinate distance to the @xmath65th source in the coordinate system centered on the @xmath61th source . if we now introduce a new radial coordinate , r_i^ , then the metric ( [ adm1 ] ) becomes ds^2 ( 1 + + ) d|s^2 . this can be compared to the schwarzschild solution ( [ schwarz ] ) in the limit @xmath67 to find the proper mass of the @xmath61th source as [ mscalar ] m_i = ( 1- ) a_i + ( 1 + ) b_i . the proper mass is again a function of every @xmath54 and @xmath55 . to determine the scalar charge for the black holes , let us define the scalar charge of a source as [ scalarcharge ] q_i _ , i n^i da , where the integral is carried out over a 2-sphere in the limit @xmath193 ( or @xmath63 ) . in this limit , the radial component of the unit vector @xmath194 is @xmath195 , while the area element is @xmath196 . substituting this into eq ( [ scalarcharge ] ) gives the scalar charge of the @xmath61th source as [ qscalar ] q_i = - 2 a_i + 2 b_i . again , this is a function of every @xmath54 and @xmath55 . for a given set of @xmath59 sources , and with a given set of @xmath54 and @xmath55 , we can now calculate @xmath89 and @xmath197 for each of them . note that the solutions given in section [ sec : sphere ] have no scalar charge , according to the definition ( [ scalarcharge ] ) . the solutions described in this appendix are valid for sources positioned at arbitrary locations on a 3-sphere . it is , however , instructive to compare them to the janis - newman - winicour solution for a static black hole in an asymptotically flat space . this is given in isotropic coordinates by @xcite ds^2 = - dt^2 + ( 1 - ) ^2 ( 1- ) ( 1 + ) ^2 ( 1 + ) ( dr^2 + r^2 ( d^2 + ^2 d^2 ) ) , with the scalar field being = 2 . a coordinate transformation @xmath198 with constant @xmath199 gives @xmath200 the conformal factor of the spatial part of the metric is then = ( 1- ) ^(1-)/2 ( 1 + ) ^(1+)/2 . if we want the horizon at @xmath130 to appear at @xmath201 in the new coordinates , we should choose @xmath202 . this gives = - ^(1-)/2 + ^(1+)/2 , where the solution is valid both inside and outside the horizon . , for the eight regularly arranged janis - newman - winicour - like black holes . the value of @xmath39 is proportional to the distance of the surface from the origin . we have chosen to taken @xmath203 , @xmath204 , @xmath205 , and @xmath152 negative to create this plot.,width=396 ] [ fig2 ] , for eight regularly arranged janis - newman - winicour - like black holes . the value of @xmath18 is proportional to the distance of the surface from the origin . here we have chosen the same parameter values as in fig . [ fig2].,width=396 ] [ fig1 ] it is now clear that the janis - newman - winicour solution corresponds to two black holes on a hypersphere , with @xmath189 and @xmath190 given by = - , = + . this prompts us to take the following parameters for these two black holes : _ 1 = , _ 2 = - , _ 1 = , _ 2 = . the proper mass and scalar charge on these black holes can then be read off from eqs . ( [ mscalar ] ) and ( [ qscalar ] ) as m_1 = m_2 = m , q_1=q_2 = 2 m . having considered the 2-mass configuration , which turns out to be isometric to a time - symmetric slice through the janis - newman - winicour solution , let us consider eight regularly arranged masses on a hypersphere ( as considered in section [ sec : sphere ] for the case @xmath24 ) . for this case we can choose @xmath206 which leads to masses and charges m_i = [ 3 ( 1 + ) + ] m , q_i = 2 ( 1 + 3 ) m . graphical representations of the values of @xmath39 and @xmath18 on @xmath201 slices are given in figs . [ fig2 ] and [ fig1 ] , respectively . using the relevant evolution equations , it is possible to determine that some parts of this space are at a minimum of expansion , while others are at a maximum . it can also be seen that ( at least ) one circle collapses all the way to zero size . therefore , although these solutions are interesting generalizations of the initial data of the janis - newman - winicour solution , they do not constitute promising cosmological models . further solutions with @xmath208 can be found by allowing @xmath209 . these solutions are of more cosmological interest but present difficulties in terms of their interpretation as a universe filled with black holes . if we take v ( ) = v_0 e^- , then eq . ( [ conformal ] ) implies , in the absence of radiation , = ^ , where @xmath39 is the conformal factor from eq . ( [ le ] ) , and @xmath189 obeys = . if @xmath210 , the solution to this equation is [ chiv ] = _ i _ i = _ i=1^n ( _ i + _ i ) . if @xmath211 , the trigometric functions in the numerators above are replaced by hyperbolic functions . in both cases the solution for the scalar field is = ( ) . if we now require that each term in the sum be smooth at @xmath212 , then we find _ ( [ chiv ] ) can now be recast as = _ i=1^n _ i . these solutions are characterized by : * complete freedom in the positions of the sources on a reference 3-sphere . * one number to be specified for each black hole ( @xmath54 ) . * a single number ( @xmath0 ) for the coupling strength of the scalar field . * a number ( @xmath213 ) that specifies the magnitude of the potential . * a number ( @xmath214 ) that parametrises the potential . while the geometry associated with these black holes is well defined , their proper mass is not . this is because the space on the far side of the einstein - rosen bridge of each black hole approaches minkowski space at a rate ( r^- ) . there is also a deficit angle in that space , making the process of defining mass highly problematic . we leave the possible further study of these solutions to future work . * acknowledgments . * we thank s. beheshti , a. graham and r. tavakol for helpful discussions and contributions . bc thanks the research center for the early universe ( resceu ) at the university of tokyo and the department of mathematics and statistics at dalhousie university and ac thanks the school of physics and astronomy at queen mary university of london , for hospitality received during this work . tc acknowledges support from the stfc and hospitality from the department of mathematics and statistics at dalhousie university . 99 g. lemaitre , _ mon . not . r. astron . soc . _ * 91 * , 490 ( 1931 ) .

effects of size and charge asymmetry between oppositely charged ions or particles on spatial inhomogeneities are studied for a large range of charge and size ratios . we perform a stability analysis of the primitive model ( pm ) of ionic systems with respect to periodic ordering using the collective variables based theory . we extend previous studies [ a. ciach et al . , phys . rev.e * 75 * , 051505 ( 2007 ) ] in several ways . first , we employ a non - local approximation for the reference hard - sphere fluid which leads to the percus - yevick pair direct correlation functions for the uniform case . second , we use the weeks - chandler - anderson regularization scheme for the coulomb potential inside the hard core . we determine the relevant order parameter connected with the periodic ordering and analyze the character of the dominant fluctuations along the-lines . we show that the above - mentioned modifications produce large quantitative and partly qualitative changes in the phase diagrams obtained previously . we discuss possible scenarios of the periodic ordering for the whole range of size- and charge ratios of the two ionic species , covering electrolytes , ionic liquids , charged globular proteins or nanoparticles in aqueous solutions and charge - stabilized colloids .

1207.1598

the study of phase diagrams of ionic systems in which the phase separation is mainly driven by electrostatic forces is of great fundamental interest and practical importance . electrolyte solutions , molten salts , ionic liquids and charge - colloidal suspensions are examples of systems with dominant coulomb interactions . the strong correlations between ions and counter - ions are also known to play an important role in determining structure and phase behavior of micelles , polyelectrolytes and proteins . over the last few decades , the phase behavior of ionic fluids has been the subject of many experimental , theoretical and simulation studies and the reviews of the state of the art in this field are available in refs . . a great deal of the research has been focused on the fluid - fluid phase separation , whereas the fluid - solid phase transition has received less attention so far . on the other hand , recent experimental , simulation and theoretical studies of room temperature ionic liquids ( rtils ) @xcite , charged globular proteins @xcite or colloidal particles @xcite reveal structural inhomogeneities of different type and spatial extent , and it becomes evident that the ordering in systems dominated by coulomb interactions has a very rich and complex nature , and is far from being understood . consequently , ordering of ions or charged particles on different length scales attracts increasing attention . the much studied simple salts or electrolytes are quite well understood . in these systems , the sizes of ions are similar . therefore , theoretical studies focused mainly on the restricted primitive model ( rpm ) where ions of the same valence are modeled by charged hard spheres of the same diameter immersed in a structureless dielectric continuum . at large enough volume fractions , transition to an ionic crystal occurs in molten salts when temperature decreases @xcite . in the ionic crystal the charge is periodically ordered and this ordering is subjected to the charge neutrality . on the other hand , charge - ordered , neutral living clusters of various sizes and life times were found in simulation studies of the fluid phase @xcite . a direct inspection of the structure in ionic systems is complicated due to a small size of the ions . however , in compliance with the law of corresponding states one my expect a similar behavior in oppositely charged colloidal particles of similar sizes under appropriate rescaling of the length and energy units . the natural length scale is the sum of radii of the anion and the cation , @xmath1 , while the energy unit is the coulomb potential between the oppositely charged ions at contact , @xmath2 , where @xmath3 is the charge of the @xmath4-type ion ( @xmath5 ) and @xmath6 is the dielectric constant of the solvent . if the non - coulomb interactions are negligible , the phase diagrams of systems consisting of spherical ions of various sizes should have the same form in terms of the volume fraction and a reduced temperature @xmath7 . indeed , simulations show that the phase diagram of charged particles with similar sizes in deionized solvent ( large debye screening length ) resembles the phase diagram of the rpm @xcite . in the above reduced units , the room temperature is very low for simple salts , whereas for colloidal particles it is high . ordering of colloids at room temperature is directly accessible to observations and vacancies in the crystal can be seen @xcite . formation of neutral aggregates in the rpm agrees with recent experimental studies of oppositely charged proteins of similar sizes @xcite . the aggregation was entirely suppressed when the electrostatic interactions had been screened by addition of a sufficient amount of salt . thus , the electrostatic interaction is essential at least for the first step of the self - assembly of oppositely charged globular proteins into aggregates . large spherical aggregates presumably indicate nucleation of a crystal . rtils differ from simple salts mainly because the sum of radii of the ions is much larger . this reduces the strength of the coulomb interactions at ion contact and , in turn , increases the room temperature in the reduced units . in addition , in many rtils the size and shape of the anion can be quite different from the size and shape of the cation . moreover , non - coulomb interactions may play an important role here . in such a case the rpm is an oversimplification , and the above discussion does not apply to the rtils . recent experimental and simulation studies on a series of imidazolium - based rtils indicate spatial inhomogeneities on the length scale @xmath8 nm @xcite . the inhomogeneous structure of liquid resembles the structure of the crystal @xcite . the origin of such nanoheterogeneity has not been convincingly explained yet , and new studies are required . in a mixture of positively and negatively charged globular proteins , the size difference between the positively and negatively charged molecules can be similar to the size difference between the anion and the cation in the rtils . recent experimental studies show that the size difference plays an important role for the assembly of proteins into aggregates . according to ref . @xcite , formation of spherical aggregates of oppositely charged proteins with overall charge near zero requires charge- and size compensation . the size difference between charged globular proteins and counterions in a solution is much larger than in the previously described cases . for example , the diameter of a lysozyme molecule is of order of @xmath9 nm , i.e. , it is 10 times larger than the counterions in an aqueous solution . the charge of small globular proteins is of order of @xmath10 , where @xmath11 is the elementary charge . scattering experiments @xcite suggest clustering of protein molecules in water . effective interactions between the lysozyme molecules were shown to have a form of short - range attraction and long - range repulsion the latter resulting from the charges on the molecules @xcite . the effective attraction between two molecules is caused by the interactions between them and the counterions attracted to both molecules , as well as by the non - coulomb forces . simulations carried out in ref . @xcite for the potential derived in ref . @xcite show that the fraction of clustered molecules and the shape of clusters strongly depend on the lysozyme volume fraction . charged nanoparticles or colloidal particles with a diameter @xmath12 nm are several orders of magnitude larger than the microscopic counterions in a solution , and their charge can be orders of magnitude larger than the charge of the counterion . in these systems , the experiments indicate the formation of colloidal crystals with large interparticle distance for small volume fractions , and a re - entrant melting for larger volume fractions @xcite . from the above description of different experimental systems it follows that spatial inhomogeneities are common , but their extent and nature still need to be comprehended and classified for different sizes and charges of the anion and the cation . in the first step , it is essential to predict the structure formation and phase diagram for various size and charge ratios for a generic model where non - coulomb interactions , deviations from a spherical shape and flexibility of molecules are neglected . the role of the above factors can be determined after comparing the results obtained within a generic model with the experiments . the generic model that allows one to predict the phase separation driven exclusively by coulomb forces is a primitive model ( pm ) . in this model , the ionic fluid is described as an electroneutral mixture of charged hard spheres immersed in a structureless dielectric continuum . the pm pair potential for two ions @xmath4 and @xmath13 at distance @xmath14 apart is @xmath15 where an ion of species @xmath4 has a diameter @xmath16 , and charge @xmath3 , @xmath17 and @xmath6 is the dielectric constant . the pm is the simplest model for all the above discussed systems . the two - component pm can be characterized by the parameters of size and charge asymmetry : @xmath18 for @xmath19 one arrives at the rpm . in the pm with large difference in the ion radii , the spatial distribution of the ions is expected to be quite different from the rpm , because the tendency for minimizing the electrostatic energy competes with the geometrical restriction on packing of spheres with different sizes @xcite . packing of large and small spheres that maximizes the entropy could lead to the formation of mesoscopic charged regions , and , on the other hand , when the electrostatic energy is minimized , periodic pattern involving voids could be formed . as a result of the competition between maximizing entropy and minimizing energy , both the charge and the number - density of ions can oscillate in space . the systematic studies of the effects of size and charge asymmetry on the periodic ordering in pms were initiated in ref . @xcite within the framework of the field theoretical description . based on a mean - field stability analysis , the authors found the boundaries of stability of the disordered phase for the whole range of @xmath0 and @xmath20 . it was shown that , besides a gas - liquid separation , in a certain portion of the phase diagram , the uniform fluid became unstable with respect to the order parameter oscillations of wavelengths @xmath21 with @xmath22 . the line in the phase diagram corresponding to the instability of the disordered phase with respect to periodic ordering is called the @xmath0-line @xcite to distinguish it from the spinodal line for which @xmath23 . the results obtained in ref . @xcite show that ( i ) the periodic ordering mainly depends on the size asymmetry ; ( ii ) the qualitative dependence on the charge asymmetry is found only for a sufficiently large size asymmetry . in this paper we continue the systematic study of the periodic ordering in asymmetric pms . we extend the previous study in several ways . the first modification concerns an approximate description of the reference hard - sphere mixture . in ref . @xcite a local - density approximation is employed for a hard - sphere free energy functional . here , we consider a non - local approximation for the reference hard - sphere fluid which is shown to lead to the percus - yevick ( py ) theory for the uniform case @xcite . in our calculations we use the lebowitz s solution of the generalized py equation @xcite . another modification concerns the regularization of the coulomb potential inside the hard core . it is worth noting that in the treatments of models with hard cores , the perturbation potential is not defined uniquely inside the hard core . here we use the weeks - chandler - andersen ( wca ) regularization scheme for the coulomb potentials @xmath24 @xcite @xmath25 instead of @xmath26 adopted in @xcite . as was shown in ref . @xcite , the simple form for @xmath24 given in eq . ( [ wca ] ) produces rapid convergence of the series of the perturbation theory for the free energy . on the other hand , the best theoretical estimates for the gas - liquid critical point of the rpm was obtained using the wca regularization scheme @xcite . finally , we extend the study of the relevant order parameter ( op ) undertaken in @xcite . following the ideas of refs . @xcite , we determine the op connected with the phase transition to an ordered phase and analyze the character of the dominant fluctuations along the @xmath0-lines associated with the periodic ordering . a theoretical background for this study is the statistical field theory that exploits the method of collective variables ( cvs ) @xcite . the theory enables us to derive an exact expression for the functional of grand partition function ( gpf ) of the model and on this basis to develop the perturbation theory @xcite . as was shown in ref.@xcite , the well - known approximations for the free energy , in particular debye - hckel limiting law and the mean spherical approximation , can be reproduced within the framework of this theory . links between this approach and the field theoretical approach @xcite were established in @xcite for the case of the rpm . our paper is organized as follows . in section 2 we give some brief background to the cvs based theory for the pm . based on the gaussian approximation of the functional of gpf we obtain the pair direct correlation functions and determine the op characterizing the periodic ordering in pms . in section 3 we study the effects of size and charge asymmetry on the periodic ordering taking into account the above - listed modifications . we discuss the results in section 4 and conclude in section 5 . we start with the general case of a two - component pm consisting of @xmath27 cations carrying a charge @xmath28 of diameter @xmath29 and @xmath30 anions carrying a charge @xmath31 of diameter @xmath32 . the ions are immersed in a structureless dielectric continuum . the system is electrically neutral : @xmath33 and @xmath34 is the number density of the @xmath4th species . the pair interaction potential is assumed to be of the following form : @xmath35 where @xmath36 is the interaction potential between the two additive hard spheres of diameters @xmath16 and @xmath37 . we call the two - component hard - sphere system a reference system . thermodynamic and structural properties of the reference system are assumed to be known . @xmath38 is the coulomb potential given in eq . ( [ wca ] ) , and hereafter we put @xmath39 . using the cv method we get an exact functional representation of gpf for the pm with size and charge asymmetry @xcite @xmath40&=&\int ( { \rm d}\rho)({\rm d}\omega)\exp\left ( -\frac{\beta}{2v}\sum_{\alpha,\beta}\sum_{{\mathbf k}}\tilde \phi_{\alpha\beta}^{c}(k)\rho_{{\mathbf k},\alpha}\rho_{-{\mathbf k},\beta}+\right.\nonumber\\ & & \left . + { \rm i}\sum_{\alpha}\sum_{{\mathbf k}}\omega_{{\mathbf k},\alpha}\rho_{{\mathbf k},\alpha}+\ln \xi_{\rm{hs}}[\bar \nu_{\alpha}-{\rm i}\omega_{\alpha } ] \right ) . \label{2.3}\end{aligned}\ ] ] in eq . ( [ 2.3 ] ) @xmath41 is the cv which describes the value of the @xmath42-th fluctuation mode of the number density of the @xmath4th species , each of @xmath43 ( @xmath44 ) takes all the real values from @xmath45 to @xmath46 ; @xmath47 is conjugate to the cv @xmath48 ; @xmath49 and @xmath50 are volume elements of the cv phase space @xmath51 and the product over @xmath52 is performed in the upper semi - space ( @xmath53 , @xmath54 ) . @xmath55 is the fourier transform of the coulomb potential . in the case of the wca regularization ( see eq . ( [ wca ] ) ) we obtain for @xmath56 @xcite : @xmath57 where the following notations are introduced : @xmath58 is the dimensionless temperature , @xmath59 , @xmath60 and @xmath61 similarly , hereafter we introduce the parameter @xmath62 @xmath63 the parameters @xmath64 and @xmath62 are more convenient than the parameters @xmath0 and @xmath20 because they vary between @xmath65 and @xmath66 . following ref . @xcite , we choose the dimensionless temperature @xmath67 given in ( [ temp ] ) and the volume fraction of all ions @xmath68 as thermodynamic variables . @xmath69 $ ] is the gpf of a two - component hard - sphere system with the renormalized chemical potential @xmath70 in the presence of the local field @xmath71 . in ( [ 2.7 ] ) @xmath72 is the dimensionless chemical potential , @xmath73 , @xmath74 is the chemical potential of the @xmath4th species , @xmath13 is the reciprocal temperature , @xmath75 is the inverse de broglie thermal wavelength . in order to develop the perturbation theory we present @xmath76 $ ] in the form of the cumulant expansion @xmath77&=&\sum_{n\geq 0}\frac{(-{\rm i})^{n}}{n!}\sum_{\alpha_{1},\ldots,\alpha_{n } } \sum_{{\mathbf{k}}_{1},\ldots,{\mathbf{k}}_{n } } { \mathfrak{m}}_{\alpha_{1}\ldots\alpha_{n}}(\bar\nu_{\alpha } ; k_{1},\ldots , k_{n})\times\nonumber\\ & & \times \omega_{{\bf{k}}_{1},\alpha_{1}}\ldots\omega_{{\bf{k}}_{n},\alpha_{n } } \delta_{{\bf{k}}_{1}+\ldots + { \bf{k}}_{n } } , \label{2.11}\end{aligned}\ ] ] where @xmath78 is the kronecker symbol . in eq . ( [ 2.11 ] ) the @xmath79th cumulant @xmath80 coincides with the fourier transform of the @xmath79-particle connected correlation function of a two - component hard - sphere system @xcite . having set @xmath81 for @xmath82 , after integration in eq . ( [ 2.3 ] ) over @xmath83 one arrives at the gaussian approximation for the functional of gpf @xcite @xmath84=\xi_{\rm{mf}}[\bar\nu_{\alpha}]\;\xi'\int(\mathrm{d}\rho ) \exp\big\{-\frac{1}{2v}\sum_{\alpha,\beta}\sum_{\bf k}\tilde{\cal c}_{\alpha\beta}(k)\rho_{{\bf k},\alpha}\rho_{-{\bf k},\beta}\big\ } , \label{ksi - g}\end{aligned}\ ] ] where @xmath85 is the gpf in the mean - field approximation and @xmath86^{-1/2}$ ] with @xmath87 being the matrix of elements @xmath88 . @xmath89 is the fourier transform of the pair direct ( vertex ) correlation function in the random phase approximation ( rpa ) @xmath90 in ( [ c - alpha - beta ] ) , @xmath91 is the fourier transform of the pair direct correlation function of a two - component hard - sphere system . it is connected with @xmath92 by the relation @xmath93 , where @xmath94 denotes the matrix of elements @xmath95 and @xmath96 is the unit matrix . in the limit @xmath97 , the @xmath98 coincides with the coefficients @xmath99 which are obtained in @xcite as a result of the local - density approximation . it is convenient to introduce cvs which describe the fluctuation modes of the total number and charge density , @xmath100 and @xmath101 , by the relations @xmath102 then , eq . ( [ ksi - g ] ) can be rewritten in terms of @xmath100 and @xmath101 as follows : @xmath84=\xi_{\rm{mf}}[\bar\nu_{\alpha}]\;\xi'\int(\mathrm{d}\rho_{n})(\mathrm{d}\rho_{q } ) \exp\big\{-\frac{1}{2v}\sum_{a , b}\sum_{\bf k}\tilde{\cal c}_{ab}(k)\rho_{{\bf k},a}\rho_{-{\bf k},b}\big\ } , \label{ksi - g1}\end{aligned}\ ] ] where @xmath103 and @xmath104 , \nonumber \\ \tilde{\cal c}_{qq}(k)&=&\frac{1}{(1+z)^{2}}\left[\tilde{\cal c}_{++}(k)+\tilde{\cal c}_{--}(k)-2\tilde{\cal c}_{+-}(k)\right ] , \nonumber \\ \tilde{\cal c}_{qn}(k)&=&\frac{1}{(1+z)^{2}}\left[\tilde{\cal c}_{++}(k)-z\tilde{\cal c}_{--}(k)+(z-1)\tilde{\cal c}_{+-}(k)\right ] \label{c - a - b}\end{aligned}\ ] ] are the density - density , charge - charge and charge - density direct correlation functions , respectively . in general , an equation for the boundary of stability of the uniform phase with respect to fluctuations is given by @xmath105 where @xmath106 denotes the matrix of elements @xmath107 ( or @xmath89 ) . the corresponding wave vector @xmath108 is determined from the equation @xcite @xmath109 the case @xmath110 corresponds to the gas - liquid - like separation @xcite . here we are interested in the @xmath0-line , the boundary of stability associated with fluctuations of the op with @xmath111 . on the @xmath0-line , the fluid becomes unstable with respect to the periodic ordering indicating that there can be a phase transition to an ordered phase . the determination of the op is the important issue in the phase transition theory of mixtures . this problem has got a consistent and clear solution within the given approach . in order to determine the op associated with the periodic ordering we follow the ideas of refs . first , we diagonalize the square form in eq . ( [ ksi - g1 ] ) by means of an orthogonal transformation @xmath112 the explicit expression for coefficients @xmath113 are given in appendix . the corresponding eigenvalues @xmath114 and @xmath115 are found to be @xmath116^{1/2}\right ) . \label{epsilon - i}\ ] ] based on the solutions of equations ( [ spinodal])-([k_b ] ) we find the eigenvalue which becomes equal to zero along the calculated @xmath0-lines for the fixed values of parameters @xmath64 and @xmath62 . we suggest that the corresponding eigenmode is connected with the relevant op . ): the dependence of eigenvalues @xmath117 and @xmath118 on the wave numbers ( @xmath59 ) . solid and dashed lines correspond to @xmath119 ( @xmath120 , @xmath121 ) and @xmath122 ( @xmath120 , @xmath123 ) , respectively . , height=226 ] our analysis shows that only one of the eigenvalues , i.e. , @xmath115 , becomes zero along the @xmath0-line of the both symmetric and asymmetric pms . the same result was obtained earlier for a mixture of neutral particles @xcite . in figs . 1 and 2 we show a typical behaviour of @xmath114 and @xmath115 under thermodynamic conditions corresponding to the points located on the @xmath0-lines associated with @xmath124 . we assume that cv @xmath125 is connected to the relevant op . based on eqs . ( [ ksi1])-([ksi2 ] ) we can determine the direction of strong fluctuations along the @xmath0-line by the relation @xmath126 where @xmath127 is the rotation angle of axes @xmath128 and @xmath129 in the plane ( @xmath130,@xmath131 ) . the case @xmath132 corresponds to the pure charge density fluctuations ( axes @xmath129 and @xmath131 coincide ) and the case @xmath133 corresponds to the pure total number density fluctuations ( axis @xmath129 coincides with axes @xmath134 , respectively ) . taking into account the formulas from appendix ( eqs . ( [ tij])-([alpha_ij ] ) ) , we have @xmath135 it is worth noting that in the long - wavelength limit @xmath136 , and one gets @xmath137 in agreement with the expected separation into homogeneous charge neutral dilute and dense phases . , @xmath119 ) : the dependence of eigenvalues @xmath117 and @xmath118 on the wave numbers ( @xmath59 ) . solid and the dashed lines correspond to @xmath120 , @xmath138 and @xmath139 , @xmath140 , respectively.,height=226 ] in ref . @xcite the analysis was similar , except that the eigenvalues denoted by @xmath141 and @xmath142 were defined directly in terms of @xmath143 in such a way that @xmath144 when @xmath145 and @xmath146 when @xmath147 . @xmath141 and @xmath142 defined in ref . @xcite reduce to the direct correlation functions for the charge and for the number density , respectively for @xmath148 ( i.e. for the rpm ) , and the corresponding eigenmodes reduce for @xmath148 to the charge and the number density waves . the advantage of the present approach is that the critical mode is associated with the same eigenvalue @xmath149 independently of @xmath150 , @xmath151 and @xmath152 . it is also more convenient to present the nature of the eigenmode in terms of the angle @xmath127 , rather than in terms of the parameter @xmath153 introduced in ref . @xcite in a way analogous to @xmath154 in eq.([theta ] ) . in this section we study the effects of size and charge asymmetry on the boundary of stability against the fluctuations with @xmath155 . to this end , we use eqs . ( [ spinodal])-([k_b ] ) taking into account eqs . ( [ coulomb_wca-1])-([coulomb_wca ] ) . in order to determine the character of the dominant fluctuations we use eq . ( [ theta ] ) . we take into account the @xmath150-dependence of the direct correlation functions of the reference system using an exact solution of the generalized py equation obtained by lebowitz @xcite . the explicit expressions for the fourier transforms of the oz partial direct correlation functions of a two - component hard sphere system are given in refs . they are too cumbersome to be reproduced here . as in ref . @xcite , we distinguish the three regimes in size asymmetry : small size asymmetry , moderate and large size asymmetry , and very large size and charge asymmetry . each regime is characterized by a typical behaviour of the @xmath0-lines . as we will see below , the @xmath64-ranges of these regimes slightly differ when compared to @xcite . we also dwell briefly on the size - symmetric case . we start with a size - symmetric pm corresponding to @xmath156 ( or @xmath157 ) . in fig . 3 the @xmath0-line associated with the wave vector @xmath124 is displayed . as is seen , it is a straight line identical to that obtained for the rpm @xcite : @xmath158 . for the wca regularization , @xmath159 and @xmath160 @xcite . -line for the transition to the ordered phase for @xmath156 . temperature @xmath67 and the volume fraction of ions @xmath151 are in dimensional reduced units defined in eqs . ( [ temp ] ) and ( [ fraction ] ) , respectively.,height=226 ] for the size - symmetric pm , the angle @xmath127 indicating the direction of the strong fluctuations is equal to zero . therefore , the @xmath0-line shown in fig . 3 is the boundary of stability of the uniform phase against the charge density fluctuations . -lines for the transition to the ordered phase for @xmath161 ( @xmath162 and @xmath163 ) . lines from the top to the bottom : @xmath122 , @xmath164 , @xmath119 , @xmath165 , @xmath166 , @xmath167 , @xmath168 . temperature @xmath67 and the volume fraction of ions @xmath151 are in dimensional reduced units defined in eqs . ( [ temp ] ) and ( [ fraction ] ) , respectively.,height=245 ] the charge asymmetry has no effect on the @xmath0-line of the size - symmetric pm . this property does not persist if the higher - order terms are taken into account in eq . ( [ 2.11 ] ) ( see refs . the effects of charge - density fluctuations on a phase behaviour of rpm ( @xmath156 , @xmath165 ) were studied in ref . @xcite using the field theoretical description . it was found that in the presence of fluctuations , the @xmath0-line disappears . instead , a fluctuation - induced first - order transition to an ionic crystal appears . interestingly , the wavelength of the charge wave is independent of the density along the liquid - crystal coexistence line @xcite , which was interpreted as the formation of vacancies in the crystal of a fixed unit cell when the density of ions decreases . we expect the similar situation to take place for the pm with the small size asymmetry . now we consider the asymmetric pms with a small size asymmetry ( @xmath169 or @xmath170 ) . this case corresponds to molten salts , electrolytes , some rtil and to oppositely charged globular protein or nanoparticle mixture in the limit of infinite screening length . the @xmath0-lines associated with the wave vectors @xmath124 are shown in fig . 4 for @xmath161 . it should be noted that the @xmath0-lines are located at the temperatures that are by an order of magnitude lower than those in ref . @xcite . in addition , the monotonously increasing behaviour of the @xmath0-line temperature , @xmath67 versus @xmath151 , is found for a sufficiently large charge asymmetry ( @xmath171 ) . for @xmath172 , @xmath67 has got a maximum in the range @xmath173 . the effect comes into prominence with an increase of @xmath64 . this differs from the previous results @xcite demonstrating a monotonous increase of @xmath67 with @xmath151 along the @xmath0-lines ( @xmath174 ) for all @xmath62 . the difference is directly related to the non - local approximation for the reference hard - sphere system adopted in the present work . as is seen from fig . 4 , the @xmath0-line temperature increases when @xmath62 increases from @xmath175 to @xmath176 which qualitatively agrees with @xcite . corresponding to the ordering of ions along the @xmath0-lines shown in fig . 4 for @xmath161 ( @xmath177 and @xmath163 ) . lines from the bottom to the top : @xmath122 , @xmath164 , @xmath119 , @xmath165 , @xmath166 , @xmath167 , @xmath168 . @xmath151 is the volume fraction of ions.,height=245 ] for a small size asymmetry , the wave numbers characterizing the period of the op oscillations in a non - uniform phase are @xmath178 and their magnitudes depend very slightly on @xmath151 ( see fig . 5 ) . the comparison with ref . @xcite implies that the magnitude of @xmath179 is mainly determined by the regularization method of the coulomb potential inside the hard core . in fig . 5 we also demonstrate the effect of charge asymmetry on @xmath179 . as is seen , @xmath179 increases with the variation of @xmath62 from @xmath176 to @xmath175 for the fixed @xmath151 which agrees with the results obtained in @xcite . along the @xmath0-lines presented in fig . 4 ( @xmath162 and @xmath163 ) . @xmath127 is measured in degrees.,height=226 ] we have calculated the angle @xmath127 showing the direction of strong fluctuations along the @xmath0-line . the results for @xmath161 are presented in fig . as is seen , @xmath127 changes around zero and this change increases with an increase of size asymmetry . as regards the charge asymmetry , the modulus of @xmath127 decreases with the variation of the charge asymmetry parameter @xmath62 from @xmath176 to @xmath175 for the fixed @xmath64 , and @xmath180 approaches zero for @xmath168 . in particular , for @xmath161 the angle @xmath127 changes continuously in the range from @xmath181 to @xmath182 for @xmath122 and from @xmath183 to @xmath184 for @xmath168 when @xmath151 is varied from @xmath185 to @xmath186 . for @xmath187 we have @xmath188 ( @xmath122 ) and @xmath189 ( @xmath168 ) in the same range of @xmath151 . in general , the charge density fluctuations are the dominant fluctuations for @xmath190 . it should be noted that for a small size asymmetry , the @xmath0-lines associated with the periodic ordering are located at much higher temperatures than the spinodals indicating phase separation into two uniform phases . summarizing , in the pm with a small size asymmetry we expect the phase transition to an ionic crystal with a compact unit cell where nearest neighbours are oppositely charged , at least when @xmath151 is sufficiently large . in particular , the @xmath191 crystal was observed experimentally for the system of oppositely charged colloids of comparable sizes @xcite . for low volume fraction compact charge - ordered clusters are expected , in agreement with experiments for oppositely charged proteins @xcite and with simulations @xcite . let us consider the case of moderate and large size asymmetry corresponding to @xmath192 ( @xmath193 ) . in fact , pms with @xmath194 demonstrate a crossover - type behaviour . the @xmath0-lines and the corresponding wave numbers for @xmath194 are shown in figs . 7 and 8 , respectively . it follows from our calculations that pms with @xmath194 and @xmath195 do not undergo an instability with respect to fluctuations with @xmath22 . this is contrary to the results obtained for pms with a small size asymmetry ( see fig . 4 ) . similar to the small size asymmetry case , the @xmath0-line temperature decreases when @xmath62 varies from @xmath176 to @xmath196 . for @xmath167 , the @xmath0-line associated with @xmath124 is located at @xmath197 slightly lower than @xmath197 at the spinodal indicating the separation in two uniform phases for the same @xmath151 . -lines for the transition to the ordered phase for @xmath194 ( @xmath177 and @xmath163 ) . temperature @xmath67 and the volume fraction of ions @xmath151 are in dimensional reduced units defined in eqs . ( [ temp ] ) and ( [ fraction ] ) , respectively . the curve for @xmath167 is located at a very low temperature and is indistinguishable in this plot.,height=226 ] as is seen from fig . 8 , the dependence of the wave numbers @xmath179 on the volume fraction @xmath151 is more prominent than that shown in fig . 5 . in particular , @xmath179 is an increasing function of @xmath151 for @xmath198 . such behavior is consistent with the decrease of the interparticle separation for decreasing average volume per particle for the same type of structure . for @xmath199 , @xmath179 first very slowly increases and then again slowly decreases when volume fraction increases . for @xmath200 , @xmath179 rapidly decreases . for very small values of @xmath151 , @xmath201 ( @xmath202 ) without regard to the charge asymmetry . corresponding to the ordering of ions along the @xmath0-lines shown in fig . 7 for @xmath194 @xmath151 is the volume fraction of ions.,height=226 ] + the variation of the direction of strong fluctuations along the @xmath0-lines ( angle @xmath127 ) is shown in fig . 9 . as is seen , the charge density fluctuations still prevail over the total number density fluctuations for @xmath203 except for a high - density region ( see the case @xmath122 ) . the fluid - solid phase coexistence in the pm with @xmath194 and @xmath165 was studied by computer simulations in ref . in particular , it was found that there is a coexistence at @xmath204 between a fluid at @xmath205 and a ( @xmath206 ) solid phase at @xmath207 . our calculations of the @xmath0-line temperature corresponding to @xmath205 yield @xmath208 ( see fig . 7 ) . at this point , the order parameter is connected to the charge density fluctuations ( @xmath209 ) suggesting the phase transition to the ionic crystal . for @xmath210 all the @xmath0-lines associated with @xmath22 demonstrate a single maximum for the volume fraction @xmath211 ( see figs . 10 , 11 and 14 ) . this implies that the periodic ordering is less favorable at higher volume fractions . this behaviour is similar to that found in @xcite but for the higher values of @xmath64 , namely for @xmath212 . it is worth noting that the @xmath0-lines lie at temperatures lower than those found in @xcite . this tendency is kept for the whole region of the variation of @xmath64 . on the other hand , the comparative analysis performed within the same regularization scheme , i.e. , the wca regularization , shows that the account of the @xmath150-dependence of the reference system correlation functions leads to a pronounced increase of the @xmath0-lines temperatures for a moderate and large size asymmetry ( see fig . such strong impact of the above - mentioned @xmath150-dependence on the magnitude of the @xmath0-lines temperature is contrary to the results found for a one - component fluid with competing attractive and repulsive interactions @xcite . along the @xmath0-lines presented in fig . @xmath127 is measured in degrees.,height=226 ] -lines for the transition to the ordered phase for @xmath122 and for two values of the size asymmetry ( @xmath213 and @xmath214 ) . solid and dashed lines correspond to a non - local and local approximations for the hard - sphere reference system , respectively . temperature @xmath67 and the volume fraction of ions @xmath151 are in dimensional reduced units defined in eqs . ( [ temp ] ) and ( [ fraction ] ) , respectively.,height=226 ] unlike ref . @xcite , our results demonstrate the dependence of @xmath211 on the both parameters @xmath64 and @xmath62 . in general , @xmath211 decreases noticeably with an increase of @xmath64 and has a non - monotonous behavior with the variation of @xmath62 . by contrast , the maximal value of the @xmath0-lines temperature slightly depends on the size asymmetry for @xmath215 . let us describe in some detail the case @xmath213 ( @xmath216 , volume ratio @xmath217 ) , characterizing moderate size asymmetry . @xmath213 can be found in a solution of charged molecules of a diameter @xmath218 nm in a presence of counterions of a diameter @xmath219 nm , or in some rtil and finally in a mixture of oppositely charged globular proteins or nanoparticles in deionized solvent . -lines for the transition to the ordered phase for @xmath213 . temperature @xmath67 and the volume fraction of ions @xmath151 are in dimensional reduced units defined in eqs . ( [ temp ] ) and ( [ fraction ] ) , respectively.,height=226 ] for @xmath213 , the wave number @xmath179 increases along the @xmath0-line from @xmath220 for @xmath221 to @xmath222 for @xmath223 except for the case of large charge asymmetry @xmath122 ( see fig . such a behaviour implies that different ordered structures can be formed with the variation of @xmath151 . for @xmath122 , the wave number @xmath220 along the @xmath0-line ( @xmath224 ) and depends only slightly on the volume fraction . for @xmath225 , @xmath179 change their trend sharply for the large values of @xmath151 ( @xmath226 ) . corresponding to the ordering of ions along the @xmath0-lines shown in fig . 11 for @xmath213 . @xmath151 is the volume fraction of ions.,height=226 ] along the @xmath0-lines presented in fig . @xmath127 is measured in degrees.,height=226 ] the dominant field is @xmath227 at small @xmath151 for all @xmath62 ( see fig . 15 ) . for @xmath228 , the character of the dominant fluctuations changes with an increase of @xmath151 in particular , @xmath127 varies from @xmath229 to @xmath230 for @xmath122 and from @xmath231 to @xmath232 for @xmath233 when the volume fraction increases from @xmath185 to @xmath186 . for @xmath164 , the dominant field is @xmath234 when @xmath235 . for @xmath172 , the angle @xmath127 takes a maximum value @xmath236 for @xmath226 and the modulus of @xmath236 increases when @xmath62 varies from @xmath237 to @xmath238 : @xmath239 for @xmath240 and @xmath241 for @xmath242 . thus , the charge density fluctuations prevail over the total number density fluctuations along the @xmath0-lines for this range of @xmath62 . in general , for the intermediate size asymmetry both the charge density and the total number density are inhomogeneous in space . -lines for the transition to the ordered phase for @xmath243 . temperature @xmath67 and the volume fraction of ions @xmath151 are in dimensional reduced units defined in eqs . ( [ temp ] ) and ( [ fraction ] ) , respectively.,height=226 ] corresponding to the ordering of ions along the @xmath0-lines shown in fig . 11 for @xmath243 . @xmath151 is the volume fraction of ions.,height=226 ] let us focus on the large size asymmetry , @xmath244 ( @xmath245 , volume ratio @xmath246 ) . for @xmath247 , we obtain @xmath220 in the region @xmath248 . the @xmath0-line and the dependence of @xmath179 on the volume fraction for @xmath243 are shown in figs . 14 and 15 . as is seen , @xmath179 first increases reaching its maximal value and then it decreases with the increase of @xmath151 . the maximum becomes more pronounced with the decrease of charge asymmetry . such a behaviour of @xmath179 is typical of the systems with @xmath249 . the variation of the angle @xmath127 along the @xmath0-lines for @xmath243 is shown in fig . 16 . for the charge asymmetry range @xmath250 , the total number density fluctuations are the dominant fluctuations along the @xmath0-lines moreover , the contribution from this type of fluctuations increases when the charge asymmetry decreases . for @xmath122 , the character of the dominant fluctuations changes continuously along the @xmath0-line from the total number density fluctuations for @xmath251 to the charge density fluctuations reaching a maximal value at @xmath252 . for @xmath253 , the dominant fluctuations are again the total density fluctuations . it should be noted that for @xmath254 the total density fluctuations play the dominant role along the whole @xmath0-line for @xmath255 . in the approximation considered , pms with @xmath244 and @xmath256 do not undergo instabilities against the fluctuations with @xmath124 . moreover , the @xmath0-lines tend to lower @xmath197 with the decrease of the charge asymmetry coming close to the spinodals associated with the phase separation into uniform phases . this implies that the phase transition to the periodic ordering in the pms with large size asymmetry becomes less favorable when the charge asymmetry decreases . in conclusion , for the pms with moderate and large size and charge asymmetry , we expect the phase transition to the colloid crystals of different structure formed by a nearly charge - neutral units for small and large volume fractions . the size of these units depends on the charge ratio and is generally larger than in the case of @xmath190 . for volume fractions near the maximum temperature at the @xmath0-line the number density in positively charged regions is different from the number density in negatively charged regions in the fluctuation that destabilizes the homogeneous phase ( see figs . 14 and 16 ) . global stability , however , may correspond to another structure . for a large size asymmetry case , the complex structures discussed above may be preempted by the phase separation in two uniform phases when the charge asymmetry decreases . along the @xmath0-lines presented in fig . @xmath127 is measured in degrees.,height=226 ] let us consider the system with a very large asymmetry in size and charge : @xmath64,@xmath257 . for @xmath258 ( @xmath259 ) and @xmath260 ( @xmath261 ) , the @xmath0-line and the corresponding wave vectors are shown in figs . 17 and 18 . as expected , the region bounded by the @xmath0-line extends to higher temperatures in this case . the @xmath0-line assumes a maximum with the coordinates @xmath262 , @xmath263 and @xmath264 . remarkably , @xmath265 is five times higher and @xmath211 is two times lover than for @xmath266 ( @xmath267 ) . on the other hand , the above - mentioned value of @xmath265 is by an order of magnitude smaller than that obtained for the same model in ref . @xcite . recall that @xmath211 has no size and charge dependence in the approximation used in the previous work . the period of the op oscillations @xmath268 persists in the region @xmath269 and increases for a higher value of the volume fraction . the behaviour of the angle @xmath127 indicating the direction of the dominant fluctuations is shown in fig . 19 . as is seen , @xmath127 changes its trend sharply from @xmath270 to @xmath271 in the region @xmath272 . then , @xmath127 remains nearly constant ( @xmath273 ) in a wide region of @xmath151 ( @xmath274 ) and again changes sharply to @xmath275 keeping this value for @xmath276 . we suggest that such a behaviour of the op indicates the re - entrant phase transition observed experimentally in the colloidal systems @xcite . -lines for the transition to the ordered phase for @xmath277 . temperature @xmath67 and the volume fraction of ions @xmath151 are in dimensional reduced units defined in eqs . ( [ temp ] ) and ( [ fraction ] ) , respectively.,height=226 ] corresponding to the ordering of ions along the @xmath0-line shown in fig . @xmath151 is the volume fraction of ions.,height=226 ] along the @xmath0-lines shown in fig . @xmath127 is measured in degrees.,height=226 ] the present analysis has shown that the size and charge asymmetry is expected to have a significant effect on the periodic ordering in the systems with dominant coulomb interactions . for the pm with a small size asymmetry ( @xmath278 ) , the trend of the @xmath0-lines in a wide region of the volume fractions is qualitatively similar to that found for the rpm . this is consistent with the behavior of op demonstrating only a small deviation from the pure charge density oscillations . in this case , we expect that the ordered phase corresponding to an ionic crystal with a compact unit cell is formed at lower temperatures and at higher volume fractions when compared to the @xmath0-lines . in particular , such a behaviour was confirmed experimentally for the system of oppositely charged colloidal particles of comparable sizes ( see ref . @xcite ) . on the other hand , we expect that charge - ordered living clusters are formed in the fluid phase with the period of the charge wave similar to the one in the crystal . it is worth noting that in the case of moderate and small charge asymmetry , the trends of the @xmath0-lines deviate from the rpm - like behaviour when the volume fraction increases . pms with @xmath279 show a crossover - type behaviour between the regime of small size asymmetry and the regime of moderate and large size asymmetry . in qualitative agreement with the results of ref . @xcite , our results predict the phase transition to the ionic crystal for the system with @xmath279 and @xmath148 . for a moderate and large size asymmetry ( @xmath280 ) , the @xmath0-lines show a single maximum @xmath265 at volume fraction @xmath211 . while @xmath265 depends slightly on the size asymmetry , @xmath211 decreases noticeably when the size asymmetry increases . in addition , @xmath265 decreases with the decrease of charge asymmetry . the nonmonotonic temperature at the @xmath0-line indiates that some kind of a crystalline phases can be stable for intermediate volume fractions , and re - entrant melting occurs at the high volume fractions ( where the temperature at the @xmath0-line decreses ) . it should be noted that both the charge density and the total number density oscillate along the @xmath0-lines and their contributions to the op vary depending on the charge asymmetry as well as on the volume fraction . note that for overall charge neutrality the number of negatively charged species is @xmath20 times larger than the number of positively charged species . for a large charge asymmetry the main contribution to the number density comes from the negatively charged species . the expected phase transition to the crystal phase of different structure is in qualitative agreement with the simulation studies which predict different crystal structure in mixtures of large and small oppositely charged spherical colloids with @xmath281 @xcite . remarkably , three of the predicted structures were also observed experimentally @xcite . our results also show that the large nearly neutral clusters may be formed in the pms with a moderate and large size asymmetry at the small volume fractions . such a behavior was observed experimentally in water solutions of the charged globular proteins @xcite . on the other hand , the appearance of instabilities in the fluid phase against the periodic ordering for a moderate size and charge asymmetry may explain to some extent the inhomogeneous structure observed in rtils . in these cases the role of non - coulomb interactions should be clarified in future studies . the trend of the @xmath0-lines found for the large size asymmetry implies that the periodic ordering becomes less favorable when the charge asymmetry decreases and the fluid - solid phase transition may be preempted by the phase separation in two uniform phases when the charge asymmetry decreases . for a very large size and charge asymmetry ( @xmath282 ) the number density fluctuations dominate except for small ranges of @xmath151 ( fig . the period of the op oscillations depends very slightly on @xmath151 in the region @xmath269 ( fig.18 ) . in this case the clusters with diameter @xmath283 are formed . each cluster is composed of a large positive ( negative ) ion surrounded by a narrow layer of compensating negative ( positive ) charges . such clusters are associated with pre - transitional ordering - they are present in the solution when the transition to the crystal is approached . stability analysis alone is not sufficient for determination of the crystalline structure . we can expect formation of the colloidal crystal with periodic distribution of particles surrounded by a cloud of counterions for very small volume fractions . it should be noted that at the point of the @xmath0-line corresponding to @xmath284 the inverse screening length is @xmath285 . the screening increases with the increase of @xmath151 , indicating the reduction of the long - range repulsion along the @xmath0-line . for @xmath286 we find @xmath287 , and the op consists of almost pure number density wave with the wavelength @xmath288 , with negligible periodic ordering of the charge density ( see fig.19 ) . we may thus conclude that nearly charge - neutral units composed of the particle surrounded by a thin layer of the neutralizing counterions are formed . crystal formation of neutral ( thus noninteracting ) units is not expected if their volume fraction is @xmath289 . thus , a re - entrant melting could be expected in this region . for @xmath290 we find decreasing @xmath127 and the op consists of both the number and the charge waves . the period of the op oscillations increases with the increase of the volume fraction . such a behaviour may indicate the ion rearrangement leading to the formation of larger clusters . in experiments a fluid bcc crystal fluid phase coexistence was found with an increase of the colloid volume fraction @xcite . the dilute fluid - dilute crystal - dense fluid - dense crystal transitions , for increasing @xmath151 , with the re - entrant melting for @xmath289 were also seen @xcite . in the latter experiment room temperature in our reduced units is @xmath291 . for @xmath286 the reduced temperature at the @xmath0-line is much higher than @xmath292 , and further studies are necessary for comparison between our theory and experiment . nevertheless , we can make an observation that crystal phases were observed for the volume fractions corresponding to the op consisting of both the number and the charge density waves , and the re - entrant melting was observed when the op consists only of the number density wave , with no charge inhomogeneity . in this paper we have used the cvs based theory to study the periodic ordering in the pms with size and charge asymmetry . we consider the gaussian approximation of the functional of the grand partition function which , in turn , leads to the free energy and the direct correlation functions in the rpa . using analytic expressions for direct correlation functions in the rpa we study the effects of the size and charge asymmetry on the instabilities of the uniform phase with respect to the periodic ordering . we determine the cv associated with the op . to this end , we diagonalize the square form of the hamiltonian and analyze the behaviour of the eigenvalues . this enables us to identify the op connected with the periodic ordering and determine the character of the dominant fluctuations along the @xmath0-lines . we extend the study undertaken in ref . @xcite by introducing several modifications . we take into account the dependence of the reference hard - sphere correlation functions on the wave vectors and adopt the wca regularization of the coulomb potential inside the hard core . while the latter leads to a significant decrease of the @xmath0-line temperatures , the former gives rise to their increase for the same regularization scheme . the combination of the above - mentioned modifications gives rise to the @xmath0-line temperatures much lower than those found in ref . a similar situation holds for the whole range of diameter- , @xmath0 , and charge , @xmath20 , ratios of the two ionic species . besides , the wca regularization scheme leads to the higher values of the wave vectors @xmath108 ( smaller @xmath293 ) associated with the @xmath0-line when compared to ref . precise values of @xmath294 can not be determined in a perturbative approach , since the regularization of the coulomb potential is not unique . according to a typical behaviour of the @xmath0-lines , we distinguish three regimes : the regime of small size asymmetry regime ( @xmath278 ) , the regime of moderate and large charge asymmetry ( @xmath280 ) and the regime of very large size and charge asymmetry ( @xmath295 ) . as is seen , the first regime is found to be narrower than that considered in the previous work . remarkably , the qualitative dependence on the charge asymmetry appears already in the first regime : for a small charge asymmetry , the @xmath0-lines @xmath67 versus @xmath151 demonstrate the departure from the monotonously increasing behaviour which becomes more evident with an increase of @xmath0 . this change in the phase diagram is directly related to the non - local approximation used for the reference hard - sphere system . unlike ref . @xcite , our results also show that the models with @xmath296 and @xmath297 ( a large charge at the smaller ion ) do not undergo the instabilities with respect to the periodic ordering in the whole region of @xmath151 . thus , we can state that the modifications introduced in this paper have led to quantitative and partly qualitative changes in the phase diagram obtained in ref . @xcite we conclude that both the size and charge asymmetry affect the periodic ordering in the systems with dominant coulomb interactions , namely : ( 1 ) for @xmath278 , the charge density oscillations dominate in the op the phase transition to an ionic crystal with a compact unit cell is expected ; ( 2 ) for @xmath298 , both the charge density and the total number density oscillate along the structural line and their contributions to the op depend on @xmath0 and @xmath20 ; ( 3 ) for a moderate and large @xmath0 , a crystalline phase can be stable for intermediate volume fractions and re - entrant melting occurs at high volume fractions , the large nearly neutral clusters may be formed at small volume fractions ; ( 4 ) for a large @xmath0 and small @xmath20 , the fluid - crystal phase transition can be preempted by the gas - liquid - like phase separations ; ( 5 ) for a very large @xmath0 and @xmath20 , the colloidal crystal with periodic distribution of particles surrounded by a cloud of counterions is expected for very small volume fractions and a re - entrant melting for the higher volume fractions . the rpa predicts only the existence of a region in which a model ionic fluid is unstable with respect to the op oscillations associated with the periodic ordering . in this context our phase diagrams indicate the pre - transitional effects . the fluctuation effects of the higher order than the second order should be taken into account in order to get the information on both the more precise location of the phase diagrams and the pattern shapes which can be formed . this will be done elsewhere . partial support by the ukrainian - polish joint research project under the agreement on scientific collaboration between the polish academy of sciences and the national academy of sciences of ukraine for years 2009 - 2011 is gratefully acknowledged . a part of this work was realized within the international phd projects programme of the foundation for polish science , co - financed from european regional development fund within innovative economy operational programme `` grants for innovation '' . 99 g. stell , j. stat . phys . * * 7**8 197 ( 1995 ) . y. levin , m.f . fisher , physica a * * 2**25 164 ( 1996 ) . w. schrer , in : d. henderson , m.holovko , a. trokhymchuk ( eds . ) , ionic soft matter : modern trends and applications , nato asi series ii , springer , dordrecht , 2004 , p. 143 . w. schrer , contributions to plasma physics , * * 5**2 , 78 ( 2012 ) . panagiotopoulos , j. phys . : condens . matter * * 1**7 , s3205 ( 2005 ) . a .- hynninen , a.z . panagiotopoulos , mol . phys . * * 1**06 , 2039 ( 2008 ) . o. russina , a. triolo , l. gotrani and r. caminiti , j. phys . * 3 * , 27 ( 2011 ) a. stradner , h. sedgwick , f. cardinaux , w.c.k . poon , s. u. egelhaaf , p. schurtenberger , nature , * 432 * , 492 ( 2004 ) . y. desfougeres , t. croguennec , v. lechevallier s. bouhallab and f. nau , j. phys . chem b * 114 * , 4138 ( 2010 ) . arora , b.v.r . tata , a.k . sood , r. kesavamoorthy , phys . lett . * 60 * , 2438 ( 1988 ) . d. el masri , t. vissers , s. badaire , j. c. p. stiefelhagen , h. rao vutukuri , p. helfferich , t. h. zhang , w. k. kegel , a. imhof and a. van blaaderen , soft matter * 8 * , 2979 ( 2012 ) j. yamanaka , h. yoshida , t. koga , n. ise , t. hashimoto , phys . lett . * 80 * , 5806 ( 1998 ) . royall , m.e . leunissen , a.p . hynninen , m. dijkstra , a. van blaaderen , j. chem . phys . * 124 * , 244706 ( 2004 ) . f.stillinger and r. lovett j. chem . phts . * 48 * , 3858 ( 1968 ) . a. ciach , o. patsahan , phys . e * 74 * , 021508 ( 2006 ) . c. vega , f. bresme , j.l.f . abascal , phys . e * * 5**4 , 2746 ( 1996 ) . hynninen , m. e. leunissen , a. van blaaderen and m. dijkstra , phys . lett . * 96 * , 018303 - 4 ( 2006 ) . j. weis , d. levesque , and j. caillol , _ j. chem . phys . _ * 109 * , 7486 ( 1998 ) . e. spohr , b. hribar and v. vlachy , j.phys . chem b , * 106 * , 2343 ( 2002 ) . romero - enrique , g. orkoulas , a.z . panagiotopoulos , and m.e . fisher , phys . lett . * * 8**5 , 4558 ( 2000 ) . q. yan , and j.j . de pablo , phys . lett . * * 8**6 , 2054 ( 2001 ) . caballero , e.g. noya , c. vega , j. chem . phys . * 127 * , 244910 ( 2007 ) . c. vega , e. sanz , j.l.f . abascal , e.g. noya , j. phys . : condes . matter . * * 2**0 , 153101 ( 2008 ) . leunissen , c.g . christova , a.p . hynninen , c.p . royall , a.i . campbell , a. imhof , m. dijkstra , r. van roij and a. van blaaderen , nature , * * 4**37 , 235 ( 2005 ) . p.m. biesheuvel , s. lindhoud , r. de vries , m. a. cohen stuart , langmuir , * 22 * , 1291 ( 2006 ) a. shukla , e. , e. di cola , s. finet , p. timmins , t. narayanan , d.i . svergun , proc . sci . * 105 * 5075 ( 2008 ) . p. kowalczyk , a. ciach , p. a. gauden and a. p. terzyk i. j. colloid and interf . , * 363 * , 579 ( 2011 ) . a. ciach , w.t . gd , and g. stell , phys . e * * 7**5 , 051505 ( 2007 ) . a. ciach and g. stell , int . b * 19 * 3309 ( 2005 ) y. rosenfield , phys . * 63 * , 980 ( 1989 ) . lebowitz , phys . rev . * 133 * , 895 ( 1964 ) . j. d. weeks , d. chandler , and h.c . andersen , j. chem . phys . * 54 * , 5237 ( 1971 ) . chandler d. , andersen h.c . , j. chem . phys . * 54 * , 26 ( 1971 ) . patsahan , condens . matter phys * 7 * , 35 ( 2004 ) . patsahan , physica a * 272 * , 358 ( 1999 ) . o.v . patsahan , t.m . patsahan , j. stat . phys . * * 1**05 , 285 ( 2001 ) . d. n. zubarev , dokl . nauk sssr * * 9**5 , 757 ( 1954 ) ( in russian ) . yukhnovsky , zh . fiz . * * 3**4 , 379 ( 1958 ) ( in russian ) . yukhnovskii , phase transition of the second order : collective variables method . world sci . co. ltd . , singapore , 1987 . o. patsahan , and i. mryglod , condens . matter phys . * 9 * , 659 ( 2006 ) . patsahan , t.m . patsahan , phys . e * * 8**1 031110 ( 2010 ) . patsahan , i.m . mryglod , t.m . patsahan , j.phys . : condens . matter * 18 * , 10223 ( 2006 ) . patsahan , i.m . mryglod , j.phys . a : math . gen . 39 ( 2006 ) l583 . n.w . ashcroft , d.c . langreth , phys . rev . * 156 * , 685 ( 1967 ) n.w . ashcroft , d.c . langreth , phys . rev . * 166 * , 934 ( 1968 ) . a. ciach , g. stell , j. mol . liq . * 87 * , 255 ( 2000 ) . patsahan , i.m . mryglod , condens . matter phys . * 7 * , 755 ( 2004 ) . caillol , j. stat . phys . * 115 * , 1451 ( 2004 ) . caillol , mol . phys . * 103 * , 1271 ( 2005 ) . a .- hynninen , a.z . panagiotopoulos , j. phys . : condens . matter * 21 * , 465104 ( 2009 ) . archer , c. ionescu , d. pini , r. reatto , j. phys . : matter * 20 * , 415106 ( 2008 ) .